When Trial Endpoints Can’t Follow Patients In Clinical Care

By Léon van Wouwe, Clinical Innovation Director, Volv Global

Approved but Not Accessible: The Evidence Gap That Delays Market Access

Regulatory approval is only the first step towards reaching patients.

In December 2025, the FDA issued a complete response letter for Sanofi’s tolebrutinib in non-relapsing secondary progressive multiple sclerosis. The drug had received Breakthrough Therapy and Priority Review designations, and the Phase III HERCULES trial had shown a 31% delay in the onset of confirmed disability progression versus placebo ^[1]. Even so, the agency concluded that a favourable benefit-risk profile could not be established for any patient subpopulation ^[2].

The core objection extended beyond safety. The FDA questioned the endpoint itself, noting that there are no widely accepted criteria for defining disability accumulation independent of relapse activity, and that Sanofi could not distinguish patients with active from non-active disease because historical MRI data had not been collected in the trial design ^[2].

Four months later, in April 2026, the EMA’s CHMP adopted a positive opinion on the same drug for adults with secondary progressive multiple sclerosis without relapses in the last two years ^[3].

Two regulatory agencies had reached different verdicts on the same drug from essentially the same evidence package. This is not an isolated case; it points to a structural problem, that the evidence that wins regulatory approval is sometimes not what sustains market access. When the two diverge, patients are caught in the middle, with therapies that were designed and approved for them but that may not be available in their own market or may not be reimbursed.

The divergence runs deeper than a single drug or a single agency, and calls into question how trial endpoints are chosen and how payers and HTA bodies evaluate them, given the gap between what trials measure and what routine healthcare data can show in the long run.

Four links, four languages, one patient

The evidence chain that connects a therapy’s clinical promise to its long-term real-world value runs through four links, each with its own vocabulary, standards of proof and institutional incentives, and each designed and operated largely independently of the others.

The first link is the patient, and it concerns what matters to someone living with a disease, from symptoms and function to independence and the ability to work or care for a child. Clinical trials rarely measure these things directly, and when they do, the instruments are often too blunt or too infrequent to capture what patients themselves would call meaningful change.

The second link is the clinician, and it concerns what a physician observes, orders and records in routine care, including diagnosis codes, lab results, imaging, prescriptions and referrals. This data accumulates in health records and claims and is vast and longitudinal, yet it is structured for billing and clinical workflows rather than for measuring therapeutic outcomes.

The third link is the trial, and it concerns the endpoints chosen to demonstrate efficacy and safety in a controlled setting, such as overall survival, proteinuria reduction or six-minute walk distance. These satisfy marketing authorisation requirements and are optimised for statistical power within a defined follow-up window.

The fourth link is the payer, and it concerns what HTA bodies and reimbursement authorities need in order to assess added therapeutic value and cost-effectiveness, including patient-relevant outcomes, comparative effectiveness, long-term safety and evidence that benefit persists in routine clinical populations.

The disconnect between the third and fourth links can now be quantified. A retrospective study of 433 benefit assessments under Germany’s AMNOG process between 2011 and 2023 found that the G-BA accepted the primary endpoints used by the EMA in only 55% of cases ^[4]. In the rest, an endpoint that had been sufficient for European marketing authorisation was not accepted as patient-relevant.

The two agencies are answering different questions. The EMA asks whether an endpoint is scientifically valid and statistically powered, whereas the G-BA asks whether it directly reflects benefit for the patient across mortality, morbidity, quality of life and safety.

How widespread is the disconnect?

The gap between regulatory approval and payer acceptance is documented across geographies, therapeutic areas and evidence frameworks. Three patterns are now clear.

The first is that HTA bodies routinely reject the primary endpoints that regulators accepted. An earlier analysis across oncological, metabolic and infectious diseases ^[5] found inconsistent acceptance of primary endpoints as patient-relevant, even where they were standard for regulatory purposes, and a comparison of HTA rejections across seven OECD countries ^[6] found cancer or orphan indications, low-quality evidence and uncertainty to be significant predictors of rejection, which are precisely the areas where surrogate endpoints and small trial populations are most common.

