Executive Summary
FDA's move toward single-trial development should be understood as an evolution in evidence architecture, not as the sudden creation of a permissive pathway. One pivotal trial has supported a large share of approvals for years, particularly in oncology, rare disease, gene therapy, and accelerated approval settings.
The June 2026 draft guidance is materially more qualified than the February 2026 declaration that one pivotal trial would become FDA's new default. It describes one adequate and well-controlled investigation plus confirmatory evidence as one available approach, then imposes a sliding evidentiary burden: the less independently confirmatory the supporting evidence, the more persuasive the pivotal trial must be.
The most likely outcome is not a universal collapse from two Phase 3 trials to one smaller Phase 3 trial. It is a bifurcated model consisting of very large, definitive outcome trials for common diseases and much smaller programs for rare or genetically defined diseases where mechanism, natural history, pharmacodynamic evidence, and longitudinal follow-up can carry more evidentiary weight.
Leadership turnover makes the categorical policy slogan less durable than the underlying legal and scientific framework. The statutory authority, historical approval practice, and revised guidance will survive the departures of Vinay Prasad and Martin Makary, but implementation is likely to remain division-specific and highly dependent on early FDA agreement.
Single-Trial Approval Was Already Mainstream
The historical record does not support a clean divide between an earlier two-trial era and a new one-trial era. An analysis of novel therapeutics approved from 2005 through 2012 found that 74 of 201 evaluable indications, or 36.8%, were supported by a single pivotal efficacy trial. Across all indications, the median was two pivotal trials and 760 patients, but the evidence base varied substantially by therapeutic area, orphan status, and regulatory pathway. 1
The direction of travel was already clear before 2026. Across three approval cohorts, the proportion supported by at least two pivotal trials fell from 80.6% in 1995 through 1997 to 60.3% in 2005 through 2007 and 52.8% in 2015 through 2017, while approvals supported only by single-group pivotal trials rose from 4.0% to 17.0%. These figures do not establish that the later approvals were weaker, but they show that replication through two conventional trials had become progressively less universal. 2
Single-trial approvals were also not confined to oncology or orphan products. A review of 27 non-oncology, non-orphan indications approved in the United States and/or European Union from 2012 through 2016 found that 85% of the pivotal trials were randomized and controlled, the median enrollment was 828, the range was 29 to 26,449, and every superiority trial met its primary endpoint at p less than or equal to 0.005. Most approvals also included supportive efficacy information outside the pivotal trial. 3
More recent studies confirm that the one-trial model is now numerically important, although estimates depend heavily on how an approval, pivotal trial, and indication are classified. A 2015 through 2023 indication-level review placed 185 of 401 included original indications, or 42%, in its principal single-pivotal-trial category. AgencyIQ, using the number of clinical studies listed in original labeling for CDER novel drugs, reported that 33 of 50 approvals in 2024, or 66%, relied on one listed clinical study. The methodologies are not interchangeable, but both indicate that single-trial evidence is no longer exceptional. 45
Representative Approvals, 2010 to 2026
The following cases are selected to show the range of evidence structures FDA has accepted, not to present an exhaustive census. They span large randomized mortality and morbidity trials, small biomarker-driven accelerated approvals, open-label gene therapy studies, and programs in which the single pivotal result was disputed internally or later failed to reproduce. Because 2026 is incomplete, the approval examples extend through 2024, while the 2026 portion of the analysis addresses policy, guidance, and leadership. 6789101112131415161718
| Product and FDA action | Pivotal trial type and scale | Principal efficacy result | Confirmatory or supportive evidence | Evidentiary lesson |
|---|---|---|---|---|
| Makena, 2011 accelerated approval | One randomized, double-blind, placebo-controlled trial; n=463 | Delivery before 37 weeks: 37.1% with treatment versus 54.9% with placebo; nominal p=0.0003 | Surrogate endpoint plus a required postapproval confirmatory trial | The confirmatory trial did not verify benefit, and FDA withdrew approval in 2023. A statistically positive single trial did not protect the approval when the surrogate and replication failed. |
