Olaparib, rucaparib and niraparib all reached patients with ovarian cancer on the strength of a tumour-response or progression-free-survival signal. All three were later withdrawn from those indications. The reason repeated itself in each case: once the overall survival data arrived, the drugs were performing no better than the comparator, and in places worse. Rucaparib's confirmatory trial, ARIEL4, put median overall survival at 19.4 months against 25.4 for the control arm. Olaparib's SOLO-3 read 29.4 against 39.4. The three indications came off the US market seven, five and two years after they went on it. [1]
Every one of those approvals was, on its own terms, a win: the endpoint did its job, the agency said yes, and patients got the drug sooner. The surrogate's science was never the problem. What broke was the assumption that an endpoint a regulator accepts is the same evidence a payer needs. It is not, and the space between those two facts is where a great deal of biotech value quietly goes to die.
So here is the thesis, where you can argue with it. Regulators and payers do not disagree about whether your endpoint is valid. They need different kinds of validity, and a pivotal trial built around the fastest regulator-friendly surrogate does not save you time. It defers a bill. That bill comes back as a market withdrawal, an HTA restriction, or years spent in managed-access limbo and it is denominated in a currency the surrogate never measured.
The reflex is to file this under "payers are tougher than regulators", which quietly assumes both sit on one scale with the payer's bar simply set higher up it. That is the wrong frame. Toughness would mean the same exam with a harsher pass mark; the payer sets a different exam.
A regulator can accept a surrogate (tumour response, PFS, a biomarker) because under an accelerated-approval-style pathway the surrogate only has to be reasonably likely to predict benefit, and accepting it lets a seriously ill patient have the drug now rather than in six years. That is a defensible trade, but it is not the payer's trade.
A payer is built to do something else. An HTA body prices a therapy against outcomes a patient actually feels and a cost-effectiveness model can consume: survival, physical function, quality of life. To that machine a regulator-accepted surrogate proves almost nothing about benefit, because it never measured the thing the model runs on.
How large is the gap? Germany's IQWiG publishes its own numerical bar for when a surrogate counts as validated. A correlation with the true endpoint of R ≥ 0.85 at the lower confidence bound is proof of validity; R ≤ 0.7 at the upper bound is proof of invalidity; everything between is "unclear" and earns, at best, a hint of a patient-relevant effect. Applied to IQWiG's own worked oncology example, tumour response as a surrogate for overall survival "remains unclear." [2] Read that again. In oncology, the most surrogate-friendly corner of medicine, the reference HTA methods institute in Europe's largest drug market does not accept tumour shrinkage as a validated stand-in for survival.
Let's be frank: this is not some fringe standard a payer invented to be difficult. The EMA's own conditional-marketing-authorisation guideline (finalised in 2016, still in effect) names the gap in plain text: the pathway exists for cases still needing results on the "longer-term, clinically most relevant efficacy endpoint... e.g. survival data vs. overall response rate," and it explicitly invites sponsors to seek "parallel scientific advice" with HTA bodies during development. [3] The regulator wrote down, in a guideline, that its yes is not the payer's yes. Sponsors keep not reading it.
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 →Here is the finding that should retire the "approve now, reimburse later" strategy on its own. Across a cohort of oncology products, a 2024 Lancet Oncology analysis found accelerated regulatory authorisation was associated with a longer time to a positive HTA decision, not a shorter one (hazard ratio 0.70, 95% CI 0.51–0.95). What actually predicted a positive payer outcome was the magnitude of proven clinical benefit, measured on the ESMO-MCBS scale, independent of which regulatory route the drug took. [4] The fast lane to approval is not a fast lane to reimbursement. It may be the slow one.
Why? Because the confirmatory data was always going to have to carry the payer case, and often it doesn't arrive, or doesn't say what was hoped. A 2024 JAMA cohort study found that of oncology accelerated approvals granted between 2013 and 2017, 41% had not shown an improvement in overall survival or quality of life in their confirmatory trials after five or more years of follow-up. [5] Broader analyses land in the same territory: a 2025 JAMA Network Open review of 102 confirmatory trials found only a third demonstrated a significant OS benefit. [6]
This is the second invoice.