EFPIA’s own analysis ^[7] names misalignment on evidence requirements as a root cause of access delays across Europe, and a 2026 review of twelve regulatory and HTA agencies ^[8] found persistent divergence in data-quality standards, methodological requirements and registry governance despite agreement on high-level principles. Without alignment, this fragmentation can lead to duplicated effort and delays in patient access.

The second pattern is that surrogate endpoints approved through accelerated pathways frequently fail to show clinical benefit in confirmatory trials. In a JAMA cohort study of 46 cancer drug indications granted accelerated approval ^[9], fewer than half demonstrated confirmed clinical benefit after more than five years of follow-up. Between 2011 and 2024, 25 cancer drugs granted accelerated approval were withdrawn from the US market ^[10], and a broader analysis across all therapy areas found a similar pattern ^[11].

The most public case is aducanumab. It received accelerated approval in 2021 for treating Alzheimer’s disease, on the basis of amyloid plaque reduction, then met a Medicare coverage restriction limiting its use to clinical trials ^[12], a negative opinion from the CHMP followed by withdrawal of its European application ^[13], and discontinuation by its manufacturer in 2024 ^[14]. The result was lost years of development and investment, and patients who were told that a therapy existed for their disease but could not access it.

Oncology is where the pattern is most visible, although it appears elsewhere too. In rare diseases the challenge is compounded by small trial populations and the difficulty of running long studies. An AMCP Partnership Forum white paper ^[15] warned that without adequate evidence FDA approval may not guarantee payer coverage, calling for an evidence ecosystem that draws on health records, registries and other post-marketing data to link surrogate endpoints to meaningful outcomes.

The third pattern is the consequence for patients and health systems. EFPIA’s 2024 W.A.I.T. report ^[16] found that the time from marketing authorisation to patient access continues to increase in Europe, with fewer than half of the medicines assessed available to patients on average, while in Latin America the average time from FDA or EMA approval to availability exceeds five years ^[17].

In response, payers across Europe increasingly use managed entry agreements, which tie reimbursement to predefined clinical outcomes or to real-world evidence generated after launch ^[18]. For orphan therapies across the EU4 and the UK, real-world evidence was pivotal for full or positive reimbursement in a substantial share of cases ^[19], and HTA reassessments tend to resolve endpoint uncertainty with trial data combined with real-world evidence, or with trial data alone, rather than with real-world evidence alone ^[20].

When payers hedge against evidence uncertainty rather than accept trial data at face value, they are telling pharma that evidence strong enough for regulators does not close the case for a reimbursement decision.

Mapping trial endpoints onto routine data in Pompe disease

If the evidence chain breaks because trial endpoints cannot be measured directly in routine care, the natural question is whether they can be approximated through proxy measures.

Findings presented at ISPOR Global 2026 in Philadelphia show how this can be done. In the largest machine learning study of Pompe disease to date, Volv Global and Sanofi showed that clinician-defined trial endpoints can be systematically mapped to routine claims data, and that machine learning can discover disease features beyond the prespecified endpoints ^[21].

Pompe disease is a rare genetic metabolic disorder in which enzyme deficiency causes progressive, irreversible damage to skeletal, cardiac and smooth muscle, including the muscles used for breathing. Enzyme replacement therapy has extended survival, so patients now live with long-term manifestations that were not fully assessed in the original trials and are not routinely captured in clinical practice.

The study drew a cohort of 3,549 patients from a large US administrative claims database covering more than 320 million patients, and applied three analytical layers.

The first layer validated the data itself. Almost every literature-based Pompe symptom was significantly more prevalent in the patient cohort than in a demographically matched control population ^[21], which supports the view that the right patients are represented and that these symptoms can be measured reliably in claims data.

The second layer mapped trial endpoints. Advanced semantic inference models mapped 46 of 67 predefined clinical endpoints to diagnosis, procedure and treatment codes in the claims data, and every mapped endpoint was significantly more prevalent in patients with Pompe disease than in controls, most clearly in lab biomarkers, pulmonary function tests and diagnostic workup ^[21]. The 21 endpoints that could not be mapped are informative too, because they mark the limits of what claims data can capture and they give HEOR teams a feasibility assessment for post-approval evidence planning.