| Entresto, 2015 traditional approval | PARADIGM-HF, randomized, double-blind, active-controlled outcomes trial; n=8,442 | Cardiovascular death or heart-failure hospitalization: hazard ratio 0.80, 95% CI 0.73 to 0.87; p<0.001, with the FDA review reporting p=0.0000002 | Pharmacologic rationale, dose-finding and early-phase data, consistency across cardiovascular outcomes, and a large internally coherent trial | One trial can provide unusually strong evidence when it is large, event-driven, clinically consequential, and statistically overwhelming. |
| Nuplazid, 2016 traditional approval | ACP-103-020, multicenter, randomized, double-blind, placebo-controlled, 6 weeks; 199 randomized and 185 in the modified intent-to-treat analysis | SAPS-PD treatment difference approximately -3 points; 95% CI -4.91 to -1.20; p=0.0014 | Three supportive studies and consistency across hallucination and delusion components, although prior efficacy studies were mixed or unsuccessful | FDA accepted a single positive study, but the mixed supportive record illustrates that additional studies can contextualize rather than replicate the pivotal result. |
| Exondys 51, 2016 accelerated approval | Small randomized placebo-controlled phase followed by open-label extension; the key biopsy analysis involved 12 patients | No approval-defining clinical p-value established clinical benefit; approval relied on increased dystrophin as a surrogate endpoint | Dystrophin biology, muscle-biopsy evidence, mechanistic plausibility, and a required confirmatory clinical trial | The approval became a prominent example of disagreement over whether a small biomarker change was reasonably likely to predict clinical benefit. |
| Luxturna, 2017 traditional approval | Open-label, randomized, controlled crossover Phase 3 trial; n=31 | Median bilateral mobility-test change of +2 light levels with treatment versus 0 in controls; p=0.001 | Phase 1 experience, RPE65 disease mechanism, objective functional testing, crossover evidence, and durability follow-up | A small trial can be persuasive when the disease mechanism is precise, the untreated course is unfavorable, and the functional effect is large and objectively measured. |
| Oxbryta, 2019 accelerated approval | HOPE, randomized, double-blind, placebo-controlled Phase 3 trial; total n=274, including 90 at the approved 1500 mg dose and 92 placebo | Hemoglobin response at week 24: 51.1% versus 6.5%; p<0.0001 | Hemolysis markers, pharmacokinetic and pharmacodynamic evidence, mechanism, and a postapproval requirement to verify clinical benefit | FDA later identified safety and postmarketing concerns, and the sponsor voluntarily withdrew the product in 2024. Biomarker certainty did not ensure durable clinical benefit-risk. |
| Relyvrio, 2022 traditional approval | CENTAUR, randomized, double-blind, placebo-controlled; n=137; 24 weeks | Prespecified ALSFRS-R analysis p=0.034; FDA-preferred joint-rank analyses yielded p=0.063 in the modified intent-to-treat population and p=0.079 in the intent-to-treat population | Open-label extension, post hoc survival analysis, external natural-history comparisons, and regulatory flexibility for ALS | The larger PHOENIX trial failed, and the sponsor initiated withdrawal in 2024. The case shows the risk of relying on a modest primary result supplemented by post hoc and externally controlled analyses. |
| Hemgenix, 2022 traditional approval | HOPE-B, open-label Phase 3 trial with within-patient lead-in control; n=54 | Annualized bleeding rate 1.9, 95% CI 1.0 to 3.4, versus 4.1, 95% CI 3.2 to 5.4, during lead-in; rate ratio 0.46, 95% CI 0.26 to 0.81; p-value unspecified in the action summary | Initial clinical investigation, preclinical studies, sustained factor IX expression, and each patient's prospective lead-in history | FDA expressly characterized the basis as one adequate and well-controlled trial with compelling clinical benefit plus supportive evidence. |
| Qalsody, 2023 accelerated approval | VALOR, randomized, double-blind, placebo-controlled Phase 3 trial; n=108, with a primary faster-progression subgroup of n=60 | Clinical primary endpoint: ALSFRS-R difference 1.2 points, 95% CI -3.2 to 5.5; p=0.97. Plasma neurofilament reduction showed a 67% difference in geometric mean ratios; nominal p<0.0001 | CSF SOD1 target engagement, neurofilament biology, mechanism, open-label extension trends, and the ongoing ATLAS confirmatory trial | FDA approved on a surrogate despite a clearly negative clinical primary endpoint, making the credibility of the biomarker-to-benefit relationship central. |
