One day. Two single-arm trials. Two opposite regulatory decisions.

On 17 September 2026, the US Food and Drug Administration approved Fayuvi. On the same day, the European Medicines Agency confirmed its recommendation to refuse Xervyteg.[1,2]

Both medicines were supported by open-label, single-arm trials. Everybody received the medicine. Neither pivotal trial randomly assigned patients to a control group.

One programme supported approval. The other did not.

This is not a simple story about FDA accepting single-arm trials while EMA rejects them. The medicines treated different diseases, measured different outcomes and were supported by different evidence packages.

But the two decisions expose the same question:

What would have happened without the medicine?

A single-arm trial cannot prove causality on its own

A single-arm trial can show that patients improved, remained stable or experienced adverse events after receiving a medicine.

It cannot show on its own whether the medicine caused those outcomes. Patients may have improved because of other treatments, supportive care or the natural course of the disease.

The missing outcome without the medicine is called the counterfactual.

In a randomised trial, the control group estimates that outcome within the same study. In a single-arm trial, it has to be supported by external evidence, such as natural history studies, previous trials, registries or routinely collected healthcare data.[3]

That evidence must represent similar patients, receiving a clearly defined alternative treatment or standard care, measured from the same starting point and using compatible outcomes.

A single-arm trial is only as persuasive as its counterfactual. The external evidence must establish what would probably have happened without the medicine and make alternative explanations for the observed improvement sufficiently unlikely.

Fayuvi used natural history to estimate the untreated outcome

Fayuvi is a one-time gene therapy for mucopolysaccharidosis type IIIA, also known as Sanfilippo syndrome type A. The disease causes heparan sulphate to accumulate in the body and leads to progressive loss of cognitive function.[1,4,5]

FDA approved Fayuvi for paediatric patients with preserved neurodevelopmental function.[1]

The pivotal study, ABT-001, enrolled 28 children. Twenty-two received the intended commercial dose, and 17 were included in the primary efficacy population.[1]

There was no randomised control group. Instead, the 17 treated children were compared with 27 untreated children from two prospective natural history studies.[1,4,5]

Those studies had followed how MPS IIIA progressed without a disease-modifying treatment. They documented a recognisable pattern of developmental gain, plateau and subsequent regression. Importantly, they also measured cognition using the same Bayley-III instrument used in the Fayuvi programme.[4,5]

The analysis compared change in cognitive raw score between 24 and 60 months of age. It adjusted for chronological age and used weighting based on baseline age to make the treated and untreated groups more comparable.[1]

The estimated mean cognitive score increased by 16.0 points with Fayuvi and fell by 7.6 points in the untreated cohort. The estimated difference was 23.5 points, with a 95% confidence interval from 17.2 to 29.9.[1]

The external patients were not a perfect substitute for randomisation. The dataset was small, and adjustment cannot remove every measured or unmeasured difference between separately recruited groups.

But FDA considered the comparison sufficiently informative for the population included in the final indication.[1]

FDA identified clinical and biological evidence pointing in the same direction

FDA’s review did not rely on the modelled comparison alone.

The review reported that analyses based on the observed cognitive scores also showed the treated and untreated trajectories separating over time. Cerebrospinal fluid heparan sulphate fell after treatment and remained lower for at least 24 months. Studies in mice showed expression of the delivered gene, activity of the missing enzyme and reductions in accumulated glycosaminoglycans.[1]

These are FDA’s stated sources of supporting evidence. FDA described the clinical study, together with confirmatory evidence, as providing substantial evidence of effectiveness.[1]

FDA did not state that any one supporting result was decisive. Our interpretation is that their value came from their alignment. Together, the external clinical comparison, the observed cognitive trajectories, the biomarker response and the animal evidence built a coherent story.

This is where biological evidence and biological plausibility can strengthen a treatment-effect claim. They do not replace the missing clinical comparison. They make the clinical result more coherent and alternative explanations less convincing.

FDA limited the claim to the population supported by the evidence

The evidence did not support every possible patient group.

FDA found the results in ten children outside the primary early-stage population difficult to interpret. Baseline disease differed, assessments were missing, numbers were very small and suitable natural history matches were limited.[1]

The approved indication therefore focuses on children with preserved neurodevelopmental function.[1]

This matters. FDA accepted uncertainty, but it did not treat uncertainty as unlimited. The final claim followed the population in which the clinical outcome and its untreated trajectory could be interpreted.

