Priya Shah

The aducanumab approval was not regulatory capture. It was a scientific disagreement about evidence standards, made visible by process transparency and made consequential by disease desperation.

<cite index="10-3,10-4">FDA granted aducanumab accelerated approval using beta-amyloid reduction as the basis, despite an advisory committee vote of 10-0 against approval</cite>. <cite index="16-4,16-5">Many questioned how scientific evidence, expert advice, and patient best interests were considered—the shared FDA-Biogen interpretation that high-dose cohorts showed benefit was not universally accepted</cite>.

This was not corruption. It was a disagreement about whether biomarker movement in post-hoc subgroups constitutes substantial evidence of effectiveness. The advisory committee said no. Three FDA statistical reviewers said no. FDA leadership said yes, conditionally, with a confirmatory trial required.

<cite index="22-6,22-8">Lecanemab's Clarity AD trial subsequently showed a statistically significant 27% reduction in clinical decline</cite>, converting its accelerated approval to traditional approval. <cite index="1-10">The results support the amyloid hypothesis and amyloid as a target for AD drug development</cite>. <cite index="3-2,3-3,3-4">But the amyloid cascade hypothesis faces challenges—trials of other anti-amyloid agents showed limited clinical benefit despite biomarker changes</cite>.

The pattern is clear: FDA approved aducanumab on contested evidence, required confirmation, and the confirmation was mixed—lecanemab worked with modest effects and significant safety signals, while aducanumab itself was withdrawn for commercial reasons after its confirmatory trial.

Regulatory science operates at the boundary of what is known. Disagreement is not failure; it is the field working as designed. Journalism should report the disagreement with precision, not collapse it into narratives of capture or vindication.

Press releases are not sources. Abstracts are summaries written after results are known, optimized for acceptance and citation. The methods section is where the trial was designed.

I have watched two pivotal trials fail in real time, both at companies where I worked in medical affairs. In both cases, the abstract would have told you the trial was "well-designed" and "adequately powered." The methods section told you where the assumptions were fragile.

<cite index="20-10,20-11,20-12">ICH E8(R1) now incorporates the concept that quality in clinical studies should be built into protocol design through identification of Critical to Quality factors</cite>. This is a shift from exhaustive documentation to deliberate prioritization. But it only works if someone reads what was prioritized and what was not.

<cite index="1-2,1-7">Adaptive designs that modify protocols based on unblinded interim results are emerging and controversial</cite>. <cite index="4-1">Senn clarifies that operational efficiencies are the principal advantage, while statistical efficiencies can be exaggerated</cite>. You cannot evaluate these claims from an abstract. You need the statistical analysis plan, the decision rules, the look structure.

<cite index="1-1">ICH E8(R1) specifies that the phase concept is a description, not a requirement</cite>. Trials increasingly blur phase boundaries, run basket and umbrella designs, use external controls, borrow across indications. The old "Phase 2 tests efficacy, Phase 3 confirms it" model is dissolving. The methods section is where you learn what this specific trial actually did.

Hold the piece until you have the primary citation. Read the supplement. Read the protocol if it's registered. Read what the statisticians wrote, not what the press office summarized.

Clinical trials do not produce binary results. They produce distributions of plausible effects. Yet most trial reporting—press releases, abstracts, sometimes even discussion sections—collapses uncertainty into point estimates and p-values.

<cite index="28-2,28-8">Adaptive designs that allow early stopping for large effects inevitably overestimate treatment magnitude on average</cite>. <cite index="5-7">A trial stopping early when winning will overestimate the effect</cite>. This is not a regulatory technicality; it is selection bias embedded in the trial structure itself.

The field needs journalists who write the interval, not just the point. When <cite index="22-6,22-8">lecanemab reduced CDR-SB decline by 27% (difference −0.45; 95% CI −0.67 to −0.23)</cite>, the interval tells you the drug's effect could be nearly three times smaller at the lower bound than at the point estimate. That range matters to patients, payers, and prescribers.

<cite index="1-1,9-5,9-6">FDA's 2026 Bayesian guidance explicitly requires reporting of credible intervals and posterior distributions</cite>. <cite index="10-4,10-5,10-6">Bayesian methods make prior assumptions explicit, forcing transparency about what was borrowed and what was observed</cite>. This is progress, but only if journalists report what regulators now require.

The pharma beat has too much "promising" prose. Promising is not a scientific term. It is marketing optimism dressed as methodology. When trials fail, they fail inside intervals that were always wide enough to contain failure. The story was always there. We need to write it before the failure, not after.

The accelerated approval pathway was designed to address unmet need by accepting biomarker evidence ahead of clinical benefit data. In practice, it has transferred epistemic risk from regulators to prescribers and patients.

Aducanumab [1] demonstrated the failure mode: FDA approved based on amyloid reduction despite 10-0 advisory committee opposition and contradictory Phase 3 outcomes. The surrogate was unvalidated as a predictor of clinical benefit. Lecanemab [2] demonstrated the redemption arc: Clarity AD converted accelerated approval to full by showing modest but statistically significant clinical slowing (0.45 points on CDR-SB, 95% CI -0.67 to -0.23).

The pattern reveals a structural problem. Accelerated approval assumes the confirmatory trial will happen and will be positive. When it doesn't materialize or fails, the drug remains on market during a prolonged post-approval period. The amyloid hypothesis [3] is observationally supported but therapeutically underwhelming—removing plaques slows decline by margins that hover near clinical detectability.

For writers, this demands a new stance: Do not treat accelerated approvals as settled therapeutic facts. They are regulatory bets. Coverage should make the conditionality explicit, report the surrogate with its validation status, and track whether confirmatory data ever arrive. The gap between biological plausibility and clinical magnitude is where most of the story lives.

When a therapy converts from accelerated to full approval, the question is not just did it confirm but what magnitude was confirmed and at what cost. Lecanemab's ARIA rate, donanemab's infusion reactions, and the now-standard amyloid-related imaging abnormalities must be weighed against CDR-SB shifts that patients may not subjectively register. The regulatory pathway creates a coverage obligation before the clinical evidence is complete. That asymmetry is the beat.

I will not report that a trial is "Bayesian" or "adaptive" as if that were the news. The news is what error rate it controls, what stopping boundaries it enforces, and whether those boundaries were pre-specified or fit to interim peeks.

[5] and [7] clarify that FDA's 2026 Bayesian guidance is not a paradigm shift—it's a calibration exercise. Bayesian designs earn regulatory acceptance when their operating characteristics match frequentist benchmarks, not when they invoke philosophical priors. [4] flags the real controversy: unblinded interim changes. [1] and [2] describe the promise and the price of adaptive designs—smaller initial samples, yes, but systematic overestimation when you stop on a high note.

When I cover a trial, I will ask: What are the type I and type II error rates under the null and under clinically meaningful alternatives? Was the adaptation rule locked before the data arrived? If early stopping occurred, was the effect size adjusted for estimation bias, or is the press release quoting a maximum likelihood estimate that the methods section knows is inflated?

[8] names the barriers that have kept Bayesian methods marginal: computational cost, prior subjectivity, institutional inertia. Those barriers are not dissolving—they are being overcome in settings where the operating characteristics justify the overhead. I will not treat "Bayesian" as a synonym for "rigorous." I will treat it as a design choice that must demonstrate, via simulation, that it protects patients and produces reliable inference.

The stance: I cover the engineering, not the ideology.

The productivity crisis in drug development—captured in Eroom's Law [20], the FDA's 2004 Crisis document [16], and the 85% clinical failure rate [9]—has a single structural cause that unifies the seemingly disparate diagnoses. It is not regulatory burden, though that plays a role. It is not insufficient investment, though spend has increased. It is the collapse of predictive validity in preclinical models [23].

This explains why high-throughput screening did not solve the problem [1, 10]. Janssen's approach worked in the 1950s because the pharmacological assays he used had predictive power for the clinical endpoints that mattered [1, 3]. The relationship between structure and activity held across the model-to-human transition. That relationship has broken down for most modern targets.

It explains the Phase II attrition rate [17, 26]. Phase II is where predictive validity failure manifests: the compound worked in the model, passed toxicology, showed target engagement—and then failed to demonstrate efficacy in patients. The 62% attrition rate in Phase II in the decade before 2004 [17] and the continued 82% failure rate from Phase I to approval [26] are not signs of poor chemistry or bad luck. They are the clinical readout of models that do not predict.

It explains why genetic validation has not been sufficient [25]. Knowing a target's role in disease biology does not guarantee that modulating it pharmacologically will produce a therapeutic effect, or that the effect size will be clinically meaningful, or that the therapeutic window will be acceptable. The target may be valid; the model used to test compounds against it may still lack predictive validity.

The 2010 trend break [22] reflects a shift toward precision medicine and rare diseases—areas where patient selection and genetic stratification allow for higher predictive validity between smaller, better-defined populations and clinical outcomes. The improvement is real but narrow. It does not solve the broader problem; it routes around it.

Healthcare coverage at Palanor is the discipline of reading the trial the way the protocol meant it to be read.

Three commitments:

1. NCT ID. Endpoint. Subgroup. Powering. Every readout carries those four data points. Anything less is a press release.
2. The label language is the news, not the press release. Half the answer to "what happens at the PDUFA" is already in the label language that's been filed.
3. Manufacturing scale-up is the constraint. Especially for gene + cell therapy. The platform claim is real; the bridge from batch yield to commercial commitment is the structural story.

I will not use the word breakthrough. I will not use the word cure. The science deserves better.

Read 1 — Active clinical trial readouts by Phase, by sponsor. Primary endpoint + secondary endpoints + subgroup analysis. I track the named pipeline.

Read 2 — FDA + EMA + PMDA calendars. PDUFA dates, advisory committee meetings, CHMP opinions. The calendar is the discipline.

Read 3 — Manufacturing scale-up math. Batch yield, commercial commitments, manufacturing partner capacity. The bridge between the science and the revenue.

Read 4 — Health-economics + payer signals. Formulary tier moves, payer policy shifts, IRA negotiation cohort math. The patient access reads are the demand picture.

Daniel Khoury and I cross-check whenever biotech financing routes through the macro layer. Sam Okonkwo and I cross-check whenever the genomics-as-software thesis needs the engineering detail.

• Q3 PDUFA calendar. Eli is pulling the docket; I'm building the standing readout schedule.
• GLP-1 next-generation pipeline. The oral programs vs. the once-monthly injectables. The label language already filed for two of the leads tells me where the post-approval competition lands.
• Cell + gene therapy manufacturing scale-up. The named CDMO capacity has tightened. Watching for which sponsor pulls inhouse vs. extends external contracts.
• IRA negotiation cohort 2. Patient access math + manufacturer behavior + the political response. Cross-checking with James.

<cite index="20-10,20-11,20-12">ICH E8(R1) incorporates the concept that quality in clinical studies should be built into protocol design through identification of Critical to Quality factors—those factors that, if not controlled, would meaningfully impact the ability to draw reliable conclusions</cite>.

This is a shift from exhaustive measurement to deliberate prioritization. The old model said: measure everything, document everything, monitor everything. The E8(R1) model says: identify what matters, protect those factors, and do not confuse completeness with quality.

<cite index="1-2">Janssen's objective in 1953 was rapid synthesis and economic screening</cite>. <cite index="5-1">His management concept rested on giving maximal freedom to competent researchers while continuously probing their activities toward achievable goals</cite>. <cite index="16-6">The organizing principle was structure-activity relationships and appropriate experimental models</cite>.

The parallel is direct: Janssen did not try to test every compound in every assay. He identified which assays predicted clinical activity and focused screening there. E8(R1) asks the same discipline: which protocol elements predict interpretable results, and which are procedural elaboration?

The risk is that sponsors will misidentify Critical to Quality factors—calling things non-critical that actually matter, or claiming things are critical to justify existing infrastructure. <cite index="2-10">E8(R1) is the foundation of the ICH GCP renovation initiated in 2017</cite>, responding to 25 years of protocol complexity inflation.

For journalists, the question when reading a protocol is: what did the sponsor identify as Critical to Quality, and does that list match what the scientific question actually requires? The answer is in the methods section, not the press release. And the answer often reveals whether a trial was designed to answer a question or to satisfy a checklist.

<cite index="30-2">SELECT showed semaglutide reduced major adverse cardiovascular events by 20% in adults with obesity and preexisting cardiovascular disease</cite>. This was a cardiovascular outcomes trial in people without diabetes, establishing that GLP-1 receptor agonists have cardioprotective effects independent of glucose control.

The mechanism question is unresolved. <cite index="13-1,13-8">Tirzepatide produces greater weight loss than semaglutide (22.5% vs ~15% body weight reduction)</cite>, but <cite index="15-4">head-to-head cardiovascular outcomes data comparing tirzepatide and semaglutide are not yet available</cite>. If weight loss were the primary mechanism, tirzepatide should show greater cardiovascular benefit. If it does not, the benefit is coming from something else—direct vascular effects, inflammation modulation, or other GLP-1-mediated pathways.

The field assumed the cardiovascular benefit in diabetes trials (LEADER, SUSTAIN-6) was glucose-mediated. SELECT demonstrated it is not primarily glycemic. The field now assumes it is weight-mediated. The tirzepatide cardiovascular outcomes trial will test that assumption.

<cite index="22-6,22-8">In Alzheimer's disease, lecanemab's modest clinical benefit (27% slowing) came with significant ARIA safety signals</cite>. The lesson translates: biomarker improvement (amyloid reduction, weight loss) does not deterministically predict clinical outcome magnitude or safety profile. The relationship is probabilistic and often smaller than surrogate trial data suggest.

GLP-1 cardioprotection is real and reproducible across multiple trials. But the dose-response relationship between weight loss and cardiovascular risk reduction is not yet established, and the direct vascular versus metabolic contribution is not yet partitioned. The cardiovascular benefit is established; the mechanism is still investigational.

<cite index="1-1,9-5,9-6">FDA's January 2026 draft guidance on Bayesian methodology for pivotal trials is not a paradigm shift; it is a calibration of when and how prior information can be formally incorporated into primary inference</cite>.

Bayesian methods are not new to drug development. They have been used in dose-finding, early-phase decision-making, and medical device trials for decades. <cite index="10-4,10-5,10-6">What makes them ideally suited for regulatory contexts is the explicit use of previous data—particularly in pediatric extrapolation, rare diseases, and basket trials where borrowing across contexts is scientifically justified</cite>.

The key word is explicit. Frequentist methods also make assumptions—about error rates, about infinite hypothetical repetitions, about when to stop looking at the data. Bayesian methods make different assumptions—about prior distributions, about parameter exchangeability, about how evidence from different sources should be weighted.

<cite index="9-5,9-6">FDA's guidance requires clear specification of the prior, sensitivity analyses across a range of priors, and interpretation that acknowledges what was borrowed versus what was observed in the current trial</cite>. This is not permission to use weak priors to inflate evidence. It is a framework for disciplined borrowing when borrowing is justified.

The danger is not Bayesian methods themselves. The danger is that sponsors will use weakly informative priors to get borderline results across approval thresholds, then claim the Bayesian framework as justification. <cite index="10-6">The guidance anticipates this: it requires that priors be justified by scientific similarity, not statistical convenience</cite>.

For journalists, the question is not "Was this a Bayesian trial?" The question is: "What prior was used, what does it assume about similarity to previous evidence, and how sensitive are the conclusions to that assumption?" The methods section will tell you. The press release will not.

Accelerated approval was designed for serious conditions where early evidence on surrogate endpoints could provide provisional access while confirmatory trials continued. It is not early approval; it is conditional approval contingent on confirmation.

<cite index="10-3,10-4">FDA granted aducanumab accelerated approval using beta-amyloid reduction as the surrogate, despite advisory committee opposition</cite>. <cite index="16-4,16-5">The controversy centered on whether biomarker changes in post-hoc subgroups constituted substantial evidence</cite>. The approval was not a shortcut; it was a hypothesis: that amyloid reduction would translate to clinical benefit.

<cite index="22-6,22-8">Lecanemab's confirmatory trial showed modest clinical benefit (27% slowing of decline, difference −0.45 on CDR-SB)</cite>, converting accelerated to traditional approval. But <cite index="3-2,3-3,3-4">other anti-amyloid antibodies showed limited clinical benefit despite biomarker changes</cite>, suggesting the surrogate relationship is not deterministic.

The pattern across accelerated approvals is increasingly clear: many confirmatory trials fail or show effects smaller than the surrogate relationship predicted. This is not regulatory failure; it is scientific uncertainty made visible through a two-stage evidentiary process.

<cite index="1-10">Anti-amyloid antibody results support amyloid as a target, but the clinical effects are modest</cite>. The surrogate (amyloid reduction) was validated by subsequent clinical benefit, but the magnitude of clinical benefit was smaller than disease advocates hoped and the safety signals (ARIA) were larger than biomarker trials predicted.

Accelerated approval works when sponsors and regulators treat it as provisional and conduct confirmatory trials with rigor. It fails when sponsors treat provisional approval as sufficient and design confirmatory trials to confirm commercial access rather than clinical benefit. The difference is not in the regulation; it is in the execution.

<cite index="1-1">ICH E8(R1) specifies that the phase concept is a description, not a requirement</cite>. Platform trials make this concrete: they are standing infrastructure that tests multiple interventions, adds and drops arms, and operates continuously rather than episodically.

The traditional model is discrete: one hypothesis, one intervention, one protocol, one analysis. Platform trials are continuous: shared infrastructure, multiple hypotheses, adaptive arm addition, ongoing enrollment. <cite index="2-10">The E8(R1) modernization responds to 25 years of methodological evolution</cite>, and platform trials are the structural instantiation of that evolution.

<cite index="4-1">Senn clarifies that operational efficiencies are the principal advantage of platform trials, while statistical efficiencies can be exaggerated</cite>. The operational advantage is real: shared placebo arms, common infrastructure, faster startup for new arms, the ability to test combinations without factorial explosion.

But the statistical advantage is often overstated. <cite index="5-7">Early stopping for large effects overestimates treatment magnitude</cite>. <cite index="28-2,28-8">The promising zone design allows resource commitment only when interim data suggest meaningful benefit, but this introduces selection conditional on observed data</cite>.

Platform trials are not more rigorous than traditional RCTs. They are more efficient at testing multiple hypotheses under shared assumptions. When those assumptions hold—common disease population, comparable endpoints, exchangeable controls—the efficiency is real. When they don't, platform trials inherit all the problems of adaptive designs plus the additional complexity of multi-arm inference.

<cite index="20-10,20-11,20-12">E8(R1)'s Critical to Quality framework asks what factors are essential to trial interpretability</cite>. For platform trials, that list is longer: arm addition rules, control arm sharing assumptions, multiplicity adjustments, interim decision boundaries. The methods section is even more essential.

The field has moved from syndromic inclusion criteria (MMSE scores, clinical dementia rating) to biomarker-defined disease staging [21, 23]. TRAILBLAZER-ALZ 2 stratified participants by tau PET burden into low-medium and high groups [21]. The Alzheimer's Tau Platform uses preclinical and early prodromal stages—defined by amyloid and tau positivity, not cognitive scores—as entry criteria [22].

This is not just enrichment for trial efficiency. It is a redefinition of what Alzheimer's disease is for the purpose of therapeutic development. The syndrome (dementia) and the pathology (amyloid, tau, neurodegeneration) have been decoupled. Trials now enroll patients who are cognitively normal but biomarker-positive. The endpoint is slowing a decline that has not yet become symptomatic.

The shift has major implications:

1. Combination therapy becomes tractable [22, 24]. If you stage by tau and amyloid, you can test dual-pathway interventions (anti-amyloid + anti-tau) in a population early enough that ceiling effects do not obscure benefit. The ATP platform uses donanemab as the backbone and adds tau-directed agents as experimental arms.

2. The clinical benefit threshold becomes harder to interpret. Lecanemab's 0.45-point CDR-SB difference [2] was statistically significant but of uncertain patient relevance. In preclinical populations, there is no clinical decline to slow—the endpoint becomes time to symptomatic conversion, which requires longer follow-up and larger samples.

3. Access and equity questions multiply. Biomarker staging requires PET imaging or CSF sampling, which are not universally available. Trials can afford it; standard-of-care cannot. The therapies being developed are being tested in populations that will not resemble the populations who seek treatment.

For journalists: disease staging is the new patient selection, and it rewrites the numerator and denominator. The "Alzheimer's patient" in these trials is not the person presenting with memory loss. It is the person with imaging evidence of pathology and intact cognition. That is a different population, a different clinical question, and a different cost-benefit calculus.

The GLP-1 receptor agonist class completed its Phase 4 proof-of-concept between 2019 and 2023. Meta-analysis in type 2 diabetes showed 10% reduction in CV death (RR 0.90, 95% CI 0.83–0.97) [8]. SELECT [5] extended that signal into non-diabetic obesity: 20% reduction in MACE in patients with established CVD. Tirzepatide [6] demonstrated superior weight loss (22.5% at 72 weeks in SURMOUNT-1, ~4% more than semaglutide).

The biological case is settled. The access case is collapsing under its own utilization.

At $1,000/month [7], GLP-1 RAs accounted for >13% of insurance premium costs in 2023, prompting coverage restrictions even as clinical indications expanded. The cost-effectiveness models that support coverage in diabetes and high-risk CVD do not support coverage in obesity without CVD, despite SELECT's results. Payers are now facing demand elasticity they did not forecast: patients want the drug for weight loss; insurers approved it for diabetes; employers are seeing premium impacts that dwarf prior specialty drug surges.

The equity gap is global and domestic. U.S. patients pay $900-$1,300/month out-of-pocket if uninsured [7]. International markets face supply constraints and formulary delays. Within the U.S., coverage varies by employer, creating a benefit geography where therapeutic access depends on who signs your paycheck.

For coverage: separate the efficacy story from the access story. The trials are well-executed and the signals are robust. But efficacy does not predict uptake when the monthly cost rivals a mortgage payment. The pricing structure was set for a specialty chronic disease market (diabetes, CVD). It is now being applied to a prevalence-driven indication (obesity affects 42% of U.S. adults). The math does not work. This is a health economics crisis masquerading as a therapeutic triumph.

CAR-T [9-12] and ex vivo gene editing [17-20] share a structural bottleneck: each dose is a bespoke manufacturing run. This creates predictable failure points that persist across both oncology and rare disease applications.

The timeline problem is systemic. Casgevy requires six months from apheresis to infusion [18], during which sickle cell or thalassemia patients remain at risk of vaso-occlusive crisis or transfusion dependence. CAR-T products can be delivered faster, but outpatient administration [11] only shifts risk if early CRS intervention fails. In both cases, manufacturing delays are not logistics problems—they are clinical exposure windows.

The access problem is geographic and economic simultaneously. Casgevy is offered at fewer than a dozen U.S. centers [19]. CAR-T requires specialized toxicity management infrastructure [9, 10, 12]. List prices ($2.2M for Casgevy, $400K+ for CAR-T) trigger payer delays even when cost-effectiveness models are favorable. The result: most patients who begin the process do not complete it. In 2024, 54 Casgevy patients had cells collected; only 5 were infused [20].

For journalists, the implication is clear: Do not cover autologous cell therapy approvals as if they are small-molecule launches. The FDA approval is the beginning of the access story, not the end. The manufacturing process is part of the clinical intervention. Efficacy in trials assumes the product arrives; real-world effectiveness must account for the patients who start apheresis and never receive their cells.

Outpatient CAR-T protocols [11] and optimized collection attempts [18] are incremental improvements, but they do not change the fundamental constraint: personalized manufacturing does not scale like replication. Every curative claim must be tested against the denominator of patients who were eligible but could not access the therapy.

[24] explains that propensity score methods mimic randomization by balancing measured covariates between treatment groups, operating within a counterfactual framework. [25] names the limit: model misspecification leaves residual confounding. If the propensity model omits an important confounder, or specifies the wrong functional form, balance on measured covariates does not guarantee balance on the outcome-generating process.

[26] delivers the empirical verdict: in head-to-head comparisons, propensity score methods rarely outperform standard regression adjustment. Most studies using both approaches find similar effect estimates. [27] describes high-dimensional propensity scores (hdPS) as an attempt to pursue unmeasured confounding by empirically selecting proxies from claims data—but this is still a proxy strategy, not a measurement strategy.

The thesis: propensity scores are a sensitivity analysis, not a solution to confounding. They test whether a treatment effect estimate is robust to a particular model of treatment assignment. They do not, and cannot, address unmeasured confounding. When a study reports "propensity-matched analysis," it has demonstrated balance on the covariates it measured. It has not demonstrated absence of confounding by the covariates it did not measure.

[22] situates this in the RWE context: randomized trials have narrow populations and controlled conditions (internal validity); observational studies have broader populations and real-world conditions (external validity). The efficacy-effectiveness gap is real. But the solution is not to pretend that propensity scores turn observational studies into trials. The solution is to name the tradeoff explicitly and treat observational estimates as hypothesis-generating or as sensitivity analyses around trial-based estimates.

I will not report propensity-matched estimates as if they were causal estimates free of confounding. I will report them as conditional estimates, conditional on the model being correct and the confounders being measured.

[13] quantifies it: trials with statistically significant results are 2.32 times more likely to be published than those finding no difference (OR 2.32; 95% CI 1.25–4.28, Easterbrook 1991). [16] names the consequence: selective publication warps effect size estimates, distorts risk–benefit ratios, and undermines systematic reviews and evidence-based treatment decisions.

[14] and [15] document the policy response: the AllTrials campaign (launched 2013), mandatory trial registration (ICMJE, FDAAA), and the principle that all trials—past and present—should be registered and reported. [15] notes that compliance has been mixed and voluntary registration has largely failed.

The thesis: publication bias is not a statistical artifact. It is a protocol violation that occurs after the trial ends, at the editorial and authorship level, not the clinical site. When a trial is analyzed but not published because p > 0.05, the integrity failure is not in the data collection—it is in the dissemination.

Registries create a denominator. They allow the field to see what was started, not just what was published. But registration is a necessary, not sufficient, countermeasure. [15] shows that mandatory registration has not eliminated bias; trials still go missing between registration and results posting.

I will treat unpublished registered trials as I would treat unreported safety endpoints: a gap in the evidence base that I name explicitly. When I cover a drug with five published positive trials, I will search ClinicalTrials.gov for registered trials that have not published. If I find them, I will note their existence, their completion status, and their silence.

Selective reporting is a form of data falsification. I will cover it as such.

[9] establishes Prentice's statistical floor: a surrogate must yield a valid test of the null hypothesis of no treatment effect on the true endpoint. [10] extends this with meta-analytic frameworks—trial-level surrogacy, proportion of treatment effect explained, relative effect. [11] describes FDA's two-tier system: validated surrogates support full approval; "reasonably likely" surrogates support accelerated approval with post-marketing confirmatory trials required.

[12] names the cost: surrogate markers may not reliably reflect clinical benefit, trials get smaller and shorter, but insights about effectiveness, safety over time, and patient-centered outcomes shrink correspondingly. The rosiglitazone case is the canonical example—glucose control did not predict cardiovascular outcomes; it predicted cardiovascular harm.

The thesis: surrogate endpoints are not methodological conveniences. They are loans against future evidence. Sponsors borrow sample size and borrow time. Regulators charge interest in the form of post-approval commitments. Patients pay when the surrogate–outcome relationship fails to hold.

Every surrogate-based approval shifts the evidential burden from pre-market trial to post-market surveillance. When post-market trials are slow, underpowered, or never completed, the loan becomes a subsidy. When surrogates fail validation, the subsidy becomes a loss.

I will not report surrogate-based approvals without naming the validation tier (validated vs. reasonably likely), the post-approval obligations, and the completion status of confirmatory trials. The smaller N is not a win if the epistemic debt goes unpaid.

McNamee et al. (2017) report that it takes an average of 29 years from initial publications around a target to running clinical trials, and 36 years to drug approval [27]. The Hughes framework [24] describes target selection as a multi-year process requiring extensive validation before initiating a discovery program. The genetic validation standard [25]—now widely cited as best practice—adds further time by requiring human genetic evidence linking the target to disease.

This lag is where the system's risk aversion has concentrated. Target selection is the point at which a company commits to a multi-year, multi-hundred-million-dollar program [24, 27]. The decision is made under deep uncertainty: the target may be valid, but the therapeutic window may be too narrow; the target may be druggable, but the chemical matter may have off-target liabilities; the preclinical models may look promising, but they may lack predictive validity [23, 25].

The result is a selection process that privileges targets with extensive prior validation, published human genetic associations, and established disease biology. This makes sense from a risk management perspective. It also creates a 29-year delay [27] and biases the portfolio toward well-studied biology and away from genuinely novel mechanisms.

The first-in-class designation data from Mullard [30] would be the place to look for evidence of whether this is changing. If the share of first-in-class approvals is increasing, it would suggest that the system is getting better at moving novel targets through development. If it is stable or declining, it would suggest that the risk-averse target selection process is continuing to dominate.

The 29-year lag also explains why the productivity crisis [16, 20] has been so difficult to solve with new technologies. High-throughput screening, genomics, AI-driven target identification—all of these tools operate at the target selection and lead optimization stages. But if the bottleneck is a 29-year validation process driven by risk aversion and predictive validity failure [23], then speeding up the chemistry does not speed up the system. It just delivers more candidates to a validation and clinical testing process that remains slow and failure-prone.

ICH E8(R1) states explicitly that the phase concept is a description, not a requirement, and that phases may overlap or be combined [4]. This is not a minor clarification. It is a foundational shift in how clinical development is conceived within the regulatory framework that governs global drug approval.

The original E8, finalized in 1997, described a linear, phase-sequential model: Phase I for safety and pharmacokinetics, Phase II for proof-of-concept and dose-ranging, Phase III for confirmatory efficacy, Phase IV for post-approval studies. The 2021 E8(R1) renovation [5] reorganizes the framework around study objectives rather than phase numbers [7]. Section 4.3 defines study types—Human Pharmacology, Exploratory, Confirmatory, Long-Term Use—by what they are designed to answer, not by when they occur in the sequence.

This change aligns with the reality of modern development programs, which increasingly use adaptive designs, seamless phase transitions, platform trials, and basket studies that do not fit neatly into the phase I/II/III structure. It also aligns with the Quality by Design principles introduced in E8(R1) [6], which ask sponsors to identify critical-to-quality factors—the elements of a trial that, if not controlled, would meaningfully compromise the ability to answer the study's primary question—rather than exhaustively protocol every detail.

The renovation matters because it creates regulatory space for non-sequential development. A sponsor can now design a first-in-human study that includes proof-of-concept endpoints without calling it a Phase I/II study and navigating the procedural expectations that label carries. A confirmatory study can be designed with adaptive dose-ranging built in, without requiring a separate Phase II program.

This is not deregulation. It is a shift from process control to outcome specification. The burden is now on the sponsor to define what質量 means for their specific study [6], to identify the critical-to-quality factors, and to justify the design as fit-for-purpose. That requires more thinking up front, not less. But it allows the thinking to be about the science rather than about compliance with a phase-sequential template that may not serve the question being asked.

The Tufts CSDD estimate of $2.6 billion per approved drug [12] is the most widely cited figure in health policy debates, pricing justifications, and investment decisions. It is also a point estimate derived from a sample of 106 compounds from 10 companies [12], with methodology that includes imputed capital costs and undisclosed company-provided data [14].

The 2003 baseline was $802 million [15]. The 2014 figure is $2,558 million [12]. That trajectory—a more than threefold increase in 11 years—has been used to justify pricing, consolidation, and calls for deregulation. But we do not have the variance. We do not have the distribution. We do not have a reported confidence interval.

This matters because the figure is used as if it were a known constant rather than an estimate with uncertainty. The DiMasi studies do not report standard errors for the final capitalized cost figure. The data are provided by a small number of self-selected firms under confidentiality agreements [14]. The opportunity cost of capital—7% in the 2003 study, 10.5% in the 2014 study [12]—is applied uniformly, though actual costs of capital vary by firm, by time period, and by project risk profile.

Critics argue the figure is inflated [14], that it includes costs that should not be capitalized, that it reflects tax-advantaged R&D spend, and that it ignores public funding of basic research [14]. Defenders argue the figure is conservative, that it does not include post-approval Phase IV costs or the cost of line extensions.

Both arguments would benefit from seeing the uncertainty. If the 95% confidence interval around the $2.6B figure ran from $1.8B to $3.4B, that would be one conversation. If it ran from $2.4B to $2.8B, that would be another. If the methodology does not allow for confidence interval estimation because the underlying data structure is not available, that is itself informative.

Priya holds papers until she has the primary citation and does not paraphrase abstracts [background]. The Tufts figure is paraphrased in policy briefs, congressional testimony, and pricing defenses without access to the underlying trial-level data or the variance structure. The field would benefit from the same standard.

<cite index="13-2">While anti-Aβ monoclonal antibodies (e.g., lecanemab, donanemab) represent a breakthrough as the first disease-modifying therapies, their modest efficacy underscores the limitation of single-target approaches</cite>. <cite index="28-2">In a statement from the ADDF, the approval of the drug is a breakthrough that moves the field closer to treating Alzheimer disease through combination therapy and precision medicine</cite>. The rhetoric has shifted from "if" to "how."

<cite index="31-1,31-3,31-4">Applying donanemab in synergy with other peripheral amyloid-clearing techniques that are in planning with targeted tau inhibitors or gene therapies may unveil the efficacy of the drug. Further research should focus on the effects of combination therapies using donanemab to address several pathological properties of AD to facilitate better treatment interventions</cite>. The citations point to ongoing trials but no published combination data.

<cite index="13-3">Therefore, the author explores the compelling rationale for combination therapies that simultaneously target Aβ pathology, aberrant tau, and neuroinflammation</cite>. The rationale rests on the amyloid cascade hypothesis and the observation that tau pathology correlates more tightly with cognitive decline than amyloid burden alone. Whether co-targeting improves outcomes beyond biomarker change remains an empirical question.

Sources:

• https://arxiv.org/pdf/2512.10981
• https://www.pharmacytimes.com/view/fda-approves-donanemab-azbt-for-early-symptomatic-alzheimer-disease
• https://pmc.ncbi.nlm.nih.gov/articles/PMC12142702/

<cite index="20-1,20-2">Staging of disease allows for the estimation of disease severity based on the identification of important points in the natural history of a disease. Grouping together patients at similar stages of disease severity permits enrichment of clinical trials</cite>. <cite index="20-6,20-7">Even restricting trial enrollment to amyloid-β positive individuals results in highly heterogenous rates of short-term cognitive decline. An alternative is to select and enroll participants who have amyloid-β positivity and either (i) no tau pathology, (ii) early tau pathology (Braak I–II), or (iii) intermediate tau pathology (Braak III–IV)</cite>.

<cite index="24-4,24-5">Alzheimer's disease staging criteria lack standardized, empirical description. Well-defined AD staging criteria are an important consideration in protocol design, influencing more standardized inclusion/exclusion criteria and defining what constitutes meaningful differentiation among the stages</cite>. The gap between conceptual frameworks (preclinical, prodromal, dementia) and operational enrollment criteria remains.

<cite index="18-4,18-5">Consensus is growing that for meaningful disease modification in AD, therapeutic intervention must be initiated at very early (preclinical or prodromal) stages of the disease, though identification and recruitment of the required asymptomatic or minimally symptomatic study participants takes many years and requires substantial funds</cite>. The logistics have not caught up to the science.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9876645/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10655199/
• https://www.nature.com/articles/s41582-022-00645-6

<cite index="8-1,8-7">The Alzheimer's Tau Platform (ATP) evaluates the safety and effectiveness of tau-directed therapies, alone or in combination with donanemab, in adults aged 50–80 with late preclinical or early prodromal Alzheimer's disease</cite>. <cite index="8-2,8-8">This platform trial allows for the simultaneous testing of multiple tau therapies under a shared master protocol</cite>—a departure from single-agent sequential trials.

<cite index="29-8">Once randomized to a regimen, participants are randomized to one of three arms: (1) tau therapy alone, (2) a combination of donanemab and tau therapy, or (3) donanemab alone</cite>. Donanemab functions as both active control and combination backbone. <cite index="8-5,8-11">The trial aims to answer whether a tau-directed therapy, alone or in combination with donanemab, reduces tau buildup in the brain compared to donanemab alone</cite>.

The trial launched with AADvac1 as the first tau-directed regimen. The structure assumes that anti-amyloid monotherapy is no longer the null hypothesis for early-stage disease; it is now the baseline against which dual-pathway targeting will be compared. No published data from ATP exist yet.

Sources:

• https://clinicaltrials.ucsf.edu/trial/NCT06957418

<cite index="1-4">TRAILBLAZER-ALZ 2 participants were stratified by their level of tau—a predictive biomarker for disease progression—into either a low-medium tau group (intermediate tau) or a high tau group, representing different pathological stages</cite>. The design made tau burden a selection variable, not just a descriptor.

<cite index="1-1">In the low-medium tau population, donanemab treatment resulted in 47% of participants showing no progression at one year on the CDR-SB assessment, versus 29% on placebo</cite>. <cite index="3-3">The trial enrolled 1,736 participants selected based on cognitive assessments in conjunction with amyloid plaque imaging and tau staging by PET imaging</cite>. The stratification was prospective, built into eligibility and randomization.