| Casgevy, 2023 traditional approval | Multinational, single-arm Phase 1/2/3 study plus long-term rollover follow-up; 44 treated and 31 evaluable for the primary efficacy outcome | 29 of 31 evaluable patients, or 93.5%, were free from severe vaso-occlusive crises for at least 12 consecutive months; one-sided 98% CI 77.9% to 100%; conventional p-value not reported | BCL11A and fetal hemoglobin mechanism, sustained fetal hemoglobin, magnitude of response relative to expected disease course, and long-term follow-up | A single-arm trial can support traditional approval when the outcome is dramatic, the disease history is well understood, and the biological intervention is directly measurable. |
| Elevidys, 2023 accelerated approval and 2024 indication expansion | Initial approval relied on micro-dystrophin evidence. Study 301 later randomized 63 patients to treatment and 62 to placebo | Study 301 missed its primary NSAA endpoint: difference 0.65 points; p=0.2441. Several timed-function secondary endpoints favored treatment | Micro-dystrophin expression, mechanistic rationale, secondary and exploratory functional endpoints, and studies 101, 102, and 103 | The decisions involved documented disagreement among FDA review functions and center-level judgment about how to weigh a failed primary endpoint against the broader evidence package. |
| Beqvez, 2024 traditional approval | Open-label Phase 3 trial using each patient's prospective baseline period as control; n=45 | Model-derived annualized bleeding rate 2.5, 95% CI 1.0 to 3.9, versus baseline 4.5, 95% CI 1.9 to 7.2; difference -2.1, 95% CI -4.8 to 0.7; met the prespecified noninferiority margin of 3.0; p-value unspecified | Phase 1/2 clinical data, preclinical evidence, factor IX activity, and long-term follow-up | FDA again expressly relied on a single adequate and well-controlled trial, showing that prospective within-patient controls can be acceptable in selected gene-therapy settings. |
What Confirmatory Evidence Actually Meant
The phrase confirmatory evidence can imply independent replication, but FDA practice has used it more broadly. Depending on the program, the confirmatory layer has included controlled trials in related indications, trials of related products, early-phase dose-response or pharmacodynamic data, mechanistic evidence, natural-history data, real-world evidence, external controls, consistency across endpoints, and postapproval verification. The June 2026 guidance places the strongest general expectation on related adequate and well-controlled trial data, while permitting weaker or different sources when the pivotal trial is itself highly persuasive or regulatory flexibility is justified. 19
| Confirmatory-evidence category | Typical regulatory contribution | Principal limitation |
|---|---|---|
| Related adequate and well-controlled trials | Provides clinical replication across a related disease, disease stage, product, or pharmacologic class | Transportability depends on similarity of pathophysiology, mechanism, endpoints, population, and treatment effect |
| Early-phase clinical evidence | Supports dose selection, target engagement, exposure-response, and consistency of effect | Often small, exploratory, unblinded, or not multiplicity-controlled |
| Mechanistic or pharmacodynamic evidence | Raises the prior probability that the pivotal result is causal, especially for monogenic disease or direct target correction | A plausible or demonstrated mechanism does not necessarily establish improvement in how patients feel, function, or survive |
| Natural-history or real-world evidence | Anchors the untreated course and can corroborate an unusually large response | Susceptible to selection, measurement, temporal, and treatment-confounding biases; data used as confirmatory evidence should be distinct from an external control used in the pivotal analysis |
| Within-trial coherence | Secondary endpoints, subgroups, biomarkers, and sensitivity analyses can make a single result more internally credible | These analyses are not independent replication and can be misleading if post hoc or affected by multiplicity |
| Postapproval verification | Allows earlier access while testing whether a surrogate or intermediate endpoint predicts clinical benefit | Delayed, infeasible, negative, or poorly enforced confirmatory studies can leave patients exposed to prolonged uncertainty |
The empirical record shows that FDA has not always identified this layer transparently. In the 2015 through 2023 review, FDA explicitly referenced confirmatory evidence for 36 of 185 approvals in the principal single-pivotal-trial category and implicitly referenced it for four more. The 40 approvals cited 99 individual sources, most commonly pharmacodynamic or mechanistic evidence. References to confirmatory evidence increased from 7% before the December 2019 draft guidance to 34% afterward, with p<0.0001. The study's classification framework does not prove that the remaining approvals lacked support, but it does expose a recurring documentation problem: the public record often does not make clear which evidence FDA treated as legally confirmatory and how much weight it carried. 4