The third layer went beyond prespecified frameworks. Data-driven analysis without prior clinical assumptions surfaced disease features that were significantly more prevalent in Pompe patients but were not among the predefined endpoints, including cardiovascular, respiratory and systemic manifestations, as well as patterns of healthcare utilisation ^[21]. The analysis was retrospective, and claims record that a test or diagnosis occurred rather than its result, so the next step could be prospective validation ^[21].

This matters because the natural history of a disease changes when therapy changes outcomes. Endpoints set when Pompe disease was rapidly fatal may no longer capture what matters to patients who now live for decades with managed disease, and the same applies wherever effective therapy shifts survival. By combining large-scale routine claims data with machine learning, the study could help close the gap between legacy endpoints and outcomes that reflect patients’ lived experience today.

For HEOR and value assessment teams, the Pompe methodology answers three questions that post-approval evidence strategies depend on, namely which trial endpoints can be tracked in routine healthcare data, which cannot, and what the data reveals that the trial framework missed. It does so at population scale, using data that already exists and adding no collection burden for clinicians or health systems.

Six new therapies, one surrogate endpoint and an unsolved diagnostic delay

IgA nephropathy is the most common primary glomerulonephritis worldwide and a leading cause of kidney failure in young adults, with up to 50% of patients progressing to kidney failure within 20 years of diagnosis ^[22]. Diagnosis requires a kidney biopsy and is routinely late, and a systematic review found that, in US data, a high proportion of patients are already in later stages of chronic kidney disease at diagnosis ^[23], which suggests referral often comes after substantial irreversible damage.

The disease had been managed conservatively for decades, and that changed abruptly. Between 2021 and 2026, the FDA approved or granted accelerated approval to at least six disease-modifying therapies (Tarpeyo, Filspari, Fabhalta, Vanrafia, Voyxact and Trutakna), most on the basis of proteinuria reduction as a surrogate endpoint. The rationale is sound, since early proteinuria reduction has been associated with a lower risk of kidney failure ^{[24, 25]}, but the association is not uniform across populations, and real-world validation in heterogeneous cohorts needs careful evaluation ^[24].

This creates a time-sensitive challenge for HEOR teams. Proteinuria reduction supported approval, but HTA bodies will ask whether it translates into delayed kidney failure, fewer dialysis starts and preserved quality of life in routine populations over decades, and the trials were not designed to generate that evidence within their follow-up windows.

Building that evidence will need real-world strategies of the kind the Pompe work demonstrates, although IgAN presents further obstacles. There was no specific ICD-10-CM code for the disease until October 2023, and because diagnosis requires a biopsy, many patients in claims and EHR databases carry non-specific nephritis or CKD codes, which means large-scale database studies could not directly find IgAN cases ^[23].

As with Pompe disease, effective therapy in IgAN is changing the natural history. Enzyme replacement therapy extended survival in Pompe and revealed manifestations that earlier trials had not captured, and in IgAN disease-modifying therapy may let patients live for decades with managed disease, changing which outcomes matter and how they can be measured.

Machine learning could help clinicians recognise IgAN-suggestive patterns in claims and EHR data ahead of biopsy referral, which could shorten the diagnostic delay. Endpoint mapping could show which proteinuria-related events and eGFR trajectories can be tracked in routine data, and unsupervised discovery could surface new manifestations as patients live longer with preserved kidney function.

IgAN therapies are approved and the HTA submissions are underway or imminent, yet the evidence infrastructure to support them is not yet in place.

Designing the evidence before the payer asks

The pattern described here is closely tied to how the pharmaceutical industry sequences its evidence generation, and it can be changed. Three shifts are needed.

First, endpoint strategy must integrate HEOR, regulatory and clinical development perspectives from Phase II onward, rather than from the point of HTA submission. The German data shows that waiting is a bet against yourself that fails nearly half the time ^[4], and when the G-BA rejects an endpoint that the EMA accepted, the result can be a non-positive benefit rating that drives price negotiations down. Retrospective evidence repair is likely to cost more than prospective endpoint alignment.

