Ask a clinical development lead what the 2023 accelerated approval reforms did, and you will usually hear some version of the same thing. The FDA cracked down. The bar went up. A surrogate endpoint that would have carried a filing five years ago now has to be stronger. The trade coverage that year all pointed the same way, and the belief has stuck.
It is a reasonable thing to believe. It is also wrong about the half that matters.
Open 21 CFR §314.510 today. The legal test for an accelerated approval surrogate is a drug whose effect on a surrogate endpoint is "reasonably likely, based on epidemiologic, therapeutic, pathophysiologic, or other evidence, to predict clinical benefit." [2] That sentence has not changed. FDORA, the accelerated approval provisions folded into the Consolidated Appropriations Act 2023, did not touch it. The evidentiary bar for what makes a surrogate acceptable at the front door is word-for-word what it was.
What changed sits on the other side of the approval: the machinery for holding you to the confirmatory trial, and the speed at which that machinery now moves. Between the 1992-2013 and 2014-2024 periods, the median time from accelerated approval to withdrawal fell from 9.5 years to 3.2 years (Jenei et al., Journal of Cancer Policy, 2025). [1]
The standard didn't move. The clock did.
That distinction is not academic. If your filing strategy is built around the tempo of the last decade, with the confirmatory trial as a slow-burn obligation and withdrawal as a distant, theoretical risk, you are planning for a reckoning that no longer arrives on that schedule.
Before any of this bites, be honest about which kind of surrogate you have, because the two are not interchangeable and sponsors conflate them constantly. So what actually separates the two?
The FDA/NIH BEST resource draws the line cleanly. A validated surrogate endpoint is supported by "clinical data providing strong evidence" that it predicts the clinical outcome, and it "can be used to support marketing approval... without the need for additional studies." [3] That means full approval, no strings: HbA1c, HIV-RNA, LDL-C, blood pressure and serum uric acid all sit in this bucket, because decades of data have tied them to outcomes that actually matter.
A "reasonably likely" surrogate is a different animal. It rests on "strong mechanistic and/or epidemiologic rationale" and nothing more, and the same resource says outright that it "sometimes fail[s] to predict an actual benefit." [3] Tumour response rate, progression-free survival in certain cancers, sputum culture conversion in TB: these are the surrogates accelerated approval exists for, and the confirmatory trial exists precisely because they sometimes lie.
Here is the category error I see repeatedly. A team treats its "reasonably likely" surrogate as though it were validated, plans as though the confirmatory trial is a formality, and is then genuinely surprised when the FDA treats a provisional approval as, well, provisional.
Could you sidestep the whole thing by getting your biomarker formally qualified? In theory, yes, and in practice almost no one does, because the numbers are brutal. As of 1 July 2025, only eight biomarkers have ever achieved full qualification through the FDA's Biomarker Qualification Program, and seven of those eight predate the modern framework the 21st Century Cures Act established in 2016 (Collins et al., Therapeutic Innovation & Regulatory Science, 2026). [4] One biomarker qualified in the nine years since. For surrogate endpoints it is starker still: just 5 of 61 accepted BQP projects have ever included a surrogate-endpoint biomarker, and none has reached qualification. [5] Median time from acceptance to a completed qualification plan runs to 47 months. Nearly four years, for a plan, not an approval.
So formal validation is not a live option for a biotech with a lead asset and a finite runway. "Reasonably likely" plus a confirmatory trial is the pathway you will actually walk. Which means the confirmatory obligation, and FDORA's new teeth around it, is the thing to design for from day one, not the thing you contextualise after a regulator starts asking pointed questions.
Let's be blunt about what FDORA actually did. The regulation has always required the sponsor to "study the drug further... with due diligence" [2] after an accelerated approval. What FDORA changed is the enforcement machinery around that duty, in three specific ways.
The before/after numbers are striking. The share of accelerated approvals with a confirmatory trial already underway at the time of approval rose from 63% to 85% (p=0.003). Median time to withdrawal fell from 9.5 to 3.2 years; median time to conversion from 4.3 to 2.3 years, both p<0.001, comparing 1992-2013 with 2014-2024. [1]
One caveat, how much of that shift FDORA itself caused is not settled. The trend predates the 2022 statute, the comparison period opens in 2014, eight years before FDORA, and the authors who documented it are careful to flag better trial design and tighter regulatory coordination as plausible contributors too. [1] I am not going to claim the statute single-handedly bent the curve. The point is narrower and firmer: the mechanisms are now on the books, and they are being used.
Which kills a comfortable belief worth naming so we can bury it: the "confirmatory formality," the idea that the confirmatory trial is a box you tick after the real win. It has not been that for a while.
Free download
The TPP Template
Target and minimally-acceptable profiles side by side, every claim linked to the evidence that has to support it.
Get the template →If you want to see the machinery run end to end, look at Pepaxto.
Melphalan flufenamide received accelerated approval on 26 February 2021 on the strength of overall response rate and duration of response in a single-arm trial. Its confirmatory Phase 3, OCEAN, then failed to confirm benefit, and it failed in the most instructive way. It met its PFS primary endpoint. But overall survival trended the wrong way, worse on the melflufen arm than on the control. In September 2022 the FDA's ODAC voted 14 to 2 against a positive benefit-risk. The FDA issued a withdrawal notice under the new FDORA procedure on 7 July 2023, and the final withdrawal decision landed on 23 February 2024, roughly seven months from notice to decision. [7]
That is the process working end to end. A confirmatory trial that met its surrogate but failed to confirm benefit, a streamlined procedure, a drug off the market inside a year of the formal notice.