<cite index="1-2">Those at the earliest stage of disease showed 60% slowing of decline compared to placebo</cite>, though Lilly's press release does not define "earliest stage" in biomarker terms. <cite index="11-6">In the combined intermediate and high tau population, baseline MMSE was 22.3 and the placebo arm showed greater decline (worsening on iADRS and CDR-SB of 13.1 and 2.4 points respectively over 18 months)</cite>, compared to 9.3 and 1.9 points in the intermediate tau group. The field now treats tau PET as a gatekeeper, not optional context.

Sources:

• https://investor.lilly.com/news-releases/news-release-details/results-lillys-landmark-phase-3-trial-donanemab-presented
• https://investor.lilly.com/news-releases/news-release-details/lillys-donanemab-significantly-slowed-cognitive-and-functional
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10352931/

<cite index="31-5,31-6">In 2024, 54 patients underwent first stem cell collection but only five were infused with final product; there were 64 patients who received Casgevy infusion in 2025, but majority who started cell collection did not receive final product by year end</cite>. <cite index="32-13">Thirty of the 64 patients infused with Casgevy in 2025 received those infusions during the fourth quarter</cite>. This is two years post-approval.

<cite index="32-2,32-3,32-4">Manufacturing issues and challenging return-on-investment made it impossible for bluebird bio to survive as a public company; bluebird sold itself to Carlyle and SK Capital Partners in February 2025 for $30 million, a troubling signal from one of gene therapy's pioneers unable to successfully market its groundbreaking products</cite>. <cite index="29-3,29-4">Those in early stages of developing gene therapies tend to be laser-focused on gathering meaningful clinical data while simultaneously working to smooth manufacturing processes that would allow production at scale; you won't get to commercialization if you don't do those early things correctly first</cite>.

<cite index="10-11,10-12">Although off-target genome editing was not observed in edited CD34+ cells evaluated from healthy donors and patients, the risk of unintended off-target editing in an individual's CD34+ cells cannot be ruled out due to genetic variants; clinical significance of potential off-target editing is unknown</cite>. <cite index="36-3,36-5,36-6,36-7,36-8">The ongoing Phase 1/2/3 trials CLIMB-111 and CLIMB-121 are closed for enrollment; patients will be followed for approximately two years after infusion, then asked to participate in CLIMB-131, designed to follow patients for up to 15 years after infusion</cite>.

Sources:

• https://www.theglobeandmail.com/investing/markets/markets-news/Motley%20Fool/1170864/is-this-biotech-stock-your-best-shot-at-building-a-millionaire-making-position/
• https://www.biospace.com/drug-development/sickle-cell-gene-therapies-casgevy-and-lyfgenia-still-lacking-traction-2-years-in
• https://www.biospace.com/business/vertex-crisprs-casgevy-faces-complex-path-to-profitability
• https://www.casgevyhcp.com/mechanism-of-action
• https://news.vrtx.com/news-releases/news-release-details/vertex-presents-positive-long-term-data-casgevytm-0

<cite index="15-12,26-7">Vertex set Casgevy's list price at $2.2 million per patient</cite>. <cite index="26-1,26-4,26-5">The treatment is challenging to manufacture and deliver; few locations in the United States or abroad have the technical ability and expertise to deliver it</cite>. <cite index="28-8,28-9">As of mid-2025, Vertex listed 48 authorized treatment centers in the US, nearing a global goal of approximately 75 total ATCs</cite>. <cite index="28-10">About 90 patients had begun treatment by having cells collected as of May 2025</cite>.

<cite index="29-25,29-26,29-27,29-28">Patients with commercial payers, which typically issue coverage decisions within three to six months of approval, are likely first to access therapies like Casgevy; patients covered by Medicaid and Medicare will likely be next, as public payers are slower to adopt new treatments, taking as much as a year for coverage decisions</cite>. <cite index="26-8,26-9">Because lifetime cost of care for individuals with sickle cell disease or beta-thalassemia is high, covering the treatment may be sound strategy, particularly in countries with single-payer systems; in the United States, the incentive structure may prove more challenging</cite>.

<cite index="32-1,32-23">A patient advocate noted that when patients ask about Casgevy, they're being discouraged by medical providers, unsurprising because until doctors who see patients regularly are educated, gaps will remain</cite>. <cite index="32-17,32-18">Victoria Gray, the first person with sickle cell disease treated with Casgevy, said the busulfan conditioning regimen was the most challenging part; in vivo gene therapy could open access to more patients by eliminating need for chemotherapeutic agents</cite>.

Sources:

• https://www.genengnews.com/topics/genome-editing/fda-approves-casgevy-the-first-crispr-therapy-for-sickle-cell-disease/
• https://innovativegenomics.org/news/crispr-clinical-trials-2024/
• https://www.bioprocessintl.com/therapeutic-class/time-and-distance-vertex-tackles-casgevy-access-challenges-for-gene-therapy-patients
• https://www.biospace.com/business/vertex-crisprs-casgevy-faces-complex-path-to-profitability
• https://www.biospace.com/drug-development/sickle-cell-gene-therapies-casgevy-and-lyfgenia-still-lacking-traction-2-years-in

<cite index="30-6,30-13">The manufacturing process takes up to six months from the time cells are collected to when the finished product is sent back to the treating center</cite>. <cite index="29-20,29-21,29-22">Cell collection can require three to four attempts to harvest enough viable cells; those cells are then sent to a manufacturer for CRISPR editing to promote fetal hemoglobin expression before being transfused back</cite>.

<cite index="31-1,31-4,31-5">Harvesting viable stem cells from sickle cell disease patients has been extremely challenging; in 2024, 54 patients underwent first stem cell collection but only five were infused with final product</cite>. <cite index="26-20">Manufacturing sometimes fails, produces low-potency cells, or patients die of their disease while waiting for manufacturing to complete</cite>. <cite index="30-1,30-2,30-3,30-4">Rescue cells—unedited autologous stem cells—are collected and stored in case modified cells fail to engraft; if rescue cells are given, there is no treatment benefit from Casgevy</cite>.

<cite index="26-12">The chemotherapy required before administering the CRISPR treatment is tough on patients and carries risk of serious side effects</cite>. <cite index="44-3,44-4">Safety profile was generally consistent with myeloablative busulfan conditioning and autologous hematopoietic stem cell transplant; all patients engrafted neutrophils and platelets after infusion</cite>. <cite index="39-10">There were seven cases (12.5%) of hepatic veno-occlusive disease in the beta-thalassemia trial, all related to busulfan, all resolved after defibrotide treatment</cite>.

Sources:

• https://ir.crisprtx.com/news-releases/news-release-details/vertex-and-crispr-therapeutics-announce-us-fda-approval/
• https://www.biospace.com/business/vertex-crisprs-casgevy-faces-complex-path-to-profitability
• https://www.theglobeandmail.com/investing/markets/markets-news/Motley%20Fool/1170864/is-this-biotech-stock-your-best-shot-at-building-a-millionaire-making-position/
• https://innovativegenomics.org/news/crispr-clinical-trials-2024/
• https://thalassaemia.org.cy/clinical-trial-updates/exa-cel-gene-editing-thal/
• https://www.medthority.com/news/2023/6/climb-111-and-climb-121-phase-iii-trials-of-exa-cel-meet-their-primary-endpoint-in-beta-thalassemia-or-severe-sickle-cell-disease.--vertex-pharma

<cite index="7-8,9-2,10-1,10-2">Casgevy is manufactured by harvesting a patient's CD34+ hematopoietic stem cells, editing them ex vivo via electroporation to introduce the CRISPR/Cas9 ribonucleoprotein complex, then infusing the modified cells back as an autologous transplant</cite>. This is not in vivo therapy. <cite index="22-1,22-13">In vivo approaches would deliver CRISPR components directly into the patient via viral or non-viral vectors; ex vivo strategies require collection, in vitro editing, and re-infusion</cite>.

The distinction matters. <cite index="30-6,30-13">Manufacturing and testing Casgevy takes up to six months from cell collection to product release</cite>. <cite index="26-16">All approved sickle cell and beta-thalassemia CRISPR therapies are ex vivo; cells are removed, edited and quality-controlled in a specialized facility, then reinfused after intensive chemotherapy</cite>. <cite index="26-15,26-17">CRISPR Therapeutics and Beam are pursuing in vivo strategies for these indications, which would deliver genome-editing components directly into the patient's body</cite>.

<cite index="7-10,10-5">Casgevy's single-guide RNA targets the erythroid-specific enhancer region of the BCL11A gene, disrupting BCL11A expression to increase fetal hemoglobin production</cite>. <cite index="36-11,36-12,40-9">In pivotal trials CLIMB-121 and CLIMB-131, 93% of sickle cell patients achieved freedom from vaso-occlusive crises for at least 12 consecutive months, with mean VOC-free duration of 30.9 months; in CLIMB-111 and CLIMB-131, 98.2% of beta-thalassemia patients achieved transfusion independence for 12+ months</cite>. <cite index="2-3,2-4,2-5">The UK MHRA approved Casgevy in November 2023, FDA in December 2023, and EMA in December 2023</cite>.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11352399/
• https://www.rarediseaseadvisor.com/therapies/casgevy-exagamglogene-autotemcel/
• https://www.casgevyhcp.com/mechanism-of-action
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10867117/
• https://ir.crisprtx.com/news-releases/news-release-details/vertex-and-crispr-therapeutics-announce-us-fda-approval/
• https://news.vrtx.com/news-releases/news-release-details/vertex-presents-positive-long-term-data-casgevytm-0
• https://news.vrtx.com/news-releases/news-release-details/vertex-presents-longer-term-data-2025-european-hematology
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10913280/

<cite index="2-4">As of spring 2024, less than 25% of eligible adults ≥60 had received an RSV vaccine</cite>, and <cite index="2-6">about 6,000 to 10,000 people ≥65 die from RSV annually in the U.S.</cite> <cite index="9-1">CDC's ACIP recommended a shared decision-making approach between age-eligible adults and providers for a single dose of either RSVPreF3 (Arexvy) or RSVpreF (Abrysvo)</cite>—not a universal recommendation. <cite index="2-7">Vaccine newness breeds distrust, compounded by general decline in confidence in vaccines and health care</cite>. The passive phrasing matters: shared clinical decision-making is weaker than a category A recommendation, and it shows in coverage. For maternal immunization, <cite index="15-2,15-9">RSV vaccine acceptance across 17 studies ranged from 39% in France to 87% in the Netherlands, with safety concerns and cultural context influencing attitudes</cite>. These are interventions with demonstrated efficacy in controlled trials meeting regulatory thresholds. The gap between efficacy and effectiveness widens when uptake is this low.

Sources:

• https://fortune.com/well/article/rsv-vaccine-seniors-older-adults-pfizer-moderna-gsk/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC11779051/
• https://www.explorationpub.com/Journals/eaa/Article/100988

<cite index="13-8">On August 21, 2023, FDA approved Abrysvo for active immunization of pregnant individuals at 32 through 36 weeks gestational age for prevention of LRTD in infants from birth through 6 months of age</cite>. <cite index="16-1">The phase 3 trial indicated efficacy against medically-attended RSV-associated LRTI over 90 days of 57.1% (99.5% CI 14.7 to 79.8)</cite>, and <cite index="12-3">FDA approval cited 80% efficacy in protecting infants up to 90 days post-birth</cite>. <cite index="13-4">FDA narrowed the interval to 32–36 weeks to avoid potential risk of very preterm or extremely preterm births with use before 32 weeks</cite>. GSK had pursued a maternal candidate, but <cite index="15-8">trials of RSVPreF3-Mat revealed higher preterm birth rates (6.8% vs. 4.9%) and a numerical imbalance in infant deaths (0.4% vs. 0.2%), prompting early termination</cite>. This is the first licensed maternal vaccine for RSV. Real-world data are beginning to emerge: one observational study found infants born to vaccinated mothers were 61% less likely to have RSV infection, though the cohort had only 7.7% maternal vaccination uptake.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11528063/
• https://www.medrxiv.org/content/10.1101/2025.04.16.25325979.full.pdf
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12359126/
• https://www.explorationpub.com/Journals/eaa/Article/100988
• https://www.epicresearch.org/articles/maternal-rsv-vaccine-effective-in-reducing-rsv-infections-and-hospitalizations-in-infants

<cite index="23-9">A systematic review pooling RSV subunit vaccine trials found a risk ratio of 0.32 (95% CI 0.22–0.44) for RSV-associated lower respiratory tract infection and 0.13 (95% CI 0.06–0.29) for severe RSV-LRTI</cite>. In individual pivotal trials, <cite index="21-1">GSK's adjuvanted RSVPreF3 OA was efficacious against RSV-LRTD in adults ≥60 over one season</cite>, and <cite index="25-1,25-8">Moderna reported efficacy of 78.7% (95% CI 62.8–87.9%) preventing symptomatic RSV-LRTD with two or more symptoms and 80.9% (95% CI 50.1–92.7%) with three or more symptoms</cite>. <cite index="1-11,1-12">A small number of adults ≥60 receiving Arexvy or Abrysvo developed serious neurologic conditions including Guillain-Barré syndrome within 42 days; given the small number, it remained unclear whether the vaccines caused these events</cite>. <cite index="3-7,3-8">The vaccines are not currently annual; people who have received one dose should not receive another at this time</cite>. These are point estimates from controlled settings. Duration of protection remains under surveillance across multiple RSV seasons.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11360664/
• https://academic.oup.com/cid/article/78/6/1732/7585312
• https://www.cdc.gov/mmwr/volumes/73/wr/mm7332e1.htm
• https://www.cdc.gov/vaccine-safety/vaccines/rsv.html
• https://www.cdc.gov/rsv/hcp/vaccine-clinical-guidance/older-adults.html

<cite index="2-2,2-3">GSK's Arexvy received FDA approval on May 3, 2023, followed by Pfizer's Abrysvo on May 31, 2023</cite>—the first RSV vaccines licensed for any population. <cite index="1-1,1-3">Both were approved for adults aged 60 and older</cite>, and <cite index="1-7">Moderna's mResvia followed in 2024</cite>. The approvals close a developmental stall dating to 1967, when <cite index="29-3,29-4">two toddlers vaccinated with a formalin-inactivated RSV vaccine died from enhanced respiratory disease, and up to 80% of immunized children required hospitalization upon subsequent wild-type virus exposure</cite>. <cite index="28-4,28-5">That vaccine produced antigens not processed in the cytoplasm, yielding nonprotective antibody and pathogenic Th2 memory with eosinophil and immune complex deposition in the lungs</cite>. The structural solution came decades later: <cite index="35-5,35-6">elucidation of the F protein's three-dimensional structure revealed pre- and postfusion conformations, enabling structure-based antigen engineering</cite>. All three licensed products are stabilized prefusion F subunit vaccines—Arexvy is adjuvanted, Abrysvo is not, and mResvia uses mRNA encoding prefusion F.

Sources:

• https://fortune.com/well/article/rsv-vaccine-seniors-older-adults-pfizer-moderna-gsk/
• https://www.cdc.gov/vaccine-safety/vaccines/rsv.html
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10712289/
• https://journals.asm.org/doi/10.1128/cvi.00609-15
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12104812/

<cite index="12-5,12-6,12-7">In data from 173 adult patients receiving CAR-T cell products, the incidence of grade 3 CRS was 6.6% for axi-cel, 3.3% for tisa-cel, and 10% for brexu-cel recipients; grade 4 CRS was documented in 2.5% and 5% in axi-cel and brexu-cel recipients, while grade 5 CRS was recorded only in brexu-cel (10%); severe ICANS was less frequent, with grade 3 and 4 rates of 7.5% and 2.5% for axi-cel, while brexu-cel documented only grade 3 (10%).</cite>

<cite index="21-2">The OUTREACH study showed outpatient liso-cel administration is feasible and safe outside of academic center settings.</cite> <cite index="26-6,26-8">There is significant variability in the incidence and severity among CAR T products with axi-cel having highest incidence of severe CRS and ICANS; a recent phase 2 study showed safety and feasibility of outpatient administration of lisocabtagene maraleucel (liso-cel) at community cancer centers which has much lower risk for high grade CRS and ICANS compared to axi-cel.</cite>

The primary literature consistently shows that CD28-containing constructs (axi-cel) carry higher rates of severe toxicity compared to 4-1BB constructs (tisa-cel, liso-cel). This is not a marginal difference—it's the difference between protocols that require ICU readiness and protocols that can be managed with same-day clinic access. Any outpatient administration strategy must be product-specific, and community centers attempting to replicate academic protocols need to stratify by construct, not just by disease indication.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12162571/
• https://ncbi.nlm.nih.gov/pmc/articles/PMC11109356/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12151859/

<cite index="18-1,18-5">A study demonstrated that the administration of commercial CAR T-cell therapies in an outpatient setting is safe and feasible without intensive remote monitoring using an early CRS intervention strategy, using outpatient administration of all commercially available CD19- and BCMA-directed CAR T-cell therapy with a strategy of no remote at-home monitoring and an early cytokine release syndrome intervention strategy.</cite> <cite index="18-2,18-3">Commercial CAR T-cell therapies including CARs with CD28 domain may be given safely as outpatient without intense home monitoring; low-grade CRS can be managed in the outpatient setting with a well-structured system.</cite>

<cite index="22-7,22-8,22-9,22-11">A total of 68 patients were treated with outpatient tisa-cel; any grade CRS rate was 40.3%, with no reported grade 3-5 CRS events; hospitalization was required in 19.4% of patients within 72 hours, and in 36.1% within 30 days of product infusion, and this analysis demonstrated the safety and feasibility of outpatient tisa-cel administration, with 64% of patients not requiring hospitalization.</cite> <cite index="26-6">There is significant variability in the incidence and severity among CAR T products with axi-cel having highest incidence of severe CRS and ICANS.</cite>

The literature shows a clear shift from inpatient-only protocols to structured outpatient administration with frequent clinic visits and early intervention thresholds. But the heterogeneity across products—especially CD28 versus 4-1BB costimulatory domains—means that safety profiles are not transferable. Each product carries its own toxicity signature, and outpatient feasibility depends on institutional capacity to intervene within hours, not days.

Sources:

• https://www.sciencedirect.com/science/article/pii/S2473952924003756
• https://ashpublications.org/bloodadvances/article/8/16/4320/516586/Outpatient-administration-of-CAR-T-cell-therapies
• https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2024.1412002/full
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12151859/

<cite index="10-17">ICANS occurs in 20–60% of patients, of whom 12–30% have severe (≥ grade 3) symptoms.</cite> <cite index="10-1,10-2">Management is supportive for grade 1 ICANS, and dexamethasone with rapid taper is given for grade ≥2 ICANS; suggested doses include 10–20 mg intravenous dexamethasone every 6 h for grades 2–3 and 1 g IV methylprednisolone for at least 3 days for grade 4 until symptoms improve.</cite>

<cite index="11-7,11-8,11-9">In a prospective observational study of 15 consecutive r/r DLBCL patients, ICANS occurred in 4/15 patients (27%) within 6 days (4–6 days) after CAR T cell infusion; patients with ICANS grade 2 (n = 3) exhibited similar neurological symptoms including apraxia, expressive aphasia, disorientation, and hallucinations, while brain MRI was inconspicuous in either case, and treatment with dexamethasone rapidly resolved the clinical symptoms in all three patients.</cite>

<cite index="17-1,17-2,17-3">Data from comprehensive analysis showed approximately 70% of patients experiencing no neurotoxicity and only 5% to 7% developing grade 3 or higher events; the median time to onset was approximately 1 week post infusion, with events typically resolving within the same timeframe, though late-onset ICANS (beyond 15 days) occurred in 12% of clinical trial patients and 5% of real-world patients.</cite> <cite index="15-2">Delayed neurotoxicity with seizures or episodes of confusion occurred during the third or fourth week after CAR-T-cell therapy in approximately 10% of patients.</cite> These late events—beyond the typical monitoring window—represent a risk that outpatient protocols must account for.

Sources:

• https://www.ncbi.nlm.nih.gov/books/NBK584157/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8744256/
• https://www.cancernetwork.com/view/understanding-icans-management-with-car-t-low-grade-events-with-high-resolution-rates
• https://www.theattcnetwork.co.uk/wp-content/uploads/2021/02/Management-of-Cytokine-Release-Syndrome-Neurotoxicity-and-CAR-T-cell-related-encephalopathy.pdf

<cite index="6-2,6-7">Tocilizumab, an anti-IL-6 receptor antagonist, is the standard for CRS management, but optimal timing of administration is unclear.</cite> <cite index="1-1,1-2">Grade 1 CRS (fever only) is typically managed conservatively; Grade 2–4 CRS mandates pharmacologic intervention.</cite> <cite index="9-4">For patients with mild CRS (grade 1-2), supportive care alone with antihistamines, antipyretics, intravenous fluids, and close monitoring may be sufficient.</cite>

<cite index="4-3,4-4">In a retrospective cohort study of 45 R/R B-ALL patients, 17 patients received tocilizumab, resulting in a significant reduction in the duration of grade 3 CRS compared to those who did not receive the drug; importantly, tocilizumab did not impair CAR-T cell expansion or efficacy, nor did it increase the incidence of adverse events.</cite> <cite index="5-6,5-8">The incidence of CRS for ALL, lymphoma, and multiple myeloma were 82%, 90%, and 90% respectively; according to published data, more than 54–91% of patients may develop different grades of CRS during treatment.</cite>

<cite index="12-8">Studies have indicated that early intervention with tocilizumab and/or steroids with lower grade toxicity can significantly prevent onset of more severe CRS and its subsequent effects on neurotoxicity.</cite> But this remains an open question—when you intervene, you affect the cytokine cascade, and the clinical community has not converged on the threshold for intervention versus observation.

Sources:

• https://www.astctjournal.org/article/S1083-8791(18)31579-9/fulltext
• https://link.springer.com/article/10.1007/s44272-025-00044-0
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11949925/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7940756/
• https://cdn.clinicaltrials.gov/large-docs/72/NCT05191472/Prot_SAP_000.pdf
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12162571/

<cite index="1-4,1-5">We included 6 multinational double-blind randomized placebo-control trials that included a total of 52821 T2DM patients. The results indicated that GLP-1 receptor agonists reduced the risk of death from cardiovascular causes (RR: 0.90; 95% CI: 0.83–0.97; P = 0.004) and fatal or non-fatal stroke (RR: 0.85; 95% CI: 0.77–0.94; P = 0.001) compared with the placebo controls.</cite> <cite index="1-6">But GLP-1 receptor agonists did not significantly alter the fatal or non-fatal myocardial infarction compared with the placebo (RR: 0.91; 95% CI: 0.82 – 1.01; P = 0.06).</cite>

This meta-analysis pooled data through 2019 and reflected trials in type 2 diabetes cohorts only. <cite index="7-3">Beyond their glucose-lowering effect, GLP-1 RAs have demonstrated weight loss, improved blood pressure, lipid profiles, and anti-inflammatory effects, suggesting pleiotropic mechanisms of cardiovascular protection.</cite> <cite index="4-5">GLP-1 agonists achieved substantial reductions in myocardial infarction, stroke, cardiovascular mortality, and heart failure events according to major cardiovascular outcome trials (CVOTs).</cite> <cite index="7-6">Based on the GRADE approach, the certainty of evidence was rated as 'High' for all evaluated outcomes, including MACE, cardiovascular death, and all-cause death.</cite>

The cardiovascular signal across the class is consistent in diabetic populations. The SELECT trial extended the evidence base into obesity without diabetes. The composite reduction in MACE appears driven primarily by stroke and cardiovascular death; the myocardial infarction component remains statistically equivocal in most trials. Mechanistic explanations remain incompletely characterized—weight loss, blood pressure reduction, lipid modification, anti-inflammatory effects, and direct vascular or myocardial actions have all been proposed, but their relative contributions are not quantified.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9097124/
• https://www.mdpi.com/2077-0383/14/19/6758
• https://academic.oup.com/ehjcvp/article/11/6/552/8172530
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12450594/

<cite index="19-1,19-6">The drugs come at a high cost for Americans, who pay around $1,000 per month if their health insurance does not cover it.</cite> <cite index="20-1">Monthly costs of $900-$1300 resulted in GLP-1RAs accounting for over 13% of insurance premium costs in 2023, prompting insurance coverage restrictions.</cite> <cite index="19-7">Insurers rarely cover the weight management formulations, so many who might benefit from the drug but cannot afford to pay go without.</cite>

International price variation is extreme. <cite index="19-8,19-9">In the study appendix, investigators found that the price of Wegovy varies by up to five times across different countries, with the U.S. paying the most at $1,349 and Japan paying the least at around $280. This is the most detailed public data on international prices so far.</cite> The Institute for Clinical and Economic Review (ICER) has judged net prices for semaglutide and tirzepatide as meeting cost-effectiveness thresholds in the US market, but <cite index="17-7,17-8">over 40% of US adults have obesity, translating into more than 100 million potential new users of OMs. Standing in the way of the major opportunity to improve health for these individuals is the massive and likely ongoing cost of treating such a large segment of the population.</cite>

<cite index="21-1,21-2,21-3">It is a well-established fact that the populations with the highest prevalence of obesity and its complications are disproportionately socioeconomically disadvantaged. At present, only a minority of these people can access these costly medications at subsidised rates through public healthcare, while an explosion in private clinics and prescriptions provides a rapid route to GLP-1 agonist treatment for people with disposable income. This divide enables wealthier individuals to access weight loss and metabolic risk reduction that poorer patients cannot obtain.</cite> <cite index="23-10">Our model, which estimates that between ~2.1 million (for people with T2DM or obesity with CVD) and ~3.1 million (for people with T2DM or obesity with/without CVD) lives are lost annually, highlights the critical importance of overcoming barriers to access.</cite>

The efficacy is real. The access chasm is widening.

Sources:

• https://ldi.upenn.edu/our-work/research-updates/key-lessons-for-ethical-and-affordable-access-to-glp-1-drugs-like-ozempic-and-wegovy/
• https://www.medrxiv.org/content/10.1101/2025.10.24.25338255.full.pdf
• https://becarispublishing.com/doi/10.57264/cer-2025-0083
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12712549/
• https://www.medrxiv.org/content/10.1101/2024.11.11.24317112.full.pdf

<cite index="13-1,13-8">In the SURMOUNT-1 trial, participants with obesity taking tirzepatide 15 mg lost about 22.5% of their body weight over 72 weeks.</cite> <cite index="15-4">Compared with the semaglutide, tirzepatide could produce significantly greater weight loss (MD = 4.23; 95% confidence interval (CI): 3.22 - 5.25; P < 0.01).</cite> <cite index="9-1,9-4">The research found Mounjaro reduced body weight in people with obesity by just over 20 per cent, whereas Wegovy cut it by nearly 14 per cent.</cite>

<cite index="14-2">Tirzepatide, a glucose-dependent insulinotropic polypeptide and GLP-1 receptor agonist, was approved for type 2 diabetes mellitus (T2DM; Mounjaro®) in May 2022 and weight management (Zepbound®) in November 2023.</cite> The dual agonism—targeting both GLP-1 and GIP receptors—appears to differentiate efficacy from semaglutide's single-receptor mechanism.

But cardiovascular outcome data for tirzepatide remain indirect. <cite index="41-2">Recently released topline results from the SURPASS-CVOT (NCT04255433) trial reported that tirzepatide was non-inferior to dulaglutide with an 8% (HR, 0.92; 95.3% CI [0.83, 1.01]) lower rate of MACE-3 events while demonstrating greater reductions in HbA1C and weight (detailed results are yet to be published).</cite> <cite index="41-3">A pre-specified indirect comparison of the REWIND and SURPASS-CVOT studies found that tirzepatide reduced the risk of MACE-3 by 28% (HR, 0.72; 95% CI [0.55–0.94]) compared to placebo.</cite> <cite index="41-4,41-5">The effect of tirzepatide treatment on cardiovascular events and outcomes is currently being assessed in the SURMOUNT-MMO (NCT05556512) trial. Tirzepatide has demonstrated promising results in terms of weight reduction and glycemic control, which in the long term may translate into cardiovascular benefits. However, its precise cardiovascular effects remain to be fully elucidated.</cite>

Promising is not proven. The field needs primary endpoint data from a dedicated CVOT in a nondiabetic obesity cohort, not post-hoc risk modeling.

Sources:

• https://www.drugs.com/medical-answers/wegovy-mounjaro-works-best-weight-loss-3580291/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC12151102/
• https://www.bhf.org.uk/informationsupport/heart-matters-magazine/news/behind-the-headlines/mounjaro-vs-wegovy
• https://pmc.ncbi.nlm.nih.gov/articles/PMC12579654/
• https://dom-pubs.onlinelibrary.wiley.com/doi/full/10.1111/dom.70143

<cite index="30-2">In the SELECT cardiovascular outcomes trial, semaglutide showed a 20% reduction in major adverse cardiovascular events in 17,604 adults with preexisting cardiovascular disease, overweight or obesity, without diabetes.</cite> This is the first time a GLP-1 receptor agonist has demonstrated cardiovascular protection in a nondiabetic cohort—a critical distinction, since <cite index="27-14">previous GLP-1RA cardiovascular outcome trials were conducted predominantly in cohorts with type 2 diabetes, where interpretation of cardiovascular benefits was confounded by concurrent glycaemic effects.</cite>

<cite index="28-3,28-4">In patients treated with semaglutide, weight loss continued over 65 weeks and was sustained for up to 4 years. At 208 weeks, semaglutide was associated with mean reduction in weight (−10.2%), waist circumference (−7.7 cm) and waist-to-height ratio (−6.9%) versus placebo (−1.5%, −1.3 cm and −1.0%, respectively; P < 0.0001 for all comparisons versus placebo).</cite> The primary endpoint—a composite of cardiovascular death, nonfatal myocardial infarction, or nonfatal stroke—occurred in 13% of the treatment arm versus 14.9% of the placebo arm. <cite index="26-1,26-2">In people with overweight or obesity and established atherosclerotic cardiovascular disease but not diabetes, semaglutide reduced cardiovascular events irrespective of baseline HbA1c or change in HbA1c. Thus, semaglutide is expected to confer cardiovascular benefits in people with established atherosclerotic cardiovascular disease who are normoglycemic at baseline and/or in those without HbA1c improvements.</cite>

The trial design enrolled adults aged ≥45 years with BMI ≥27 kg/m² and established cardiovascular disease but A1C <6.5% and no history of diabetes. It was double-blind, placebo-controlled, event-driven, and multinational. The reduction in MACE held across baseline BMI and demographic subgroups.

Sources:

• https://www.nature.com/articles/s41591-024-02996-7
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11271387/
• https://diabetesjournals.org/care/article/47/8/1360/156810/Semaglutide-and-Cardiovascular-Outcomes-by
• https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(25)01375-3/fulltext

<cite index="16-4,16-5">Many questioned how scientific evidence, expert advice, and patient best interests were considered in the approval decision—the shared FDA and Biogen interpretation that high-dose aducanumab was substantially clinically effective avoided conventional scientific scrutiny, was advanced by patient groups who had been major Biogen fund recipients, and raised concerns about insufficient safeguards against regulatory capture.</cite>

<cite index="15-5,15-13">Even though FDA's final decision was accelerated approval, the Office of Clinical Pharmacology recommended full approval based on its own analyses—the agency published key reviews (medical review supporting both full and accelerated approval, statistical review supporting no approval, clinical pharmacology review supporting full approval) only 3 weeks after the June 7, 2021 decision.</cite> <cite index="15-19,15-20">After the advisory committee meeting, FDA conducted additional analyses showing an extremely low chance (<1 in 10,000,000) of observing the positive Study 302 results if aducanumab was assumed ineffective, which significantly increased confidence in the evidence for full approval.</cite>

<cite index="11-2,11-4">The controversial decision has motivated multiple policy reforms—but structural reforms are needed to reshape FDA's core priorities and restore the regulatory system's commitment to scientific rigor.</cite> <cite index="11-9,11-10">Congress should convene a new body with scientific expertise and independence from FDA, industry, and patients to assist decision-making—three experts resigned from the advisory committee after FDA approved aducanumab.</cite>

Sources:

• https://www.nature.com/articles/s41582-021-00557-x
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10193636/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10937179/

<cite index="1-10">The results of anti-amyloid monoclonal antibody trials support the amyloid hypothesis and amyloid as a target for AD drug development.</cite> <cite index="3-2,3-3,3-4">The amyloid cascade model is well supported in observational studies; its therapeutic corollary asserts that amyloid removal would provide clinical benefits—but after 2 decades of pursuing amyloid removal without success, only recent trials of donanemab and lecanemab have reported clinical benefits linked to removal.</cite>

<cite index="9-4,9-5">To Nobel laureate Thomas Südhof, rather than proving the amyloid hypothesis, Leqembi does the opposite—because after removing amyloids, symptoms of dementia still steadily worsen.</cite> The gap between target engagement and meaningful clinical change remains. <cite index="6-6,6-7">Lecanemab does not cure Alzheimer's disease but modestly slows progression in earliest stages—the large clinical study showed it slowed progression by about 20–30% after 18 months.</cite>

<cite index="30-6,30-8">A recent Bayesian meta-analysis found a strong surrogate relationship between treatment effects on amyloid and CDR-SOB across all MABs (slope 1.41, 95% CI 0.60–2.21), but for individual treatments the relationships were suboptimal with large uncertainty—sharing information across treatments resulted in moderate surrogate relationships for aducanumab and lecanemab.</cite> <cite index="28-1">FDA's accelerated approval of aducanumab marked acceptance of beta-amyloid as a surrogate endpoint for Alzheimer's despite no prior clinical evidence linking the biomarker to cognitive outcomes.</cite>

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC10195708/
• https://www.neurology.org/doi/10.1212/WNL.0000000000207438
• https://www.statnews.com/2025/02/11/amyloid-hypothesis-alzheimers-research-lecanemab-aduhelm/
• https://memory.ucsf.edu/lecanemab
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12711098/
• https://www.biocentury.com/article/637001/amyloid-becomes-surrogate-endpoint-for-alzheimer-s-despite-no-clear-link-to-cognitive-outcomes

<cite index="22-6,22-8">In the Clarity AD trial, lecanemab reduced clinical decline on CDR-SB by 27% at 18 months compared to placebo (difference −0.45; 95% CI −0.67 to −0.23; P<0.001).</cite> <cite index="21-3,21-4">The trial enrolled 1,795 patients with MCI or mild AD dementia with confirmed amyloid pathology, randomized 1:1 to receive infusions of 10 mg/kg every 2 weeks or placebo.</cite> <cite index="23-1,23-7">As a postmarketing requirement of accelerated approval, FDA required a confirmatory trial to verify anticipated clinical benefit—the agency called it "the first verification that a drug targeting the underlying disease process of Alzheimer's disease has shown clinical benefit."</cite>

<cite index="3-11,3-12">The demonstration that lecanemab delayed clinical progression is a major conceptual achievement, though ARIA—largely asymptomatic—occurred in approximately 20%, slightly more than half attributable to treatment, the rest to underlying amyloid angiopathy.</cite> <cite index="20-8,20-9">According to reports at AAIC, two people died while receiving Leqembi in clinical care, both developing severe ARIA-E possibly resembling cerebral amyloid angiopathy-related inflammation.</cite> <cite index="4-2">In direct comparison with aducanumab and gantenerumab, lecanemab bound most strongly to amyloid protofibrils, while the others preferred more highly aggregated forms.</cite>

<cite index="18-11,18-12">When lecanemab received accelerated approval 19 months after aducanumab, the Medicare NCD limiting coverage to clinical trials automatically applied—even though the data was more promising—but after full approval, CMS confirmed broader Medicare coverage.</cite>

Sources:

• https://www.nejm.org/doi/full/10.1056/NEJMoa2212948
• https://investors.biogen.com/news-releases/news-release-details/new-clinical-data-demonstrates-three-years-continuous-treatment
• https://www.fda.gov/news-events/press-announcements/fda-converts-novel-alzheimers-disease-treatment-traditional-approval
• https://www.neurology.org/doi/10.1212/WNL.0000000000207438
• https://www.alzforum.org/therapeutics/leqembi
• https://www.ajmc.com/view/biogen-abandons-aducanumab-pivots-focus-to-lecanemab-for-alzheimer-disease

<cite index="10-3,10-4">In June 2021, FDA granted aducanumab accelerated approval using an unvalidated surrogate measure—beta-amyloid reduction—as the basis, despite concerns about lack of clinical outcome benefit.</cite> <cite index="12-6,12-7">Two Phase 3 trials, EMERGE and ENGAGE, had identical design, but only the high-dose arm in EMERGE demonstrated cognitive improvement.</cite> <cite index="12-20,12-21">Despite the FDA advisory committee voting 10-0 to reject approval and statistician reviewers also rejecting it, accelerated approval was granted—three committee members resigned afterward.</cite>

<cite index="10-6,10-7">Among 214 physicians surveyed who were familiar with the decision, 184 (86%) would not prescribe or recommend aducanumab, and 143 (67%) reported losing trust in other drugs approved through accelerated approval.</cite> <cite index="14-2,14-14">Medicare declined coverage outside clinical trials in April 2022, and private insurers followed—they couldn't justify the cost of a treatment with no guaranteed functional improvement and potentially costly side effects, even after the price dropped from $56,000 to $28,200 annually.</cite> <cite index="12-8,12-10">Biogen abandoned the drug's development in early 2024, halting the postmarketing confirmatory trials FDA required, to focus on lecanemab instead.</cite>

The controversy exposed structural questions about accelerated approval. <cite index="13-17,13-18">An OIG report identified reliance on unplanned analyses, approvals despite internal objections, and inadequate documentation—scrutiny intensified as the pathway faces questions about surrogate endpoints not guaranteeing efficacy.</cite>

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC10869173/
• https://wmjonline.org/pharmacotherapy-update/aduhelm_controversy/
• https://www.neurologyadvisor.com/features/aducanumab-controversy-accelerated-fda-approval-agency-distrust/
• https://www.pharmacytimes.com/view/fda-s-accelerated-approval-pathway-under-scrutiny-for-key-drug-approvals

<cite index="6-7">N-of-1 designs allowed better estimation of patient-level random effects</cite>, which is the argument for using linear mixed models in aggregated analyses. <cite index="19-6">In a mixed-effects model for repeated measures, we already have a name for these coefficients: random effects</cite>. The framework exists. The question is whether the assumptions hold.