Second, the feasibility of measuring trial endpoints in real-world data should be assessed prospectively, as part of clinical programme design. The Pompe study shows this is achievable at population scale ^[21], by mapping clinician-defined endpoints to routine claims and EHR data and making explicit which can be tracked, which cannot, and where the data reveals disease features beyond the trial framework. That assessment belongs before Phase III protocol lock.

Third, machine-learning-driven discovery should complement prespecified endpoint frameworks, because the natural history of a disease changes when therapy changes outcomes. The cardiovascular, respiratory and systemic manifestations that emerged from the Pompe analysis were not among the predefined endpoints ^[21], yet they are the kind of evidence that long-term HTA reassessments and outcome-based reimbursement agreements will demand. In IgAN, where six new therapies may alter the disease trajectory, the same question is already urgent. What will these patients look like in ten years, and are we measuring the right things to know?

None of this needs to wait for better data infrastructure, because the data already exists. More than 320 million patients are represented in the US claims database used in the Pompe study alone, and electronic health records across Europe and beyond capture routine care every day. What is missing is the analytical layer that connects trial endpoints to these routine data, validates the connection and surfaces what the trials did not look for.

For HEOR and Value & Access teams, payers increasingly expect real-world evidence to sustain market access, so the question is whether the evidence infrastructure is in place before the payer asks for it, or whether it is assembled under pressure, after the fact, to defend a product whose endpoint strategy was designed for a different audience.

The patients who need these therapies are better served when the evidence is assembled before it is needed.

About the author

Léon van Wouwe is Clinical Innovation Director at Volv Global, a Swiss healthcare AI company that uses machine learning to map clinical trial endpoints into routinely collected real-world healthcare data and to find undiagnosed patient populations.

About the poster

The Pompe disease research described in this article was presented at ISPOR Global 2026 with Sanofi (Guso et al., ISPOR 2026, Poster RWD22).  You can download the poster here.

Links:

• Volv Global proprietary machine learning methodology
• Volv Global inTrigue – Finding more patients
• Volv Global for Value & Access, HEOR
• Volv Global for Clinical Development
• Volv Global Disease Directory: Pompe Disease
• Volv Global Poster: Real-World Endpoint Mapping and Identification of Evolving Phenotypes of Pompe Disease using Machine Learning
• Press release: Machine Learning Maps and Discovers Clinical Endpoints in Pompe Disease Using Real-World Data