Approval on a surrogate is market access bought on credit. The payer holds the note, and it is written in survival and quality of life, the very things the pivotal trial was designed not to measure. When the note comes due it is expensive for everyone: one 2025 BMJ Medicine analysis attributes roughly $20.1 billion in US Medicare spending to accelerated-approval cancer indications, at about $263,371 per life-year gained. [7]
Let me kill a tempting idea first, because I went looking for it and it is not there. There is no single, fused endpoint that satisfies a regulator and a payer at once, no clever composite that threads both needles, and neither the published literature nor the trial registries have one to point to. Anyone selling you a "dual-purpose endpoint" as a discrete object is selling you something that does not exist.
But what is real, and sitting in the public record, is endpoint hierarchy: a regulatory-grade surrogate and a payer-grade patient-relevant outcome, built into the same protocol from inception.
Look at how real pivotal trials register. KEYNOTE-240 put progression-free survival and overall survival as genuine co-primary endpoints from the start: one bar for each audience, in one trial. [8] The sarcoma trial ANNOUNCE registered overall survival as primary alongside the fullest patient-relevant battery you could ask for, the EQ-5D-5L utility index, the EORTC QLQ-C30 and the Brief Pain Inventory. [9] Across a set of fifteen pivotal trials, the pattern is almost mechanical: every single-arm study registered ORR as its sole primary, and every randomised confirmatory trial registered a survival endpoint. The design tells you which bar it was built to clear.
Now the trial that shows why "we will add it later" fails. In the melflufen study HORIZON, quality-of-life instruments were bolted on by protocol amendment partway through, and at the data cutoff only 23 of 157 patients had QoL data, added too late to interpret and therefore added for nothing. [10] That is the whole argument compressed into one trial: the payer-relevant endpoint is a decision that expires when the protocol locks.
But here's the thing: building the payer-relevant endpoint in doesn't guarantee a good answer, and pretending otherwise would be a sales pitch. ANNOUNCE was well-designed by exactly the standard I am advocating, and olaratumab still failed to show a survival benefit, becoming the first conditional marketing authorisation the EMA ever revoked. [11] Collecting the endpoint doesn't promise a win; you run the trial to find out honestly, rather than build a reimbursement case on a foundation no regulator ever required to be true.
I have watched a team learn this the expensive way. Years ago I worked with a small oncology company that locked its pivotal around ORR alone, cleanly, on regulatory advice, and then sat down with HTA reviewers who wanted a utility measure and a survival read the protocol had never been built to produce. Nobody had done anything wrong. They had just never been asked the second question until it was too late to answer it.
The strongest counterargument is a good one, so here it is at full strength. For a lean biotech, accelerated approval on a surrogate is the entire point. You reach the market, book revenue, and fund the confirmatory work from sales. Loading the pivotal with overall survival and a PRO battery adds patients, follow-up and cost you do not have. And Europe's conditional-marketing-authorisation record is comparatively calm: of 30 CMAs granted over ten years, 11 converted to standard authorisation and only 2 were withdrawn, both influenza vaccines pulled for commercial rather than clinical reasons. [12] So why load the pivotal?
Three answers. First, the two endpoints are rarely mutually exclusive to measure. You are usually not adding a second trial, you are adding an instrument and a longer follow-up window to the trial you are already funding, and an EQ-5D-5L costs a rounding error against a market withdrawal. Second, "fund it from sales later" is itself the gamble, given that roughly 40% of these confirmatory trials never verify the benefit. Third, do not borrow Europe's benign CMA record to price the oncology accelerated-approval risk; they are different pathways with different odds, and the PARP inhibitors at the top of this piece are what the risky one looks like when the note comes due.