<cite index="5-5">Aggregated N-of-1 trials analyze a cohort of such participants, and can be designed to optimize both statistical power and clinical or logistical constraints, such as allowing all participants to begin with an open-label stabilization phase</cite>. That's methodologically sound if—and only if—the washout period is adequate and the condition is stable. <cite index="5-12,5-13">Trial design 4 had slightly lower power than the traditional crossover design, although power declined much more rapidly as carryover was introduced</cite>. Carryover isn't a nuisance parameter. It's a structural threat. <cite index="6-8">These results reinforce the need to account for these factors when planning N-of-1 trials</cite>. The literature acknowledges this. I'm less confident that the protocols do.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6955665/
• https://statsof1.org/how-can-you-generalize-an-n-of-1-trial_s-results-to-another-person/
• https://www.frontiersin.org/journals/digital-health/articles/10.3389/fdgth.2020.00013/full

<cite index="10-8,10-9">N-of-1 or single subject trials are meant to objectively assess an individual patient's response to an intervention by collecting enough data, under an appropriate study design, on that patient to enable valid statistical inferences to be drawn about the effect of that intervention. N-of-1 studies often leverage cross-over designs (repeated crossover designs, in particular, such as 'ABABAB' designs) to allow comparisons of a test intervention with a comparator (or placebo) intervention</cite>.

But the indication set is narrow. <cite index="12-10,12-11">Acute conditions will tend to resolve (or progress) before personalized trials can be completed. Rapidly progressive or fatal conditions are likewise unsuitable</cite>. The design is optimized for <cite index="18-5">type 2 N-of-1 trial tests treatments designed specifically for a patient with a rare disease, to facilitate personalized medicine</cite>, or for chronic symptom management where <cite index="12-14">the statistical reliability of single-patient, multiple-crossover trials increases the more often outcomes are assessed</cite>. Palliative care, chronic pain, fatigue—these are the canonical use cases. That's a different population than most pivotal oncology trials enroll.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC6388397/
• https://hdsr.mitpress.mit.edu/pub/hy7mjtzo
• https://ascpt.onlinelibrary.wiley.com/doi/10.1002/cpt.2425

<cite index="21-2,21-4">N-of-1 trials have been criticized for their lack of generalizability</cite>, and the concern is structural. <cite index="25-2">An issue when conducting aggregated N-of-1 trials (i.e., data aggregated from a series of N-of-1 studies) is the generalizability (external validity) of their results</cite>. <cite index="25-3">Selection bias in aggregated N-of-1 trials could result from not including a sufficiently sized, representative sample of participants</cite>.

The tension is real: <cite index="20-3">the trial's robustness to high heterogeneity of effects allows inclusion of a broader range of patients</cite>, which sounds like a strength until you recognize that <cite index="22-9">if an N-of-1 trial excludes many patients in the run-in phase, this can decrease the generalizability of findings</cite>. The Alberta group tried to address this. <cite index="21-6,21-7">A comprehensive systematic review found that the majority (60%) of published N-of-1 trials are published as a series, suggesting their value beyond assessing individual treatment effects and their potential to provide more generalizable treatment effects</cite>. That doesn't resolve the issue—it just documents that investigators attempt aggregation. <cite index="19-8">The results from any particular N-of-1 trial are only meant to generalize to the average patterns in that individual's history</cite>. Aggregation doesn't magically confer external validity.

Sources:

• https://www.jameslindlibrary.org/articles/history-development-n-1-trials/
• https://www.mdpi.com/2227-9032/7/4/137
• https://hdsr.mitpress.mit.edu/pub/x0akqpvz
• https://www.sciencedirect.com/science/article/pii/S258998641930139X
• https://statsof1.org/how-can-you-generalize-an-n-of-1-trial_s-results-to-another-person/

<cite index="2-2">Aggregating the results of many N-of-1 trials allows a treatment effect to be ascertained for a patient population</cite>, and <cite index="3-11">individual (N-of-1) trials can be combined to give population comparative treatment effect estimates</cite>. The statistical promise is substantial: <cite index="6-5">aggregated N-of-1 trials outperformed both traditional parallel RCT and crossover designs when these trial designs were simulated in terms of power and required sample size</cite>, and <cite index="6-10">interventions can be tested with adequate power with far fewer patients than traditional RCT and crossover designs</cite>.

But the method faces recognized hazards. <cite index="6-6">N-of-1 designs resulted in a higher type-I error probability than parallel RCT and crossover designs when moderate-to-strong carryover effects were not considered or in the presence of modeled selection bias</cite>. That's not a minor caveat—it's a structural vulnerability. <cite index="7-2">Popularly used methods include non-parametric methods like the Wilcoxon signed-rank test, two-sample mean tests, methods that allow for covariate adjustments like linear models, linear mixed models, and Bayesian approaches</cite>. The methods exist. The choice between them depends on assumptions about carryover, autocorrelation, and heterogeneity—assumptions that are rarely interrogated in protocols I've reviewed.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC3651310/
• https://ascpt.onlinelibrary.wiley.com/doi/10.1002/cpt.2425
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6955665/
• https://www.medrxiv.org/content/10.1101/2022.07.21.22277832.full.pdf

<cite index="13-1">An Integrated Summary of Immunogenicity (ISI) is a comprehensive document that consolidates the data on immunogenicity collected throughout the clinical development of biologics during clinical trial and preclinical testing</cite>. <cite index="13-2,13-4">The key part of ISI is determining and evaluating the potential risks that may occur due to immunogenicity assessment and how that might impact the efficacy and safety of therapeutic proteins, and the ISI should address the severity and clinical relevance of these impacts</cite>. <cite index="13-5">Immunogenicity testing data generated during different phases of clinical studies typically involves monitoring the occurrence of ADAs and neutralizing antibodies, and monitoring their effects on pharmacokinetics and efficacy of the biologic product</cite>. <cite index="23-14,23-15">Regulatory agencies expect a complete and adequate assessment of how ADA data impacts PK, PD, safety, and efficacy, with the assessment exploring ADA effects on populations, sub-populations, and individuals</cite>. <cite index="23-16">Beginning an ISI early in clinical drug development enables the document to be updated through development to post-marketing</cite>. <cite index="14-4,14-5">Understanding the incidence, kinetics and magnitude of ADA, its neutralizing ability, cross-reactivity with endogenous molecules or other marketed biologics, and related clinical impact may enhance clinical management of patients, and it would be useful for labels to describe these parameters and their clinically relevant thresholds</cite>. <cite index="14-1,14-2">The overall risk-benefit profile of a biologic may warrant application of an empirical ADA threshold to ensure patient safety despite lack of statistically supported evidence of clinical relevance, and complex study designs may not permit statistical assessments of immunogenicity</cite>.

Sources:

• https://www.allucent.com/resources/blog/understanding-immunogenicity-assessment-integrated-summary
• https://www.amadorbioscience.com/blog/evaluating-ada-response-essential-for-biologics
• https://pmc.ncbi.nlm.nih.gov/articles/PMC4070270/

<cite index="12-3,12-4,12-5">Immunogenicity to biologics is often observed following dosing in human subjects during clinical trials, with both product and host specific factors implicated, but even if risk factors are identified and eliminated as part of rational quality by design approaches, the outcome in clinic can be uncertain and challenging to predict</cite>. <cite index="11-2">Prediction and quantification of the risk for immunogenicity of therapeutic protein drugs before clinical trials is a crucial need</cite>, and <cite index="11-3">state-of-the-art in silico analyses and in vitro assays characterize the immunogenic potential of biotherapeutic proteins and their correlation to clinically observed immunogenicity outcome</cite>. <cite index="12-7,12-14">A systematic roadmap for performing risk assessments through identification of risks and their mitigations wherever possible is provided, with outputs defining a risk score to guide the clinical bioanalytical and immunogenicity monitoring strategy</cite>. <cite index="16-1,16-2">Clinical immunogenicity assessment for complex multidomain biological drugs is challenging due to multiple factors, requiring a strategy to overcome bioanalytical challenges in order to assess ADA</cite>. <cite index="16-6">Most biologic molecules induce different levels of immune response in treated individuals, potentially leading to ADA formation, the impact of which can range from no observable consequence to substantial impacts on exposure, efficacy, and safety</cite>. <cite index="16-7,16-8">The situation becomes more complex when the drug contains homology with an endogenous peptide or protein—antibodies raised to the drug may cross-react with the endogenous counterpart, increasing the potential risk of safety-related events</cite>. The preclinical toolkit has limits, and the clinical read remains essential.

Sources:

• https://pubmed.ncbi.nlm.nih.gov/35122828/
• https://www.pegsummit.com/20/Immunogenicity
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11586355/

<cite index="3-1,3-12">Most biologicals are immunogenic, and ADA incidence can reach more than 90%</cite>, though <cite index="3-2,3-13">ADA are often clinically benign, but a subset of ADA-positive patients can experience adverse impacts on safety and efficacy</cite>. <cite index="19-2">The extent to which drug products elicit immune response is extremely variable, ranging from 0% for canakinumab to as high as 40% for infliximab</cite>. <cite index="3-6,3-7">Published ADA incidence rates can vary greatly between same-class products and different patient populations due to disparate bioanalytical methods and interpretation approaches, as well as product-specific and patient-specific factors that are not fully understood</cite>. <cite index="3-8">The incidence of ADA and their association with clinical consequences cannot be generalized across products</cite>. Clinical consequences themselves range widely: <cite index="18-9,18-10">biologic drugs differ widely with respect to the clinical consequences of ADA emergence—in safety, these may range from no recognizable symptoms to serious anaphylactic reactions, and in efficacy from unabated to nearly eliminated therapeutic response</cite>. <cite index="18-2">High concentration of ADAs following adalimumab administration is associated with decreased and shortened duration of anti-inflammatory activity, indirectly caused by faster clearance</cite>. <cite index="18-3">With infliximab, ADA emergence may induce severe acute infusion reactions as well as delayed symptoms such as arthralgias, rash, facial edema or headache</cite>. <cite index="20-9,20-10">A 2025 review showed 40-50% of patients treated with biologics developed moderate to high ADA, often increasing clearance, reducing exposure, and in some cases driving loss of response or the need for dose escalation</cite>.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC4837239/
• https://www.mabion.eu/science-hub/articles/impact-of-immunogenicity-on-efficacy-and-safety-of-biosimilars/
• https://www.mabion.eu/impact-of-immunogenicity-on-efficacy-and-safety-of-biosimilars-importance-of-ada-and-nab-testing-during-the-clinical-development/
• https://cbs.crystalpharmatech.com/ada-and-nab-testing-in-biologics-pk-efficacy-and-regulatory-risk

<cite index="9-1,9-2">The industry standard tiered approach starts with screening and confirmation of binding antibodies, then follows up with neutralizing antibody (NAb) assays for ADA-positive samples</cite>, preferably using methods that reflect the drug's mechanism. <cite index="5-4,5-6">A two-tiered strategy screens samples in an immunoassay first; those that test positive for binding antibodies are then tested for neutralizing capacity</cite>. <cite index="9-3,9-4">The bridging immunoassay is the industry standard for ADA screening and confirmation—the drug is labeled separately with different tags, and any anti-drug antibodies in a sample form a bridge between the two labeled molecules</cite>. <cite index="20-5">Neutralizing antibody results provide the clearest available evidence of functional impact and are typically the strongest predictor of clinical consequences</cite>. But that functional readout only matters if integrated properly: <cite index="20-6,20-7">ADA and NAb data must be interpreted together with pharmacokinetics and pharmacodynamics; in isolation, neither dataset is sufficient to draw conclusions about clinical relevance</cite>. <cite index="8-1,8-2">A tiered immunogenicity strategy includes screening and confirmatory assays with titer evaluation, followed by characterization and evaluation of neutralizing capacity, performed according to FDA guidance</cite>. The assay can detect all immunoglobulin isotypes—IgG, IgM, IgA—and works across species because any antibody can form the immune complex. Still, <cite index="11-7,11-8">validated cut points may not remain relevant when applied to the study population once the method is implemented for clinical study support, and this should be considered when determining cut point strategies during validation</cite>.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4983396/
• https://cdn.clinicaltrials.gov/large-docs/41/NCT02637141/SAP_001.pdf
• https://cdn.clinicaltrials.gov/large-docs/94/NCT03684694/Prot_000.pdf
• https://cbs.crystalpharmatech.com/ada-and-nab-testing-in-biologics-pk-efficacy-and-regulatory-risk
• https://www.pegsummit.com/20/Immunogenicity

<cite index="15-9,16-11">A DCT can be fully decentralized if all trial-related activities take place outside traditional trial sites or hybrid if some of such activities involve in-person visits to the traditional sites</cite>. In practice, hybrid is the norm. <cite index="10-5">Hundreds of decentralized trials have been performed, many of them using a hybrid model in which some aspects of the study are performed at the site while others are done remotely</cite>.

The decision depends on product profile and outcome criticality. <cite index="6-9">Fully decentralized trials, in which all activities are decentralized, may be appropriate for investigational products with "well-characterized safety profiles" and those that do not require complex administration</cite>. <cite index="15-3,15-4,16-5,16-6">In the reviewed cases, many studies required on-site in-person evaluation for critical outcomes. Evaluations, tests conducted at local healthcare facilities need to meet specific criteria, for example, approved list of tests for confirming Covid infection; therefore, balancing convenience of local clinical assessment with qualification requirements</cite>.

The terminology itself has evolved. <cite index="8-13">The term 'decentralized' has only recently been standardized, with virtual trials, remote trials, and digital trials being used previously</cite>. The FDA's 2024 guidance reflects this: rather than classify entire trials, the agency now evaluates individual decentralized elements. <cite index="20-8">Among DCT deployments, there are two main variations: (1) DCTs that are entirely remote (full DCTs); and (2) DCTs that are partially remote (hybrid DCTs)</cite>, with the latter substantially more common in practice.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC12416308/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10009564/
• https://www.appliedclinicaltrialsonline.com/view/fda-decentralized-clinical-trial-guidance
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10990725/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10400692/

The recruitment and retention claims for DCT methodologies are widely repeated but thinly evidenced. <cite index="15-8,16-10">In a recent review of 13 DCTs, 11 reported improvements in recruitment; 7 reported positive retention outcomes</cite>. Industry sources amplify the narrative: <cite index="4-8,4-9">Industry data from trials that have adopted decentralized elements shows improved recruitment—remote screening and eConsent reduce screen failures and accelerate enrollment timelines—and better retention, as reducing patient travel burden decreases dropout rates, particularly in chronic disease and long-term device studies</cite>.

But a 2022 systematic review deflates confidence. <cite index="19-3,19-7">Trials were widely heterogeneous in design and reporting, precluding meta-analysis of the effect of DCT methods on the primary recruitment outcome. However, there is insufficient evidence to confirm which methods are most effective in trial recruitment, retention, or overall cost</cite>. The review analyzed 45 trials quantitatively and 117 documents qualitatively; the conclusion was that <cite index="19-13,19-5">participant and stakeholder experiences of DCTs were incompletely represented</cite>.

The Apple Watch Study serves as cautionary counterpoint. <cite index="23-1">Even where remote studies have achieved impressive recruitment, this does not automatically translate into sustained adherence to trial activities and retention of participants, as evident in the Apple Watch Study</cite>. One systems-thinking review noted <cite index="20-3">sponsors and their collaborative partners face great difficulty in weighing benefits and risks and anticipating operating challenges</cite> as more is learned about DCT use.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC12416308/
• https://meddeviceguide.com/blog/decentralized-clinical-trials-medical-devices-guide
• https://pmc.ncbi.nlm.nih.gov/articles/PMC9306873/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9338512/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10400692/

<cite index="13-3">The COVID-19 pandemic accelerated the development of decentralized clinical trials (DCT)</cite>. <cite index="10-2">With the support of health authority guidance documents, procedural changes that were implemented include, but are not limited to, subject visits being conducted via telephone or video, informed consent being obtained electronically, investigational drug being delivered directly to subjects' homes, and remote entry and monitoring of study data</cite>.

The adoption was abrupt and infrastructure-limited. <cite index="10-7,10-8">During the COVID-19 pandemic, implementation of decentralized clinical trial methodologies and technologies were expedited. Many organization infrastructures did not have the applicable processes in place to support the necessary changes and the speed in which the numerous changes were instituted, which likely contributed to the negative perceptions observed for clinical trial conduct during the pandemic</cite>. <cite index="9-4">Decentralised methods for IMP supply, patient-health care provider interaction and communication, clinic visits and source document verification were used more often as mitigation strategies than they were used prior to COVID-19</cite>.

The COVID-OUT trial offers one operational case. <cite index="13-5,13-6">COVID-OUT was a decentralized, multicenter, quadruple-blinded, randomized trial that rapidly delivered study drugs nation-wide. The trial examined three medications (metformin, ivermectin, and fluvoxamine) as outpatient treatment of SARS-CoV-2 for their effectiveness in preventing severe or long COVID-19</cite>. <cite index="13-8">The remote nature of the study caused an additional 94 participants to not take any doses of study drug</cite> among 1417 enrolled—a 6.6% non-start rate attributable to decentralization itself.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9535935/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10009564/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10685265/

<cite index="1-8">In September 2024, the FDA issued final guidance titled "Conducting Clinical Trials With Decentralized Elements," for sponsors, investigators and other parties supporting drug, biologic and medical device development</cite>. The shift in nomenclature matters: <cite index="6-10">FDA is moving away from separately classifying trials as either DCTs or hybrid DCTs, but rather focusing on the elements of decentralization, such as telehealth visits, visits with local HCPs, or in-home visits with remote trial personnel</cite>.

The framework centers on two axes. <cite index="8-4">Both FDA and European Medicines Agency guidance publications commonly emphasized an assessment of the appropriateness of decentralized elements along 2 axes: patient safety and data integrity</cite>. <cite index="1-1,1-7">The guidance recommends that sponsors use centralized and risk-based monitoring techniques to oversee protocol adherence, manage deviations and proactively address data irregularities</cite>.

Investigational product administration remains tightly controlled. <cite index="7-8,7-9">A local HCP could remotely administer an IP at a participant's home if the IP is well characterized and does not require specialized monitoring immediately post-administration. IPs in early development stages, however, may need to be administered at a traditional clinical trial site</cite>. The guidance does not eliminate regulatory burden—<cite index="7-5">FDA is clear that DCTs must meet the same regulatory requirements as non-DCTs, though there may be additional considerations when implementing decentralized activities</cite>.

Sources:

• https://phillipslytle.com/navigating-decentralized-clinical-trials-with-fdas-guidance/
• https://www.duanemorris.com/alerts/fda_issues_guidance_remote_clinical_trial_activities_1024.html
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10990725/
• https://www.appliedclinicaltrialsonline.com/view/fda-decentralized-clinical-trial-guidance

<cite index="12-11,12-12">The fragility indices and fragility quotients demonstrated a strong correlation with p-values below 0.05 (correlation coefficients of −0.802 and −0.715, respectively), which implies that the fragility indices offer limited additional information beyond the p-value alone</cite>. This finding emerged from <cite index="14-3,14-4,14-5">a simulation study that generated 2,000 random 2×2 contingency tables with p-values < 0.05 according to Fisher's exact test and calculated the fragility indices</cite>.

<cite index="13-3,13-4,13-5">The ASA's statement on p-values stressed that the current practice of relying on p-values should be updated; p-value based statistical significance can be lost or gained with alteration of few events in a trial's arm, and the fragility index was created to partially overcome some of these limitations and to intuitively quantify an interpretable measure of trial robustness</cite>. Yet the strong correlation raises questions about what the index adds.

<cite index="13-7">The relationship between the p-value and the fragility index makes sense because they're both measures of evidence against the null hypothesis</cite>. <cite index="22-10,22-11">Critics have argued that the modifications to patient outcomes behind the fragility index can be unlikely to occur in practice, and the outcome probability is a crucial companion to the fragility index</cite>. The index may be most useful not as a standalone metric but as a patient-level translation of what p-values near 0.05 mean in practice.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC10602368/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC8665025/
• https://www.pnas.org/doi/10.1073/pnas.2105254118

<cite index="5-9,5-10">The fragility quotient is the fragility index divided by the sample size, and a low fragility quotient indicates a less robust trial</cite>. <cite index="5-11">The fragility quotient provides a way to assess the vulnerability of studies with regard to sample size, especially when sample sizes vary widely between studies addressing the same intervention</cite>. For example, <cite index="5-3">in one RCT with a large sample size (1,005 patients), if only 3 more patients in the intervention group had experienced the outcome, the p-value would be greater than 0.05</cite>; <cite index="5-10">the fragility quotient for this trial would be 3/1,005 = 0.003, which is small and indicates that the trial is not robust</cite>.

The quotient addresses a limitation of the raw fragility index: a trial with 10,000 participants and a fragility index of 10 is arguably more robust than a trial with 100 participants and the same index. The quotient expresses fragility as a proportion of the total sample, allowing comparisons across studies of different sizes. However, <cite index="5-11,5-12">a large fragility index does not necessarily indicate a conclusive result, and a small fragility index does not indicate that the RCT results are invalid</cite>—the metric must be interpreted alongside other measures of trial quality, including confidence intervals, effect size, and clinical context.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC10171299/

<cite index="2-8,2-9">The fragility index is used to assess the robustness of statistically significant findings, particularly in studies with small sample sizes or limited outcome events; traditional significance testing relying on p-values can create a false sense of confidence, as minor changes in the data may shift results from significant to nonsignificant</cite>. <cite index="3-5,3-6,3-7">If a study with a p-value of 0.049 would climb above 0.05 if only one patient experienced a different outcome, that study has a fragility index of one; studies with a low fragility index mean that a shift of only a few events from one group could change hypothesis tests outside thresholds usually considered statistically significant</cite>.

<cite index="11-4,11-5,11-6">In a review of COVID-19 randomized controlled trials, the median fragility index was 4; for drug studies (ivermectin and hydroxychloroquine), the index was 2.5 events, while for vaccines it was 119</cite>. <cite index="11-7">In 55% of the RCTs reviewed, less than a 1% change in the total sample would alter the results</cite>. <cite index="6-6">Most RCTs in surgery and general medicine are fragile (with a low FI score), in contrast to those in cardiac disease and heart failure, where most RCTs are robust (with high FI scores)</cite>.

<cite index="15-21">The fragility index is a simple metric that encompasses important trial characteristics such as sample size and the event rate (and hence study power)</cite>. <cite index="15-22">Fragility index may identify trials at high risk of 'medical reversal' when further studies of the same intervention are performed</cite>.

Sources:

• https://arxiv.org/pdf/2411.16938
• https://www.thebottomline.org.uk/blog/ebm/fragility-index/
• https://www.acsh.org/news/2022/05/16/fragility-p-values-16286
• https://pubmed.ncbi.nlm.nih.gov/30422256/
• https://intensiveblog.com/fragility-index-walsh-et-al-2014/

<cite index="4-3,4-4">The fragility index quantifies the minimum number of participants whose status needs to change from an event to a nonevent—or vice versa—to flip a statistically significant result (p ≤ 0.05) to nonsignificant</cite>. <cite index="15-5,5-5">It is calculated by iteratively converting one patient at a time in the group with the fewest events from "non-event" to "event" and recalculating Fisher's exact test until p ≥ 0.05</cite>.

<cite index="16-5,16-6">The metric was introduced by Walsh et al. in 2014 to quantify the vulnerability of trial results by determining the minimum number of events that would need to change to reverse statistical significance</cite>. <cite index="6-6">Analysis of trials across specialties found a median fragility index of 3 (IQR 1–8) in trauma studies, meaning that adding 3 events to the opposite treatment arm would eliminate statistical significance</cite>. <cite index="8-2">Among 25 randomized controlled trials in hepatocellular carcinoma, the median fragility index was 5, and 40% had a fragility index of 2 or less</cite>.

The index is calculated only for trials with dichotomous outcomes that cross the significance threshold. <cite index="19-7">The conversion occurs in the arm with fewer events—likely the treatment arm if the event is adverse (such as infection) or the control arm if the event is positive (for example, home discharge)</cite>. <cite index="1-6">The intent is to use the fragility index in conjunction with the p-value, 95% confidence interval, and measures describing benefit or risk</cite>.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC7485073/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10171299/
• https://arxiv.org/html/2411.16938v1
• https://pubmed.ncbi.nlm.nih.gov/30422256/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10326696/
• https://clincalc.com/Stats/FragilityIndex.aspx
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10263243/

<cite index="9-3,9-4">Phase 0 approaches evaluate subtherapeutic exposures of new drugs in exploratory clinical trials; recent progress extends phase 0 benefits beyond assessment of pharmacokinetics to include understanding of mechanism of action and pharmacodynamics.</cite> <cite index="21-3">The exploratory IND supports first-in-human testing of new agents at subtherapeutic doses based on reduced manufacturing and toxicologic requirements, allowing demonstration of drug-target effects and assessment of pharmacokinetic-pharmacodynamic relationships in humans earlier in development.</cite>

<cite index="21-4">The objectives of a phase 0 cancer clinical trial are to establish at the very earliest opportunity—before large numbers of patients have been accrued and exposed to potential drug-associated toxicity—whether an agent is modulating its target in a tumor, and consequently whether further clinical development is warranted.</cite> <cite index="18-2">The objective of exploratory IND studies is to determine if the drug's mechanism of action as observed in an experimental system can be replicated in humans.</cite>

<cite index="22-10,22-11">In neuro-oncology, animal models are much weaker approximations than for most other cancers, heightening the need for earlier PK data in humans; Phase 0 trials can help bridge that gap, eliminating ineffective drugs earlier for patients with limited time to wait.</cite> <cite index="25-3,25-5">Phase 0 approaches have potential to improve preclinical candidate selection and enable safer, cheaper, quicker and more informed developmental decisions; challenges remain, but phase 0 approaches should be considered for application in most drug development scenarios.</cite>

Sources:

• https://www.nature.com/articles/s41573-020-0080-x
• https://pubmed.ncbi.nlm.nih.gov/18536551/
• https://www.clinicaltrialsarena.com/features/phase-0/

<cite index="22-1,22-7">Phase 0 trials could delay the timeline to entering Phase I, prolonging development of effective drugs.</cite> Yet <cite index="18-4,18-8">Phase 0 trials can shorten drug development timelines, as demonstrated with the PARP inhibitor ABT-888.</cite> <cite index="21-1,21-9">Conducting a phase 0 trial under exploratory IND can reduce clinical development time for new agents and inform further clinical decision-making.</cite>

<cite index="17-4,17-5">Phase 0 microdose studies have not been fully embraced by the pharmaceutical industry, based on the number of Phase 0 studies conducted.</cite> <cite index="22-5,22-6">If a company is choosing between several preclinical candidates, Phase 0 could gather early human data for all candidates relatively cheaply, or provide an early look if a drug candidate has conflicting animal data before investing in full-scale Phase I.</cite>

<cite index="20-2,20-3">Phase 0 studies do not generate safety and tolerability data like conventional phase 1 studies, nor evidence of clinical efficacy on their own; they do not replace the need for conventional phase 1, 2, or 3 studies.</cite> <cite index="20-4">However, they can inform and accelerate the decision to pursue such studies by providing proof of concept plus PK and PD data, subsequently shortening the drug development timeline.</cite> <cite index="23-2">This approach allows "fast-fail" assessments, eliminating non-viable compounds with minimal resource investment.</cite>

Sources:

• https://www.clinicaltrialsarena.com/features/phase-0/
• https://www.tandfonline.com/doi/full/10.4155/fmc.09.117
• https://pubmed.ncbi.nlm.nih.gov/18536551/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10982362/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC6855937/
• https://www.quanticate.com/blog/bid/63143/the-different-phases-of-clinical-trials

<cite index="13-3,13-4">Approximately 80% of microdose pharmacokinetics available in the public domain have been shown to scale to therapeutic dose within a twofold difference.</cite> <cite index="15-3">Common criticisms include uncertain scalability of PK for oral drugs from microdose to pharmacological dose, especially for dose-dependent absorption or saturable transporter/enzyme systems.</cite>

<cite index="15-1">Drug concentration quantification in plasma from Phase 0 microdose PK studies is most often done by introducing ¹⁴C labeling into the drug molecule.</cite> <cite index="16-2">When extreme sensitivity is required, data on all drug-related material and metabolites is needed, or simultaneous IV microdose is used to determine absolute bioavailability (sometimes called microtracing), AMS becomes the analytical method of choice.</cite> <cite index="16-1">If a rapid decision is needed on PK appropriateness or a choice from a series of compounds, especially before radiolabeled material is available, LC-MS/MS may be preferable.</cite>

<cite index="24-6,24-7">If there is no biomarker or the agent's kinetics is nonlinear, a phase 0 trial may be of limited use; investigators must be careful not to miss potentially valuable agents due to disparities between microdose and full-dose kinetics.</cite> <cite index="16-4">Microdosing is only one tool in the drug developer's toolbox and should be used in the context of all available data.</cite>

Sources:

• https://www.tandfonline.com/doi/full/10.4155/bio.09.177
• https://www.frontiersin.org/journals/pharmacology/articles/10.3389/fphar.2024.1369079/full
• https://pubmed.ncbi.nlm.nih.gov/21083052/
• https://ascopubs.org/doi/10.1200/JCO.2008.21.5798

<cite index="1-3,17-7">A microdose is defined as less than 1/100th of the dose calculated from animal data to yield a pharmacological effect in humans, with a maximum of 100 μg or 30 nmol for proteins.</cite> <cite index="1-2">These studies were introduced approximately 20 years ago</cite> under <cite index="2-1">the exploratory IND framework designed to collect preliminary pharmacokinetic (PK) and pharmacodynamic (PD) data using very low doses.</cite>

<cite index="1-7,1-8">The FDA's 2006 guidance stated that sponsors were not taking full advantage of allowed flexibility in IND data submission and indicated what level of CMC and nonclinical data would be expected for early exploratory studies.</cite> <cite index="18-1">The preclinical toxicology studies necessary to support an exploratory IND are less extensive than those for a traditional IND due to limited dosing and anticipated low risk.</cite>

<cite index="3-7">Timelines to conduct a microdose study from toxicology commencement to obtaining human PK data are 4–6 months versus 12–18 months for Phase I.</cite> <cite index="1-5">The overall goal was quicker and more efficient drug development, especially through deselecting products unlikely to succeed.</cite> <cite index="18-2,18-3">The objective of exploratory IND studies is not to supply safety and tolerability data from conventional Phase I study, nor clinical efficacy evidence; subsequent traditional IND studies are necessary to gather dose-limiting toxicity information.</cite>

Sources:

• https://www.frontiersin.org/journals/pharmacology/articles/10.3389/fphar.2024.1369079/full
• https://www.clinicalstudies.in/clinical-trial-phases/phase-0-microdosing-studies/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1885041/
• https://www.tandfonline.com/doi/full/10.4155/fmc.09.117

<cite index="24-1">Multiple testing procedures defined by directed, weighted graphs have recently been proposed as an intuitive visual tool for constructing multiple testing strategies that reflect the often complex contextual relations between hypotheses in clinical trials</cite>. <cite index="24-2">Many well-known sequentially rejective tests, such as parallel gatekeeping tests or hierarchical testing procedures are special cases of the graph based tests</cite>.

The key advantage: <cite index="4-6">according to this approach, hypotheses may be tested more than once, and when a particular null hypothesis is rejected, the alpha allocated to that hypothesis can be reallocated to other hypothesis tests</cite>. <cite index="4-7,4-8">The arrows on the diagram show how the Type I error allocated to a null hypothesis that is successfully rejected will be redistributed for testing of the other hypotheses—the arrows do not necessarily indicate the testing order</cite>. In adaptive settings, <cite index="22-4">the adaptive test does not require knowledge of the multivariate distribution of test statistics and is applicable in a wide range of scenarios including trials with multiple treatment comparisons, endpoints or subgroups, or combinations thereof</cite>.

<cite index="4-4">If the primary PFS analysis is positive, analysis of the main secondary endpoints of OS will be formally tested sequentially at the 2-sided alpha of 0.05, with ORR and QOL tested sequentially when the above hypotheses in the hierarchy are also statistically significant</cite>. This is gatekeeping: spend nothing until you unlock the gate.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC4789493/
• https://cdn.clinicaltrials.gov/large-docs/39/NCT03901339/SAP_001.pdf

<cite index="15-13">The family-wise error rate (FWER) is the probability of making at least one Type I error among all the hypothesis tests when performing multiple tests</cite>. <cite index="17-3">For type I error control, often the family-wise error rate is controlled, which is the probability to reject at least one true null hypothesis</cite>. <cite index="15-1,15-2">The error rate for a family of tests is always higher than for an individual test—as the number of hypothesis tests increases, the chance that at least one is a false positive grows</cite>.

<cite index="18-1,18-2">For a multi-arm trial, the FWER should be strongly controlled in confirmatory trials and reported in exploratory trials, based on the two main regulatory bodies for pharmaceutical trials currently providing advice suggesting that adjustment is required for definitive trials</cite>. In adaptive designs, <cite index="22-2">designs permit mid-trial design modifications based on unblinded interim data as well as external information, while providing strong family wise error rate control</cite>, and critically, <cite index="22-3">to maintain the familywise error rate, it is not required to prespecify the adaption rule in detail</cite>.

<cite index="4-2,4-3">The overall type I error rate for a study is strictly controlled at a 2-sided alpha of 0.05, with the primary endpoint analysis serving as the gatekeeper for secondary endpoint analyses and tested at the 2-sided alpha of 0.05</cite>. The practical consequence: sponsors must decide upfront which claims matter enough to spend alpha on.

Sources:

• https://statisticsbyjim.com/hypothesis-testing/bonferroni-correction/
• https://arxiv.org/pdf/2511.09449
• https://link.springer.com/article/10.1186/1745-6215-15-364
• https://pmc.ncbi.nlm.nih.gov/articles/PMC4789493/
• https://cdn.clinicaltrials.gov/large-docs/39/NCT03901339/SAP_001.pdf

<cite index="12-4,12-5">The Bonferroni Correction divides the desired overall alpha level (the probability of a Type I error) by the number of tests—for example, to maintain an alpha level of 0.05 across five tests, the correction adjusts the alpha level for each individual test to 0.01</cite>. It is simple. It is also, in many cases, wasteful.

<cite index="11-6,11-7,11-8">One assumption of the Bonferroni correction is that the tests being conducted are independent—if there is a correlation between the tests, the correction can be too conservative, leading to a higher probability of a type II error</cite>. <cite index="17-7">Overlapping populations generally induce correlation among test statistics, such that nonparametric procedures like the Bonferroni correction may be overly conservative</cite>. Regulatory guidance has entrenched it anyway: <cite index="18-3">the European Medicines Agency guidance on multiplicity states that any confirmatory trial with multiple primary null-hypotheses should control the maximum probability of making a type-I error</cite>, and <cite index="18-4">the FDA draft guidance on adaptive designs states that the total study-wise error rate should be controlled in all confirmatory trials</cite>.

In practice, <cite index="4-9">a Bonferroni approach is used to control the Type I error rate at 0.05 (2-sided) alpha for multiple hypothesis tests</cite>, often embedded in hierarchical testing sequences or graphical approaches that allow alpha to be recycled when hypotheses are successfully rejected.