What the June 2026 Guidance Changes
The revised draft guidance consolidates FDA's effectiveness framework and makes one scientifically rigorous adequate and well-controlled investigation plus confirmatory evidence a central, expressly described route to substantial evidence. It is a draft Level 1 guidance, is nonbinding, and will replace the 1998 effectiveness guidance only when finalized. It revises the December 2019 draft rather than simply converting the February 2026 policy announcement into an operational command. 19
The pivotal concept is an evidentiary tradeoff. FDA states that the strength of the trial's design, conduct, analysis, and results will determine the strength of confirmatory evidence required. One route combines an adequate and well-controlled trial with strong clinical confirmatory evidence, ordinarily related controlled trial data. The other route relies on a highly persuasive pivotal trial and permits the early-phase information that justified proceeding to that trial to serve as the confirmatory layer. 19
A highly persuasive trial is not defined by p<0.05 alone. FDA expects a representative, multicenter, clinically relevant design; a contemporary control and supportive care; a meaningful endpoint, preferably irreversible morbidity or mortality when feasible; adequate power; a clinically important effect with a narrow enough confidence interval; supportive prespecified secondary endpoints; consistency across important subgroups; comprehensive follow-up; minimal missing data; and robustness to plausible analytical assumptions. When the prior probability of effectiveness is low, the common one-sided 0.025 threshold may be insufficient to control false-positive risk. 19
The guidance also preserves substantial flexibility. In serious, life-threatening, severely debilitating, rare, or high-unmet-need settings, FDA may accept more residual uncertainty, an externally controlled design, a smaller trial, or even a significance level higher than one-sided 0.025 when the criterion is prespecified, scientifically justified, and agreed with the Agency. Mechanistic evidence may be confirmatory when disease biology is well understood and the intervention directly corrects the major driver, while separate natural-history or reliable real-world data may corroborate outcomes that would be highly unlikely without treatment. 19
The most important limiting language concerns safety. A trial package that is sufficient for effectiveness may still be too small or too short to support a favorable benefit-risk determination. FDA states that a larger or longer trial, or an additional trial, may be needed to characterize safety. The practical result is that the number of pivotal efficacy trials may fall without a commensurate reduction in total clinical exposure, development duration, or postmarketing obligations. 19
The February Declaration and the June Text Are Not the Same Policy
In February 2026, Prasad and Makary wrote that one adequate and well-controlled study plus confirmatory evidence would become FDA's new default for marketing authorization. They argued that modern mechanistic knowledge, biomarkers, intermediate endpoints, and other components of clinical credibility can substitute for automatic replication. Their statistical illustration contrasted a 250-in-10,000 false-positive probability from one test at the 0.05 level with approximately 6 in 10,000 when two independent positive trials are required. They also projected lower development cost and faster access, while reserving the right to require additional studies for nonspecific mechanisms, labile or surrogate outcomes, and deficient trials. 20
The June guidance does not repeat the phrase new default. It presents one trial plus confirmatory evidence, more than one adequate and well-controlled investigation, and scientifically justified extrapolation from known effectiveness as parallel approaches. It repeatedly states that sufficiency depends on the strength of the evidence, clinical context, ethics, feasibility, and prior FDA agreement. This is a meaningful institutional narrowing of the public slogan, even though it still favors more efficient development and provides a clearer route for one-trial programs. 1920