References

1. Sanofi (2024) ‘Press release: tolebrutinib designated Breakthrough Therapy by the FDA for non-relapsing secondary progressive multiple sclerosis’, 13 December.
2. Manalac, T. (2026) ‘Sanofi’s MS drug tripped up by toxicities, unclear benefit, rejection letter reveals’, BioSpace, 6 January.
3. European Medicines Agency (2026) Cenrifki (tolebrutinib): summary of opinion, CHMP, 23 April 2026.
4. Müller, G. et al. (2025) ‘Primary endpoint acceptance and its association with benefit ratings and market access in German health technology assessments: a retrospective cohort study’, The Lancet Regional Health Europe, 55, article 101382. doi: 10.1016/j.lanepe.2025.101382.
5. Staab, T. et al. (2016) ‘Inconsistent approaches of the G-BA regarding acceptance of primary study endpoints as being relevant to patients: an analysis of three disease areas: oncological, metabolic, and infectious diseases’, BMC Health Services Research, 16, article 651. doi: 10.1186/s12913-016-1902-8.
6. Papadopoulos, F. et al. (2025) ‘Why drugs fail health technology assessment: a comparative analysis of health technology assessment rejections across seven OECD countries’, Health Economics, Policy and Law, 20(3), pp. 264-283. doi: 10.1017/S1744133125100212.
7. EFPIA and Charles River Associates (2024) The root causes of unavailability of innovative medicines and delays in access: shortening the wait. Brussels: European Federation of Pharmaceutical Industries and Associations.
8. Sagkriotis, A. (2026) ‘Advancing real-world evidence harmonization: lessons from the UK, EMA and global policy frameworks’, Journal of Comparative Effectiveness Research, 15(3), article e250183. doi: 10.57264/cer-2025-0183.
9. Liu, I.T.T., Kesselheim, A.S. and Cliff, E.R.S. (2024) ‘Clinical benefit and regulatory outcomes of cancer drugs receiving accelerated approval’, JAMA, 331(17), pp. 1471-1479. doi: 10.1001/jama.2024.2396.
10. U.S. Food and Drug Administration (2026) Withdrawn: cancer accelerated approvals.
11. Borin, M. et al. (2025) ‘OP74 Accelerated approval of drugs in the USA: a comparative analysis of United States Food and Drug Administration and Brazilian regulatory decisions’, International Journal of Technology Assessment in Health Care, 41(S1), p. S34. doi: 10.1017/S0266462325101293.
12. Centers for Medicare & Medicaid Services (2022) National coverage determination (NCD) 200.3: monoclonal antibodies directed against amyloid for the treatment of Alzheimer’s disease.
13. European Medicines Agency (2022) Withdrawal of application for the marketing authorisation of Aduhelm (aducanumab). Question-and-answer document, 22 April.
14. Biogen (2024) ‘Biogen to realign resources for Alzheimer’s disease franchise’, press release, 31 January.
15. Academy of Managed Care Pharmacy (2024) Addressing evidence gaps in the FDA accelerated approval: payer perspectives. AMCP Partnership Forum white paper. Alexandria, VA: AMCP.
16. EFPIA and IQVIA (2025) EFPIA Patients W.A.I.T. Indicator 2024 survey. Brussels: EFPIA.
17. IQVIA (2025) FIFARMA Patient W.A.I.T. Indicator 2025. IQVIA Value & Access white paper.
18. Fortrea (2026) Outcome-based agreements in Europe: opportunities, challenges, and future directions. Fortrea info sheet, 18 March.
19. Pushkarna, D., Kaur, R. and Fazeli, M.S. (2025) ‘Role of real-world evidence (RWE) in pricing and reimbursement decisions for rare disease in health technology assessment (HTA) submissions across EU4+UK’, poster HTA293, ISPOR Europe 2025, Glasgow, 9 to 12 November. Abstract in Value in Health, 28(12), p. S463. doi: 10.1016/j.jval.2025.09.2076.
20. Jaksa, A. et al. (2025) ‘Use of real-world evidence in health technology reassessments across 6 health technology assessment agencies’, Value in Health, 28(6), pp. 898-906. doi: 10.1016/j.jval.2025.02.012.
21. Guso, E. et al. (2026) ‘Real-world endpoint mapping and identification of evolving phenotypes of Pompe disease using machine learning’, poster RWD22, ISPOR 2026, Philadelphia, 17 to 20 May. Abstract in Value in Health, 29(S6).
22. Caster, D.J. et al. (2023) ‘Clinicopathological characteristics of adult IgA nephropathy in the United States’, Kidney International Reports, 8(9), pp. 1792-1800. doi: 10.1016/j.ekir.2023.06.016.
23. Kwon, C.S. et al. (2021) ‘A systematic literature review of the epidemiology, health-related quality of life impact, and economic burden of immunoglobulin A nephropathy’, Journal of Health Economics and Outcomes Research, 8(2), pp. 36-45. doi: 10.36469/001c.26129.
24. Lorenzi, M. et al. (2025) ‘Association between surrogate endpoints and clinical outcomes in immunoglobulin A nephropathy: a systematic literature review’, Advances in Therapy, 42(10), pp. 4824-4865. doi: 10.1007/s12325-025-03331-3.
25. Maillard, N. et al. (2026) ‘Early change in proteinuria as a surrogate endpoint in studies of IgA nephropathy: an updated patient-level meta-analysis and discussion of appropriate methodology’, Advances in Therapy, 43(1), pp. 200-213. doi: 10.1007/s12325-025-03402-5.

Photo by Dorrell Tibbs on Unsplash