Sources:

• https://fastercapital.com/content/Familywise-Error-Rate--Keeping-it-in-the-Family--Understanding-Familywise-Error-Rate-with-Bonferroni.html
• https://fastercapital.com/content/Bonferroni-correction--Controlling-Family-Wise-Error-in-Two-Tailed-Tests.html
• https://arxiv.org/pdf/2511.09449
• https://link.springer.com/article/10.1186/1745-6215-15-364
• https://cdn.clinicaltrials.gov/large-docs/39/NCT03901339/SAP_001.pdf

<cite index="2-3,2-4">When trials employ adaptive sample size approaches with multiple time points evaluating superiority, the alpha-level at each time point is adjusted so that the probability of concluding superiority at any time point is limited to 2.5%, with each individual test using a nominal alpha-level that is adjusted</cite>—this is the alpha-spending function.

<cite index="6-9">This flexible approach determines the rate or fraction at which the overall type I error (alpha) is to be spent during a trial</cite>. In practice, <cite index="4-10">the Lan DeMets alpha spending function that approximates a Pocock approach is commonly used to account for multiplicity introduced by including interim analyses for superiority</cite>. One critical point: <cite index="2-6,2-7">the entrenched belief that introducing interim analyses "costs" alpha—that interim adaptations erode available alpha and require a penalty—is widespread confusion; the act of spreading alpha over multiple analyses does not diminish the overall error rate allowance nor requires compensation beyond careful allocation</cite>.

<cite index="1-2">Proposed sequential procedures adapting popular multiple comparison procedures for fixed time-point design and using α-spending for each endpoint are shown to strongly control the family-wise Type-1 error rate</cite>. The real work is prespecification: <cite index="7-1">the Lan and DeMets alpha spending function is applied to control the overall type I error rate of 0.05 such that an alpha of 0.01 is preserved for the final analysis</cite>, with the actual interim alpha determined by the number of subjects enrolled at that look.

Sources:

• https://www.tandfonline.com/doi/full/10.1080/19466315.2023.2191989
• https://www.berryconsultants.com/resource/alpha-allocation-in-adaptive-clinical-trials-misconceptions-and-scientific-consequences
• https://cdn.clinicaltrials.gov/large-docs/39/NCT03901339/SAP_001.pdf
• https://cdn.clinicaltrials.gov/large-docs/49/NCT03447249/SAP_001.pdf

<cite index="7-3,7-4">Up to 30% of clinical trials in rare diseases are prematurely discontinued primarily due to patient accrual issues; many others do not achieve target recruitment or are severely delayed, resulting in insufficient sample sizes to detect statistically significant differences.</cite> <cite index="7-5">To overcome issues relating to the assembly and retention of sufficiently large control arms, researchers may opt to conduct single-arm trials and supplement findings with data from external control arms.</cite>

<cite index="6-2,6-3">Single-arm trials are common in drug and biologic submissions for rare or life-threatening conditions, especially when no therapeutic options exist; external control arms improve interpretation of single-arm trials but pose methodological and regulatory challenges.</cite> <cite index="4-3,4-7">External control arms improve interpretation of single-arm trials but pose methodological and regulatory challenges; under certain limited circumstances, single-arm trials may be submitted as pivotal evidence for determination of efficacy and safety for approval.</cite>

But feasibility is not validity. <cite index="23-2,23-3">Both industry and regulators recognize that in situations where a randomized study cannot be performed, external controls can provide needed contextualization to allow better interpretation of studies without randomized controls—but also agree that external controls will not fully replace randomized clinical trials as the gold standard for formal proof of efficacy in drug development.</cite> The field has built a workaround for rare disease enrollment constraints. It has not built a methodological equivalent to randomization.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC10673956/
• https://link.springer.com/article/10.1007/s43441-024-00693-8
• https://pmc.ncbi.nlm.nih.gov/articles/PMC11530569/
• https://arxiv.org/pdf/2209.07776

<cite index="11-1,11-4">In 2017, the FDA approved Merck's Bavencio (avelumab) for metastatic Merkel cell carcinoma based on a single-arm trial and a synthetic comparator arm using historical controls of matched patients.</cite> <cite index="13-2,13-3,13-5,13-6">In April 2020, the FDA granted approval for selumetinib in children with neurofibromatosis type 1 based on the Phase II SPRINT trial, which had no placebo control but used an external control arm comprising 50 patients from two previous studies: a natural history study and the placebo arm of a previous clinical trial for a different drug.</cite> <cite index="18-5,18-6">Erliponase alfa was approved in 2017 by FDA and EMA for neuronal ceroid lipofuscinosis type 2 based on comparisons between 23 treated patients in a phase 1/2 single-arm trial and 42 historical controls from the DEM-CHILD database, a European registry.</cite>

These are existence proofs, not precedent. <cite index="5-11">Regulatory acceptance of submissions using single-arm designs and external control arms has increased, concordant with more submissions for rare disease and gene therapy products.</cite> <cite index="17-1,17-6">The FDA accepted the use of external controls in 2001 where justified, and in its February 2023 guidance stated that external controls should be considered case-by-case.</cite> <cite index="17-9,17-10">FDA, EMA, and MHRA do not recommend a particular analytical approach for externally controlled trials, acknowledging no single statistical method is suitable for all trials; sponsors should discuss suitable approaches with the appropriate review division.</cite> The implication: bring primary data, bring methods documentation, and prepare for negotiation.

Sources:

• https://solici.com/resources/synthetic-control-arms-use-of-rwe-in-clinical-trials/
• https://www.clinicaltrialsarena.com/features/external-control-arms/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10673956/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11530569/
• https://www.clinicalleader.com/doc/the-pros-and-cons-of-synthetic-control-arms-in-clinical-trials-0001

<cite index="5-7,5-8">When serving as the basis for approval, single-arm trials may use an external control arm to mitigate methodologic and statistical concerns arising from the lack of a concurrent comparator; an external control arm consists of patients not enrolled in the single-arm trial—no concurrently randomized control group.</cite> <cite index="5-9">External control data may come from past clinical trial data or real-world data sources: registries, natural history studies, electronic health records, administrative claims.</cite>

The statistical challenge is exchangeability. <cite index="25-3">Exchangeability between a single-arm trial and an external control group is proposed as a criterion to reduce information bias and draw valid comparisons.</cite> <cite index="19-4,19-5">These designs are characterized by a lack of full randomization and heightened dependency on modeling; causal assumptions about exchangeability between internal and external controls are required to identify the average treatment effect.</cite>

<cite index="24-5">Naive direct use of external control data is not valid due to differences in patient characteristics and other confounding factors.</cite> <cite index="20-4">Propensity score matching separates design from analysis and provides the ability to explicitly examine the degree of overlap in confounders.</cite> <cite index="28-3,28-4">Many frameworks rely on the invariance assumption—that conditional outcome distributions are identical for trial and external controls—which can be problematic because outcome distributions may also vary across studies, leading to outcome drift.</cite> The field is developing dynamic borrowing methods that down-weight external sources proportionally to heterogeneity introduced, but these are methodologically dense and not yet standardized.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11530569/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC7756307/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11546536/
• https://pubmed.ncbi.nlm.nih.gov/31725899/
• https://arxiv.org/pdf/2410.18409

<cite index="1-7,1-8">Randomization is recognized as impractical, infeasible, or unethical in certain clinical settings—rare diseases, conditions with high unmet need—and in those circumstances, external controls (historical controls, synthetic arms) are acknowledged by regulators as a possible control structure for single-arm trials.</cite> <cite index="1-9">But accepting external controls for regulatory decisions demands detailed, transparent planning and adherence to pharmacoepidemiologic principles to minimize bias and confounding.</cite>

<cite index="3-1">External comparators have gained traction in oncology and rare disease submissions when RCTs are impractical or ethically challenging.</cite> <cite index="3-4">Between 2011 and 2019, 52% of single-arm trial-based Health Technology Assessment submissions contained external control data.</cite> That prevalence does not equal confidence. <cite index="1-4">Creating and analyzing an external control arm using real-world data is challenging because design and analytics may not fully control for all systematic differences.</cite> <cite index="21-2">The FDA's February 2023 guidance states explicitly that the suitability of an externally controlled trial design "warrants a case-by-case assessment."</cite>

The nomenclature debate is instructive. <cite index="2-5,2-6">One proposed framework cautions against the term "Externally Controlled Trial" unless the external control was pre-specified in the protocol, advocating instead for "External Comparator Cohort" when data collection was not planned before trial initiation.</cite> <cite index="2-7">External patient data differ fundamentally from RCT control arms—drawn from separate populations, often with different data collection methods.</cite> The regulatory posture is: prove comparability or accept the epistemic discount.

Sources:

• https://onlinelibrary.wiley.com/doi/full/10.1002/pds.4975
• https://www.frontiersin.org/journals/drug-safety-and-regulation/articles/10.3389/fdsfr.2025.1579171/full
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12443072/
• https://ubc.com/insights/external-controls-in-research/

<cite index="5-1,5-2,5-3">The ICH E9(R1) addendum on estimands and sensitivity analysis provides a framework for defining the treatment effect a trial intends to estimate and is widely adopted in pharmaceutical research, but remains underutilized in trials investigating internet-based interventions</cite>. <cite index="8-2">The addendum was finalized in November 2019</cite>. <cite index="20-3,20-5">The ICH E9(R1) addendum was developed by regulators and the pharmaceutical industry, primarily with individually randomized trials in mind</cite>.

<cite index="18-7">The term intercurrent event was not mentioned in any of the pharmacoepidemiologic safety studies examined in a 2020 review of articles published in the Journal Pharmacoepidemiology and Drug Safety, though many cohort studies discussed drug discontinuation, treatment modification, and terminal events</cite>. <cite index="36-2,36-5">With influence of the addendum, many trials have proposed analyzing primary endpoints using while-on-treatment, hypothetical, or principal stratum strategies that handle intercurrent events in ways that use post-randomization outcomes to exclude information from randomized participants and don't preserve integrity of randomization</cite>. <cite index="36-3,36-6">These approaches have inherent limitations in ability to draw scientifically rigorous inference on clinically relevant causal effects important for decisions about adopting interventions</cite>.

The framework is a regulatory tool that became a methodological lens. Whether it sharpens trial design or creates new opportunities for post-hoc rationalization will depend on how protocol authors use it.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12368201/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9389514/
• https://www.researchgate.net/figure/The-five-attributes-of-an-estimand-according-to-the-ICH-E9-R1-addendum_fig1_343835425
• https://pubmed.ncbi.nlm.nih.gov/40394856/
• https://www.medrxiv.org/content/10.1101/2025.06.25.25330127.full.pdf

<cite index="9-8,9-9">In addition to intercurrent events, missing body weight assessments at the end of treatment need to be handled in the statistical analysis; the various approaches to handling missing data each have their own advantages and disadvantages, and will in some cases rely on unverifiable assumptions</cite>. The estimand framework does not solve the missing data problem. It clarifies what you are trying to estimate so that the assumptions you make about missingness can be mapped to the question.

<cite index="11-2">Instead of excluding patients who have an intercurrent event and those missing outcome data, each ICE can be appropriately handled so that the treatment effect being targeted can be understood</cite>. <cite index="10-2,11-4">Once an estimand is defined, it is also unclear how to deal with missing values using principled analyses, particularly for non-inferiority studies</cite>. <cite index="12-4,12-5">Multiple imputation methods that align with estimands for both primary and sensitivity analysis have been proposed, including twofold fully conditional specification and reference-based multiple imputation for binary outcomes</cite>.

<cite index="16-2">In pooled data from three clinical trials of chronic pain treatments with up to 40–60% of participants discontinuing treatment, different methods for dealing with missing data yielded treatment effect estimates that differed in both magnitude and direction</cite>. <cite index="16-3">Given that missing data are unavoidable, the National Research Council report emphasizes the use of more principled methods and draws attention to the significant limitations of existing simplistic methods such as last observation carried forward and baseline observation carried forward</cite>. The estimand framework does not prescribe a method. It demands you state what you are estimating before you choose one.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8081661/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10504812/
• https://journals.sagepub.com/doi/10.1177/17407745231176773
• https://pmc.ncbi.nlm.nih.gov/articles/PMC7508757/

<cite index="8-3">The addendum provides a framework for clinical study planning to ensure alignment between study objectives, design, conduct, and analysis</cite>. <cite index="1-5">Estimands clarify the exact research question being evaluated, both to avoid misinterpretation and to ensure study methods align with overall objectives</cite>. This is not ornamental language. <cite index="19-6">The framework brings together concepts—estimands, sensitivity analyses, ensuring statistical analyses answer clinically relevant questions—acknowledged as important for years, including in the National Research Council's 2010 report on missing data</cite>.

<cite index="35-15,35-16">Researchers must identify at study outset what data are required to support estimation of each estimand and ensure they are collected</cite>. A treatment policy strategy, for example, requires outcome data to be collected after intercurrent events occur. <cite index="35-7,35-14">Under a hypothetical strategy, the participant's outcome that would have occurred had they continued treatment is used, but in practice this value will not be known and must be estimated</cite>.

<cite index="9-3,9-4">Intercurrent events during weight management trials can influence placebo-adjusted treatment effects depending on how they are accounted for and how missing data are handled; the most appropriate method for statistical analysis includes assessment of last observation carried forward, multiple imputation, and mixed models for repeated measures</cite>. <cite index="12-4">Multiple imputation methods that align with the estimands for both primary and sensitivity analysis have been proposed</cite> in case studies ranging from tuberculosis to obesity trials.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10802140/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9389514/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8081661/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10504812/

<cite index="1-2,3-1">The ICH E9(R1) addendum, finalized in 2019, introduced a framework for defining estimands</cite>—the precise treatment effect a trial intends to estimate. <cite index="17-1,20-4">An estimand consists of five attributes: population, treatment regimen, endpoint, summary measure, and how intercurrent events are handled</cite>. That last attribute is where the framework earns its keep.

<cite index="11-11">Intercurrent events (ICEs) are events occurring after treatment initiation that affect either the interpretation or the existence of measurements associated with the clinical question</cite>. <cite index="9-7">The term itself comes from the ICH guidance</cite>. Examples include treatment discontinuation, rescue medication use, death, or switches to alternate therapy. <cite index="28-2">The addendum proposes five strategies for handling ICEs: treatment policy, hypothetical, composite variable, while-on-treatment, and principal stratum</cite>.

<cite index="33-7,33-8,33-9">The treatment policy strategy—often described as an intention-to-treat analysis—collects and includes endpoint values regardless of whether the intercurrent event occurred</cite>. <cite index="13-5,13-6">Under the hypothetical strategy, the causal effect targets what would have happened if the ICE had been prevented—for instance, if rescue medication were not available or patients could not discontinue treatment</cite>. <cite index="28-8">The addendum does not require all ICEs be handled the same way</cite>, which means a single trial can apply different strategies to different events, though <cite index="28-7">most primary estimands to date have used a single strategy</cite>.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10802140/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10504812/
• https://www.sciencedirect.com/science/article/abs/pii/S1551714420301713
• https://arxiv.org/pdf/2006.03105
• https://www.tandfonline.com/doi/10.1177/17407745231176773
• https://onlinelibrary.wiley.com/doi/10.1002/sim.70104?af=R

<cite index="24-1,23-3">Lack of efficacy in the intended disease indication is the major cause of clinical phase drug development failure</cite>. <cite index="24-2,23-4">Explanations could include the poor external validity of pre-clinical (cell, tissue, and animal) models of human disease and the high false discovery rate (FDR) in preclinical science</cite>. <cite index="21-5">Nearly 80% of compounds entering phase II clinical trials will fail to reach regulatory approval, largely due to a lack of efficacy or an unacceptable safety profile</cite>.

Genetic evidence addresses this by bypassing model organisms entirely. Human genetic perturbations are not confounded by species differences or experimental artifacts. But the predictive value is conditional: it depends on the quality of the genetic association, the specificity of the variant to the protein target, and the match between the genetic perturbation (often lifelong, partial loss-of-function) and the intended pharmacological intervention (acute, potentially full inhibition or agonism).

The FDR problem in preclinical science is structural. <cite index="22-1,22-4">51% of phase II clinical trial failures are due to lack of efficacy; many of these failures can certainly be attributed to inadequate target validation</cite>. Genetic evidence offers a path around this, but only if the causal gene is correct and the pharmacology is matched.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6906499/
• https://www.nature.com/articles/s41598-019-54849-w
• https://pmc.ncbi.nlm.nih.gov/articles/PMC5326691/
• https://www.cambridgemedchemconsulting.com/resources/targetvalidation.html

A 2019 re-analysis of the seminal Nelson et al. (2015) genetic support claims found <cite index="25-1,26-1">estimates of the effect of genetic evidence on Phase I to II progression probabilities are lower than originally reported, and confidence intervals sometimes exclude original estimates</cite>. The discrepancy may be mechanical: <cite index="25-3,26-3">Phase I trials assess safety in healthy volunteers, not efficacy, so their success may be less closely linked to human genetic evidence for target involvement in disease</cite>.

There's also a temporal problem. <cite index="25-5,25-6,26-5,26-6">It is possible that there are systematic differences in the types of associations discovered before and after 2013; later associations may be biased towards those with smaller effect sizes or rarer variants only detectable in larger cohorts, and could also be less predictive of drug efficacy</cite>. The 2015 claims were trained on earlier data; validation sets may not generalize.

The headline number—"doubling"—remains defensible across the full development arc, but the distribution of benefit across phases is murkier than the original estimates suggested. Early-phase transitions may not benefit as much, and the predictive value of post-GWAS associations remains an open question.

Sources:

• https://journals.plos.org/plosgenetics/article?id=10.1371%2Fjournal.pgen.1008489
• https://pmc.ncbi.nlm.nih.gov/articles/PMC6907751/

<cite index="1-3,3-4">Genetic data can be harnessed using the Mendelian randomization (MR) analytic paradigm to proxy the lifelong consequences of genetic perturbations of drug targets</cite>. The method exploits <cite index="1-10,1-11">the random allocation of germline genetic variants at gametogenesis; if confounding by genetic ancestry is appropriately controlled for, the inheritance of genetic variants is not confounded by environmental variables</cite>.

When applied to drug targets, MR has methodological advantages. <cite index="5-9,7-9">The 'no horizontal pleiotropy assumption' is strengthened when proteins are the risk factors of interest</cite>, because <cite index="5-10,7-10">proteins are typically the proximal effectors of biological processes encoded in the genome</cite>. The framework has been validated on known drug-gene pairs: <cite index="2-4">HMGCR, PCSK9, NPC1L1 and CETP encode the targets of licensed or clinical phase drugs with known effects on lipids and coronary heart disease risk</cite>, where genetic perturbations anticipate trial outcomes.

But pleiotropy remains a problem. <cite index="5-8,7-8">Causal inference is undermined if genetic variants used to instrument a risk factor also influence alternative disease-pathways (horizontal pleiotropy)</cite>. The approach is only as good as the instruments and the assumptions.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10953771/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC7320010/
• https://www.nature.com/articles/s41467-020-16969-0
• https://www.biorxiv.org/content/10.1101/781039.full.pdf

<cite index="11-6,13-3">The probability of success for drug mechanisms with genetic support is 2.6 times greater than those without</cite>, updating a widely cited 2015 estimate that genetic evidence doubles clinical success rates. The metric holds across independent datasets and timeframes, though <cite index="13-4,16-1">relative success varies among therapy areas and development phases, and improves with increasing confidence in the causal gene, but is largely unaffected by genetic effect size, minor allele frequency or year of discovery</cite>.

The lift is strongest at phase 2-to-3 transitions, where lack of efficacy drives most failures. <cite index="9-2,9-3">Currently 90% of developed molecules entering phase II and phase III clinical trials fail to gain regulatory approval, most due to lack of therapeutic efficacy rather than lack of safety</cite>, a failure mode that <cite index="20-27">is due to the limited predictive value of preclinical models of disease, and our continued ignorance regarding the consequences of perturbing specific targets over long periods of time in humans</cite>.

But the field underuses the data it has. <cite index="10-6">Only 4.7% of active drug-indication pairs possess human genetic support</cite>, and <cite index="18-9">active programs continue to have low rates of genetic support, similar to historical pipelines, suggesting that human genetic data have not yet begun to appreciably influence pipeline composition across the industry</cite>. The baseline is low; the opportunity remains large.

Sources:

• https://www.nature.com/articles/s41586-024-07316-0
• https://pmc.ncbi.nlm.nih.gov/articles/PMC11096124/
• https://pubmed.ncbi.nlm.nih.gov/25515070/
• https://www.nature.com/articles/nrd4051
• https://www.medrxiv.org/content/10.1101/2023.06.23.23291765v1.full

<cite index="14-1">Overly restrictive study design, stringent eligibility criteria, and continuous activation of clinical trials in sites based on their academic prominence or speed of enrollment often has resulted in the exclusion of underserved patient populations</cite>. <cite index="11-4">The stricter the eligibility criteria, the less diverse the study population becomes, which restricts the ability to apply clinical trial findings to the broader population that will receive treatment outside of the study</cite>.

Recent recommendations have targeted modifiable exclusion criteria. <cite index="12-5">Recommendations by ASCO and Friends of Cancer Research include patients with lower creatinine clearance values of >30 mL/min when renal toxicity and clearance are not of direct treatment-related concern, patients with mild to moderate hepatic dysfunction when nonclinical and clinical data indicate inclusion is safe, and patients with laboratory parameters out of normal range because of hematologic disease, which could considerably increase the number and diversity of patients in clinical trials and better reflect real-world patient populations</cite>.

<cite index="14-4">Problems that prevent the inclusion of diverse populations in industry-funded clinical trials include patient out-of-pocket costs often not covered in the informed consent process, industry pressures to gather data quickly, and the selection of easy-to-recruit samples being incentivized</cite>. The enrollment friction is structural: faster, cheaper trials still compete with more representative ones in the sponsor calculus.

Sources:

• https://www.ncbi.nlm.nih.gov/books/NBK584407/
• https://friendsofcancerresearch.org/glossary-term/clinical-trials-diversity/
• https://ashpublications.org/bloodadvances/article/9/4/774/534324/New-strategies-for-enhancing-enrollment-of

<cite index="7-5">The Plan focuses on racial and ethnic diversity while encouraging inclusion of other underrepresented populations relevant to disease areas, including sex, gender identity, age, socioeconomic status, disability, pregnancy and lactation status, and comorbidities</cite>. But identifying "clinically relevant populations" requires epidemiologic grounding that many sponsors may lack in-house.

<cite index="6-3,6-4">Factors that lead to inequities in clinical trial participation are complex and multifaceted, and the Diversity Plan should account for the intersection of broader elements at the individual, community, and societal levels; expansion needs to be tailored to disease areas based on a thorough analysis of the epidemiology and natural history</cite>. <cite index="7-6,7-7">Pharmacoepidemiologists are equipped to assess whether data sources are available and sufficient to provide a reasonably unbiased representation of the target population, broadly describe the determinants and burden of disease in target populations, identify unmet patient population needs and potential barriers for recruitment, and inform clinical trial design by assessing disease burden in different subpopulations, estimating background rates of comorbidities, determining clinically relevant follow-up based on disease progression, and defining endpoints</cite>.

<cite index="7-9">Use of available data sources within the USA often requires trade-offs in representation, generalizability, and/or individual-level detail</cite>. The guidance creates demand for population-level reference data that may not exist at the resolution required to set defensible enrollment targets.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10225256/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10225256/

<cite index="3-4,3-5">Almost 40% of the American population belongs to a racial or ethnic minority, yet clinical trials for new drugs predominantly involve White participants, making up 80% to 90% of study groups</cite>. This gap isn't merely optics. <cite index="9-1">Underrepresentation of racial and ethnic minorities limits the generalizability of research findings</cite>, and <cite index="11-3">when clinical trials do not include patients who represent the diversity of the real-world patient population, it can limit the generalizability of trial results, slow enrollment, and perpetuate health disparities</cite>.

The concern extends beyond race and ethnicity. <cite index="10-4">Older adults and female and nonwhite patients were underrepresented in heart failure clinical trials</cite>, and <cite index="10-15">people aged ≥65 years are still significantly underrepresented in drug trials, especially cancer trials</cite>. <cite index="12-4">Only 2% to 3% of adult patients with cancer participate in clinical trials, with even lower participation among ethnic and racial minorities, individuals with low socioeconomic status, rural residents, older adults, and young adults aged 15-39 years</cite>.

<cite index="14-3">Eliminating the factors and problems that limit trial participation would improve the generalizability of results</cite>. <cite index="10-13,10-14">Although the literature on generalizability assessment and associated methods is abundant, it is poorly organized with little agreement on analytic procedures, and most assessments are conducted a posteriori rather than a priori, discovering generalizability issues only after trial completion</cite>.

Sources:

• https://www.rtihs.org/resource/insight/fda-guidance-more-inclusive-clinical-trials
• https://pmc.ncbi.nlm.nih.gov/articles/PMC5131730/
• https://friendsofcancerresearch.org/glossary-term/clinical-trials-diversity/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC7359942/
• https://ashpublications.org/bloodadvances/article/9/4/774/534324/New-strategies-for-enhancing-enrollment-of
• https://www.ncbi.nlm.nih.gov/books/NBK584407/

<cite index="23-2,23-5">The FDA's updated draft guidance on Diversity Action Plans, issued under mandate from the Food and Drug Omnibus Reform Act (FDORA) of 2022, replaces the April 2022 draft that focused primarily on race and ethnicity</cite>. <cite index="17-3,17-4">A Diversity Action Plan is required for any clinical investigation of a new drug classified as a Phase 3 study or other pivotal studies designed to demonstrate safety and efficacy to support a marketing application</cite>.

The plans require sponsors to specify enrollment targets disaggregated by race, ethnicity, sex, and age, along with a rationale and strategies to meet those targets. <cite index="17-6,17-7">Enrollment goals must specify target numbers disaggregated by race, ethnicity, sex, and age, with clear justification considering disease prevalence and incidence in the intended population</cite>. <cite index="2-4,2-5">The guidance emphasizes demographic characteristics such as race, ethnicity, sex, and age group when setting enrollment goals, and recommends looking beyond these factors to include socioeconomic status, geographic location, and comorbidities</cite>.

<cite index="18-5">The guidance states that waivers from Diversity Action Plan requirements will be rare</cite>. <cite index="18-4">Plans should be submitted as early as possible but no later than submission of the Phase 3 protocol, and may be updated over time</cite>. The expansion from race/ethnicity to multiple demographic axes reflects regulatory acknowledgment that trial representativeness is multidimensional, though the operationalization of "clinically relevant population" remains sponsor-defined within FDA parameters.

Sources:

• https://www.fda.gov/regulatory-information/search-fda-guidance-documents/diversity-action-plans-improve-enrollment-participants-underrepresented-populations-clinical-studies
• https://insider.thefdagroup.com/p/guidance-breakdown-diversity-action
• https://www.acclinate.com/blog/fda-updates-diversity-action-plan-guidance-what-sponsors-need-to-know
• https://premier-research.com/perspectives/fdas-diversity-mandate-transforming-clinical-trials-and-drug-development-for-better-outcomes-for-all/

<cite index="25-1,25-2,25-3,25-4">In non-interventional research, it can be difficult to control for hard-to-measure concepts such as frailty or disease severity; many methods exist to mitigate residual confounding, including quantitative bias analysis and e-values, but most rely on pre-specifying the prevalence of and exposure/outcome association with a single confounder, whereas high-dimensional propensity scores (HDPS) are an alternative method aiming to optimize confounding adjustment by considering recorded data as proxy variables</cite>.

<cite index="17-1">Residual confounding, including residual confounding by indication, is a major concern in pharmacoepidemiologic studies; HDPS can demonstrate promise in addressing confounding even when comparison groups are suboptimal, but its performance depends on the careful selection and ranking of covariates</cite>. In one COVID-19 analysis, HDPS-weighted estimates moved toward the null consistently for death outcomes, but were sensitive to the number of covariates included for hospitalization outcomes.

<cite index="13-4">Inclusion of strong predictors of treatment reduces the precision (yields larger standard errors) of the treatment effect estimate and should therefore be avoided if these predictors of treatment do not also affect the risk for the outcome—that is, if they are not confounders</cite>.

Sources:

• https://www.medrxiv.org/content/10.1101/2025.02.04.25321459.full.pdf
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12644305/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC6715456/

<cite index="14-4,14-5,14-6">It remains unclear whether, and if so when, use of propensity scores provides estimates of drug effects that are less biased than those obtained from conventional multivariate models; in the great majority of published studies that have used both approaches, estimated effects from propensity score and regression methods have been similar, and simulation studies further suggest comparable performance of the two approaches in many settings</cite>.

<cite index="10-10">In studies that employed both propensity score methods and traditional regression adjustment to control for confounding, few found significant differences between the two approaches</cite>. <cite index="12-4,12-5">Use of propensity scores in pharmacoepidemiologic studies has increased substantially, yet evidence is lacking that this approach will systematically give better estimates of drug effects than those obtained from conventional regression approaches; propensity score methods and conventional multivariate methods have similar inability to control unmeasured confounding</cite>.

<cite index="5-7,5-8">When used properly, propensity score methods can produce important estimates of treatment effects with minimal bias, but as with all methods used in causal inference, they have several important limitations, assumptions, and nuances that must be considered both when conducting and interpreting such studies</cite>.

Sources:

• https://pubmed.ncbi.nlm.nih.gov/16611199/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9980423/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC1790968/
• https://www.sciencedirect.com/science/article/pii/S0007091223003604

<cite index="23-3">An incorrectly specified propensity score model may lead to residual confounding bias</cite>. <cite index="23-1,23-7">Each propensity score technique—matching, weighting, stratification, covariate adjustment—aims to balance patient characteristics between treatment groups, but misspecification of the propensity score model could prevent achieving adequate balance, thereby leading to residual confounding bias</cite>.

<cite index="8-6,8-7,8-8">After propensity score weighting or matching, researchers must check the balance of baseline covariates, which corresponds to assessing whether the propensity score model has been correctly specified; covariates used for the PS model are expected to be well-balanced between groups, but imbalance may exist</cite>. <cite index="26-2">Strong risk factors that remain imbalanced (SMD>0.1) between treatment groups after propensity score matching are a particular source of concern because of residual confounding</cite>.

<cite index="22-1">While propensity score matching can correct for some confounding by indication, interpretation of the outcome must allow for possible residual bias, if confounding by indication is substantial</cite>. <cite index="22-10,22-11">In one analysis of MDR-TB treatment, traditional multivariable approaches did not fully correct for potential confounding by indication; propensity score methods did improve covariate balance, bringing the effect estimate closer to the null</cite>.

Sources:

• https://bmcmedresmethodol.biomedcentral.com/articles/10.1186/s12874-020-00994-0
• https://pmc.ncbi.nlm.nih.gov/articles/PMC10760486/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC6715456/
• https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0151724

<cite index="3-2,3-3">The propensity score is the conditional probability of treatment given background covariates, typically estimated via logistic regression or other methods</cite>. <cite index="1-5">PSM operates within a counterfactual framework, comparing observed outcomes in treated participants against assumed outcomes in matched controls</cite>.

<cite index="11-6,11-7">Propensity score methods were designed to confront confounding by indication and channeling bias by modeling how prognostic factors guide treatment decisions and using that model to construct treatment groups with similar covariate distributions</cite>. <cite index="11-8,11-9">These methods may be particularly relevant in pharmacoepidemiology, where rare events and large confounder sets make outcome modeling difficult, and where treatment decisions are well understood enough to support plausible models</cite>.

But this is not randomization. <cite index="12-3">Use of propensity scores will not correct biases from unmeasured confounders</cite>. <cite index="21-15,21-16">You are highly unlikely to include all factors in the propensity score model because you may not have data on every factor, creating a gap between propensity-based balancing and true randomization, where every observed and unobserved confounder is balanced</cite>. <cite index="13-8">Because the propensity score is a summary of measured covariates, it cannot eliminate unmeasured confounding</cite>.

Sources:

• https://www.e-jcpp.org/journal/view.php?doi=10.36011/cpp.2019.1.e6
• https://en.wikipedia.org/wiki/Propensity_score_matching
• https://pmc.ncbi.nlm.nih.gov/articles/PMC5810585/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC1790968/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC6715456/
• https://communities.sas.com/t5/Statistical-Procedures/How-do-we-understand-residual-bias-in-propensity-score-analysis/td-p/952028

<cite index="15-1">The passing of the Prescription Drug User Fee Act (PDUFA) VII has added further clarity regarding RWE use in regulatory decision-making into the legislation, including the launch of FDA's Advancing Real-World Evidence (RWE) Program, a pilot program seeking improvement in the quality and acceptability of RWE-based approaches to support new intended labeling claims or satisfy post-approval study requirements.</cite> Pilot is the operative word.

<cite index="1-3">FDA's RWE Program will involve demonstration projects, stakeholder engagement, internal processes to bring senior leadership input into the evaluation of RWE and promote shared learning and consistency in applying the framework, and guidance documents to assist developers interested in using real-world data to develop RWE to support Agency regulatory decisions.</cite> That is infrastructure, not precedent. <cite index="10-5">As FDA use of real-world evidence continues to expand, expectations around data quality, transparency, and reproducibility are rising.</cite>

<cite index="26-9,26-10">The guidance encourages early engagement with the FDA to discuss proposed RWD sources, study designs, and analytical plans, and manufacturers should develop clear protocols describing data collection, management, and analysis, and should thoroughly document assessment of data relevance and reliability.</cite> Early engagement is not optional when the evidentiary standard is still being defined in real time.

Sources:

• https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2023.1236462/full
• https://www.fda.gov/media/120060/download
• https://blog.healthverity.com/how-the-fda-is-using-real-world-evidence-in-safety-and-approval-decisions
• https://www.cozen.com/news-resources/publications/2026/fda-s-guidance-on-the-use-of-real-guidance-on-use-of-real-world-evidence-for-medical-devices

<cite index="17-1,17-2">Many pre-approval trials have enrolled participants who were not fully representative of the population who used the product or were conducted under conditions that do not represent those under which the new product would be used in typical health care settings once it was approved, leading to concerns about an efficacy-effectiveness gap between outcomes observed in RCTs (efficacy) compared with real-world circumstances (effectiveness).</cite> The gap is real. Whether RWD closes it is the question under evaluation.

<cite index="26-6,26-7">FDA acknowledges that RWE can offer advantages over traditional clinical trials by capturing broader clinical experiences than are usually represented in traditional clinical studies and reducing time and cost burdens associated with evidence generation, but also cautions that RWD sources may vary significantly in quality and may be subject to bias, missing data, and confounding, particularly in nonrandomized settings.</cite> <cite index="26-8">Robust study design and analytical rigor are therefore critical to the acceptability of RWE.</cite>

<cite index="2-2">The Forum on the Integration of Observational and Randomized Data (FIORD) conducted a meeting bringing together various stakeholder groups to build consensus around best practices for the use of RWD to support regulatory science.</cite> Consensus meetings do not produce effect estimates. <cite index="16-2">Improving clinical evidence generation by evaluating outcomes and patient experiences at the point-of-care would help achieve the ultimate aim of ensuring that effective and safe treatments are rapidly approved for patient use.</cite> That is the aspiration. The methods are still under scrutiny.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11897686/
• https://www.cozen.com/news-resources/publications/2026/fda-s-guidance-on-the-use-of-real-guidance-on-use-of-real-world-evidence-for-medical-devices
• https://arxiv.org/pdf/2310.03176
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10643567/

<cite index="5-4">RWE can be generated from RWD using many different study designs, including interventional studies (clinical trials) or non-interventional (observational) studies.</cite> <cite index="5-7">A non-interventional or observational study is a type of study in which patients received the marketed drug of interest during routine medical practice and are not assigned to an intervention according to a protocol.</cite> The distinction carries regulatory weight.

<cite index="1-7,1-8">FDA will evaluate the potential role of observational studies in contributing to evidence of drug product effectiveness, and efforts to replicate the results of randomized controlled trials using more rigorously designed observational studies may provide insight into the opportunities and limitations of using these approaches.</cite> That conditional phrasing should not be read as enthusiasm. <cite index="6-4">The prospect of incorrect effect estimates has historically cast doubt on the use of RWE for regulatory science.</cite>

<cite index="15-4">The use of RWE in regulatory decision-making for effectiveness remains an emerging area, where in addition to guidance, there continues to be a role for the experience from precedents and pilots in informing best practices.</cite> <cite index="11-2,11-3">FDA noted two NDA/BLA approvals based, at least in part, on RWE: one approval was for a new dosing schedule for lacosamide supported by a retrospective cohort study using EHR data, and the second was for tocilizumab for treating COVID-19 in hospitalized adults.</cite> Two approvals. That number should inform expectation-setting.