The difference matters operationally. A default can be treated as the starting presumption, with a second trial requiring justification. The guidance instead creates a structured negotiation in which the sponsor must justify why one trial is sufficient and how the confirmatory package offsets the missing independent replication. The likely practical standard is therefore not one trial unless FDA objects, but one trial when the division agrees that the full evidence architecture is adequate before the pivotal study begins. 19
The Core Controversy Is Independence, Not Trial Count
Two trials offer something that a large p-value margin cannot fully replace: an independent opportunity to test whether the effect survives different sites, investigators, operational errors, patient mix, calendar time, and unrecognized design bias. A single trial can reduce random error through sample size, but it can remain vulnerable to a systematic flaw shared by every patient and endpoint. FDA's guidance acknowledges this problem by noting that even precisely replicated trials can share design limitations, while nonidentical trials may be more persuasive because an unknown flaw is less likely to affect both. 19
The debate therefore turns on whether confirmatory evidence is genuinely independent and outcome-relevant. Related controlled trials can provide real corroboration. By contrast, target engagement, a biomarker response, a secondary endpoint from the same dataset, an open-label extension, or a post hoc natural-history comparison may strengthen a causal narrative without independently reproducing the clinical effect. The June guidance allows these sources, but generally only when disease biology is unusually clear, the pivotal trial is highly persuasive, or the clinical context warrants flexibility. 194
Makena, Oxbryta, and Relyvrio are cautionary cases for different reasons. Makena's confirmatory trial did not verify the expected benefit. Oxbryta produced a strong hematologic response but was later voluntarily withdrawn after emerging safety and postmarketing concerns changed the benefit-risk assessment. Relyvrio's modest and analytically fragile pivotal evidence was not reproduced in the larger PHOENIX trial. These cases do not show that single-trial approval is categorically unsound, but they demonstrate that statistical significance, biological plausibility, surrogate movement, and post hoc support are not interchangeable with replicated clinical benefit. 62111221223
Exondys 51 and Elevidys show a separate institutional controversy: how much decisional authority should center leadership exercise when review disciplines disagree about the adequacy of the efficacy evidence. In both programs, the public record documents substantial internal disagreement over surrogate interpretation, clinical endpoints, or the totality of evidence. The 2026 guidance supplies a framework for scientific judgment, but it cannot eliminate differences in how reviewers value a biomarker, a failed primary endpoint, secondary outcomes, or unmet need. 9161719
External experts also questioned the process by which the February policy was announced, including the absence of a detailed implementation framework and uncertainty about safety and postmarket data. Those concerns are partly answered by the June draft, which is substantially more technical and explicit, but they remain relevant because a nonbinding guidance cannot by itself ensure consistent interpretation across review divisions. 2419
Leadership Turnover Changes the Policy's Durability
The two officials who publicly framed one trial as FDA's new default were no longer at the Agency by the time the revised draft guidance appeared. Reuters reported that Prasad left FDA at the end of April 2026, and that Makary resigned as Commissioner on May 12, 2026, with Kyle Diamantas becoming acting Commissioner. 2526
Their departures weaken the political force of the February announcement, particularly its categorical default framing and its prediction of a surge in development. A successor Commissioner or center leadership team can alter review culture, implementation priorities, and the degree of tolerance for mechanistic or postmarket evidence without changing the statute. The revised guidance's more balanced wording is therefore more likely to persist than the stronger rhetoric of the NEJM article. This is an inference from the timing, wording, and leadership transition rather than an announced reversal by FDA. 19202526
The durable element is the framework, not the personalities. Congress has permitted one adequate and well-controlled investigation plus confirmatory evidence since 1997, FDA has used the route for decades, and the 2026 draft provides a detailed analytical structure that career reviewers can apply. Leadership can influence how readily divisions accept a one-trial proposal, but it would take more than personnel turnover to return practice to an automatic two-trial presumption. 19124
Forecast: How Large Will the Single-Trial Model Become?