Xervyteg exceeded a historical benchmark but could not isolate its effect

Xervyteg contained pooled faecal microbiota from healthy donors. It was intended for adults with gastrointestinal acute graft-versus-host disease after corticosteroids and ruxolitinib had not worked well enough.[2]

The pivotal ARES trial was open label and had no control group. The published analysis included 66 treated patients. At day 28, 62% had a gastrointestinal response, with a 95% confidence interval from 49% to 74%. This was statistically above a prespecified historical response threshold of 22%.[6]

The public EMA summary describes the application population as 67 adults. It concluded that the main study could not reliably show whether Xervyteg was safe and effective or determine the size of any benefit.[2]

EMA highlighted three connected problems. There was no comparison group, the study was open label and patients received other treatments at the same time.[2]

Its conclusion was direct:

“It was difficult to determine whether the observed effects were due to the medicine or to other factors.”[2]

The refusal does not show that Xervyteg was ineffective. It shows that EMA did not consider the available evidence sufficient to attribute the observed responses to Xervyteg or establish their magnitude.

The difference was the evidence around the single arm

The public sources allow a direct comparison of the evidence structures. They do not support a direct comparison of the medicines themselves.

Component	Fayuvi	Xervyteg
Pivotal evidence	Open-label, single-arm study. The primary efficacy analysis included 17 treated children.[1]	Open-label, single-arm study. EMA described 67 adults in the application; the published ARES analysis included 66 treated patients.[2,6]
Estimate of what would happen without the medicine	Patient-level longitudinal data from 27 untreated children in two prospective natural history studies.[1,4,5]	A prespecified historical response threshold of 22%. EMA stated that the main study had no comparison group.[2,6]
Main clinical result	An estimated 23.5-point advantage in cognitive raw score change from 24 to 60 months, with a 95% confidence interval from 17.2 to 29.9.[1]	A 62% gastrointestinal response rate at day 28, with a 95% confidence interval from 49% to 74%, compared with the 22% threshold.[6]
Other supporting evidence	Observed cognitive trajectories, sustained reduction in cerebrospinal fluid heparan sulphate and animal evidence supporting the mechanism.[1]	Responses were also reported in the earlier HERACLES study, alongside biological changes in responders. Its authors acknowledged that delayed effects of earlier treatments could not be excluded.[7]
Regulatory interpretation	FDA considered one adequate and well-controlled study plus confirmatory evidence sufficient for traditional approval in a defined paediatric population.[1]	EMA concluded that the evidence could not reliably separate Xervyteg’s effect from other treatments or the disease course, and confirmed its recommendation to refuse.[2]

The clearest contrast is not single arm versus single arm. It is a longitudinal untreated cohort versus a fixed historical percentage, and whether the external evidence adequately addressed other explanations for what happened.

A historical percentage is not automatically a counterfactual

The 22% threshold gave the ARES analysis a reference point. But a fixed percentage does not show the outcomes of comparable patients receiving clearly defined alternative care at the same time and measured in the same way.

It also does not automatically carry forward the uncertainty in the source estimate, differences in patient prognosis or changes in clinical practice.[3]

A separate retrospective study called CHRONOS later described 59 patients treated after corticosteroids and ruxolitinib had failed. It reported a day-28 gastrointestinal response rate of 37%, with a 95% confidence interval from 25% to 51%.[8]

CHRONOS was descriptive. It included heterogeneous third-line treatments and did not make a formal adjusted comparison with ARES. The public EMA record does not establish that it was submitted as an external control or rejected by the agency.[2,8]

Its value here is narrower. It shows why an expected outcome cannot be reduced to one universal historical response rate. The estimate can change with the patients, care pathway, outcome assessment and data source.

Concurrent treatment left another credible explanation

The counterfactual for Xervyteg was not simply “no Xervyteg”. It also had to specify what other treatment patients would receive.

EMA explicitly highlighted concurrent therapies as a reason the effect could not be isolated.[2] The earlier HERACLES publication had similarly acknowledged that delayed benefits from previous treatments could not be excluded.[7]

This is particularly difficult in a severe condition where patients receive several treatments and supportive care while their health can change quickly.

An external comparison must therefore align more than the disease name. It needs a common starting point, similar previous treatment, compatible concomitant care, comparable outcome definitions and sufficient information on the factors that predict response and survival.[3,9]

Biological plausibility can strengthen a comparison, not replace it

The Xervyteg programme also produced biological evidence. HERACLES reported changes in microbiome richness and other biological measures among responders.[7]

That can make a treatment effect more plausible. But an association within treated patients cannot establish what would have happened without treatment. The biological change could be a cause of improvement, a consequence of improvement or related to another part of care.

The same principle applies to Fayuvi. Its biomarker and animal results strengthened the interpretation because they sat alongside an external clinical comparison. They were not a substitute for it.[1]

Build the counterfactual as early as possible

Ideally, external evidence planning starts while the pivotal trial can still be changed. The search may reveal which outcomes are reported, whether patient-level data are accessible, which prognostic factors need collecting and whether the same starting point and follow-up can be created.[3,9,10]

But the end of the pivotal trial is not the end of the opportunity.

If the trial is already complete, the next best step is to map the available natural history, trial, registry and real-world evidence. Then run a formal feasibility assessment and finalise the external-comparison protocol and statistical analysis plan before estimating the comparative result.