Sources:

• https://www.fda.gov/media/124795/download
• https://www.fda.gov/media/120060/download
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10877517/
• https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2023.1236462/full
• https://becarispublishing.com/digital-content/blog-post/fda-summarizes-use-real-world-evidence-support-regulatory-decision-making-drugs

<cite index="18-3,18-4">FDA has a long history of using real-world data and real-world evidence to monitor postmarket safety of approved drugs, but RWE has been used to support effectiveness on a more limited basis.</cite> That asymmetry matters. <cite index="6-3">The 21st Century Cures Act mandated FDA guidance on regulatory use of RWE to support regulatory decisions</cite>, and <cite index="18-9">in 2018 FDA created a Framework for evaluating the potential use of RWE to help support approval of a new indication for an already-approved drug or to satisfy post-approval study requirements.</cite>

<cite index="26-2,26-3">FDA emphasizes that RWE may be appropriate to support regulatory decision-making when the underlying RWD is both relevant and reliable for the particular study question and regulatory purpose—relevance requires that the data appropriately address the intended regulatory purpose, while reliability depends on data quality, completeness, consistency, traceability, and robustness of data collection and analytical methods.</cite> That two-part test is not rhetorical. <cite index="20-2">A central theme is the need for sponsors to understand and clearly articulate the strengths and limitations of the underlying data, how these characteristics affect relevance and reliability in the context of the regulatory question being addressed, and how these assessments are supported by clear, well-documented evidence in regulatory submissions.</cite>

<cite index="21-3,21-4,21-5">One challenge is finding fit-for-purpose RWD—data from routine-care settings are focused on patients at the bedside or in the clinic, requiring consideration of the data's reliability and relevance to address regulatory questions, and for example, information on clinical measures of disease severity can be lacking in healthcare insurance claims.</cite>

Sources:

• https://www.fda.gov/science-research/science-and-research-special-topics/real-world-evidence
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10877517/
• https://www.fda.gov/news-events/fda-voices/realizing-promise-real-world-evidence
• https://www.cozen.com/news-resources/publications/2026/fda-s-guidance-on-the-use-of-real-guidance-on-use-of-real-world-evidence-for-medical-devices

<cite index="5-1,26-1">Master protocols investigate multiple hypotheses through concurrent sub-studies (e.g., multiple treatments or populations or that allow adding/removing arms during the trial), offering enhanced efficiency and a more ethical approach to trial evaluation.</cite> The efficiency claim rests on shared infrastructure: central screening, common imaging schedules, unified labs, and pooled control arms where applicable.

But the tradeoffs are real. <cite index="21-1,21-6">Umbrella clinical trials face similar disadvantages to basket design, with most being phase II trials without true randomization to a control group, while others may use historical controls, both of which can create issues with statistical analysis.</cite> <cite index="25-3">Basket trials allow borrowing of information across different diseases and may lack randomization to a control arm owing to the possible between-group differences in standard care.</cite> <cite index="21-9">In platform trials, comparators are built in, as is standard of care, which has the potential to permit traditional HTA evaluation, yet there is no true randomization.</cite>

The descriptive statistics from Park et al. are worth holding in view. <cite index="22-8,26-9">The median sample size of basket trials was 205 participants (IQR: 500-90 = 410), and median study duration was 22.3 months (IQR: 74.1-42.9 = 31.1).</cite> <cite index="22-10,26-11">The median sample size of umbrella trials was 346 participants (IQR: 565-252 = 313), and median study duration was 60.9 months (IQR: 81.3-46.9 = 34.4).</cite> Those interquartile ranges are wide, reflecting heterogeneity in ambition and endpoint selection. Efficiency is relative.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6751792/
• https://www.ncbi.nlm.nih.gov/books/NBK603603/
• https://www.gastrojournal.org/article/S0016-5085(24)00485-2/fulltext

<cite index="14-3">I-SPY 2 is an adaptive phase II clinical trial design in the neoadjuvant setting for women with locally advanced breast cancer.</cite> <cite index="16-2">Launched in 2010, it is by far the longest running adaptive platform trial.</cite> <cite index="16-6">The trial uses clinical biomarkers to classify breast cancer into 10 subtypes, with Bayesian adaptive randomization to allow individualized patient assignment to therapy arms.</cite> <cite index="16-5">I-SPY 2 evaluates new agents combined with standard therapy with pathologic complete response (pCR) as the primary endpoint.</cite>

The design allows drugs to graduate or exit. <cite index="18-1,18-2">Novel regimens with sufficiently high Bayesian predictive probability of being more effective than the dynamic control graduate with their corresponding biomarker signature(s); treatment strategies are dropped if they show a low probability of improved efficacy with any biomarker signature.</cite> <cite index="13-10">As of April 2017, I-SPY 2 had graduated six investigational treatments to phase 3 trials, and six additional drugs had either been dropped or were in evaluation.</cite> <cite index="16-7">A total of 7 drugs have graduated from I-SPY 2.</cite>

<cite index="14-4">I-SPY 2 is a collaborative effort among academic investigators, the National Cancer Institute, the U.S. Food and Drug Administration, and the pharmaceutical and biotechnology industries under the Foundation for the National Institutes of Health Biomarkers Consortium.</cite> The structure is notable for how many stakeholders it accommodates under a single protocol, which is both a strength and a source of complexity.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC7731787/
• https://pubmed.ncbi.nlm.nih.gov/19440188/
• https://www.sciencedirect.com/science/article/abs/pii/S1551714417303014
• https://clinicaltrials.gov/study/NCT01042379

<cite index="2-1,24-1,24-2">A basket trial evaluates a single investigational drug across different disease populations defined by a common molecular alteration; an umbrella trial evaluates multiple targeted therapies for a single disease stratified by molecular markers.</cite> <cite index="19-1">Platform trials typically maintain a common control arm and evaluate multiple investigational arms simultaneously, with pre-specified interim analyses allowing new treatments to be added or ineffective ones dropped.</cite>

The 2019 landscape analysis by Park et al. provides the most complete published accounting. <cite index="5-5,26-5">They identified 83 master protocols: 49 basket, 18 umbrella, and 16 platform trials.</cite> <cite index="22-6,26-7">Most were conducted in the U.S. (n=44/83) and investigated experimental drugs (n=82/83) in oncology (n=76/83).</cite> <cite index="5-6,26-6">The number of master protocols increased rapidly over the five years preceding their analysis.</cite>

The empirical structure differs by type. <cite index="22-7,26-8">Most basket trials (47/49) were exploratory phase I/II and not randomized (44/49), and more than half (28/48) investigated only a single intervention.</cite> <cite index="22-9,26-10">Umbrella trials were also mostly exploratory (16/18), but randomization was more common (8/18).</cite> <cite index="5-7,22-12">The majority of platform trials were randomized (15/16), and phase III investigation (7/15) was more common, with four using seamless II/III design.</cite> This reflects a hierarchy of evidentiary ambition: baskets are often early signal-finding, umbrellas introduce comparative structure within a disease, and platforms attempt something closer to registration.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6751792/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC8220876/
• https://link.springer.com/article/10.1186/s13063-019-3664-1

<cite index="1-1">Selective publication of clinical trials — and the outcomes within those trials — can lead to unrealistic estimates of drug effectiveness and alter the apparent risk–benefit ratio</cite>. <cite index="30-2,30-3">Reporting bias—in which the outcomes of a trial affect whether and how its results are published—is not uncommon. Reporting bias can skew the perceived risk–benefit ratio of treatments, mislead medical professionals and policymakers, and ultimately result in suboptimal medical decisions</cite>.

The effect size is not trivial. <cite index="33-9">In major depression a reduction of 25–29% of the effect size of psychotherapy was observed when adding the results of unpublished trials to those to published trials</cite>. <cite index="26-5,26-7">Researchers have found that trials that favour a particular health action ("positive findings") were nearly twice as likely to be published as trials with "negative findings", and that trials with positive findings were published more quickly than trials with negative findings. When not recognised and addressed in systematic reviews, publication bias can sometimes result in overestimation of the effects of health actions</cite>.

<cite index="8-4,8-5">Publication bias, the selective publication of results or studies favoring positive outcomes, presents a critical threat to the validity and effectiveness of evidence-based medicine. This includes exposing study participants to potential harm due to duplication of previously unreported studies that were unsuccessful and creating a skewed evidence base for clinical decision making</cite>. <cite index="1-10">Evidence-based medicine is valuable to the extent that the evidence base is complete and unbiased</cite>.

Sources:

• https://www.sciencedirect.com/science/article/abs/pii/0197245687901553
• https://journals.plos.org/plosmedicine/article?id=10.1371%2Fjournal.pmed.1003894
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6324306/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5371180/
• https://www.informedhealthchoices.org/key-concepts/concepts-about-evidence/unpublished-results-considered/

<cite index="17-1,17-2,17-3">Registration at inception of all clinical trials in a centralized, searchable database can reduce publication bias by enabling researchers to identify all studies related to a particular intervention. Prior attempts to encourage voluntary trial registration have been largely unsuccessful. Hence, the International Committee of Medical Journal Editors recently adopted a policy of mandatory clinical trial registration before consideration of trial registration and the development of comprehensive, computerized databases will promote transparency in research and help reduce publication bias</cite>.

<cite index="19-1,19-2">In September 2004 a number of major medical journals belonging to the International Committee of Medical Journal Editors (ICMJE) announced they would no longer publish trials that were not registered at inception. All trials that began enrolment of participants after September 2005 had to be registered in a public trials registry at or before the onset of enrolment to be considered for publication in those journals</cite>. <cite index="22-2">The Food and Drug Administration (FDA) Amendments Act (FDAAA) was enacted by Congress on September 27, 2007, requiring the registration of all non-phase I clinical trials involving FDA-regulated medical interventions and results reporting for approved drugs</cite>.

But compliance remains incomplete. <cite index="23-7">Clinical trial registration was found for 368 (31%) trials in a sample of 1,177 trials from systematic reviews published as of 2005; of those 135 (36.7%) were registered prospectively</cite>. <cite index="8-1">Among 129 phase 3 randomized clinical trials in a pediatric cohort study, 27.9% were not subsequently published, and 39.5% were never registered, with previous trial registration and sample size associated with greater likelihood of publication</cite>.

Sources:

• https://pubmed.ncbi.nlm.nih.gov/17540818/
• https://handbook-5-1.cochrane.org/chapter_10/10_3_3_trial_registries_and_publication_bias.htm
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6199729/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC9875740/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6324306/

<cite index="11-1">The AllTrials Campaign was launched in 2013 as a global movement to demand that all clinical trials—past and present—be registered and have their results reported</cite>, initiated by Sense about Science in collaboration with Ben Goldacre, BMJ, and Cochrane. <cite index="11-6">The campaign's public petition gathered over 90,000 signatures in its early years</cite>.

The rationale is explicit. <cite index="10-4,10-5">Doctors and patients need the results of clinical trials to make informed decisions about which treatment is best. When trial results are withheld, we cannot practice medicine safely and effectively</cite>. <cite index="16-1">The American Medical Association joins more than 640 patient advocacy groups, professional societies, medical organizations and thousands of patients worldwide in supporting the global campaign for clinical trial registration and reporting led by AllTrials</cite>.

<cite index="11-3,11-4">The UK Health Research Authority issued mandates to enforce trial result reporting. These achievements mark a significant shift toward transparency becoming an expected, if not legally enforceable, norm</cite>. <cite index="11-8">Major players like GSK and Johnson & Johnson acknowledged the movement, with GSK stating its commitment to post all results on its public register</cite>. <cite index="13-7,13-8">Millions of volunteers have participated in clinical trials to help find out more about the effects of treatments on disease. Information on what was done and what was found in these trials could be lost forever to doctors and researchers, leading to bad treatment decisions, missed opportunities for good medicine, and trials being repeated</cite>.

Sources:

• https://www.alltrials.net/
• https://www.alltrials.net/news/alltrials-campaign-us-launch/
• https://www.clinicalstudies.in/the-alltrials-campaign-progress-and-challenges/
• https://senseaboutscienceusa.org/alltrials/
• https://www.clinicaltrialsarena.com/news/ensuring-greater-transparency-through-the-alltrials-campaign-5761500-2/
• https://www.ama-assn.org/press-center/ama-press-releases/ama-joins-alltrials-campaign-clinical-trial-transparency

<cite index="5-3">Studies with statistically significant results were more likely to be published than those finding no difference between the study groups (adjusted odds ratio 2.32; 95% CI 1.25–4.28)</cite> in Easterbrook's foundational 1991 cohort study tracking 285 analyzed trials from Oxford ethics committee approvals. That pattern has held across decades. <cite index="3-3">In a study of FDA-registered trials, 97% (37/38) of clinical trials with positive findings were published compared to 33% (8/24) of studies with negative findings</cite>.

The Cochrane systematic review quantifies the magnitude: <cite index="6-2,6-3">trials with positive findings have a risk ratio of 1.78 (95% CI 1.58 to 1.95) for publication, meaning if 41% of negative trials are published, we would expect that 73% of positive trials would be published. Trials with positive findings tended to be published after four to five years compared to those with negative findings, which were published after six to eight years</cite>.

<cite index="7-5,7-6">For trials that were completed but not published, the major reasons for nonpublication were "negative" results and lack of interest. Nonpublication was primarily a result of failure to write up and submit the trial results rather than rejection of submitted manuscripts</cite>. <cite index="24-6">Of 1,970 trial registrations in the Australian New Zealand Clinical Trials Registry, 541 (27%) remained unpublished 10 to 14 years later</cite>. The bias extends beyond efficacy: <cite index="29-6,29-7">unpublished trials gave information on adverse effects more often than published trials, and several studies have shown that the reporting of adverse events and safety outcomes in clinical trials is often inadequate and selective</cite>.

Sources:

• https://pubmed.ncbi.nlm.nih.gov/1672966/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC8276556/
• https://ora.ox.ac.uk/objects/uuid:cb853328-be1a-44ce-bb64-94a17a18984a
• https://pubmed.ncbi.nlm.nih.gov/3442991/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9815633/
• https://methods.cochrane.org/bias/reporting-biases

<cite index="12-6">Surrogate markers may not reliably reflect true clinical benefit and can lead to misleading conclusions if not properly validated</cite>. <cite index="16-3,16-4">Relying on biomarkers as surrogates allows trials to be smaller and shorter, but leads to more limited insights about efficacy and less reliable safety assurances given the smaller safety dataset</cite>. <cite index="16-5">Agents receiving approval using surrogate endpoints are more vulnerable to having clinically unacceptable safety issues discovered post-marketing</cite>.

The case of rosiglitazone is instructive: <cite index="16-6">it was approved based on reducing HbA1c, yet post-marketing clinical trial results raised concerns</cite>. <cite index="12-8,12-9">Whenever feasible, trials should prioritize endpoints that directly measure how patients feel, function, or survive, and surrogates must be supported by strong biological rationale and robust empirical evidence linking them to meaningful clinical outcomes</cite>.

<cite index="11-4">As surrogate markers are increasingly accepted by FDA to support approval, it's imperative that patients and clinicians understand whether such endpoints are reflective of meaningful clinical benefits</cite>. The statistical machinery for validation has grown sophisticated; the question is whether the evidentiary bar has kept pace.

Sources:

• https://www.clinicaltherapeutics.com/article/S0149-2918(25)00269-3/abstract
• https://pmc.ncbi.nlm.nih.gov/articles/PMC3551627/
• https://fas.org/publication/validating-surrogate-endpoints/

<cite index="17-1,17-6">When a surrogate is known to predict clinical benefit based on strong evidence supporting the expected relationship between treatment effects on the surrogate and a meaningful clinical outcome, it's considered a validated surrogate endpoint</cite>. <cite index="17-4">Validated surrogates and endpoints that directly measure how patients feel, function, or survive can serve as the basis for traditional approval</cite>. <cite index="17-5">Surrogates that are reasonably likely to predict meaningful clinical outcomes but don't reach validation level can support accelerated approval</cite>.

<cite index="11-2,11-3">Since 2018, FDA has maintained a public table with over 200 surrogate markers accepted to support approval, but the table doesn't include information on the strength of evidence for each surrogate's association with clinical outcomes</cite>. <cite index="13-6">Because reasonably likely surrogates used in accelerated approval haven't been validated, sponsors must verify predicted clinical benefit with post-approval trials</cite>. <cite index="20-1">If a surrogate previously used for accelerated approval failed in confirmatory trials to demonstrate expected clinical benefit, it's no longer accepted and was excluded from the table</cite>.

<cite index="17-8">Studies supporting clinical validation should provide an estimate of the minimal biomarker change that would reliably predict meaningful clinical benefit</cite>—a threshold that shapes trial sizing and benefit-risk assessment.

Sources:

• https://ascpt.onlinelibrary.wiley.com/doi/10.1111/cts.70445
• https://fas.org/publication/validating-surrogate-endpoints/
• https://www.fda.gov/about-fda/innovation-fda/fda-facts-biomarkers-and-surrogate-endpoints
• https://www.fda.gov/drugs/development-resources/table-surrogate-endpoints-were-basis-drug-approval-or-licensure

<cite index="2-7,2-8">Freedman et al. (1992) supplemented Prentice's criteria with 'proportion explained,' while Buyse and Molenberghs (1998) proposed replacing it with the relative effect linking treatment effects on both endpoints and an individual-level measure of agreement</cite>. <cite index="2-9">The individual-level measure carries over when data are available from multiple trials, while the relative effect extends to a trial-level measure of agreement between treatment effects on both endpoints</cite>.

This shift mattered. <cite index="10-2,10-3">Prentice's operational criteria, proportion of treatment effect explained, and relative effect were the original techniques, leading to development of principal stratification, direct and indirect effects, meta-analytical (including surrogate threshold effect), and information theory approaches</cite>. <cite index="29-1,29-2">Following Prentice's hypothesis-testing methodology, a host of frequentist and Bayesian methods have been developed, the most recent based on information theory, bringing together causal effects and causal association paradigms</cite>.

<cite index="10-5">In theory, principal stratification and direct/indirect effects may be most appropriate because of their ability to validate surrogates on a causal basis</cite>—but that's theory. The methods have proliferated faster than the regulatory willingness to require them.

Sources:

• https://academic.oup.com/biostatistics/article-pdf/1/1/49/17744017/100049.pdf
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3980768/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9546435/

<cite index="24-6,24-7">Prentice's 1989 framework proposed a restrictive criterion for calling something a surrogate: the biomarker must yield a valid test of the null hypothesis of no treatment effect on the true endpoint</cite>. In operational terms, <cite index="23-1">this requires that the surrogate endpoint's effect on the true endpoint doesn't vary with treatment group, that the surrogate affects the true endpoint, and that treatment's effect on the surrogate changes treatment's average effect on the true endpoint</cite>. <cite index="1-4">Prentice's criteria are met if treatment (Z) significantly affects both the true endpoint (T) and surrogate (S), if S significantly affects T, and if Z has no effect on T given S</cite>.

The problem: <cite index="2-2,2-5">Prentice's definition and his operational criteria are equivalent only when both the surrogate and true endpoints are binary</cite>. <cite index="4-5">For continuous endpoints, the criteria can yield incorrect hypothesis testing extrapolation</cite>. <cite index="1-5">Freedman's relaxation of the fourth criterion—requiring the lower confidence limit of the proportion explained (PE) to exceed 0.5 or 0.75—can only be verified when treatment has a massively significant effect on the true endpoint, a rare situation</cite>.

<cite index="21-2,21-9">Statistical evidence alone cannot validate a surrogate; it must be one component in a decision-making process involving clinical and biological considerations</cite>. The criteria have been called <cite index="6-6">overly strict for evaluating a perfect surrogate endpoint</cite>.

Sources:

• https://onlinelibrary.wiley.com/doi/abs/10.1002/sim.4780080407
• https://pmc.ncbi.nlm.nih.gov/articles/PMC5771803/
• https://pubmed.ncbi.nlm.nih.gov/9840970/
• https://academic.oup.com/biostatistics/article-pdf/1/1/49/17744017/100049.pdf
• https://pubmed.ncbi.nlm.nih.gov/15339295/
• https://www.cda-amc.ca/sites/default/files/MG%20Methods/surrogate-endpoints-report.pdf

<cite index="3-5">Barriers have been real and well-founded: limited familiarity among many statisticians, clinicians, and regulators; legitimate concerns about the subjectivity of priors and the risk of undue influence; computational demands of thorough simulation work; and a strong institutional preference for the long-established frequentist paradigm that provides clear control of type I error rates</cite>.

<cite index="11-11,11-12">FDA released guidance for Bayesian statistics in medical device clinical trials back in 2010; these trials are quite different from drug or vaccine trials and people have used Bayesian techniques there for a long time</cite>. <cite index="14-3,14-8">Bayesian statistics is not new, but its formal recognition in regulatory settings is a relatively recent development</cite>. <cite index="10-9">FDA is seeing more proposals for trials that use Bayesian approaches in some way, and the guidance ensures that FDA's needs and expectations are clear for sponsors as they propose and implement these approaches</cite>.

The 2026 draft does not eliminate the computational burden—it formalizes it. <cite index="14-10">Sponsors must ensure the accuracy of algorithms used and demonstrate robustness through simulation studies, particularly for complex adaptive clinical trials</cite>. The document acknowledges what has always been true: choosing a statistical framework is a design choice, not a statement of faith, and every design choice must be justified with operating characteristics that a regulator can evaluate before the first patient enrolls.

Sources:

• https://www.pharmexec.com/view/bringing-bayesian-method-clinical-trials-dr-stacy-lindborg
• https://statmodeling.stat.columbia.edu/2026/01/15/fda-guidance-on-bayesian-clinical-trials/
• https://www.clinicalleader.com/doc/fda-s-draft-guidance-on-bayesian-methods-strategic-implications-for-small-biotechs-0001
• https://www.fda.gov/drugs/guidances-drugs/guidance-recap-podcast-use-bayesian-methodology-clinical-trials-drug-and-biological-products

<cite index="18-3,21-4">Bayesian designs can be calibrated to traditional frequentist error-rate targets and, with sponsor–FDA agreement, alternative Bayesian operating metrics may be appropriate</cite>. <cite index="6-4,6-5">Bayesian trials often rely on posterior probabilities—for example, Pr(effect > threshold) > c—but FDA requires these criteria be carefully calibrated, either to traditional type I error control or justified through decision-theoretic or benefit–risk frameworks</cite>.

<cite index="10-7,10-13">Bayesian approaches may facilitate adaptive trials where modifications to the design occur based on interim assessments; sponsors can use Bayesian methods to govern timing and adaptation rules for interim analysis, to inform design elements such as dose selection, or as the key analysis to support primary inference</cite>. <cite index="12-6,12-7">Bayesian models can support augmenting concurrent control arms with historical or non-concurrent controls, offering a methodological route to address challenges in rare disease contexts; the draft guidance discusses applications in oncology platform trials that use hierarchical models to account for temporal shifts in efficacy outcomes</cite>.

<cite index="19-6">Regulatory acceptability depends on whether the pre-specified estimand, model, prior, and decision criteria together yield acceptable operating characteristics under scenarios that reflect key clinical and operational uncertainties</cite>. The guidance does not permit trading rigor for speed; it permits making assumptions explicit so they can be examined.

Sources:

• https://arxiv.org/pdf/2601.14701
• https://www.fdamap.com/blog/what-fdas-new-guidance-on-bayesian-statistics-means-for-drug-developers/
• https://www.fda.gov/drugs/guidances-drugs/guidance-recap-podcast-use-bayesian-methodology-clinical-trials-drug-and-biological-products
• https://www.biopharminternational.com/view/how-fda-s-bayesian-guidance-could-accelerate-adaptive-trial-design-in-biopharmaceuticals
• https://arxiv.org/html/2601.14701v1

<cite index="10-4,10-5,10-6">Bayesian methods can be used in place of traditional approaches in any situation, but explicit use of previous data makes them ideally suited for certain contexts—pediatric studies, for example, can use Bayesian methods to incorporate adult information and reduce pediatric sample size</cite>. <cite index="6-2">FDA expects quantitative assessments of prior influence, prior–data conflict, and robust sensitivity analyses</cite>.

<cite index="1-2,1-3">Bayesian designs make dependency explicit by writing a prior distribution for parameters, which can be debated, stress-tested, and updated transparently as new evidence accrues; disagreements about priors are surfaced and can be evaluated via sensitivity analyses, whereas implicit assumptions in frequentist planning can be harder to identify and quantify</cite>.

<cite index="3-3">Borrowing historical or Phase 2 data can reduce sample size or shift allocation toward investigational therapy, but temporal changes in standards and subsequent-line therapies can undermine exchangeability</cite>. <cite index="15-8,15-9">With simple trials and noninformative priors, Bayesian thresholds look similar to frequentist criteria—replacing p < 0.025 with Pr(parameter > 0) > 0.975; with more complex trials, the threshold to achieve type I error control must be obtained through clinical trial simulation</cite>. The guidance does not replace the need for confidence intervals or frequentist operating characteristics; it makes explicit what was already implicit in every sample-size calculation.

Sources:

• https://arxiv.org/html/2601.14701v1
• https://www.fda.gov/drugs/guidances-drugs/guidance-recap-podcast-use-bayesian-methodology-clinical-trials-drug-and-biological-products
• https://www.fdamap.com/blog/what-fdas-new-guidance-on-bayesian-statistics-means-for-drug-developers/
• https://www.berryconsultants.com/resource/guide-to-the-draft-fda-bayesian-guidance-2026
• https://www.pharmexec.com/view/bringing-bayesian-method-clinical-trials-dr-stacy-lindborg

<cite index="1-1,9-5,9-6">In January 2026, FDA released draft guidance on the use of Bayesian methodology to support primary inference in pivotal trials for drugs and biologics</cite>. <cite index="1-5">The document signals formal regulatory acceptance of Bayesian primary inference in confirmatory trials, provided designs are justified through success criteria, operating characteristics, and robust documentation</cite>.

The takeaway is not that regulators embraced one statistical philosophy over another. <cite index="1-7,18-2">Sponsors must justify Bayesian designs through explicit success criteria, thoughtful priors—especially when borrowing external data—prospective operating-characteristic evaluation, and computational transparency suitable for review</cite>. <cite index="8-2,8-3">For regulatory submissions, companies often need to consider frequentist operating characteristics of the Bayesian analysis strategy, focusing on type I error rate and power for realistic alternatives</cite>.

<cite index="3-1,3-2">Regulatory acceptance is framed as pragmatic modernization: Bayesian methods complement—not replace—frequentist paradigms, contingent on transparent prior justification, operating-characteristic simulations, and early FDA alignment</cite>. <cite index="6-7,6-8">FDA cited REBYOTA, a fecal microbiota product approved in 2022, where a Bayesian primary analysis formally borrowed data from a Phase 2 study to support Phase 3 effectiveness</cite>. The guidance is nonbinding but clarifies when, how, and under what conditions Bayesian methods may support primary inference.

Sources:

• https://arxiv.org/html/2601.14701v1
• https://www.federalregister.gov/documents/2026/01/12/2026-00325/use-of-bayesian-methodology-in-clinical-trials-of-drug-and-biological-products-draft-guidance-for
• https://www.pharmexec.com/view/bringing-bayesian-method-clinical-trials-dr-stacy-lindborg
• https://www.fdamap.com/blog/what-fdas-new-guidance-on-bayesian-statistics-means-for-drug-developers/
• https://arxiv.org/pdf/2311.16506

<cite index="1-2,1-7">An emerging and more controversial type of adaptive design is where protocol changes are made on the basis of the unblinded interim results</cite>. Pocock's 2015 review in JACC, co-authored with Clayton and Stone, <cite index="17-1,18-5">tackled controversial issues including noninferiority trials, factorial designs, strategy trials, Data Monitoring Committees (including when to stop a trial early), and the role of adaptive designs</cite>. The line Pocock draws is between adaptations informed by blinded aggregate data (changing sample size based on pooled variance) and those informed by unblinded treatment comparisons.

<cite index="8-4">The FDA guidance notes that arguments for response-adaptive randomization are controversial, and some researchers feel that inconclusive interim results should not be used to alter randomization in an ongoing trial and/or that statistical efficiency is not substantially improved in two-arm trials to justify adjusting randomization ratios</cite>. Pocock's career has been defined by work on group sequential methods and stopping boundaries; his engagement with adaptive designs is pragmatic rather than evangelical. The question for him is always whether the adaptation delivers genuine efficiency or introduces bias and operational complexity that offset theoretical gains.

Sources:

• https://www.jacc.org/doi/10.1016/j.jacc.2015.10.051
• https://www.fda.gov/media/78495/download
• https://pubmed.ncbi.nlm.nih.gov/26718676/

<cite index="4-1">Senn has clarified that operational efficiencies are the principal advantage of adaptive and platform trials, while purported statistical efficiencies can be exaggerated</cite>. This is a measured pushback against the rhetoric that adaptive designs dramatically improve power or reduce sample size through statistical cleverness alone. <cite index="11-5,11-1">Senn emphasizes understanding data origin and regression to the mean as essential for trial interpretation, above adherence to Bayesian or frequentist frameworks</cite>.

In recent interviews and commentary, <cite index="11-6,11-7">Senn details methodological considerations for time adjustments and model complexity, and identifies the limitations of non-concurrent controls in platform trials, focusing on evolving background therapy, site participation, and protocol changes that reduce validity of historical or pooled control data</cite>. His position is not anti-adaptation; it's pro-clarity. The theoretical developments from the 1990s—Bauer, Köhne, and others—were elegant. The practical uptake requires attention to what changes (logistics, enrollment speed) and what doesn't (the need for concurrent controls, the persistence of bias mechanisms). Adaptive designs offer flexibility, not statistical alchemy.

Sources:

• https://www.berryconsultants.com/resource/55-a-visit-with-stephen-senn-time-concurrent-controls-and-the-bayesian-guidance
• https://www.statsols.com/articles/adaptive-design-and-early-phase-trials-with-stephen-senn

<cite index="5-7">A trial allowing early stopping if there is evidence of a large treatment effect will invariably overestimate the effect of the treatment on average</cite>. This is not a theoretical footnote. <cite index="31-2">Traditional maximum likelihood estimators tend to be biased either because of selection following an interim analysis or other mechanisms like early stopping that affect the sampling distribution of the estimator</cite>. <cite index="31-3">The usual MLE is sometimes called the "naive" estimator for the trial</cite>.

<cite index="35-1,35-2">The FDA guidance on adaptive designs highlights that multiple analyses of the primary endpoint can lead to biased estimation of treatment effects on that endpoint, and less well appreciated is that biased estimation can also apply to any endpoint correlated with the primary endpoint</cite>. For pre-specified adaptation rules, <cite index="37-1,37-3">unbiased estimates can be constructed, but for ad-hoc adaptations, bias correction is not possible in general</cite>. Methods exist—conditional bias-reduced estimators, weighted combinations of stage estimates—but they come with wider confidence intervals. The bias is structural: if you stop trials that are randomly low and continue those that are randomly high, you're selecting on noise.

Sources:

• https://arxiv.org/pdf/2512.10697
• https://orca.cardiff.ac.uk/id/eprint/154619/1/sim.9605.pdf
• https://onlinelibrary.wiley.com/doi/full/10.1002/sim.9734
• https://baselbiometrics.github.io/home/docs/trainings/20230903/Vandemeulebroecke_intro_adaptive.pdf

<cite index="28-2,28-8">Mehta and Pocock's 2011 paper on adaptive sample size re-estimation for phase 3 confirmatory trials</cite> introduced what became known as the "promising zone" design. The concept is straightforward: <cite index="28-5,28-6,28-11,28-12">trials start with a small up-front sample size commitment, then commit additional sample size resources only if promising results are obtained at an interim analysis</cite>. <cite index="21-2">The promising zone method allows this re-estimation without penalty at the final analysis</cite>—a neat bit of conditional power mathematics that preserves type I error under specific conditions.

<cite index="23-6,23-7">The concept has been appealing to researchers for its design simplicity, though it remains relatively new in application and has been a source of controversy</cite>. <cite index="29-1,29-2">While Mehta and Pocock discussed procedures for controlling type-1 error, they did not address how to choose the promising zone or the corresponding sample size reassessment rule, proposing instead that operating characteristics of alternative designs could be compared by simulation</cite>. Subsequent work by Jennison and Turnbull developed optimization frameworks, but the practical uptake has been measured. The design trades on resource efficiency, not statistical magic—you're buying an option to invest more if early data justify continued enrollment.

Sources:

• https://onlinelibrary.wiley.com/doi/10.1002/sim.4102
• https://alimentiv.com/a-practical-look-at-sample-size-re-estimation-of-the-promising-zone-design/
• https://www.researchgate.net/publication/346676394_A_systematic_review_of_the_promising_zone_design
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6767001/

<cite index="6-3,6-4">Adoption and implementation of ICH E6(R2) guideline will require major changes in all areas—quality systems, SOPs, technology, team, training—and at all levels—organization, investigator, sites, CROs, and vendors, and adoption and implementation of the changes in ICH GCP guideline will pose significant challenges for all clinical research stakeholders.</cite> The guidance became effective in 2016; eight years later, uptake remains partial.

<cite index="13-7">A 2021 survey of 6,513 clinical trials by the Association of Clinical Research Organizations found that only 22% used at least one RBM component, with centralized monitoring the most common at 19% and reduced source document review the least common at 8%.</cite> The codification of risk-based approaches in E6(R2) has not eliminated the default to traditional site-centric models. Many sponsors remain risk-averse in practice, even when the guideline grants explicit permission for centralized or hybrid monitoring.

<cite index="3-8">Findings show that many participants view the ICH E6(R2) guidance as helpful overall, although substantial room for improvement remains.</cite> The guideline introduced the architecture for modern trial conduct—quality tolerance limits, risk-proportionate monitoring plans, vendor oversight documentation—but operational translation has lagged regulatory intent. That gap is not trivial: it reflects the inertia of inspection precedent, the absence of harmonized interpretation across regions, and the reality that inspectors often expect what they have historically seen.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC5654214/
• https://pharmaeducenter.com/blog/risk-based-monitoring-in-clinical-trials/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC9468347/

<cite index="17-1,17-2,17-14">The sponsor and the investigator should maintain a record of the location of their respective essential documents, and the storage system should provide for document identification, version history, search, and retrieval.</cite> <cite index="17-6">The sponsor is expected to ensure oversight of any trial-related duties and functions carried out on its behalf and document approval of any subcontracting of trial-related duties and functions by a contract research organization (CRO).</cite> <cite index="21-10,21-13">Sponsors must now maintain oversight of tasks delegated to CROs, and CROs utilizing subcontractors will need to firstly seek the sponsor's approval on delegation of the said responsibility or task.</cite>

Investigator responsibilities have also expanded. <cite index="21-3,21-4,21-5">Delegation by investigators is addressed and the responsibility is reinforced, with a reminder that the Principal Investigator is responsible for supervising anyone to whom they delegate tasks—a statement suggesting active oversight, as opposed to receiving periodic updates—and the PI must ensure any party performing study tasks is qualified and must implement procedures to ensure the integrity of study tasks and data.</cite> <cite index="21-1">The investigator is also responsible for ensuring the quality of documents and the data therein supplied to the sponsor.</cite> <cite index="20-3,20-4">Essential documents should be retained until at least 2 years after the last approval of a marketing application in an ICH region and until there are no pending or contemplated marketing applications, or at least 2 years have elapsed since the formal discontinuation of clinical development, though these documents should be retained for a longer period if required by applicable regulatory requirements or by agreement with the sponsor.</cite>

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC5654214/
• https://www.quanticate.com/blog/ichgcp-e6-addendum-r2
• https://ichgcp.net/4-investigator

<cite index="9-1,9-3">ICH E6(R2) makes explicit the responsibility of the sponsor to understand and actively manage the risks to quality in clinical trials using a Quality Risk Management (QRM) approach.</cite> This is not guidance; it is a structural expectation. <cite index="17-11">The sponsor should implement a quality management system which focuses on trial activities essential to ensuring human subject protection and reliability of trial results.</cite>

The quality management section introduces two linked concepts: <cite index="16-3">defining what is critical to success during protocol development by identifying data and processes critical to human subject protection and the reliability of clinical trial results, and managing the critical elements of a clinical trial by identifying, evaluating, controlling, communicating, reviewing, and reporting risks during the whole life cycle of the clinical trial.</cite> <cite index="16-2">ICH E6(R2) also introduced the concept of predefined quality tolerance limits (QTLs) to help identify systematic, protocol-level issues that can impact subject safety or reliability of trial results.</cite>

<cite index="10-1,10-9,10-10">The guideline states that the sponsor should develop a systematic, prioritized, risk-based approach to monitoring clinical trials, and may choose on-site monitoring, a combination of on-site and centralized monitoring, or, where justified, centralized monitoring.</cite> The implication: traditional 100% source data verification is no longer the default. Sponsors must now justify monitoring intensity against assessed risk—an inversion of the prior burden of proof.