The most defensible forecast starts from the observed baseline rather than the February rhetoric. Depending on methodology, recent evidence places single-pivotal-trial approvals at roughly 42% of included original indications from 2015 through 2023 and 66% of CDER novel drugs in 2024. Because indication-level and label-study counts answer different questions, no single percentage should be treated as the definitive current rate. 45
Regulatory Impact's base-case estimate is that approximately 60% to 70% of CDER novel-drug approvals could continue to rely on one principal pivotal trial during 2027 through 2029. This is an analytical scenario, not an FDA forecast. It assumes that the revised guidance is finalized in substantially similar form, career review divisions use it selectively, and the mix of orphan, oncology, genetic, and accelerated-approval programs remains an important driver of annual approvals. 1954
| Scenario for 2027 to 2029 | Estimated share of CDER novel approvals relying on one principal pivotal trial | Conditions that would produce the scenario | Expected program shape |
|---|---|---|---|
| Restrained implementation | 50% to 60% | New leadership deemphasizes the default rhetoric; divisions require replication for uncertain mechanisms, subjective endpoints, modest effects, or weakly validated surrogates | More conventional two-trial programs outside rare disease and oncology; single-trial use remains concentrated in high-unmet-need settings |
| Base case | 60% to 70% | Guidance is finalized with its current sliding-scale framework; sponsors secure early agreement; confirmatory evidence is planned prospectively | One large persuasive trial plus early-phase, mechanistic, related-indication, or natural-history support; safety exposure may still require additional cohorts or studies |
| Expansion case | 70% to 80% | Successor leadership preserves a strong one-trial presumption and divisions accept early-phase or mechanistic evidence broadly | Faster reduction in nominal Phase 3 trial count, greater reliance on postmarket verification, and wider use beyond orphan and oncology programs |
The more consequential prediction concerns scale, not count. The guidance is likely to reduce the number of studies designated pivotal faster than it reduces total registrational enrollment. For common diseases with variable outcomes or modest treatment effects, the single trial may need to be larger, more geographically representative, more event-rich, and more statistically conservative than either of two conventional trials. Entresto's 8,442-patient outcome study is the clearest model: one trial, but not a small program. 1973
At the other end of the distribution, genetically defined and rare-disease programs will continue to use dozens rather than thousands of patients when the intervention directly addresses disease biology and the observed outcome is difficult to explain by bias or natural variation. Casgevy, Luxturna, Hemgenix, and Beqvez illustrate different versions of that model. The result will be a wider dispersion of pivotal-trial size, with fewer medium-sized replicated programs and more very large definitive trials or very small biologically anchored programs. 1510131819
Implications for Development Strategy
A sponsor should not begin with the assertion that one trial is FDA's default. The stronger approach is to define the proposed evidentiary architecture before the end-of-phase 2 meeting: what makes the pivotal trial adequate and highly persuasive, what independent or complementary evidence will confirm it, what contradictory evidence exists, and why a second adequate and well-controlled trial would add less value than the proposed alternative. FDA specifically advises discussion before initiation of the single pivotal trial. 19
The statistical strategy should be calibrated to prior probability and consequence. A conventional two-sided 0.05 result may be legally sufficient in some programs but strategically weak when the mechanism is uncertain, the endpoint is subjective, the expected effect is modest, or the confirmatory evidence is not independent. Sponsors should consider powering for a more persuasive confidence interval and p-value, protecting distinct secondary endpoints, prespecifying subgroup and sensitivity analyses, and minimizing missing data rather than relying on a threshold-crossing primary analysis alone. 19
Confirmatory evidence should be treated as a planned workstream, not a collection of favorable observations assembled at submission. The strongest packages will prospectively specify the relationship between the pivotal endpoint and supporting clinical, pharmacodynamic, mechanistic, natural-history, or real-world evidence; define the sources' independence; establish data quality and analytical methods; and explain negative or discordant findings. The public record from 2015 through 2023 suggests that clearer identification of confirmatory evidence remains an opportunity for both sponsors and FDA. 194
Safety planning should remain separate from the efficacy-trial count. Chronic-use products, therapies with class concerns, and interventions with delayed risks may still need broad exposure, longer follow-up, registries, or additional controlled data even when one trial establishes effectiveness. A development program that saves a pivotal efficacy trial but creates an inadequate safety database is not streamlined; it is incomplete. 19
Finally, sponsors should model the downside of confirmatory failure before selecting the pathway. Accelerated approval and other evidence-flexible approaches can create substantial postmarket obligations and reputational exposure if the surrogate fails, the larger trial is negative, or the benefit-risk profile changes. Makena, Oxbryta, and Relyvrio show that the strategic value of earlier approval depends on the feasibility, timing, and credibility of the evidence that follows. 212223
Conclusion
FDA is moving toward a system in which the unit of evaluation is the evidence package rather than the number of nominally pivotal trials. That change is real, but it is less categorical than the phrase one-trial default suggests.
The strongest future programs will not merely delete a confirmatory Phase 3 trial. They will relocate certainty into a more persuasive pivotal design, more credible prior and supporting evidence, stronger internal coherence, and a feasible plan to characterize safety and resolve residual uncertainty.
Leadership turnover may moderate the pace and breadth of implementation, but it is unlikely to reverse the underlying trend. Single-trial programs are already common. The next regulatory contest will concern which programs deserve that structure, what evidence can legitimately substitute for replication, and how quickly FDA acts when postapproval evidence does not confirm the original judgment.