The data source and method should be chosen because they answer the clinical question credibly, not because they produce the most favourable result.[9]

This work should establish:

• the patients and treatment setting the comparison must represent;
• what patients would receive without the medicine;
• a common starting point for follow-up;
• compatible outcomes and assessment times;
• access to the prognostic factors and treatment-effect modifiers needed for adjustment; and
• how uncertainty and alternative assumptions will be tested.

These questions matter again at reimbursement. An external comparison accepted by a regulator may still not compare the medicine with the treatment funded in a particular health system. NICE’s real-world evidence framework similarly emphasises transparent planning, data suitability and methods that address the decision problem.[11]

Evidax TRACE maps the relevant published studies, natural history evidence, outcomes and potential adjustment variables. Evidax Comparison Blueprints then establish which external comparisons the available data can credibly support, the appropriate method and the limits of the resulting claim.

The purpose is not to make every single-arm result look comparative. It is to establish whether the missing counterfactual can be estimated well enough to support the decision.

Take home

• A single-arm trial shows what happened after treatment. External evidence is needed to judge what treatment caused.
• A single-arm trial is only as persuasive as its counterfactual.
• Fayuvi was supported by a longitudinal untreated cohort, aligned clinical outcomes and independent biological evidence. FDA still restricted the indication to the population it could interpret.
• Xervyteg exceeded a historical response threshold, but EMA concluded that other treatments and the disease course remained credible explanations for the observed responses.
• A historical percentage is not automatically a control group. The patients, treatment pathway, starting point, outcomes and uncertainty must align.
• Biological evidence can strengthen treatment attribution. It cannot replace a credible clinical comparison.
• Build the external evidence as early as possible. If the pivotal trial is complete, assess feasibility and prespecify the external comparison before estimating its result.

References

1. US Food and Drug Administration. Summary Basis for Regulatory Action: FAYUVI. 17 September 2026. https://www.fda.gov/media/195198/download

2. European Medicines Agency. Refusal of the marketing authorisation for Xervyteg (allogeneic faecal microbiota, pooled): re-examination confirms refusal. EMA/210795/2026. 18 September 2026. https://www.ema.europa.eu/en/documents/smop-initial/questions-answers-refusal-marketing-authorisation-xervyteg-allogeneic-faecal-microbiota-pooled-re-examination-confirms-refusal_en.pdf

3. European Medicines Agency. Reflection paper on establishing efficacy based on single-arm trials submitted as pivotal evidence in a marketing authorisation application. EMA/CHMP/458061/2024. 9 September 2024. https://www.ema.europa.eu/en/documents/scientific-guideline/reflection-paper-establishing-efficacy-based-single-arm-trials-submitted-pivotal-evidence-marketing-authorisation-application_en.pdf

4. Shapiro EG, Nestrasil I, Delaney KA, et al. A prospective natural history study of mucopolysaccharidosis type IIIA. Journal of Pediatrics. 2016;170:278-287. https://doi.org/10.1016/j.jpeds.2015.11.079

5. Wijburg FA, Aiach K, Chakrapani A, et al. An observational, prospective, multicenter, natural history study of patients with mucopolysaccharidosis type IIIA. Molecular Genetics and Metabolism. 2022;135:133-142. https://doi.org/10.1016/j.ymgme.2021.12.002

6. Malard F, Sanz J, Mediavilla C, et al. MaaT013 for ruxolitinib-refractory acute graft-versus-host disease with gastrointestinal involvement: results from the ARES phase III trial. Blood. 2025;146(Suppl 1):817-818. https://doi.org/10.1182/blood-2025-817

7. Malard F, Loschi M, Huynh A, et al. Pooled allogeneic faecal microbiota MaaT013 for steroid-resistant gastrointestinal acute graft-versus-host disease: a single-arm, multicentre phase 2 trial. eClinicalMedicine. 2023;62:102111. https://doi.org/10.1016/j.eclinm.2023.102111

8. Clausen J, Pérez Simón JA, Carré M, et al. Clinical outcomes of third-line therapy for acute graft-versus-host disease with gastrointestinal involvement after steroids and ruxolitinib failure. Bone Marrow Transplantation. 2026;61:577-583. https://doi.org/10.1038/s41409-026-02825-0

9. US Food and Drug Administration. Considerations for the design and conduct of externally controlled trials for drug and biological products: draft guidance for industry. February 2023. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/considerations-design-and-conduct-externally-controlled-trials-drug-and-biological-products

10. US Food and Drug Administration. Rare diseases: natural history studies for drug development. Draft guidance for industry. March 2019. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/rare-diseases-natural-history-studies-drug-development

11. National Institute for Health and Care Excellence. NICE real-world evidence framework. Corporate document ECD9. Published 23 June 2022. https://www.nice.org.uk/corporate/ecd9