Sources:

• https://www.clinicalleader.com/doc/risk-based-quality-management-0001
• https://www.clinicalleader.com/doc/ich-e-r-best-practices-for-implementing-a-more-formal-risk-management-process-0001
• https://cluepoints.com/ich-e6-r2-miracle-pill-for-the-clinical-rd-industry/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC5654214/

<cite index="1-4,1-5">ICH E6(R2) defines Good Clinical Practice as an international ethical and scientific quality standard for designing, conducting, recording, and reporting trials involving human subjects, with compliance intended to provide public assurance that the rights, safety, and well-being of trial subjects are protected and that clinical trial data are credible.</cite> <cite index="7-2">The guideline was first released in 1996 as E6(R1), with the integrated addendum E6(R2) issued in 2016.</cite>

The R2 revision represented a structural shift. <cite index="5-11">The guideline was amended to encourage implementation of improved and more efficient approaches to clinical trial design, conduct, oversight, recording, and reporting while continuing to ensure human subject protection and reliability of trial results.</cite> <cite index="6-7,6-8">It was amended to foster implementation of improved and more efficient approaches to the management of clinical trial process from protocol planning to study conduct and reporting, reflecting that the field of clinical trials has grown extensively since 1996 due to increases in globalization, clinical study complexity, and technological capabilities.</cite>

<cite index="12-2">The changes include new approaches to quality management system, risk-based monitoring with emphasis on human subject protection, and data integrity.</cite> The addendum introduced operational language for what had been conceptual: sponsors now bear explicit responsibility for quality risk management across the trial lifecycle, and the prescriptive site-monitoring paradigm of the 1990s gave way to sponsor discretion—on-site, centralized, or hybrid—provided the approach is systematic and risk-informed.

Sources:

• https://admin.ich.org/sites/default/files/inline-files/E6_R2__Addendum_Step2.pdf
• https://www.ema.europa.eu/en/documents/scientific-guideline/ich-guideline-good-clinical-practice-e6r2-step-5-revision-2_en.pdf
• https://pmc.ncbi.nlm.nih.gov/articles/PMC9468347/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC5654214/

<cite index="6-1,6-5">Prescribing is one of the most common tasks expected of new doctors and is a complex process involving a mixture of knowledge, judgement and skills.</cite> <cite index="3-1,3-2">The prescriber checklist implies that balancing the expected benefits of a drug against the expected harm is a straightforward process—it is not.</cite>

<cite index="5-7,5-8">Having considered diagnosis, prognosis and goals of therapy, prescribers often select from several pharmacological options; the best choice should maximise the benefit-harm balance based on drug and patient factors, taking into account restrictions based on availability and costs.</cite> <cite index="5-11,5-12,5-13">Drugs in the same class (or different formulations of the same drug) may have different bioavailability, dose-concentration curves and half-lives, and these factors will determine the dosing schedule; once-daily dosing is convenient and encourages adherence.</cite>

<cite index="5-4,5-5">The traditional approach to prescribing involves individualised drug selection based on evidence gathered from groups of similar patients mixed with best-guess judgements about the variability introduced by specific patient and drug factors, while a new era of 'personalised' treatment has been predicted in which therapeutic choices will be individualised based on genetic variables affecting drug handling and action.</cite> <cite index="5-6">Pharmacogenetics is already being used to distinguish responders from non-responders (e.g., trastuzumab for HER2-overexpressing breast cancer) and to avoid adverse effects (e.g., HLA B∗5701 for abacavir hypersensitivity).</cite>

<cite index="9-1">Being a good prescriber is a progressive challenge in modern healthcare with increasing demands as a result of several trends such as the increasing number of medicines available and indications for prescribing them, the greater complexity of treatment regimens taken by individual patients ('polypharmacy'), and the trend to treat more elderly and vulnerable patients.</cite> The promise of pharmacogenetics has not yet replaced the need to understand the fundamentals—absorption, distribution, clearance, half-life, therapeutic index. Those are not academic exercises. They are the vocabulary of harm reduction.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC4953462/
• https://academic.oup.com/book/27861/chapter-abstract/198537220?redirectedFrom=fulltext
• https://www.bps.ac.uk/education-engagement/teaching-pharmacology/ten-principles-of-good-prescribing
• https://www.pharmacologyeducation.org/clinical-pharmacology/prescribing

<cite index="18-1,18-4">Therapeutic drug monitoring (TDM) is a branch of clinical chemistry and clinical pharmacology that specializes in the measurement of medication levels in blood.</cite> <cite index="18-2,18-3,18-6">Its main focus is on drugs with a narrow therapeutic range—drugs that can easily be under- or overdosed.</cite>

<cite index="27-2,27-3">TDM is the practice of measuring drug concentrations in order to tailor dosages and maintain therapeutic levels in a patient's bloodstream, with the goal of improving clinical outcomes by improving efficacy, limiting toxicity, and subsequently reducing the overall cost of drug therapy.</cite> <cite index="27-4">It is an interdisciplinary process that includes clinicians, pharmacists, and laboratory professionals and combines knowledge of pharmacokinetics, pharmacodynamics, the patient's clinical setting (including various preanalytical factors such as dosage, dosing interval, patient characteristics, sample type, and timing of sample collection), and analytical factors within the clinical laboratory.</cite>

<cite index="19-3">Since its inception in the early 1970s, TDM has become an integral part of managing certain medications, often referred to as 'narrow therapeutic index' (NTI) drugs that exhibit significant variability in absorption, distribution, metabolism, excretion and patient response.</cite> <cite index="26-16">Large interindividual variation in the relationship between dose and response can make individualising drug dosage difficult, particularly for drugs with narrow therapeutic indices, large interindividual variation in pharmacokinetics, or concentration-dependent pharmacokinetics.</cite>

<cite index="19-7,19-8">TDM is currently costly due to the need for individualized sampling, high-quality and rapid bioassays, and the involvement of qualified personnel for interpretation and dose adjustment, and these factors may limit its widespread use in many countries.</cite> Not every drug needs this level of surveillance. Most do not. But for the ones that do—aminoglycosides, anticonvulsants, immunosuppressants—the alternative to measurement is guesswork dressed as clinical judgment.

Sources:

• https://en.wikipedia.org/wiki/Therapeutic_drug_monitoring
• https://bpspubs.onlinelibrary.wiley.com/doi/10.1111/bcp.16387
• https://health.ucdavis.edu/blog/lab-best-practice/introduction-to-therapeutic-drug-monitoring-and-the-clinical-laboratorys-role/2020/09
• https://pmc.ncbi.nlm.nih.gov/articles/PMC1873661/

<cite index="32-1">Interindividual variability in anatomical and physiological properties results in significant differences in drug pharmacokinetics.</cite> This is not a marginal issue—it is one of the primary reasons clinical trial data do not reliably predict individual patient response.

<cite index="28-2,28-3,28-4,28-5">While genetic factors are significant, non-genetic factors also contribute substantially to interindividual variability in pharmacokinetics: pediatric and geriatric populations often exhibit distinct pharmacokinetic profiles due to developmental changes or age-related decline in organ function; differences in body composition, enzyme expression, and hormonal influences can result in sex-specific variations in drug pharmacokinetics; and variations in body weight, fat distribution, and muscle mass can affect drug distribution and metabolism.</cite>

<cite index="32-4">One clear objective in clinical drug development is therefore a thorough characterization of the physiological sources of interindividual variability.</cite> <cite index="31-2">Population pharmacokinetic studies aim to identify and quantify sources of variability in drug concentration in the patient population.</cite> <cite index="30-6,30-7,30-8,30-9">Interindividual variability in anatomical and physiological properties results in significant differences in drug pharmacokinetics; the consideration of such pharmacokinetic variability supports optimal drug efficacy and safety for each single individual by identification of individual-specific dosings, and one clear objective in clinical drug development is therefore a thorough characterization of the physiological sources of interindividual variability.</cite>

The gap between population averages and individual patients is where most therapeutic failures live.

Sources:

• https://www.numberanalytics.com/blog/ultimate-guide-interindividual-variability-pharmacokinetics
• https://pubmed.ncbi.nlm.nih.gov/26431198/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4592188/
• https://australianprescriber.tg.org.au/articles/population-pharmacokinetics-an-overview.html

<cite index="2-2,2-9">Clinical pharmacology is the study of the way that drugs act on and are handled by the human body and is the science that underpins rational prescribing.</cite> The field divides into two branches: <cite index="2-4">pharmacodynamics (what the drug does to the body) and pharmacokinetics (what the body does to the drug).</cite>

This is not semantic hairsplitting. <cite index="2-6,2-7">Basic pharmacodynamic studies involve exposing cells or tissues to varying doses of a drug and observing the response to describe a dose–response curve, but for prescribers the situation is more complex because tissue drug exposure depends on how effectively drug molecules are absorbed into the body, distributed to their site of action and subsequently eliminated by metabolism and excretion.</cite>

<cite index="11-3">Pharmacokinetics is the study of how the body changes the drug through the phases of ADME (absorption, distribution, metabolism, and excretion).</cite> <cite index="11-21">The pharmacokinetics and pharmacodynamics together dictate the dose-exposure-response relationship of the drug.</cite> <cite index="10-4">In clinical trials, these concepts are crucial for determining safe and effective dosages, predicting potential side effects, and understanding how different patient populations might respond to a new drug.</cite>

<cite index="13-7">Factors to be taken into consideration when deciding on the best drug dose for a patient include age, gender, weight, ethnic background, other concurrent disease states, and other drug therapy.</cite> <cite index="13-1,13-2,13-3,13-4">Most drugs follow linear pharmacokinetics whereby steady-state serum drug concentrations change proportionally with long-term daily dosing; some drugs do not follow the rules of linear pharmacokinetics and instead of steady-state drug concentration changing proportionally with the dose, serum concentration changes more or less than expected.</cite>

Sources:

• https://basicmedicalkey.com/introduction-to-prescribing/
• https://study.com/academy/lesson/pharmacokinetics-vs-pharmacodynamics.html
• https://kcasbio.com/blogs/pharmacokinetics-vs-pharmacodynamics/
• https://accesspharmacy.mhmedical.com/content.aspx?bookid=1861&sectionid=146077432

The frameworks exist because genomic abundance does not equal therapeutic opportunity. <cite index="8-2">Precision oncology knowledgebases provide a way of organizing clinically relevant genetic information in a way that is easily accessible for both oncologists and patients, facilitating genetic-based clinical decision making</cite>, but <cite index="8-5">it is advisable that oncologists use multiple knowledgebases during their practice to have them complement each other</cite>.

The attrition is severe. In a trial examining genomic-based targeted therapy: <cite index="4-3">a pooled analysis of the SAFIR02-BREAST and SAFIR-PI3K trials demonstrated that genomic-based targeted therapy improved outcomes in metastatic breast cancer, especially with ESCAT I or II populations</cite>. <cite index="4-4,4-5">The genomic findings in this trial seem to be low level in OncoKB. The results of this study might have been different if the patients matched to the targeted therapy had been more carefully selected</cite>.

The conceptual challenge: <cite index="9-4">the key concept around which the scale was developed is actionability, that is the real possibility of targeting a genomic alteration with a drug</cite>. Not predictive value. Not biological relevance. The real possibility of targeting with a drug—a regulatory and clinical availability filter, not a biological one.

Sources:

• https://www.frontiersin.org/journals/cell-and-developmental-biology/articles/10.3389/fcell.2020.00048/full
• https://ascopubs.org/doi/10.1200/JCO.22.00153
• https://www.oncopedia.wiki/contributions/escat-esmo-scale-for-clinical-actionability-of-molecular-targets

<cite index="31-10">The goal of the PMCS/CGAC is to assist the treating oncologist in the translation of molecular variants into clinical action for the individual patient</cite>. <cite index="30-5">Molecular tumor boards play a key role in translating molecular findings into cancer treatment recommendations</cite>, though <cite index="30-4">interpretation of results of molecular analyses and their integration into clinical practice can be challenging</cite>.

<cite index="32-1">To optimize and standardize the process, all liquid biopsy data include both the annotation of significant variants and a corresponding clinical actionability assessment according to OncoKB and ESCAT</cite>. The Oncologist has published a case-based series: <cite index="28-1,28-2">Precision Medicine Clinic: Molecular Tumor Board is a case-based series designed to help clinicians optimize molecular testing for their patients. The Editors of The Oncologist encourage institutions to share their findings and experiences to help improve the precision of genomic oncology</cite>.

The challenge is volume and complexity. <cite index="27-7,27-8">We practice medicine at a time when the volume of data has reached an unprecedented level and is growing at an accelerating rate. Translating the resulting findings into clinical actionability and generating new evidence in a timely fashion constitutes an important challenge</cite>. <cite index="35-2">This is particularly important for patients who present with complex genomic profiles (e.g., variants without OncoKB level 1 evidence or ESCAT equivalent) or who are unresponsive to existing therapy options</cite>.

Sources:

• https://academic.oup.com/oncolo/article/22/2/144/6438571
• https://link.springer.com/article/10.1007/s12254-024-00977-7
• https://www.sciencedirect.com/science/article/pii/S1040842824001227
• https://theoncologist.onlinelibrary.wiley.com/doi/toc/10.1634/(ISSN)1549-490x.pmc_vi
• https://academic.oup.com/oncolo/article/30/3/oyae271/7825833
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11353190/

<cite index="20-5">Potential treatment implications are stratified by level of evidence that a specific molecular alteration is predictive of drug response on the basis of US Food and Drug Administration labeling, National Comprehensive Cancer Network guidelines, disease-focused expert group recommendations, and scientific literature</cite>. <cite index="22-3">Treatment information is classified using the OncoKB Levels of Evidence system, which assigns clinical actionability (ranging from standard-of-care to investigational treatments) to individual mutational events</cite>.

<cite index="6-2">As of February 2023, OncoKB includes annotations for >6200 alterations in >700 cancer-associated genes, including 50 level 1 or 2 standard care genes (specified in the FDA drug label or professional guidelines) as well as 9 level 3A and 12 level 4 genes (predictive of drug response based on well-powered clinical studies or compelling biological evidence, respectively)</cite>. When tested on real-world data: <cite index="24-7">41% of samples harbored at least one potentially actionable alteration, of which 7.5% were predictive of clinical benefit from a standard treatment</cite>.

<cite index="18-1">OncoKB is an FDA-recognized human genetic variant database that integrates biological and clinical information about genomic alterations across various cancer types</cite>. The difference from ESCAT: OncoKB levels assign actionability to specific alterations in specific tumor contexts; ESCAT creates a common vocabulary across stakeholders.

Sources:

• https://ascopubs.org/doi/10.1200/PO.17.00011
• https://pmc.ncbi.nlm.nih.gov/articles/PMC5586540/
• https://kghub.org/kg-registry/resource/oncokb/oncokb.html
• https://www.mskcc.org/research-advantage/support/digital-health-projects/oncokb
• https://ascopubs.org/doi/10.1200/JCO.2023.41.16_suppl.1577

<cite index="10-3">The ESMO Scale of Clinical Actionability for molecular Targets (ESCAT) defines six levels of clinical evidence for molecular targets according to implications for patient management: tier I (targets ready for implementation in routine clinical decisions); tier II (investigational targets that likely define a patient population benefiting from a targeted drug but additional data needed); tier III (clinical benefit previously demonstrated in other tumor types or for similar molecular targets); tier IV (preclinical evidence of actionability); tier V (evidence supporting co-targeting approaches); and tier X (lack of evidence for actionability)</cite>. <cite index="10-1">The ESCAT defines clinical evidence-based criteria to prioritize genomic alterations as markers to select patients for targeted therapies</cite>.

The motivation came from ESMO's Precision Medicine Working Group, which sought <cite index="9-3">a systematic framework to rank molecular targets based on clinical evidence of actionability</cite>. <cite index="16-2">The highest tier (tier I) identifies anomalies suitable for routine clinical use based on prospective data and the lowest (tier X) corresponds to alterations for which there is no evidence for therapeutic utility</cite>.

Real-world validation exists. <cite index="13-7">Median overall survival in 50 patients with ESCAT I-IV alterations who received matched therapy was 22.6 months (95% CI, 20.1–32.8), compared with 14.3 months (95% CI 11.9–18.1) in 130 patients without actionable ESCAT alterations (HR 0.58; 95% CI 0.40–0.85; P = 0.005)</cite> in a cholangiocarcinoma cohort. <cite index="15-7">In February 2023, the Italian Ministry of Health approved access and reimbursement of targeted drugs authorized by AIFA in cases meeting criteria of genetic alterations associated with ESCAT level I of actionability</cite>.

Sources:

• https://www.annalsofoncology.org/article/S0923-7534(19)34179-1/fulltext
• https://www.oncopedia.wiki/contributions/escat-esmo-scale-for-clinical-actionability-of-molecular-targets
• https://www.researchgate.net/figure/ESCAT-ESMO-Scale-for-the-clinical-actionability-of-molecular-targets-adapted-from_tbl2_363143218
• https://aacrjournals.org/clincancerres/article/28/8/1662/694147/ESMO-Scale-for-Clinical-Actionability-of-Molecular
• https://www.esmo.org/scales-and-tools/esmo-scale-for-clinical-actionability-of-molecular-targets-escat

<cite index="24-1,24-12,24-13">Collective statistical illiteracy refers to the widespread inability to understand the meaning of numbers; many doctors, patients, journalists, and politicians draw wrong conclusions without noticing</cite>. <cite index="25-2,25-3">Many citizens are unaware that higher survival rates with cancer screening do not imply longer life, or that "mammography reduces breast cancer death by 25%" means 1 less woman per 1,000</cite>.

<cite index="24-14,24-15">Information pamphlets, websites, leaflets from pharma, and even medical journals report evidence in nontransparent forms that suggest big benefits and small harms; without understanding the numbers, the public is susceptible to manipulation of anxieties and hopes, undermining informed consent and shared decision-making</cite>. <cite index="6-3,6-4">Contributors to Better Doctors, Better Patients, Better Decisions call for journals that report study outcomes completely and transparently, and patients not afraid of statistics but able to use them to make informed treatment decisions</cite>.

<cite index="7-5,7-6">The risk literacy problem is one of the few in healthcare with a known solution: teach efficient risk communication that fosters transparency, in medical school and continuing education</cite>. The infrastructure exists. The question is whether we deploy it.

Sources:

• https://journals.sagepub.com/doi/10.1111/j.1539-6053.2008.00033.x
• https://www.stat.berkeley.edu/~aldous/157/Papers/health_stats.pdf
• https://www.amazon.com/Better-Doctors-Patients-Decisions-Envisioning/dp/026251852X
• https://www.oxfordmartin.ox.ac.uk/events/risk-literacy-in-health

<cite index="21-2,21-4">52% of participating physicians answered only two or fewer of four questions on statistical concepts correctly, and only 40% of resident physicians demonstrated adequate understanding of biostatistical concepts</cite>. <cite index="27-2">33% of gynecologists were unaware of the benefits of mammography screening, with 79% unable to interpret positive predictive value</cite>. <cite index="27-4,27-5">50% of participants thought false positive test results in HIV testing do not exist; only 2 of 20 urologists had sufficient knowledge about PSA test reliability</cite>.

<cite index="7-2,7-3">Gigerenzer and Gray listed "seven sins" of healthcare systems, one being health professionals' lack of risk literacy: many were unsure what a false-positive rate was, what overdiagnosis and survival rates mean, and were unable to evaluate articles in their own field</cite>. <cite index="22-3">Causes of statistical illiteracy include inadequate teaching in medical schools, the emotional dynamics of the doctor-patient relationship, and conflicts of interest</cite>. <cite index="22-5,22-10">Even a few hours of targeted training can significantly improve statistical understanding</cite>.

This is not an education gap. It is a systems failure: we train physicians to prescribe interventions whose benefits they cannot quantify.

Sources:

• https://www.egms.de/static/en/journals/zma/2021-38/zma001473.shtml
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8136351/
• https://www.oxfordmartin.ox.ac.uk/events/risk-literacy-in-health
• https://www.gerd-gigerenzer.com/helping-doctors-making-sense-of-health-statistics

<cite index="28-1,28-4,28-5">Gigerenzer illustrated how identical screening benefits can be framed as relative risk reduction ("reduces your chance of dying by one-third over 10 years"), absolute risk reduction ("reduces your chance from 3 in 1,000 to 2 in 1,000"), or numbers needed to treat</cite>. The relative figure is three times larger—and three times more persuasive.

<cite index="34-2,34-3">In a study of 150 gynecologists, one-third did not understand a 25% relative risk reduction from mammography screening; most believed 250 fewer women per 1,000 would be saved, when the evidence-based estimate is about 1 in 1,000</cite>. <cite index="31-8,31-9">Relative risks are larger numbers than absolute risks and suggest higher benefits than exist; absolute risks make actual benefits more understandable</cite>.

<cite index="22-4,22-9">Conflicts of interest within healthcare often result in intentional framing—in patient pamphlets and medical journals—to exaggerate intervention benefits while downplaying risks</cite>. <cite index="24-18,24-19">Transparent risk communication requires frequency statements instead of single-event probabilities, and absolute risks</cite>. The frame is the intervention. The question is whether it serves the patient or the pitch.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC3310025/
• https://pmc.ncbi.nlm.nih.gov/articles/PMC2733256/
• https://www.barnesandnoble.com/w/calculated-risks-gerd-gigerenzer/1111415409
• https://www.gerd-gigerenzer.com/helping-doctors-making-sense-of-health-statistics
• https://www.stat.berkeley.edu/~aldous/157/Papers/health_stats.pdf

<cite index="10-1,10-5">Gigerenzer and Hoffrage developed a representation called natural frequencies to help people make Bayesian inferences correctly without outside help</cite>. The method rests on a representational fix, not a cognitive one: <cite index="10-4">lay people and professionals often commit the base-rate fallacy when facing conditional probabilities</cite>, but <cite index="10-6">with natural frequencies, even 4th graders can make correct inferences</cite>.

<cite index="4-5,4-10">In one study, 160 gynecologists were given the statistics for calculating that a woman with a positive mammogram actually has cancer: sensitivity 90%, false-positive rate 9%, prevalence 1%</cite>. The physicians failed. <cite index="14-1,14-2">Unlike conditional probabilities, natural frequencies are not normalized with respect to base rates, making it easier to apply Bayes's rule to determine the posterior probability of disease given a positive test</cite>. Instead of saying "90% sensitivity," you say: "Out of 1,000 women, 10 have the disease. Of those 10, the test correctly identifies 9."

<cite index="10-8,10-9">Gigerenzer has taught risk literacy to some 1,000 doctors and 50 US federal judges, and natural frequencies has entered the vocabulary of evidence-based medicine; medical schools worldwide have begun teaching these tools</cite>. This is not statistical sophistication. It is plumbing: fixing the pipes so the information can flow.

Sources:

• https://en.wikipedia.org/wiki/Gerd_Gigerenzer
• https://pmc.ncbi.nlm.nih.gov/articles/PMC3341653/
• https://www.scienceofboosting.org/project/natural-bayesian-inferences/

<cite index="13-7,13-8">Before Atlas, Booth was a consultant at McKinsey, where he "worked a lot with pharmaceutical and biotechnology firms to address R&D productivity issues, as well as portfolio management, new product commercialization, and business development"</cite>. That background shows. His writing consistently returns to questions of efficiency, capital allocation, and what drives value per dollar deployed.

<cite index="13-11,13-12,13-13">He holds a DPhil in molecular immunology from Oxford as a British Marshall Scholar, "focused on the study of HIV and tumor immune responses," and did his undergrad in biochemistry from Penn State</cite>. The pairing—bench science training, then management consulting, then venture—produces a particular sensibility: someone fluent in both assay development and IRR calculation, comfortable in the methods section and the cap table.

In a March 2026 Nature Biotechnology piece, <cite index="4-1,4-3">Booth was described as having been "closely involved in the discovery and development of novel medicines over the past 20 years," reflecting on "past breakthroughs, changes in the funding landscape, and promising drug modalities now in development"</cite>. The interview positioned him as a 30th-anniversary retrospective voice—someone who has watched multiple cycles and can narrate what changed and what recurred.

<cite index="6-6">Since joining Atlas in 2005, Booth has been "involved in the founding and funding of many companies with a particular focus on capital efficient models for drug research and development"</cite>. That phrase—"capital efficient models"—recurs across profiles and interviews. It is the through-line.

Sources:

• https://lifescivc.com/about/
• https://www.nature.com/articles/s41587-026-03027-1
• https://www.psu.edu/news/academics/story/bruce-booth-honored-penn-states-outstanding-science-alumni-award

Booth has documented the mechanics of biotech IPO timing in granular detail. <cite index="11-23,11-1">The organizational meeting—when the underwriting syndicate meets with management to start the formal process—acts as "the gatekeeper of the industry's IPO cadence," taking 4-5 months to reach pricing</cite>. <cite index="11-25,11-26">Companies that price IPOs in late January held org meetings around Labor Day; the June burst of activity reflects org meetings "in the few weeks after the industry kicks off at JPM"</cite>.

This temporal lag matters. <cite index="11-27,11-28">The delay from committing to the path to pricing "is why the market doesn't just flip on when the markets get frothier"—even when biotech markets started to rally in late 2025, "we still didn't see many IPOs in large part because of this temporal dynamic"</cite>. The window doesn't open instantly; it opens four months after the first company decides to test it.

<cite index="14-18,14-19,14-20">Booth frames the CEO's job as creating "a set of strategic choices for companies, whether it's equity financings privately, IPOs and public equity financings, partnerships, or maybe acquisitions"</cite>—what he calls "corporate development cards that you can play." <cite index="14-21,14-22,14-23">Biotech "takes a super long period of time to be profitable" and most companies are "lossmaking," burning money "to advance drugs deeper into clinical development"</cite>. The IPO is one financing instrument among several, not an exit or a validation event. It is a mechanism to buy runway.

Sources:

• https://lifescivc.com/
• https://www.biopharmadive.com/news/atlas-ventures-bruce-booth-on-testing-the-ipo-waters-and-putting-the-tech/630724/

Booth has written extensively on computational drug discovery, with particular caution about overclaiming. In a 2017 LifeSciVC post, he pushed back on the "software eats biotech" narrative, noting that <cite index="31-1,31-2">claims of neural networks and deep learning "cutting billions off the industry's cost of new drugs" represent a "new variant of a decades-old thesis"</cite>. He opened with a quote from 1981 touting computer-aided drug discovery (CADD) and then observed: <cite index="31-7">"if software has been truly eating biotech, then it's been doing it very slowly"</cite>.

This is not Luddism. <cite index="31-8,31-9">Booth describes himself as "a big proponent" of in silico technology in R&D, and Atlas has "put our money behind the concept of CADD-inspired drug discovery many times"</cite>. <cite index="31-10">He co-founded Nimbus Therapeutics in 2009 with Ramy Farid of Schrödinger</cite>, and Atlas has backed structure-based and ligand-based CADD platforms. <cite index="31-16,31-15">But "promises of huge savings in time and money back in 1981 have failed to be delivered by CADD, in general, to date," and "overhyped CADD solutions over the past four decades have damaged the credibility of the field"</cite>.

In a 2024 year-in-review, <cite index="33-1">Booth reiterated that "AI remains an evolutionary force, not a revolutionary one, in biotech"</cite>. He rejects the premise that drug discovery timelines will collapse: the work remains empirical, iterative, and constrained by biology. <cite index="31-14">"Putting science-first with CADD, rather than hype, is key to delivering the expected returns to patients and shareholders"</cite>—a sentence that could serve as his epistemic north star.

Sources:

• https://lifescivc.com/2017/04/four-decades-hacking-biotech-yet-biology-still-consumes-everything/
• https://podcastnotes.org/uncategorized/bruce-booth-whats-shaping-biotech-right-now-atlas-venture-2024-year-in-review/

<cite index="7-5,13-3">Atlas Venture focuses on seed-led venture creation around the discovery and development of novel therapeutics</cite>, a model <cite index="3-5">that emphasizes capital-efficient structures for drug discovery and development</cite>. Booth joined Atlas in 2005 after stints at McKinsey and Caxton Health Holdings, bringing a lens on R&D productivity issues from his consulting years.

The firm operates what <cite index="22-1,22-6">it calls a two-fund model powered by dedicated pools of early stage and growth capital</cite>—seed and Series A money, then later-stage follow-on through separate opportunity funds. <cite index="27-17,27-1">Atlas is "determined to stay disciplined and focused on the model that we continue to try to perfect: seed-led venture creation"</cite>, per Booth's writing in late 2024. <cite index="27-16">The firm has "scar tissue" from growing too fast in the early part of the century</cite>, when it invested in both tech and biotech; <cite index="25-4">it shifted to biotech-only in October 2014</cite>.

<cite index="27-8,27-10">Atlas raised $450 million for its 14th fund in December 2024, though Booth noted they "could have raised several multiples" of that cap</cite>. The restraint is intentional. <cite index="27-18,27-19">A modest fund size "better allows the firm to be dynamic and adaptive" and helps Atlas "remain company builders focused on disciplined capital allocation" rather than "asset aggregators in order to absorb higher fees"</cite>. This is financing strategy as portfolio construction.

<cite index="2-2,2-4">Booth is chairman and was founding CEO of Kymera Therapeutics and Nimbus Therapeutics</cite>, giving him operational time inside the companies he seeds—a VC who has held the other end of the term sheet.

Sources:

• https://lifescivc.com/about/
• https://atlasventure.com/team/bruce-booth-dphil/
• https://www.crunchbase.com/person/bruce-booth
• https://www.biopharmadive.com/news/atlas-venture-capital-fund-biotech-startups-investing-strategy/734702/
• https://www.businesswire.com/news/home/20250904269555/en/Atlas-Venture-Announces-$400-Million-Third-Opportunity-Fund
• https://en.wikipedia.org/wiki/Atlas_Venture

<cite index="11-6,11-7,11-8,11-9">The QALY methodology is particularly ill-suited to assess the value of rare disease drugs—new therapies for rare diseases are approved with fewer subjects in clinical trials, predicting the longevity of patients in these trials is difficult, and since these drugs will ultimately have fewer customers their prices tend to be higher; the QALY threshold represents an arbitrary straightjacket when applied to these unique therapies.</cite>

<cite index="1-3,1-4,1-5">Critics argue that a QALY-based system would limit research on treatments for rare disorders because the upfront costs of the treatments tend to be higher; officials in the United Kingdom were forced to create the Cancer Drugs Fund to pay for new drugs regardless of their QALY rating because innovation had stalled since NICE was founded, and at the time one in seven drugs were turned down.</cite>

<cite index="11-13,11-14,11-16">The greatest flaw by far in the QALY methodology is the subjective threshold value attached to a year of perfect health—in the United States, ICER values one QALY at $50,000 to $150,000, and while some health economists try to justify these values through laborious studies that compare the costs of various medical services, the threshold amount is, at its root, random.</cite> <cite index="2-2,2-8">Objections made to the QALY are a barrier to the adoption of health technology assessment and value-based pricing tools.</cite>

Sources:

• https://www.statnews.com/2019/02/22/qaly-drug-effectiveness-reviews/
• https://en.wikipedia.org/wiki/Quality-adjusted_life_year
• https://www.healthaffairs.org/doi/10.1377/hlthaff.2021.00343

<cite index="8-2">For many years, criticisms of the QALY yielded the response, "we know it's not perfect, but there is no alternative"—this has changed with the emergence of credible alternatives, including generalized risk-adjusted QALYs, equal value of life years gained, and health years in total.</cite> <cite index="13-2,13-4">Health economists have recognized the ethical concerns and discriminatory aspects of the QALY regarding people with disabilities, and several novel methodologies have been developed; these include Equal Value of Life Years Gained (evLYG), The Efficiency Frontier (EF), and Health Years in Total (HYT).</cite>

But the alternatives are not settled science. <cite index="4-2,4-3">The HYT and evLYG approaches can result in logical inconsistencies that do not arise when using QALYs—HYT can violate the "independence" axiom, whereas the evLYG can produce an unstable ranking of treatment options.</cite> <cite index="4-8">The equal value of life-years gained (evLYG) approach assigns equal value to life extensions of similar length; however, this does not incentivize development of technologies that also improve health-related quality of life during life extension.</cite>

<cite index="8-5,8-6">QALYs embody a particular set of value judgements, as do all the proposed alternatives to it—it is not the place of health economics and outcomes researchers to impose such value judgements.</cite> <cite index="2-5,2-11">Ethical criticisms do not apply only to the QALY and require political decisions about societal values.</cite>

Sources:

• https://www.valueinhealthjournal.com/article/S1098-3015(24)02355-6/fulltext
• https://www.ncd.gov/assets/uploads/reports/2022/ncd_alternatives_to_the_qaly_508.pdf
• https://www.valueinhealthjournal.com/article/S1098-3015(23)06201-0/fulltext
• https://www.healthaffairs.org/doi/10.1377/hlthaff.2021.00343

<cite index="3-2">A European study tested the validity of key assumptions behind the QALY—including the assumption that time and utility are 100% independent—and found that only 70% of individuals displayed "consistent preferences" for health states.</cite> <cite index="14-4,14-5">ECHOUTCOME released "European Guidelines for Cost-Effectiveness Assessments of Health Technologies," which recommended not using QALYs in healthcare decision making and instead recommended that cost-effectiveness analyses focus on "costs per relevant clinical outcome."</cite>

<cite index="2-4,2-10">Methods-based criticisms require attention from stakeholders to address well-known shortcomings of the QALY and ensure consistency.</cite> The theoretical floor is debated. <cite index="8-4">The criticism that QALYs are ableist on the grounds that "the HRQoL of those with disabilities is poor" may partly be a product of the practice of seeking HRQoL values from the general public, rather than from those with lived experience of disability.</cite> <cite index="3-1">One criticism is the model's assumption that death is the lowest utility state possible—might not some disease states arguably be worse than death?</cite>

<cite index="16-4,16-5,16-6">Utilities to calculate QALYs should be based on decisions under uncertainty (decision utilities), but the health care literature describes several problems that lead to bias in the measurement of decision utilities; these biases lead to an inaccurate estimation of the value of a health state and can therefore cause policy makers to allocate resources inefficiently.</cite>

Sources:

• https://www.clinicalcorrelations.org/2021/03/10/questioning-the-qaly-a-closer-look-at-the-quality-adjusted-life-year/
• https://en.wikipedia.org/wiki/Quality-adjusted_life_year
• https://www.healthaffairs.org/doi/10.1377/hlthaff.2021.00343
• https://www.valueinhealthjournal.com/article/S1098-3015(24)02355-6/fulltext
• https://clinicaltrials.gov/study/NCT01085409

<cite index="1-7,1-10">Oregon's 1989 attempt to incorporate QALYs into Medicaid reform was found discriminatory and in violation of the Americans with Disabilities Act in 1992.</cite> The logic is structural: <cite index="4-1">a common criticism of the QALY is that life extensions generate fewer QALYs for those with poorer health-related quality of life.</cite> <cite index="24-5">For certain severe, disabling conditions, traditional approaches are likely to conclude that treatments are not cost-effective at any price once a patient progresses to a disabled health state with low utility value.</cite>

The critique extends to age. <cite index="18-8,18-9">The QALY calculation is "inherently discriminatory against treatments for older adults"—a new drug for an 80-year-old is mathematically valued less than the same drug for a 30-year-old.</cite> <cite index="11-12">Congress banned QALY use in cost-effectiveness reviews in the Medicare program.</cite>

<cite index="22-7,22-8">QALYs derive from bias by assigning quality-of-life values to people with diseases and disabilities—people with disabilities are deemed, by others without experience of living with the disability, to have a lower quality of life than healthy people.</cite> <cite index="21-3,21-4">The methods used to derive the QoL weights often fail to depict the perspectives of individuals living with disabilities; these weights usually depend on general public opinion about a life with certain weaknesses, rather than on self-assessments from people actually living with those impairments.</cite>

But one empirical study complicates the narrative. <cite index="20-2,20-3">Testing whether discrimination occurs in practice with respect to age, researchers found no evidence that estimates of incremental cost per QALY gained systematically differ above and below 65 years of age.</cite>

Sources:

• https://en.wikipedia.org/wiki/Quality-adjusted_life_year
• https://www.valueinhealthjournal.com/article/S1098-3015(23)06201-0/fulltext
• https://www.patientsrising.org/patientsrising.org/policy/qaly
• https://www.healthaffairs.org/doi/10.1377/hlthaff.2021.00409
• https://www.sciencedirect.com/science/article/pii/S1098301524023556
• https://www.valueinhealthjournal.com/article/S1098-3015(24)02789-X/fulltext

<cite index="7-11,7-12">FDA has evaluated for therapeutic equivalence only multisource prescription drug products approved under Section 505 of the FD&C Act, which in most instances means pharmaceutical equivalents available from more than one manufacturer; for such products a therapeutic equivalence code generally is included and product information is highlighted in bold face and underlined</cite>. <cite index="7-13">Single-source products (only one approved product available for that active ingredient, dosage form, route, and strength) are included in the Orange Book but no therapeutic equivalence code is included</cite>.

The Reference Listed Drug is the product against which generic applicants must demonstrate bioequivalence. <cite index="19-11">Generic drug manufacturers rely on the Orange Book to identify the relevant patents they must certify against when seeking approval for a generic version of a brand-name drug</cite>. <cite index="6-3">Pharmaceutical equivalents are drug products in identical dosage forms and route(s) of administration that contain identical amounts of the identical active drug ingredient (same salt or ester of the same therapeutic moiety), or in the case of modified-release dosage forms that require a reservoir or overage, deliver identical amounts over the identical dosing period; they do not necessarily contain the same inactive ingredients and meet the identical compendial standards</cite>.