This decision is made once, and it is unmade the moment the protocol is final. So before that, three questions:
None of this is exotic. It is the ordinary discipline of a lean integrated evidence plan: deciding, before you spend the money, which evidence each audience actually needs, then building one trial that answers both. The alternative is to optimise beautifully for the regulator's yes, and then open the second invoice.
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.
[1] Shahzad M, et al. (2024). "Regulatory histories of recently withdrawn ovarian cancer treatment indications of 3 PARP inhibitors in the US and Europe: lessons for the accelerated approval pathway." J Pharm Policy Pract. PMID: 38841118. https://pubmed.ncbi.nlm.nih.gov/38841118/
[2] IQWiG. Rapid Report A10-05: "Validity of surrogate endpoints in oncology." Institut für Qualität und Wirtschaftlichkeit im Gesundheitswesen (IQWiG). https://www.iqwig.de/en/projects/a10-05.html
[3] European Medicines Agency. "Guideline on the scientific application and the practical arrangements necessary to implement Commission Regulation (EC) No 507/2006 on the conditional marketing authorisation." EMA/CHMP/509951/2006 Rev.1 (FINAL). https://www.ema.europa.eu/en/documents/scientific-guideline/guideline-scientific-application-and-practical-arrangements-necessary-implement-commission-regulation-ec-no-5072006-conditional-marketing-authorisation-medicinal-products-human-use-falling_en.pdf
[4] Kanavos P, Visintin E, Angelis A. (2024). "Use of the ESMO-Magnitude of Clinical Benefit Scale to guide HTA recommendations on coverage and reimbursement for cancer medicines: a retrospective analysis." Lancet Oncology. PMID: 39637889. https://pubmed.ncbi.nlm.nih.gov/39637889/
[5] Liu ITT, Kesselheim AS, Cliff ERS. (2024). "Clinical benefit and regulatory outcomes of cancer drugs receiving accelerated approval." JAMA. PMID: 38583175. https://pubmed.ncbi.nlm.nih.gov/38583175/
[6] Tibau A, et al. (2025). "Factors in Time to Full Approval or Withdrawal for Anticancer Medicines Granted Accelerated Approval by the FDA." JAMA Network Open. PMID: 40136298. https://pubmed.ncbi.nlm.nih.gov/40136298/
[7] Naci H, et al. (2025). "Costs and benefits of early access to new cancer drugs through the US Food and Drug Administration's accelerated approval pathway: retrospective observational study and economic evaluation." BMJ Medicine. PMID: 41431468. https://pubmed.ncbi.nlm.nih.gov/41431468/
[8] Merck Sharp & Dohme. "KEYNOTE-240: Pembrolizumab versus best supportive care in previously treated advanced hepatocellular carcinoma." ClinicalTrials.gov: NCT02702401. https://clinicaltrials.gov/study/NCT02702401
[9] Eli Lilly and Company. "ANNOUNCE: A study of doxorubicin plus olaratumab in participants with advanced or metastatic soft tissue sarcoma." ClinicalTrials.gov: NCT02451943. https://clinicaltrials.gov/study/NCT02451943
[10] Oncopeptides AB. "HORIZON: A study of melflufen and dexamethasone in relapsed/refractory multiple myeloma." ClinicalTrials.gov: NCT02963493. https://clinicaltrials.gov/study/NCT02963493
[11] Schuster Bruce C, et al. (2019). "The use of validated and nonvalidated surrogate endpoints in two European Medicines Agency expedited approval pathways: A cross-sectional study of products authorised 2011–2018." PLoS Medicine. PMID: 31504034. https://pubmed.ncbi.nlm.nih.gov/31504034/
[12] European Medicines Agency. "Conditional marketing authorisation: report on ten years of experience at the European Medicines Agency." EMA/471951/2016. https://www.ema.europa.eu/en/documents/report/conditional-marketing-authorisation-report-ten-years-experience-european-medicines-agency_en.pdf