That said, Exkivity shows the other way it plays out. Mobocertinib took accelerated approval in September 2021; its confirmatory trial, EXCLAIM-2, crossed its pre-specified futility boundary, with median PFS of 9.6 months in both arms (HR 1.04, 95% CI 0.77-1.39, P=.803). Takeda did not wait to be pushed. It voluntarily withdrew the drug worldwide on 2 October 2023, ahead of any forced FDORA proceeding. [8]
An accelerated approval is a licence with a countdown attached, the timer running from the day you file. Five years ago you could not really hear it ticking; now you can.
Here is the beat most sponsors underweight, and it is the sharpest lesson in the whole file.
Sotorasib took accelerated approval in May 2021 on ORR and duration of response. Its confirmatory trial, CodeBreaK 200, then hit its primary endpoint, PFS, and a secondary endpoint, ORR, against docetaxel. On a naive reading, that is a win, the FDA did not read it that way. The trial showed no overall survival benefit, and it carried conduct problems, unbalanced dropout and crossover that muddied the PFS result. In October 2023 ODAC voted 10 to 2 that the PFS finding could not be reliably interpreted. [9] The FDA issued a Complete Response Letter on the full-approval supplement on 26 December 2023. Sotorasib's accelerated approval remains in place, now tied to a fresh postmarketing requirement due no later than February 2028. [9]
Read that sequence again. The confirmatory trial met its primary endpoint and the drug still did not convert. That is the risk I would call winning the confirmatory trial and losing the approval, and it turns up when the confirmatory endpoint is itself another short-horizon surrogate, PFS standing in for OS, and the trial's conduct gives the agency room to doubt it.
Compare that with what "enough" actually looks like.
Mirvetuximab soravtansine converted cleanly. Its confirmatory Phase 3, MIRASOL (NCT04209855, N=453), showed a 33% reduction in the risk of death and a 35% reduction in the risk of progression versus chemotherapy. Full approval followed on 22 March 2024. [10] Tarlatamab did the same, and did it on the hardest endpoint of all: its confirmatory trial, DeLLphi-304, hit its primary endpoint of overall survival outright (13.6 vs 8.3 months, HR 0.60, P<.001), with a PFS improvement alongside it (4.2 vs 3.2 months, HR 0.72, P<.001). It converted to full approval on 19 November 2025. [11]
The pattern is not subtle. The confirmatory trials that convert are the ones that answer the question the surrogate left open, whether patients actually lived longer, rather than re-running the same surrogate ambiguity at greater expense. Across antibody-drug-conjugate trials, the trial-level association between the OS and PFS hazard ratios (R²=0.79) is far stronger than the OS-ORR association (R²=0.47). [12] Not every surrogate predicts survival equally well, and the confirmatory trial is where that gap gets exposed in public.
The honest counterargument to all of this runs roughly as follows. This is oncology-specific noise. The FDA always had the authority to pull an accelerated approval, so the enforcement trend is a story about shifting culture, not a statute. And the clean conversions prove the pathway works fine for genuinely good drugs, so a sponsor with a strong surrogate and a real effect has nothing new to fear.
Grant the first half. I already have: the 9.5-to-3.2-year shift in time to withdrawal is only partly about the statute, and I won't pretend otherwise. But the objection misreads where the risk actually lives. FDORA did not invent withdrawal. The real problem is that the base rate of these approvals is far shakier than the conversion headline suggests. Of 46 cancer-drug accelerated approvals granted between 2013 and 2017 with at least five years of follow-up, 63% had converted to regular approval, but only 43%, twenty of the forty-six, had actually demonstrated clinical benefit in a confirmatory trial; 22% were withdrawn (Liu, Kesselheim & Cliff, JAMA, 2024). [13] And across 167 accelerated approval indications, the strongest single predictor of withdrawal was low clinical benefit at the time of the accelerated approval itself (OR 4.63, 95% CI 1.50-14.33; Tibau et al., EClinicalMedicine, 2025). [14] A faster, more certain enforcement process laid over a pathway with that base rate is exactly why the filing case has to anticipate the reckoning. It does not matter who or what caused the speed-up. The speed-up is real, and the fragility underneath it always was there.
So what do you actually do differently on the Monday you start assembling the package? Five things.
A little while back I worked on an early-phase rare-disease programme where the surrogate looked wonderful and the team had built the entire filing narrative around it, with no serious plan for what happened if the confirmatory readout disappointed. When we finally sat down and war-gamed the failure scenario, the mood in the room changed within about ten minutes. It is a deeply uncomfortable exercise. It is also the single most useful hour you can spend before a surrogate-based filing, because the alternative is improvising that same conversation in public, in front of an advisory committee, on the agency's timetable rather than your own.
This is the anticipatory, cross-functional work an integrated evidence plan exists to force, and it is the work most teams defer until the confirmatory result is already staring at them. Deciding at filing how you will read, report and respond to a result you don't yet have is exactly the minimum-viable-evidence discipline that separates a plan from a hope.
InovaSight earns its keep at exactly this point: it shows how comparable surrogates have actually fared at the confirmatory stage before you commit to yours. For the fuller cautionary version of what happens when a surrogate-driven approval outruns its evidence, our study in failure on aducanumab is worth the detour.
One last thing, and it is the part that should leave every sponsor slightly uneasy. FDORA passed at the end of 2022. Three and a half years on, the FDA's own guidance on what "underway" actually means, the practical definition of the single most consequential new requirement, remains a draft, issued 7 January 2025 and still not finalised. The broader December 2024 draft guidance on accelerated approval is also still draft. [15] Sponsors are being held to a faster, harder, more public reckoning under an operational standard the agency has not yet finished writing down.
Plan for the reckoning anyway. The clock is already running.
Get the monthly digest
The 5 things evidence leads need to know each month: regulatory moves, RWE developments and what they mean in practice. No pitch, one email a month.