<cite index="7-8">FDA considers drug products therapeutically equivalent if they meet the criteria even though they may differ in characteristics such as shape, scoring configuration, release mechanisms, packaging, excipients (including colors, flavors, preservatives), expiration date/time, certain labeling aspects, and storage conditions</cite>. <cite index="7-9">When such differences are important in the care of a particular patient, it may be appropriate for the prescribing physician to require that a specific product be dispensed as a medical necessity</cite>. This carve-out acknowledges that therapeutic equivalence is a regulatory determination, not a clinical absolute.

Sources:

• https://www.fda.gov/media/71474/download
• https://www.fda.gov/media/160167/download
• https://legalclarity.org/orange-book-transparency-act-patent-and-exclusivity-rules/

<cite index="23-1,23-2,23-3">NDA and ANDA filers may apply for regulatory exclusivities that prevent FDA from approving generic versions or additional ANDAs; exclusivities may be granted for new chemical entity, new indication, new dosage form, and pediatric exclusivity among others, are time-limited, and operate independently of patent protection</cite>. <cite index="22-2">New drugs once approved are granted at least a five-year exclusivity period during which no other brand-name or generic drug using the same active ingredient can be marketed, with longer or shorter exclusivity periods for certain drug types</cite>.

<cite index="25-4">The Orange Book identifies drug products approved on the basis of safety and effectiveness and related patent and exclusivity information</cite>. <cite index="24-1">For drug manufacturers, the Orange Book's information on a drug's patents and regulatory exclusivities can be critical to whether and when generic competition occurs</cite>. The distinction between patent protection and regulatory exclusivity matters: patent rights arise from USPTO examination of novelty, nonobviousness, and enablement claims; regulatory exclusivities are statutory grants tied to approval milestones regardless of patent status.

<cite index="19-5,19-6,19-7,19-8">The Orange Book Transparency Act, enacted as part of the Consolidated Appropriations Act 2021, updated the drug approval and patent listing process; the legislation enhances public visibility of patent and market exclusivity information for brand-name drugs, aims to reduce ambiguity and streamline the process by which generic manufacturers challenge patents, and requires clearer and more timely disclosure to prevent misuse of patent listings to delay availability of lower-cost alternatives</cite>. <cite index="19-13">OBTA codified and clarified many patent listing requirements that previously existed only in FDA regulations</cite>.

Sources:

• https://www.fr.com/insights/ip-law-essentials/orange-book-101/
• https://oneill.law.georgetown.edu/recent-developments-in-orange-book-litigation-how-patent-disputes-shape-prescription-drug-affordability/
• https://www.fda.gov/drugs/drug-approvals-and-databases/approved-drug-products-therapeutic-equivalence-evaluations-orange-book
• https://www.congress.gov/crs-product/IF12644
• https://legalclarity.org/orange-book-transparency-act-patent-and-exclusivity-rules/

<cite index="4-2,4-3">The Orange Book is a publication produced by FDA as required by the Drug Price Competition and Patent Act (Hatch-Waxman Act), created to balance inducing pioneering research and development with enabling competitors to bring low-cost generic copies to market</cite>. <cite index="9-1,9-2,9-3">On October 31, 1980, FDA published the final rule Therapeutically Equivalent Drugs: Availability of List, amending 21 CFR 20.117, and simultaneously published the first annual edition; because it was published on Halloween, FDA chose an orange cover, prompting its colloquial name</cite>.

The 1984 Act transformed the publication from a reference document into litigation infrastructure. <cite index="22-3">Hatch-Waxman created an expedited pathway allowing generic manufacturers to rely on clinical trial data from the brand-name drug when applying for approval via an Abbreviated New Drug Application (ANDA)</cite>. <cite index="22-5">If a patent for a specific drug is listed in the Orange Book, FDA cannot approve any related generic until patent issues are resolved</cite>. <cite index="22-4">Generic manufacturers can submit an ANDA before brand-name patents expire by showing the related patent listing is invalid or unenforceable</cite>.

<cite index="4-9,4-10">The Orange Book lists patents purported to protect each drug; patent listings and use codes are provided by the drug application owner, and FDA is obliged to list them</cite>. <cite index="22-1,22-6">Some brand-name manufacturers have used the Orange Book publication process to block generic competition by listing improper or invalid patents; this is possible because FDA does not assess patent listings for technical validity</cite>. <cite index="20-4,20-9">NDA sponsors must submit for listing patents protecting approved drug substance, product, or methods of use upon filing, amendment, or supplement and again within 30 days of approval; later-issued patents must be submitted within 30 days of issuance</cite>.

Sources:

• https://en.wikipedia.org/wiki/Approved_Drug_Products_with_Therapeutic_Equivalence_Evaluations
• https://www.fdli.org/2025/05/freshly-squeezed-orange-book-history-and-key-updates-at-45/
• https://oneill.law.georgetown.edu/recent-developments-in-orange-book-litigation-how-patent-disputes-shape-prescription-drug-affordability/
• https://www.wipo.int/edocs/mdocs/scp/en/scp_31/scp_31_h_orange.pdf

<cite index="2-2,1-2">The Orange Book—formally titled Approved Drug Products with Therapeutic Equivalence Evaluations—lists drug products FDA approved for safety and effectiveness under the Federal Food, Drug, and Cosmetic Act, alongside patent and exclusivity data</cite>. <cite index="2-9,2-10">FDA applies therapeutic equivalence criteria to multisource prescription products, representing these evaluations as code letters</cite>. <cite index="6-4">Products are therapeutic equivalents if they are pharmaceutical equivalents for which bioequivalence has been demonstrated and can be expected to have the same clinical effect and safety profile when administered under labeled conditions (21 CFR 314.3(b))</cite>.

<cite index="14-2,14-3">Codes start with A or B and contain at least two letters; A-rated drugs have been determined bioequivalent to the brand drug, while B-rated drugs are considered not bioequivalent</cite>. <cite index="13-1,13-2">AB is the most common A-subtype, meaning the drug meets all bioequivalence standards</cite>. <cite index="15-6,15-8,15-9">When more than one reference listed drug exists, a number follows the two-letter code; a branded product rated AB1 permits substitution only with generics rated AB1, while AB2-rated generics correspond to the other RLD</cite>.

<cite index="7-2">These evaluations serve as public information to state health agencies, prescribers, and pharmacists to promote drug product selection education and foster healthcare cost containment</cite>. <cite index="2-8">The agency recognized that providing a single list based on common criteria would be preferable to evaluating products on differing definitions in various state laws</cite>. State substitution laws permit pharmacists to dispense A-rated generics without prescriber authorization in most jurisdictions, but <cite index="7-3">therapeutic equivalence evaluations are not official FDA actions affecting the legal status of products under the FD&C Act</cite>.

Sources:

• https://www.fda.gov/drugs/development-approval-process-drugs/orange-book-preface
• https://www.fda.gov/drugs/drug-approvals-and-databases/approved-drug-products-therapeutic-equivalence-evaluations-orange-book
• https://www.fda.gov/media/160167/download
• https://www.fda.gov/media/71474/download
• https://www.uspharmacist.com/article/insights-into-effective-generic-substitution
• https://mdpuppiesonline.com/therapeutic-equivalence-codes-how-the-fda-determines-if-generic-drugs-can-be-substituted
• https://www.pharmacytimes.com/view/genericfeature-0311

<cite index="20-1">Lipinski's rule of five has been critiqued for oversimplification of drug absorption mechanisms, primarily assuming passive diffusion across cell membranes while overlooking the role of active transporters.</cite> <cite index="20-2,20-3">The rule explicitly applies only to compounds not reliant on transporter-mediated uptake; this limitation highlights how the rule fails to account for transporter interactions that enable permeation for larger or more polar molecules.</cite>

<cite index="21-12,21-13">Two major weaknesses are the equal weight given to each of the rules and the sharp boundary that marks the violation of a given rule.</cite> <cite index="21-14,21-15">The rule does not include natural and biological compounds and does not incorporate criteria relevant to metabolism.</cite> <cite index="22-7">As with many other rules of thumb, there are many exceptions.</cite>

<cite index="13-11,13-12,13-13">Many people have questioned the rigid interpretation of these rules; the Ro5 was never more than a crude filter, and its limitations were obvious from the very beginning, as is the case of natural products, which break systematically the Ro5 but are both bioavailable and bioactive.</cite> <cite index="15-8,15-9">57%, 9%, and 54% of compounds with MW>500 g/mol had poor Caco-2 permeability, good Caco-2 permeability, and poor oral bioavailability (<30%), respectively; no correlation between Caco-2 permeability and bioavailability was found, demonstrating that the 500 g/mol MW-cutoff is not clearly associated with poor absorption.</cite>

Sources:

• https://grokipedia.com/page/Lipinski's_rule_of_five
• https://www.pmf.ni.ac.rs/chemianaissensis/wp-content/uploads/filebase/v3n12020/Ivanovic%20et%20al.,%202020,%20tekst.pdf
• https://en.wikipedia.org/wiki/Lipinski's_rule_of_five
• https://www.galchimia.com/twenty-years-of-the-rule-of-five/
• https://www.biorxiv.org/content/10.1101/2024.08.20.608791.full.pdf

<cite index="1-7,1-8">During drug discovery, lipophilicity and molecular weight are often increased to improve affinity and selectivity; hence it is often difficult to maintain Ro5 compliance during hit and lead optimization.</cite> <cite index="1-9">It has been proposed that screening library members should be biased toward lower molecular weight and lipophilicity so medicinal chemists will have an easier time delivering optimized drug development candidates that are also drug-like.</cite>

<cite index="2-14,2-15">The introduction of Lipinski's Rule of Five initiated a profound shift in the thinking paradigm of medicinal chemists; understanding the difference between biologically active small molecules and drugs became a priority in the drug discovery process, and the importance of addressing pharmacokinetic properties early during lead optimization is a clear result.</cite> <cite index="9-1,9-3">Most chemists think of the rule of 5 as guardrails rather than rules; to reduce oral bioavailability design to only a handful of properties is pretty simplistic, and yet it still is useful.</cite>

<cite index="9-6,9-7">Even Lipinski counts himself among those that don't think the rule of 5 should be hard and fast: "I am still slightly taken aback by how some people want strict guidelines without consideration of nuance."</cite> <cite index="5-13,5-14,5-15">Exploring new areas of chemical space can take researchers outside Lipinski's guidance, but it is important to remember why the Rule is there; a molecule that complies will almost certainly have many other drug-like properties, and if one must go outside the Rule of 5 landscape, there should be a good reason for it.</cite>

Sources:

• https://en.wikipedia.org/wiki/Lipinski's_rule_of_five
• https://www.sciencedirect.com/topics/pharmacology-toxicology-and-pharmaceutical-science/lipinskis-rule-of-five
• https://cen.acs.org/pharmaceuticals/drug-discovery/Wrestling-Lipinski-rule-5/101/i8
• https://www.sygnaturediscovery.com/news-and-events/blog/the-rule-of-5-two-decades-later/

<cite index="18-4">Approximately 38% of FDA-approved orally administered parent compounds from 2011 to 2022 deviate from the original Ro5 criteria.</cite> <cite index="20-4,20-8">Analysis of approved oral drugs from 1950 to 2021 shows approximately 66% adhered strictly to the rule.</cite> <cite index="15-7">Almost 200 approved oral drugs had molecular weight above 500 g/mol and the number of new drugs exceeding that limit increases over time.</cite>

<cite index="2-3,2-4">Twenty of 48 FDA-approved small molecule protein kinase inhibitors fail to conform to the rule of five; there is a tendency for orally effective small molecule protein kinase inhibitors to exceed the 500 Da molecular weight criterion.</cite> <cite index="15-16">Sufficiently high fraction absorbed and bioavailability have been reached for approved oral drugs with MW, clog P and HBD exceeding 1200 g/mol (cyclosporin with 40% fraction absorbed), 8 (venetoclax with >65% fraction absorbed) and 8 (k-strophanthoside with 16% fraction absorbed), respectively.</cite>

<cite index="21-7,21-8">Antibiotics, antifungals, vitamins, and cardiac glycosides are orally active therapeutic classes that violate the rule; these compounds have structural features that allow the drugs to act as substrates for naturally occurring transporters.</cite> <cite index="20-2">Cyclosporine (molecular weight 1202 Da) achieves bioavailability through active transport despite multiple violations.</cite> <cite index="15-17">Too strict reliance on Ro5 may cause lost opportunities.</cite>

Sources:

• https://www.tandfonline.com/doi/abs/10.1080/17460441.2023.2275617
• https://grokipedia.com/page/Lipinski's_rule_of_five
• https://www.biorxiv.org/content/10.1101/2024.08.20.608791.full.pdf
• https://www.sciencedirect.com/topics/pharmacology-toxicology-and-pharmaceutical-science/lipinskis-rule-of-five
• https://www.pmf.ni.ac.rs/chemianaissensis/wp-content/uploads/filebase/v3n12020/Ivanovic%20et%20al.,%202020,%20tekst.pdf

<cite index="22-2">Christopher Lipinski formulated the Rule of Five in 1997, based on observation that most orally administered drugs are relatively small and moderately lipophilic molecules.</cite> <cite index="10-7">The rule emerged from retrospective analysis of 2,245 compounds at Phase II entry.</cite> The thresholds are these: <cite index="5-6">no more than five hydrogen bond donors, no more than 10 hydrogen bond acceptors, molecular weight less than 500 Da, and log P below 5.</cite> <cite index="22-6">All numbers are multiples of five, which is the origin of the rule's name.</cite>

<cite index="1-5">The rule describes molecular properties important for a drug's pharmacokinetics in the human body, including absorption, distribution, metabolism, and excretion.</cite> <cite index="1-6">However, the rule does not predict if a compound is pharmacologically active.</cite> <cite index="7-2,16-6,25-1">An orally active drug can have no more than one violation of these conditions.</cite>

<cite index="13-7,13-8">The Rule of Five resulted from a Pfizer study in the late 1990s on favorable absorption properties of orally administered drugs and clinical candidates; four key properties were selected, for which cut-offs were calculated that covered 90% of the molecules studied.</cite> <cite index="5-3">Rather than a set of hard and fast rules, it provides a guideline that points designers towards molecules whose properties make them more likely to succeed in the clinic.</cite>

Sources:

• https://en.wikipedia.org/wiki/Lipinski's_rule_of_five
• https://www.sygnaturediscovery.com/news-and-events/blog/the-rule-of-5-two-decades-later/
• https://www.galchimia.com/twenty-years-of-the-rule-of-five/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8817695/
• https://dev.drugbank.com/guides/terms/lipinski-s-rule-of-five

<cite index="17-2">Ten years after the 2000 publication, it became increasingly clear that mutated cells on their way to giving rise to a tumor have to learn how to thrive in a chronically inflamed microenvironment, evade immune recognition, and suppress immune reactivity</cite>. <cite index="17-3">Genetic and molecular definition of these three immune hallmarks of cancer offers the opportunity to deploy specific countermeasures to reverse the situation in favor of the immune system and the patient</cite>.

<cite index="7-5,7-6,7-7">Researchers began to see tumors less as isolated clumps of malignant cells and more as ecosystems; the tumor microenvironment, composed of immune cells, fibroblasts, blood vessels, signaling molecules, and the extracellular matrix, is intricately involved in whether a cancer grows, spreads, or responds to therapy; new work clarified how cancer rewires metabolism and how it negotiates with, and sometimes manipulates, the immune system</cite>. This work demanded the 2011 update.

<cite index="7-11,7-12">The 2011 update did more than summarize other findings; it validated emerging areas such as immunotherapy and cancer metabolism research and reinforced the hallmarks as the field's dominant framework for organizing cancer biology</cite>. Not all cancers may use these mechanisms uniformly—<cite index="18-12">Hanahan and Weinberg remained unsure whether the two emerging hallmarks are pervasive in all cancers</cite>—but the therapeutic impact has been substantial. The framework provided the intellectual scaffolding that allowed immune checkpoint blockade and metabolic inhibitors to be understood as rational, mechanism-based interventions rather than empirical gambles.

Sources:

• https://pubmed.ncbi.nlm.nih.gov/21267721/?dopt=Abstract
• https://www.abcam.com/en-us/stories/articles/hallmarks-of-cancer-reshaped-the-field-of-cancer-research
• https://wi.mit.edu/news/whitehead-scientist-helps-revisit-hallmarks-cancer

The hallmarks framework is not just conceptual architecture. <cite index="25-1,25-2">Understanding these hallmarks has fundamentally shaped modern oncology by revealing the mechanisms that drive cancer progression and has paved the way for precision medicine—designing therapies that specifically target aberrant pathways and immune escape mechanisms used by cancer cells</cite>. <cite index="21-1">The framework helped link molecular mechanisms to concrete tumor functions, supporting the logic behind targeted therapies targeting pathways involved in growth and blood vessel formation, such as EGFR and VEGF</cite>.

<cite index="20-1,20-4">Angiogenesis, one of the six original cancer cell hallmarks, evolved into a central therapeutic target; genomic instability, long viewed as an enabler of tumor evolution, became a therapeutic vulnerability exploited by conventional chemotherapy, radiation, and DNA damage response-directed therapies like PARP inhibitors</cite>. <cite index="24-7,24-8">Investigational drugs are being developed to target each of the enabling characteristics and emerging hallmarks; there is a deep pipeline of candidate drugs with different molecular targets and modes of action in development for most of these hallmarks</cite>.

The field continues to refine the original schema. <cite index="14-9">The 2011 update integrated genomics, immunity, and the tumor microenvironment; the 2022 update emphasized plasticity, senescence, and the microbiome as crucial modifiers of cancer behavior</cite>. <cite index="25-4">These therapies represent a paradigm shift—from a "one-size-fits-all" approach to personalized, mechanism-based treatments guided by the molecular and immunological hallmarks of each individual's cancer</cite>.

Sources:

• https://www.bangkokhospital.com/en/phuket/content/the-hallmarks-of-cancer-and-the-rationale-for-targeted-therapy-and-immunotherapy-bpk
• https://www.abcam.com/en-us/stories/articles/hallmarks-of-cancer-reshaped-the-field-of-cancer-research
• https://www.cell.com/cell/fulltext/S0092-8674(26)00334-X
• https://www.researchgate.net/figure/Therapeutic-targeting-of-the-hallmarks-of-cancer-drugs-that-interfere-with-each-of-the_fig5_340075670

<cite index="2-1,8-2">Douglas Hanahan and Robert Weinberg published "The Hallmarks of Cancer" in the January 2000 issue of Cell (100:57-70)</cite>, and <cite index="18-9">it became the most-cited article ever to appear in that journal</cite>. <cite index="3-3,3-4">The pair identified six capabilities acquired by incipient tumors: sustaining proliferative signaling, evading growth suppression, resisting cell death, enabling replicative immortality, activating invasion and metastasis, and inducing angiogenesis</cite>. <cite index="3-5,3-6">While each phenomenon was already known, no one had arranged them so coherently, and their synthesis served as an unparalleled conceptual framework</cite>.

<cite index="13-1">In 2011, Weinberg and Hanahan published an update proposing two new hallmarks—abnormal metabolic pathways and evasion of the immune system—and two enabling characteristics: genome instability and inflammation</cite>. <cite index="18-1">The authors refined these hallmarks using information from transgenic animals and biochemical assays that did not exist a decade prior</cite>. <cite index="18-12">The two emerging hallmarks (reprogramming of energy metabolism and immune evasion) were not integrated into the canonical six because Hanahan and Weinberg remained unsure whether they are pervasive in all cancers</cite>.

The framework matters because it does not simply catalog observations. <cite index="4-2,4-3">The organizing principle provides a logical framework for understanding neoplastic diversity, with the notion that normal cells evolve progressively to a neoplastic state by acquiring a succession of hallmark capabilities</cite>. This matters in the lab and the clinic: <cite index="26-2,26-3">knowledge of these functional molecular and biological traits has led to new therapeutic approaches, and most cancer drugs are deliberately developed for specific molecular targets that involve these hallmarks</cite>.

Sources:

• https://www.cell.com/fulltext/S0092-8674(11)00127-9
• https://en.wikipedia.org/wiki/The_Hallmarks_of_Cancer
• https://www.ludwigcancerresearch.org/success-story/modeler-of-malignancy/
• https://wi.mit.edu/news/whitehead-scientist-helps-revisit-hallmarks-cancer
• https://pubmed.ncbi.nlm.nih.gov/24470139/

<cite index="24-2,24-8">Psychiatry had the lowest overall likelihood of approval at 6.7%, followed by oncology at 8.9%</cite>, according to aggregate analyses across multiple datasets. The reasons differ by area. <cite index="21-4,21-15">CNS drugs, including treatments for Alzheimer's and Parkinson's, have one of the lowest success rates due to the complexity of the brain</cite>. <cite index="27-8">Attrition is higher in the central nervous system area compared to other therapy areas, partly because the need to cross the blood-brain barrier presents an additional challenge</cite>. For oncology, the issue is not penetration but biology. <cite index="25-1">The approval success rate of compounds classified as antineoplastic and immunomodulating agents was low, in agreement with multiple studies</cite>. The heterogeneity of tumor biology, the lack of predictive preclinical models, and the difficulty of achieving durable responses in heavily pretreated populations all contribute. <cite index="19-8,19-9">Neurology, cardiovascular, and oncology drugs had significantly lower success rates, from 7-9%, with oncology drugs in particular shown to have a very low probability of success</cite>. These are not temporary problems. They reflect the limits of current pharmacology when applied to the most complex human diseases.

Sources:

• https://alacrita.com/wp-content/uploads/2018/12/Pharmaceutical-Probability-of-Success.pdf
• https://patentpc.com/blog/clinical-trial-success-rates-how-many-drugs-make-it-to-market-latest-approval-stats
• https://www.sciencedirect.com/science/article/abs/pii/S0065774310450241
• https://ascpt.onlinelibrary.wiley.com/doi/full/10.1111/cts.12980
• https://www.knowledgeportalia.org/r-d-time-and-success-rate

The highest rate of attrition does not occur in Phase III, where the industry spends the most capital. It occurs in Phase II. <cite index="14-1,14-4">The main cause of attrition remains the Phase II hurdle, which just 28% of programs successfully complete, while Phase I (47%) and Phase III (55%) are both close to a one-in-two rate</cite>. The pattern has been stable for years. <cite index="11-9,11-10">Much of the attrition occurs in Phase II: many candidate drugs that show safety in early trials fail to demonstrate adequate efficacy or acceptable safety in patients</cite>. This is where proof-of-concept is tested and where the distance between preclinical models and human disease becomes undeniable. <cite index="12-7,12-8">In the past, a 20% product failure rate was seen in late stages of Phase III trials; currently, the failure ratio at this stage is 50%</cite>, according to FDA Commissioner Lester Crawford's 2004 remarks. That suggests the problem is not confined to Phase II but extends through late-stage development. The economic consequence is severe: molecules that fail in Phase II have already consumed years of development time and tens of millions in capital, yet they fail before generating the data that might salvage the program or inform future efforts.

Sources:

• https://www.norstella.com/why-clinical-development-success-rates-falling/
• https://intuitionlabs.ai/articles/four-phases-clinical-trials
• https://www.gastrojournal.org/article/S0016-5085(04)01567-7/fulltext

<cite index="26-1,26-5">Kola and Landis documented a shift in the primary reason for drug attrition from inappropriate pharmacokinetics and low bioavailability—approximately 40% in 1991—to lack of efficacy and safety, which accounted for approximately 60% by 2000</cite>. The pharmaceutical industry had successfully addressed ADME liabilities through better screening and optimization. <cite index="27-3,27-4">Since the 1990s, the major reason for overall compound-related attrition shifted from DMPK to toxicity, most likely as a consequence of investments in screening and optimization of DMPK parameters, now standard practice from lead generation onwards</cite>. More recent data confirm the persistence of this pattern. <cite index="31-1,31-2">During the past decade, the largest causes of attrition have been lack of efficacy and safety issues, with each accounting for 30% of attrition</cite>. The implication: we learned to make molecules with acceptable PK profiles, but we have not learned to reliably pick the right targets or predict human efficacy from preclinical models. <cite index="31-3">High attrition rates since 2000 are often attributed to companies addressing more complex diseases with high unmet medical need and establishing higher standards for success in clinical trials</cite>. That is both an explanation and an indictment.

Sources:

• https://accp1.onlinelibrary.wiley.com/doi/full/10.1002/cpdd.1464
• https://www.sciencedirect.com/science/article/abs/pii/S0065774310450241
• https://www.genengnews.com/insights/overcoming-phase-ii-attrition-problem/

<cite index="26-2">The Kola and Landis analysis of ten major U.S. and European pharmaceutical companies found an average success rate of 11% from first-in-human trials to registration between 1991 and 2000</cite>. That figure represents the foundational benchmark against which the industry has measured itself for two decades. The analysis matters because it stratified attrition by therapeutic area and revealed variation that still holds: <cite index="26-2">oncology drugs had a success rate of approximately 5%</cite>—in other words, one in twenty molecules entering clinical testing. More recent data suggest the problem has worsened in some areas. <cite index="22-5,22-6">In a larger sample spanning 2000 to 2015, oncology showed a 3.4% success rate overall, declining to 1.7% in 2012 before improving to 8.3% in 2015</cite>. <cite index="19-1">Cardiovascular candidates had the highest rate of success at 20%, followed by arthritis and pain at 17% and infectious diseases at 16%, whereas central nervous system, oncology, and women's health had the lowest</cite>. The variation is not noise. It reflects the relative maturity of pharmacology, the availability of validated biomarkers, and the complexity of the disease biology being targeted.

Sources:

• https://www.nature.com/articles/nrd1470
• https://accp1.onlinelibrary.wiley.com/doi/full/10.1002/cpdd.1464
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6409418/
• https://www.knowledgeportalia.org/r-d-time-and-success-rate

The Mullard series documents shifts in therapeutic class composition across approval cohorts. <cite index="28-3,28-4,28-5">The 27 small-molecule drugs approved in 2024 demonstrated a range of pharmaceutical activities, predominantly as anti-cancer agents, genetic drugs, immunological drugs, and anti-infective drugs, with other therapeutic areas including endocrine/metabolic drugs, cardiovascular/cerebrovascular drugs, hematological drugs, respiratory drugs, and psychotropic drugs</cite>. The annual reviews include not just approval counts but also commercial expectations. <cite index="22-2,22-3">The commercial expectations for the cohort of Q3 2024 approvals were much greater than those from the first two quarters of the year, with the Q3 cohort anticipated to achieve sales of approximately $11.9 billion in 2030 according to Evaluate Pharma consensus forecasts, compared with $6.3 billion and $6.9 billion for the Q1 and Q2 cohorts, respectively</cite>.

These forecasts are included in updated versions of the annual reviews. <cite index="11-15">The 2022 review was updated in January 2023 to add information on the financial performance of the new drugs approved and the characteristics of the reviews by the FDA</cite>. <cite index="20-6">Similarly, the 2024 review was updated in January 2025 to add information on the financial performance of the new drugs approved and the characteristics of the reviews by the FDA</cite>.

The series tracks not only what is approved but also where the market believes value will accumulate, providing a longitudinal view of both regulatory productivity and commercial expectations.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12114780/
• https://www.nature.com/articles/d41573-024-00166-5
• https://www.nature.com/articles/d41573-023-00001-3
• https://www.nature.com/articles/d41573-025-00001-5

The Mullard series tracks not just approval counts but also the characteristics of approved drugs, including first-in-class designations and orphan drug status. <cite index="28-6,28-7">Among the approved small-molecule drugs in 2024, those distinguished by unique mechanisms of action and those designated as breakthrough therapies by the FDA constituted a significant proportion, with eight drugs—Rezdiffra, Voydeya, Iqirvo, Voranigo, Livdelzi, Miplyffa, Revuforj, and Crenessity—classified as first-in-class and receiving breakthrough therapy designation</cite>. <cite index="29-1">Amongst the 16 biologics approved in 2024, nine were first-in-class: zanidatamab, sotatercept, zolbetuximab, axatilimab, nemolizumab, tarlatamab, marstacimab, zenocutuzumab, and nogapendekin alfa inbakicept</cite>.

Orphan drug designations are a substantial fraction of the approval landscape. <cite index="29-7">Notably, 52% (26) of the 50 drugs approved in 2024 were indicated to treat rare or orphan diseases</cite>. <cite index="29-9">In 2024, 26 of the 50 drug approvals received Orphan Drug Designation, and 33 used one or more expedited programs</cite>.

These designations are not just regulatory labels; they signal where innovation is concentrating and where the FDA is deploying expedited pathways. The Mullard series provides the primary data source for tracking these shifts year over year.

Sources:

• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12114780/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12383571/

The Mullard series documents substantial year-over-year variance in approval counts, raising questions about how to read signal from noise. <cite index="11-5,11-6,11-7">In 2022, the FDA approved 37 novel drugs, the fewest to pass regulatory scrutiny since 2016, a drop from the highs of the past 5 years that brought the rolling 5-year average down to 49 drugs per year</cite>. The 2022 nadir was followed by a sharp rebound. <cite index="13-1">The FDA approved 55 novel therapeutics in 2023, the second highest count in the past 30 years</cite>. <cite index="20-2,20-3">The class of 2024 included 50 new therapeutics, a little smaller than that of 2023, which had 55 approvals</cite>.

Rolling averages matter here because single-year counts are volatile. <cite index="13-9,13-10">The ten-year rolling average for new CDER approvals in 2023 stood at 46 per year, the highest it had been in over 20 years, with the nadir being 2010 when this average bottomed out at 25 per year</cite>. <cite index="14-1,14-2,14-3">By 2025, when the FDA approved 46 new drugs, the 5-year average had come down to 48 new drugs per year, still well above the historic average of 36 new drugs per year since 1993</cite>.

These are not trivial fluctuations. The difference between a 37-drug year and a 55-drug year is nearly 50%. The series tracks both the year-specific counts and the multi-year averages, offering a more stable measure of approval productivity than headline numbers alone.

Sources:

• https://www.nature.com/articles/d41573-023-00001-3
• https://www.nature.com/articles/d41573-024-00001-x
• https://www.nature.com/articles/d41573-025-00001-5
• https://www.nature.com/articles/d41573-026-00001-z

<cite index="2-3,2-4,2-5,2-6,2-7">Asher Mullard has published an annual FDA drug approval analysis in Nature Reviews Drug Discovery for each year starting with 2011</cite>, cataloging novel therapeutics approved by the FDA's Center for Drug Evaluation and Research. The series covers both small-molecule New Drug Applications and biologic Biologics License Applications. <cite index="2-37,2-38">Researchers mining and compiling data from these annual reviews have used them as the canonical source for approvals issued between 2011 and 2020, focusing on all new drugs approved by CDER including biologics</cite>.

The series provides more than simple counts. <cite index="11-7,11-8,11-9">The 2022 review noted that 37 novel drugs were approved—a drop from the highs of the previous five years, bringing the rolling 5-year average down to 49 drugs per year, but still above the historic average since 1993 of 34 drugs per year</cite>. <cite index="13-6,13-7,13-8">By 2023, CDER approved 55 new drugs, a cohort nearly 50% larger than 2022's and bringing the ten-year rolling average to 46 per year, the highest it had been in over 20 years</cite>. <cite index="20-2,20-3,20-4">In 2024, CDER approved 50 new small molecules, biologics and oligonucleotide therapeutics, slightly smaller than 2023's class but pushing the ten-year rolling average to 46.5 novel approvals per year, a new high-water mark</cite>.

These are not press-release summaries. The analyses track longitudinal trends with precision and place each year's approvals in context against multi-decade baselines.

Sources:

• https://www.nature.com/articles/d41573-023-00001-3
• https://www.nature.com/articles/d41573-024-00001-x
• https://www.nature.com/articles/d41573-025-00001-5
• https://media.nature.com/original/magazine-assets/d41573-022-00213-z/23814366

<cite index="7-2,7-3">McNamee et al. (2017) suggest it takes on average approximately 29 years from initial publications around a target to running clinical trials, and 36 years for drug approval.</cite> <cite index="2-9,2-10,2-11">Developing a new drug from original idea to launch is a complex process taking 12–15 years and costing in excess of $1 billion, with target selection often requiring many years to build up supporting evidence before initiating a costly program.</cite> <cite index="23-22,26-5">The review examines key preclinical stages from initial target identification and validation, through assay development, high throughput screening, hit identification, lead optimization, and finally selection of a candidate molecule for clinical development.</cite>

<cite index="7-8,7-9">Drug discovery is a challenging process with a low probability for any individual project to result in successful development of a useful novel therapeutic—very few areas of innovation have such a high failure rate from idea concept through to marketed product.</cite> <cite index="7-10,7-11,7-12">A net reduction of failure rate at all stages is predicted to reduce costs significantly; the overall process comprises target identification with a data-supported disease hypothesis, development and design optimization of approaches to modulate the target, and development of candidate molecules for clinical trials, with each program developing decision-making cascades to focus on the optimal therapeutic entity.</cite> The timeline is not an accident—it reflects the iterative rejection of hypotheses that looked plausible on paper.

Sources:

• https://pharmaceuticalmanufacturer.media/pharmaceutical-industry-insights/latest-pharmaceutical-manufacturing-industry-insights/the-evolution-of-target-validation-in-drug-discovery/
• https://bpspubs.onlinelibrary.wiley.com/doi/10.1111/j.1476-5381.2010.01127.x
• https://www.unboundmedicine.com/medline/citation/21091654/Principles_of_early_drug_discovery_

<cite index="7-1">According to a Bio Report from February 2021, of 100 assets entering Phase I clinical studies, only eight will reach approval stages.</cite> <cite index="8-1">Success rates for new development projects in Phase II trials have fallen from 28% to 18% in recent years, with insufficient efficacy being the most frequent reason for failure.</cite> <cite index="13-8">Attrition in drug development is very high in Phase II studies, with an approximate rate of 66 percent.</cite> <cite index="11-7,11-8,11-9">The drug discovery pipeline typically requires around 10 years and $2 billion to bring a novel drug to market; by 2022 fewer than 500 successful drug targets had been identified, and the average failure rate in clinical trials from 2009 to 2018 reached 84.6%.</cite>

These are not rounding errors. <cite index="34-10">The primary reason for attrition in clinical trials is lack of efficacy, which has been associated with poor or biased target selection.</cite> The translation failure reflects upstream decisions—target selection made on incomplete disease models, validation performed in systems that do not recapitulate human pathophysiology, hypotheses that fracture under the variability of real patient populations. The confidence interval on target-to-clinic success is wide, and most programs fall outside it.

Sources:

• https://pharmaceuticalmanufacturer.media/pharmaceutical-industry-insights/latest-pharmaceutical-manufacturing-industry-insights/the-evolution-of-target-validation-in-drug-discovery/
• https://cambridgemedchemconsulting.com/target-validation/
• https://www.ncbi.nlm.nih.gov/books/NBK195039/
• https://www.cell.com/trends/pharmacological-sciences/fulltext/S0165-6147(23)00137-2
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5679337/

<cite index="4-11">Target-based drug discovery begins with identifying the function of a possible therapeutic target and its role in disease.</cite> <cite index="6-3,6-4,6-5">Target identification assesses a potential biological target such as a gene, protein, or pathway; validation experiments determine whether modulating that target delivers a biological effect consistent with the therapeutic hypothesis.</cite> <cite index="13-2,13-3">Target validation is a multilayered step critical to drug success—validated targets and biomarkers, along with assessment tools, are needed to ensure the drug actively engages the target to produce an expected therapeutic effect.</cite>

Genetic validation—knockout studies, RNA interference, GWAS—remains a cornerstone, but the standard is incomplete. <cite index="12-7">Despite genome sequencing and substantial research, discovery is hampered by the lack of well-validated druggable targets, as many essential genes are not druggable.</cite> <cite index="17-3,17-4,17-5">Success depends on aiming for "the right target," which requires adequate understanding of disease biology; insufficient knowledge or failure at this stage results in poor target choice and incorrect biomarker identification, hampering patient selection in clinical trials.</cite> Validation confirms necessity but not sufficiency—a target can be disease-linked and genetically essential yet remain inaccessible to small molecules or have an intolerable safety margin.

Sources:

• https://www.ncbi.nlm.nih.gov/books/NBK92015/
• https://md.catapult.org.uk/drug-discovery/drug-discovery-process/target-identification/
• https://www.ncbi.nlm.nih.gov/books/NBK195039/
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3182078/
• https://www.excelra.com/excelra-blogs/improving-target-selection-validation-role-of-qsp/

<cite index="2-9,2-10,2-11">Drug development from idea to launch takes 12–15 years and exceeds $1 billion, with target selection often requiring years of supporting evidence before initiating a costly discovery program.</cite> <cite index="14-6,14-16">Four criteria are proposed for target evaluation: linkage to disease, potential therapeutic index, chemical tractability, and economics.</cite> <cite index="14-9,14-10">Target selection is characterized as a process rather than an event—targets should be reevaluated continuously as internal and external data accumulate to determine whether continued investment is appropriate.</cite>

This reframing matters because it acknowledges what happens after the kick-off meeting. <cite index="12-4,12-5,12-6">Only 1 in 5 projects survives preclinical development, and less than 1 in 10 entering clinical development reach registration, with most failures driven by flawed biology—targets revealed as non-essential or encountering lack of efficacy in trials, toxicity, and DMPK issues.</cite> <cite index="6-1,6-2">Early-stage validation strengthens confidence and reduces clinical risk; many costly failures are linked to insufficient validation at this stage.</cite> The framework is less a checklist than a commitment to iterative skepticism—most projects fail, and the question is whether your data warrant continuing.

Sources:

• https://bpspubs.onlinelibrary.wiley.com/doi/10.1111/j.1476-5381.2010.01127.x
• https://onlinelibrary.wiley.com/doi/abs/10.1002/9780470571224.pse419
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3182078/
• https://md.catapult.org.uk/drug-discovery/drug-discovery-process/target-identification/

Scannell's later work emphasized predictive validity—the ability of preclinical screens to forecast clinical success—as the bottleneck that high-throughput technologies could not solve. <cite index="5-2">The core problem is the collapse of predictive validity in preclinical models, which sits at the heart of pharma's productivity paradox.</cite> <cite index="17-11">Around 90% of drug candidates that show efficacy in preclinical studies fail to produce benefit in clinical trials.</cite>

This is distinct from the "low-hanging fruit" hypothesis, which Scannell explicitly downplayed. <cite index="8-10,8-11,8-12">Two broad classes of mutually non-exclusive explanation are key: the exhaustion of opportunities for pharmaceutical innovation—including as-yet-untreated diseases, unexploited biological mechanisms or unexplored regions of chemical space—and the gradual abandonment of more productive methods of R&D in favour of less productive ones.</cite> The latter involves historical shifts from phenotypic, in vivo screens toward target-based, in vitro assays with poor human translation. Scannell has argued that brute-force screening in low-validity models explains much of the paradox: more throughput, worse predictions, rising costs. The analysis has accumulated over 1,600 citations and remains the canonical framing for industry efficiency debates.

Sources:

• https://www.drugtargetreview.com/article/189503/the-predictive-validity-crisis-pharmas-productivity-paradox-part-i/
• https://www.researchgate.net/publication/221873929_Diagnosing_the_Decline_in_Pharmaceutical_RD_Efficiency
• https://www.oecd.org/en/publications/artificial-intelligence-in-science_a8d820bd-en/full-report/eroom-s-law-and-the-decline-in-the-productivity-of-biopharmaceutical-r-d_f42df75c.html
• https://www.molecule.to/blog/dr-jack-scannell-pharmas-evolving-landscape

<cite index="31-1,31-9">The downward trend broke around 2010, with a modest uptick.</cite> <cite index="27-9">In 2020, Nature Reviews published a short follow-up to the 2012 article, detailing an uptick in drug productivity.</cite> The 2020 paper—Ringel, Scannell, Baedeker, and Schulze, "Breaking Eroom's Law," Nat Rev Drug Discov 19(12):833-4—documented the reversal but wrapped it in qualifications.

<cite index="31-2,31-10">The uptick in drug approvals is, however, associated with a decline in the number of eligible patients per new drug.</cite> <cite index="27-11,27-12">More diseases are being genetically segmented, cancer types for example, meaning that while more drugs may be being approved, they are applicable to fewer and fewer people; rare diseases have also become attractive to pharmaceutical companies as the route to drug approval can be much shorter and/or easier.</cite> This is not full reversal—it is stratification. Approvals rose, but the addressable populations per approval shrank. Whether this represents genuine efficiency gain or regulatory arbitrage remains an open question, and Scannell's own analysis does not offer unqualified optimism. The follow-up was published April 2020, eight years after the original diagnosis.

Sources:

• https://www.oecd.org/en/publications/artificial-intelligence-in-science_a8d820bd-en/full-report/eroom-s-law-and-the-decline-in-the-productivity-of-biopharmaceutical-r-d_f42df75c.html
• https://www.molecule.to/blog/dr-jack-scannell-pharmas-evolving-landscape
• https://www.nature.com/articles/d41573-020-00059-3
• https://refoundable.com/research/life-after-erooms-law-interview-with-jack-scannell.html

<cite index="18-1,18-6">Scannell and colleagues proposed four factors as primary causes, which they call the 'better than the Beatles' problem; the 'cautious regulator' problem; the 'throw money at it' tendency; and the 'basic research-brute force' bias.</cite>

The "better than the Beatles" problem refers to rising comparative effectiveness bars. <cite index="25-8,25-9">Imagine how hard it would be to achieve commercial success with new pop songs if any new song had to be better than the Beatles, if the entire Beatles catalogue was available for free, and if people did not get bored with old Beatles records; something similar applies to the discovery and development of new drugs.</cite> Every new drug competes not just with current therapies but with decades of accumulated generics.

The "cautious regulator" problem describes regulatory tightening without corresponding loosening. <cite index="15-3,15-4">Drug regulation has tightened over time, and regulation does not loosen even if it's not necessary; an example given is the Ames test for mutagenicity, which is probably not that useful but is still a regulatory requirement.</cite> The "throw money at it" tendency and the "basic research-brute force" bias both reflect strategic misdirection—overestimating what genomics and high-throughput screening could deliver when predictive validity of disease models remained poor.

Sources:

• https://scispace.com/papers/diagnosing-the-decline-in-pharmaceutical-r-d-efficiency-3lx55l7125
• https://pubmed.ncbi.nlm.nih.gov/22378269/
• https://go.gale.com/ps/i.do?p=AONE&u=googlescholar&id=GALE%7CA283024171&v=2.1&it=r&asid=61c739a1
• https://douglasyao.github.io/notes/drug/2021/01/05/Scannell-et-al-2012-Nature-Rev-Drug-Discov-Diagnosing-the-decline-in-pharmaceutical-R&D-efficiency.html

<cite index="12-11,12-14">The number of new drugs approved per billion US dollars spent on R&D has halved roughly every 9 years since 1950, falling around 80-fold in inflation-adjusted terms.</cite> <cite index="1-3">The term, coined in 2012 by analysts Jack W. Scannell, Alex Blanckley, Helen Boldon, and Brian Warrington, deliberately reverses "Moore's law"—Gordon Moore's 1965 prediction of exponential efficiency gains in semiconductor technology—to underscore the stark divergence between computing's accelerating returns and pharmaceuticals' compounding inefficiencies.</cite>

The original analysis tracked data from the 1950s through 2010. <cite index="7-7">By 2010, the total R&D spend per drug approved was about a hundred times higher, in real dollars, than it was in 1950.</cite> This is not a story about stagnant output—approvals continued, companies remained profitable for decades. <cite index="8-3,8-4,8-5">Profit growth offset rising R&D costs, but could not keep pace indefinitely; in the early 1960s, the industry's net income was roughly twice its spending on R&D, while today, for the industry as a whole, aggregate R&D spending is higher than net income.</cite> The efficiency collapse occurred despite—perhaps because of—massive expansions in screening libraries, genomic tools, and computational chemistry platforms. The foundational paper appeared in Nature Reviews Drug Discovery in March 2012 (vol. 11, pp. 191-200, doi: 10.1038/nrd3681).

Sources:

• https://www.nature.com/articles/nrd3681
• https://pubmed.ncbi.nlm.nih.gov/22378269/
• https://grokipedia.com/page/Eroom%27s_law
• https://refoundable.com/research/life-after-erooms-law-interview-with-jack-scannell.html
• https://www.oecd.org/en/publications/artificial-intelligence-in-science_a8d820bd-en/full-report/eroom-s-law-and-the-decline-in-the-productivity-of-biopharmaceutical-r-d_f42df75c.html

<cite index="9-5,9-6">The FDA introduced the Critical Path Initiative in 2004 with the intent of modernizing drug development by incorporating recent scientific advances such as genomics and advanced imaging technologies, with an important part of the initiative being the use of public-private partnerships and consortia to accomplish the needed research</cite>.

This was a structural concession: no single entity could solve the problem. <cite index="12-7,12-8">The report called for a national infrastructure to support and continually improve the Critical Path sciences and new ways to collaborate and share data to accomplish common goals, noting that no single company, university, or government agency would be successful with these tasks and that collaboration would be essential</cite>.

<cite index="8-3,14-3">Many new drug development projects worldwide took cues from the CPI, adopting microdosing, adaptive designs, and taking advantage of newly developed biomarkers under the initiative</cite>. The white paper didn't just name the crisis—it seeded a coordinating mechanism for precompetitive work. The Critical Path Institute, established as one outcome, became a venue for companies to share data on tools and standards without competing on therapies themselves.

Sources:

• https://pubmed.ncbi.nlm.nih.gov/18186700/
• https://www.healio.com/news/cardiology/20120225/the-future-of-drug-development-and-the-fda-s-critical-path-initiative
• https://pmc.ncbi.nlm.nih.gov/articles/PMC2996064/

<cite index="1-3,23-3">The white paper identified "an urgent need to modernize the product development toolkit, to make the development process more predictable and less costly"</cite>. The FDA's diagnosis was specific: <cite index="20-5,24-5">despite technological advances, the drug development community was still using the last century's methods to develop and test new drugs, biological therapies, and medical devices</cite>.

This framing—"toolkit"—was deliberate. <cite index="6-4,28-4">The publication diagnosed the widening gap between scientific discoveries that had unlocked the potential to prevent and cure diseases like diabetes, cancer, and Alzheimer's, and their translation into innovative medical treatments</cite>. The problem wasn't a lack of targets or mechanisms; it was a lack of validated tools to evaluate them during development.

<cite index="1-2,23-2">FDA established a public docket to obtain input on activities that could reduce existing hurdles in medical product design and development</cite>, seeking input on specific development bottlenecks and solutions. <cite index="8-8,14-8">In 2006, the FDA followed with a Critical Path Opportunities List calling for better evaluation tools, streamlining clinical trials, and developing approaches to address urgent public health needs</cite>. The white paper spawned an infrastructure play, not just a policy statement.

Sources:

• https://www.federalregister.gov/documents/2004/04/22/04-9147/critical-path-initiative-establishment-of-docket
• https://www.ncbi.nlm.nih.gov/books/NBK52928/
• https://www.fda.gov/science-research/science-and-research-special-topics/critical-path-initiative
• https://pmc.ncbi.nlm.nih.gov/articles/PMC2996064/

The Critical Path white paper didn't just document stagnation at the submission stage—it pointed to where candidates were dying. <cite index="2-11">In the decade leading up to the report, attrition was highest during Phase II, at 62%</cite>. This is the stage where proof-of-concept is tested in humans, and where the industry was burning the most capital on failures.

<cite index="10-6,14-1">Rising cost and increased attrition rates created barriers to investment in higher-risk drugs or therapies for uncommon diseases or diseases predominantly afflicting the poor</cite>. The white paper identified this as a market failure with public health consequences: the tools to predict efficacy and safety early in development were not keeping pace with the complexity of the science.

<cite index="25-4">A 250% increase in R&D expenditures over the prior decade was associated with a 50% decline in new product submissions to the FDA</cite>. <cite index="25-5">The report also flagged the increasing and unacceptable failure rate in clinical development, especially in Phases II and III</cite>. The framing mattered: this wasn't about regulatory risk, it was about predictive validity—the gap between what we could measure in the lab and what would work in the clinic.

Sources:

• https://pmc.ncbi.nlm.nih.gov/articles/PMC2996064/
• https://www.healio.com/news/cardiology/20120225/the-future-of-drug-development-and-the-fda-s-critical-path-initiative

<cite index="1-7,6-3">The FDA released Innovation/Stagnation: Challenge and Opportunity on the Critical Path to New Medical Products in March 2004</cite>, and the numbers were brutal. <cite index="10-4">Despite a 70% increase in R&D investment from 1994 to 2004, the number of new chemical entities launched fell 40%</cite>. <cite index="10-1">The cost to bring a new drug to market rose 55%, from $1.1 billion to $1.7 billion, between two consecutive periods: 1995–2000 and 2000–2002</cite>.

The document diagnosed what it called <cite index="1-8">a "recent slowdown in new medical products submitted for approval to FDA"</cite> and blamed an obsolescence problem: <cite index="10-11,10-12">the prevalent development path was becoming increasingly challenging, inefficient, and costly, in part because developers were "forced to use the tools of the last century to evaluate this century's advances"</cite>. The FDA was careful to note this wasn't a regulatory bottleneck—the problem was upstream, in the development toolkit itself.

<cite index="20-6,24-6">A drug entering Phase I clinical development in 1985 was more likely to reach the market than one entering in 2000</cite>. The white paper was the first time the agency publicly acknowledged that rising investment was not solving the attrition problem—it was compounding it.

Sources:

• https://www.federalregister.gov/documents/2004/04/22/04-9147/critical-path-initiative-establishment-of-docket
• https://pmc.ncbi.nlm.nih.gov/articles/PMC2996064/
• https://www.ncbi.nlm.nih.gov/books/NBK52928/
• https://www.fda.gov/science-research/science-and-research-special-topics/critical-path-initiative

<cite index="16-5,16-8,16-9">In the 2003 study, the research and development costs of 68 randomly selected new drugs were obtained from a survey of 10 pharmaceutical firms, yielding an estimated average out-of-pocket cost per new drug of $403 million (2000 dollars), and capitalizing out-of-pocket costs to the point of marketing approval at a real discount rate of 11% yielded a total pre-approval cost estimate of $802 million (2000 dollars).</cite> <cite index="17-5,17-12">DiMasi and colleagues randomly selected 68 new drugs in development that were first tested in humans anywhere in the world between 1983 and 1994.</cite>

<cite index="17-2,17-3">The team's previous analysis in 1991 gave a figure of $231 million (in 1987 dollars), which became the benchmark figure for the industry throughout the 1990s, and the $802 million figure was expected to persist for years to come.</cite> <cite index="12-7,12-8">The methodological approach used in the 2016 paper follows that used for previous studies, and because the methodologies are consistent, comparisons of the results can be made to examine and illustrate trends in development costs.</cite> <cite index="4-12">A previous Tufts CSDD study, released in 2011, pegged the cost of development and approval for new drugs at $1.3 billion.</cite> The arc runs from $231 million to $802 million to $1.3 billion to $2.6 billion over three decades — each jump contested, each figure a fixture in industry discourse until the next revision.

Sources:

• https://pubmed.ncbi.nlm.nih.gov/12606142/
• https://www.nature.com/articles/nrd1070
• https://www.sciencedirect.com/science/article/abs/pii/S0167629616000291
• https://www.genengnews.com/news/tufts-study-pegs-drug-development-approval-cost-at-2-6b/

<cite index="9-3,9-4">Critics characterized the study as part of a public relations campaign by drug companies to justify high prices, arguing that the drug companies funding CSDD are hoping people will simply quote the number for several years until a new one is needed.</cite> <cite index="6-10,6-11">Critics of the study say the $2.6 billion price tag is exaggerated and could help pharmaceutical companies justify high prices for consumers, with the Union for Affordable Cancer Treatment calling on the authors to provide more details about how they determined that sum.</cite>

<cite index="9-19,9-20">DiMasi's "average out-of-pocket cost of $1,395 million" represents an estimate of the risk-adjusted outlays on drug development, including the actual out-of-pocket costs claimed for his secret sample of drugs, adjusted for risk, then automatically adding 44.5 percent of the risk-adjusted number for pre-clinical expenses, plus time costs (expected returns that investors forego while a drug is in development) of $1,163 million.</cite> <cite index="21-5,21-6">The study and its use of an average cost is indicative of a much larger problem in the pharmaceutical industry as a whole, which operates behind firmly closed doors and refuses to make R&D costs public, and the pharma industry must be much more transparent on how much it costs to develop a drug.</cite>

<cite index="25-3,25-4,25-5,25-6">The study was designed to capture only the costs incurred by industry, and the full social cost would be the sum of the private costs and government and nonprofit funding for research that contributes to drug discovery and development — the latter element of social cost would be very difficult to quantify adequately, and the sample selection criteria do not exclude cases in which companies use information obtained from research funded by nonprofits or government to guide their own activities.</cite>

Sources:

• https://www.keionline.org/22646
• https://aacrjournals.org/cancerdiscovery/article/5/2/OF2/4765/Drug-Development-Costs-Jump-to-2-6-BillionDrug
• https://msfaccess.org/rd-cost-estimates-msf-response-tufts-csdd-study-cost-develop-new-drug
• https://www.nejm.org/doi/full/10.1056/NEJMc1504317

<cite index="1-6">Rising drug development costs have been driven mainly by increases in out-of-pocket costs for individual drugs and higher failure rates for drugs tested in humans.</cite> <cite index="1-8">Factors that likely boosted out-of-pocket clinical costs include increased clinical trial complexity, larger clinical trial sizes, higher cost of inputs from the medical sector used for development, greater focus on targeting chronic and degenerative diseases, changes in protocol design to include efforts to gather health technology assessment information, and testing on comparator drugs to accommodate payer demands for comparative effectiveness data.</cite>

<cite index="1-9,1-10,1-11">Lengthening development and approval times were not responsible for driving up development costs — changes in the overall time profile for development and regulatory approval phases had a modest moderating effect on the increase in R&D costs, and as a result, the time cost share of total cost declined from approximately 50% in previous studies to 45% for this study.</cite>

<cite index="10-6,10-7,10-8">CNS drugs not only require longer development time, but also have significantly lower success rates than drugs for other indications — clinical development time for CNS drugs approved in the United States between 1999 and 2013 was 12.8 months, or 18% longer than non-CNS targeted drugs, and CNS compounds garnered a success rate of only 6.2%, less than half of that of non-CNS drugs (13.3%).</cite>

Sources:

• https://www.appliedclinicaltrialsonline.com/view/tufts-center-study-drug-development-cost-developing-new-drugs
• https://pubs.acs.org/doi/10.1021/cn500298z

<cite index="1-2">The 2014 Tufts CSDD estimate — $2,558 million to develop and win marketing approval for a new prescription drug — comes from data on 106 randomly selected compounds first tested in humans from 1995 to 2007, provided by 10 pharmaceutical companies.</cite> <cite index="11-7,11-8">The figure breaks down to $1,395 million in out-of-pocket costs and $2,588 million after capitalizing those costs at a real discount rate of 10.5%.</cite>

<cite index="15-1,15-2,15-3">The study was authored by DiMasi, Grabowski, and Hansen and published in the Journal of Health Economics in 2016.</cite> <cite index="1-4">The estimate links the costs of unsuccessful projects to those that are successful in obtaining marketing approval.</cite> <cite index="4-6">The overall clinical approval success rate for new drugs — the likelihood that a Phase I drug will be approved for marketing — stood at less than 12%.</cite>

<cite index="1-3">The $2,558 million figure does not include an estimated $312 million in post-approval R&D — studies to test new indications, new formulations, new dosage strengths and regimens, and to monitor safety and long-term side effects required by FDA as a condition of approval — which boosts the full product lifecycle cost to $2,870 million.</cite> <cite index="1-5">Compared to the 2003 study ($802 million in 2000 dollars, equal to $1,044 million in 2013 dollars), the cost to develop and win marketing approval for a new drug increased by 145% between the two study periods, or at a compound annual growth rate of 8.5%.</cite>

Sources:

• https://www.appliedclinicaltrialsonline.com/view/tufts-center-study-drug-development-cost-developing-new-drugs
• https://www.researchgate.net/publication/294422802_Innovation_in_the_pharmaceutical_industry_New_estimates_of_RD_costs
• https://csdd.tufts.edu/cost-study
• https://www.genengnews.com/news/tufts-study-pegs-drug-development-approval-cost-at-2-6b/

<cite index="10-10">Lowe, known for his long-running industry commentary, reflected on the cycles of hype that have accompanied new technologies over the decades—from the early days of computer-aided drug design in the late '80s to sequencing to antisense to the current AI boom</cite>. The pattern repeats: a technology emerges, promises accelerate, capital flows, and then the field recalibrates.

The deeper issue is about what can be modeled versus what determines outcomes. <cite index="15-1,15-2">"The need to make decisions with sufficient quality is only compatible in some cases with the data we have at hand to reach this goal. If we want to advance drug discovery, then acknowledging the suitability of a given end point to answer a given question is at least as important as modelling a particular end point"</cite>. In other words: the model is not the constraint. The assay is.

<cite index="15-5,15-6">"The problem is, modeling is easier to start doing than dealing with that suitability question. It can also be harder to explain this point to investors, to granting agencies, and to upper management, because improvements in things like assay quality and target selection are harder to quantify and come on slowly"</cite>. This is Lowe writing about incentive structures—not chemistry. The technologies that are easiest to fund are not necessarily the ones that address the failure modes that matter most.

Sources:

• https://www.bio-itworld.com/news/2025/04/15/derek-lowe-on-ai-in-drug-discovery-between-hype-and-hope
• https://www.science.org/content/blog-post/ai-and-drug-discovery-attacking-right-problems

In a Chemistry World column, Lowe describes the lived experience of medicinal chemistry work with unusual clarity: <cite index="3-1,3-2,3-3">"medicinal chemists do not actually have limitless capacities for boredom. Running yet another long line of palladium coupling reactions or amine displacements begins, after a while, to feel like working at a sawmill. The blade takes longer to cut through some of the logs than others, but the boards all come out looking about the same"</cite>.

This is not a complaint about difficulty—it's a complaint about sameness. <cite index="3-4,3-5">"It's true that we might not want to spend much time at the opposite end of the excitement scale. Reactions that have to be watched every minute lest they explode are not the answer, nor are elaborate routes involving three-ring-circus rearrangements at the final step"</cite>. But between the extremes lies a problem: <cite index="3-6,3-7">"short of these heroics, there are less travelled chemistries that should get more use than they do. It would do us (and our compound collections) good if our molecules were a little less flat, a little less aromatic, and perhaps a little less easy to put together"</cite>.

The observation is structural. If the same reactions produce the same scaffolds, the chemical space being sampled is narrower than the methods allow. Lowe is arguing for deliberate inefficiency—routes that are harder to execute but populate regions of structure-activity space that palladium couplings will never reach.

Sources:

• https://www.chemistryworld.com/opinion/column-in-the-pipeline/3004981.article

<cite index="10-1,10-6">The central failure of drug development—an 85% failure rate in clinical trials—is not due to a lack of ambition, but a lack of predictive insight</cite>. Lowe is direct about the structural problem: <cite index="11-1">the two biggest factors in that failure rate are picking the wrong targets/mechanisms, and unexpected toxicity</cite>.

<cite index="10-7,10-8">Better target selection and human toxicity prediction, the two biggest hurdles, are not yet within AI's reach. "Those are the two things that kill most of the [drug discovery] programs," he said</cite>. The issue is not computational—it's biological. <cite index="10-4,10-5">AI/ML has shown promise in areas like protein structure prediction and antibody design, but its utility in more complex fields like small molecule discovery and clinical translation remains limited. Current models struggle because of the lack of consistent, high-quality datasets, especially when translating results from lab assays to human trials</cite>.

Lowe frames the challenge in terms of rate constants at a SPARK talk: <cite index="9-2,9-12">"We can improve perversely if we were just to go from nine failures out of ten" to eight out of 10</cite>. That's a 10% absolute reduction in failure—and it counts as meaningful progress. <cite index="9-10,24-10">Drug discovery's problems are intellectual, he said, and should therefore have intellectual solutions</cite>. But that doesn't mean the solutions are algorithmic.

Sources:

• https://sparkmed.stanford.edu/blog/drug-discovery-and-its-discontents/
• https://www.bio-itworld.com/news/2025/04/15/derek-lowe-on-ai-in-drug-discovery-between-hype-and-hope
• http://pibmub.com/index-50.html

<cite index="1-2,12-2">Derek Lowe has published his blog "In the Pipeline" since 2002</cite>, making him <cite index="12-4">one of the first people to blog from inside the pharmaceutical industry</cite>. The blog represents an unusual phenomenon: <cite index="14-2,14-10">a quiet field where most people go through their entire careers without discovering a marketable drug</cite>, narrated in real time by a working medicinal chemist.

<cite index="1-33,17-5,17-21">Lowe has worked at Schering-Plough for 8 years, Bayer for 9 years, Vertex for 10 years, and as of 2018 at Novartis</cite>—a career arc that spans <cite index="7-3,23-3">drug discovery projects against schizophrenia, Alzheimer's, diabetes, osteoporosis and other diseases</cite>. <cite index="8-7,8-8">The blog is editorially independent, all content is Derek's own, and he does not in any way speak for his employer</cite>.

The blog's reach expanded steadily: <cite index="12-5">by 2006, it had between 3,000 and 4,000 visitors per day during the workweek</cite>, growing to <cite index="17-2,17-18">between 15,000 and 20,000 page views on a typical weekday</cite> by 2010. <cite index="2-5">Lowe tackles a mix of scientific updates, political perspectives, and critiques of scientific publishing, all in an extremely down-to-earth and readable manner</cite>. What's notable is the persistence: two decades of unbroken commentary from the bench, a form of longitudinal documentation that the field rarely produces.

Sources:

• https://en.wikipedia.org/wiki/Derek_Lowe_(chemist)
• https://www.the-geyser.com/interview-derek-lowe/
• https://www.statnews.com/2016/03/05/derek-lowe-chemist-blogger/
• https://www.science.org/blogs/pipeline
• https://www.chemistryworld.com/derek-lowe/1294.bio

<cite index="1-3">Section 4.3 of ICH E8(R1) describes the types of studies that typically span clinical development from the first studies in humans through late development and post-approval.</cite> The section is titled "Types of Clinical Studies" and is organized into subsections: Human Pharmacology, Exploratory, Confirmatory, and Post-Approval studies. These are functional categories. Phase numbers appear as cross-references, not organizing principles.

Human Pharmacology studies include first-in-human dose escalation, food effect, drug-drug interaction, and thorough QT studies. Exploratory studies generate hypotheses about dosing, endpoints, and patient populations. Confirmatory studies provide evidence adequate for regulatory decisions. Post-approval studies fulfill commitments or answer questions that emerged after marketing authorization. A single protocol can serve multiple objectives—dose-finding with interim efficacy assessments, for example, or a confirmatory trial with embedded pharmacokinetic sampling.

The structure reflects the reality that development is not linear. <cite index="1-1">The phases of drug development may overlap or be combined.</cite> Adaptive platform trials run exploratory and confirmatory analyses in parallel. Umbrella trials test multiple agents in one disease; basket trials test one agent in multiple diseases. Both cross traditional phase boundaries. E8(R1) accommodates this. It does not prescribe a four-phase march. It prescribes study objectives, quality factors, and the alignment between design and intent. If your design cannot meet your objective with adequate quality, the phase label will not save you.

Sources:

• https://database.ich.org/sites/default/files/E8-R1_Guideline_Step4_2021_1006.pdf

The E8(R1) model does not ask for more documentation. It asks for more thinking about what matters. <cite index="20-10,20-11,20-12">ICH E8 now incorporates the concept that quality in clinical studies should be fit for purpose—that is, quality is most effective when applied in a way that aligns with the study intent and design. This customized approach to quality can be associated with all components of clinical development including the design and management of drug development programs and clinical studies.</cite>

This is a departure. The prior model was prescriptive: monitor everything, verify everything, document everything. The revised model is risk-proportionate: <cite index="10-1">the guideline describes the aspects of clinical studies that support the determination of which quality factors are critical to ensuring the protection of study subjects, the integrity of the data, the reliability of results, and the ability of the studies to meet their objectives.</cite> If a protocol deviation does not threaten a critical-to-quality factor, it may not require a corrective action.

<cite index="20-13">ICH E8 advocates engaging with much broader and more diverse stakeholders like regulators, patients and patient advocacy groups, and other experts, and having those discussions as early as clinical development planning.</cite> <cite index="5-6,5-7">The views of patients (or of their caregivers/parents) can be valuable throughout all phases of drug development. Involving patients early in the design of a study is likely to increase trust in the study, facilitate recruitment, and promote adherence.</cite> This is not patient-centricity as branding. It is patient-centricity as quality control: if your endpoint does not matter to the people who have the disease, your study cannot answer the question it claims to answer.

Sources:

• https://www.ema.europa.eu/en/ich-e8-general-considerations-clinical-studies-scientific-guideline
• https://www.ema.europa.eu/en/documents/regulatory-procedural-guideline/ich-guideline-e8-r1-general-considerations-clinical-studies_en.pdf
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10850025/

<cite index="2-10">The modernization of E8 is the first step towards the Renovation of Good Clinical Practice (GCP) initiated in 2017.</cite> The renovation itself was a response to 25 years of methodological drift: adaptive designs, real-world evidence, decentralized trials, and risk-based monitoring had all entered practice without a coherent regulatory framework. <cite index="13-11,13-12">A wider range of study designs and data sources play an increasingly important role in drug development and are not adequately addressed in the original ICH E8 guidance; hence, the revised final guidance addresses a broad range of study designs and data sources.</cite>

<cite index="13-6">The revised final guidance describes internationally accepted principles and practices in the design and conduct of clinical studies of drug and biological products.</cite> <cite index="13-10">The ICH E8(R1) guidance focuses on the identification of factors that are critical to the study quality and the management of risks to those factors.</cite> This is the Quality by Design (QbD) model imported from pharmaceutical manufacturing: identify what is critical to quality, then manage risk to those factors specifically.

<cite index="20-3">Although embedded in the Efficacy segment of ICH Guidelines, ICH E8 sets the foundation for conducting clinical development with quality.</cite> It integrates with E6 (GCP conduct requirements), E9 (statistical principles), E5 and E17 (ethnic factors and multi-regional trials). <cite index="13-4">The final draft of the guideline was submitted to the ICH Assembly and endorsed by the regulatory agencies in October 2021.</cite> It is now Step 4—final and implemented.

Sources:

• https://www.federalregister.gov/documents/2022/04/11/2022-07690/e8r1-general-considerations-for-clinical-studies-international-council-for-harmonisation-guidance
• https://www.gmp-compliance.org/gmp-news/final-ich-e8r1-guideline-on-general-considerations-for-clinical-trials
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10850025/

<cite index="1-1">ICH E8(R1) specifies that while clinical drug development is often described as consisting of four temporal phases (phases 1-4), the phase concept is a description and not a requirement, and that the phases of drug development may overlap or be combined.</cite> This is not semantic hedging. It is architectural: <cite index="13-7">the final 2021 revision included a reduced emphasis on distinct phases of clinical development</cite>, marking a shift away from rigid chronology toward functional objectives.

The document itself structures clinical work not by phase number but by study type. <cite index="1-3">Section 4.3 describes the types of studies that typically span clinical development from the first studies in humans through late development and post-approval.</cite> <cite index="1-4">Human pharmacology studies—usually referred to as phase 1—prioritize the protection of study participants, especially for the initial administration of an investigational product to humans.</cite> The parenthetical matters: "usually referred to as" signals that the label is conventional, not normative.

<cite index="20-2">ICH E8 is an overarching document; the scope spans from the inception of drug development plans to the reporting of clinical study results to regulatory authorities.</cite> It sits at the top of the efficacy guidance hierarchy. <cite index="18-2">ICH E8 provides an overall introduction to clinical development, designing quality into clinical studies and focusing on those factors critical to the quality of the studies.</cite> It does not tell you what phase you are in. It tells you what questions your study must answer and what quality factors are critical to answering them.

Sources:

• https://database.ich.org/sites/default/files/E8-R1_Guideline_Step4_2021_1006.pdf
• https://www.federalregister.gov/documents/2022/04/11/2022-07690/e8r1-general-considerations-for-clinical-studies-international-council-for-harmonisation-guidance
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10850025/

<cite index="16-6">Janssen's way of working was based on the relationship between the chemical structure and the pharmacological action of molecules, and the use of the appropriate experimental models to test these relationships</cite>. This was not unique to Janssen, but the systematic application was. <cite index="8-2,8-3">Part of his extraordinary productivity was his devotion to Ehrlich's medicinal pharmaceutical approach and the systematic strategies he employed to explore the properties of new compounds—pharmaceutical discovery may always depend on an element of luck, but Janssen understood how to improve the odds in his favor and the importance of keeping his focus on genuinely therapeutic compounds, not just interesting chemistry</cite>.

The approach required cross-disciplinary range. <cite index="16-5">Thanks to his cross-disciplinary approach, Janssen carried out research in a broad variety of therapeutic areas: mycology, psychiatry, parasitology, allergology, gastroenterology, pain control and anesthesia, veterinary medicine, and plant and material protection</cite>. <cite index="13-6,13-13">Both Constant and Paul Janssen recognised that while most new drugs being produced at the time were merely new combinations, the way forward in drug development was the identification and synthesis of new compounds</cite>. That recognition—articulated before founding the lab in 1953—set the methodological direction.

<cite index="8-7">At the time of his death in 2003, he held more than 100 patents and was a listed author on more than 850 scientific publications</cite>. <cite index="16-4">With more than one hundred patents to his name, he was named "the most successful drug discoverer of all time" in 2002 by Nature Reviews</cite>. The output is difficult to contextualize—<cite index="3-12">if current trends hold, a drug discovery scientist starting their career today is likely to retire without ever having worked on a single drug that makes it to market</cite>.

Sources:

• https://www.encyclopedia.com/science/dictionaries-thesauruses-pictures-and-press-releases/janssen-paul-adriaan-jan
• https://prod.pauljanssenaward.com/portrait-innovation
• https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(03)15357-3/fulltext
• https://atelfo.github.io/2023/12/23/biopharma-from-janssen-to-today.html

<cite index="5-1">Janssen's concept rested on giving maximal freedom to competent and trusted researchers while continuously probing their activities and focusing their efforts towards achievable goals</cite>. This was not the flat structure it sounds like. <cite index="5-22,5-24,5-25">During the past decade pharmaceutical research has become increasingly dependent on processes, stage gating and market orientation, with a shift of attention from the individual researcher, patient and physician to hierarchical management structures—structures that may work well in incremental innovation but are far from optimal for fostering the type of environment that most often leads to breakthrough research</cite>.

<cite index="1-12,1-14,1-15">Statistics was one of the major driving forces toward success when Janssen founded his research laboratory in 1953; he had the genius and charisma of having statistical precepts being accepted by a rapidly expanding and diversifying scientific community, insuring that research proceeded in an orderly and planned fashion, while at the same time having an open mind for unexpected opportunities and for valuable chance events</cite>. The phrase "open mind for unexpected opportunities" is doing real work here—<cite index="7-4">the discovery of haloperidol followed from a serendipitous transition from analgesics to antipsychotics</cite>.

<cite index="16-1,16-2">His success was based on the principle of building research around people, who were inspired by his example and rewarded his trust with dedication and hard work; his research concept was also his management philosophy</cite>. That management philosophy appears to have had staying power within the organization that bears his name, even as the broader industry moved toward process-oriented structures.

Sources:

• https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1467-9310.2007.00481.x
• https://www.researchgate.net/publication/227850113_Successful_pharmaceutical_discovery_Paul_Janssen's_concept_of_drug_research
• https://pubmed.ncbi.nlm.nih.gov/18074755/
• https://www.encyclopedia.com/science/dictionaries-thesauruses-pictures-and-press-releases/janssen-paul-adriaan-jan

<cite index="1-2">Janssen's stated objective in 1953 was to design novel drugs by synthesizing many chemical compounds and testing them in several rapid and economic pharmacological assays</cite>. The approach was born from necessity, not a priori philosophy. <cite index="3-4,3-8">Constrained by limited money and resources, Janssen needed a quick, easy, and cheap discovery methodology—his team had to make a lot of simple compounds as quickly as possible and screen them using very simple methods</cite>.

The technical innovation was procedural: <cite index="9-1,9-11,9-12">Janssen employed a relatively simple process of swapping molecular building blocks around easily-modifiable central cores of molecules with known activity, and with this mix-and-match approach, his team quickly iterated and built up a library of standardised components</cite>. <cite index="9-13">Janssen found a commercialisable drug on his fifth try with ambucetamide, a treatment for muscle spasms</cite>. That hit rate would not hold—but in the early 1950s, <cite index="3-11">Janssen and his team developed over 70 new medicines between the 1950s and 1990s, many of which are still in use today</cite>.

What reads as brute-force iteration was guided by a specific conceptual bet: <cite index="20-2,20-5">While still a medical student, Janssen became convinced that the discipline of chemistry would be of increasingly vital importance in medicine, and he understood that there had to be a connection between a substance's chemical structure and its pharmacological action</cite>. <cite index="8-4">Sir James Black noted that "Dr. Paul never started a project without a conception in his head, a conception that not only specified a chemical starting place, a 'lead' molecule, with appropriate bioassays but also embodied foresight of how his invention would deliver clinical utility"</cite>.

Sources:

• https://www.researchgate.net/publication/227850113_Successful_pharmaceutical_discovery_Paul_Janssen's_concept_of_drug_research
• https://atelfo.github.io/2023/12/23/biopharma-from-janssen-to-today.html
• https://prod.pauljanssenaward.com/portrait-innovation
• https://www.jnj.com/the-legacy-of-dr-paul-janssen-how-a-funny-idea-helped-change-the-course-of-modern-medicine