Why “rare disease” is a regulatory label, not a patient reality

Regulators define a rare disease by how many people have it, and the answer depends on who is counting. One review of Bayesian methods in rare disease cites about one in 1,500 people in the US, about one in 2,500 in Japan, and fewer than one in 2,000 in Europe. Definitions usually depend on prevalence thresholds and vary between locations.

So, the same condition can count as rare in one system and not in another, even though its biology has not changed (Adachi, 2023; Abozaid, 2025; Krishna, 2026). Hence its label as a regulatory category. This label is not a description of what living with the disease, or running a trial for it, actually involves. It is only useful for incentives and regulation, and a weak starting point for understanding how patients move through diagnosis, care, and trial participation.

Patients complain of long diagnostic journeys, fragmented care, travel, lost work, and strain on their families (Mease, 2023; Faye, 2024; Witt, 2023). A prevalence figure captures none of these. For trial design, heterogeneity is as important as rarity. Conditions that share the same label can differ in age of onset, progression, severity, and diagnostic pathway, so, prevalence tells little about inclusion criteria, site strategy, or outcome selection (Adachi, 2023; Phillips, 2024).

A working group of the American Statistical Association's Biopharmaceutical Section lists slow progression, poorly understood natural history, varied phenotypes and genotypes, and a lack of suitable surrogate endpoints as recurring problems. Using the label as design shorthand can also insert false assumptions. Rigid criteria, treatment-naïve requirements, and standard templates can shrink pools that are already small (Adachi, 2023; Khazen, 2025).

Suzanne Vyvoda, senior vice president of development operations at Nkarta Inc, thinks the people who run studies should be involved much earlier. "Too often, operations is handed a finished protocol and told to make it work," she says. "The teams accountable for running a study in the real world should be shaping it while it is still being designed."

Her argument moves the central question from how many patients exist to how many are diagnosed, reachable, and eligible when the study needs them.

Finding the patients

Prevalence can tell a sponsor that a population is small, but not where it is. Some patients are concentrated at a few specialist centres. Some have no confirmed diagnosis yet. Others are treated outside the centres where research takes place.

Vyvoda gives an example. Take a progressive paediatric disease with a few hundred diagnosed children in a country. An enrolment target of 40 or 50 looks demanding but achievable against that total. But most of those children were diagnosed after a long journey and are looked after at five or six centres of excellence, so there is no wide net to cast. The whole trial rests on a handful of sites and on reaching families who may live several states away. Travel, accommodation, time off work, and repeated assessments all affect whether a family can take part, and in a small study the loss of a single participant can change the evidence that comes out of it.

"The trial has to go to them," Vyvoda says.

Patients make the same point. They describe travel as a major burden, and some say trial procedures do not fit around daily life or work (Mease, 2023; Mellerio, 2022; Baffour, 2026). A survey of 1,436 parents across 34 European countries found that most were strongly willing to accept decentralised trial options, wherever the study was run.

Trust matters as much as logistics. In interviews about a start-up delay in a trial for pantothenate kinase-associated neurodegeneration, some participants felt researchers had different interests from theirs, and the two groups disagreed about how much background information was needed. Honest, transparent communication should be adjusted to what participants want.

For Vyvoda, this means working with patient advocacy groups, placing sites where patients already receive care, and considering decentralised or hybrid approaches where they suit the disease. The trial should be designed around how patients reach care, not around where the study happens to be run.

Endpoints that fit the disease

Endpoint selection is where the regulatory label separates from clinical reality. Meaningful endpoints need good natural-history knowledge, which is often incomplete in rare diseases, and with samples this small the choice carries unusual weight (Mellerio, 2022; Khazen, 2025; Cox, 2018). Progression can be slow or highly variable, so an endpoint measured over a fixed window can miss a real effect. Borrowing endpoints from common diseases also carries its own risk. For example, in epidermolysis bullosa, time to wound closure can miss the clinically central problem of recurrent breakdown (Mellerio, 2022).

Patients want outcomes that reflect daily life, and recent methodological work argues for composite endpoints and patient-reported outcomes when a single endpoint misses multidimensional benefit (Mease, 2023; Verbeeck, 2023). Expert sessions held by an ECNP working group on neuropaediatric rare diseases found the same difficulties across conditions and proposed a shared "backbone" approach that lets each disease team tailor its assessments and study design.

Real-world data can help where randomisation may prove difficult. Three FDA case examples (cerliponase alfa, asfotase alfa, and uridine triacetate) show registry, chart-review, and literature data supporting regulatory decisions. An external control group must be comparable with the treated group, and that selection bias and confounding are major concerns without randomisation. The disease trajectory, the available evidence, and the endpoint all have to shape the design from the onset.

Start with the patient

Registries, longitudinal natural-history data, and patient-reported data can show where patients are diagnosed and treated, and support endpoint development (Adachi, 2023; Cribbs, 2025). Small populations leave little or no room for avoidable barriers, so engagement has to go beyond commenting on a finished protocol. In an ethnographic study of idiopathic pulmonary fibrosis at a UK specialist centre, interviews with 32 patients and 19 staff revealed that some patients found trial information confusing and trial practices frustrating, and saw no chance to influence design. Working through contract research organisations and patient organisations leaves out these patients who do not engage them.

Site selection should start from patients, instead of investigators: where they are diagnosed and treated, and which centres have the expertise and relationships to support a study. The protocol then comes after this. "The place to start is the disease and the patients, not the protocol template," Vyvoda says.

This also means factoring in differences between patients, instead of leaving them out; choosing endpoints that can be measured reliably and considering external or real-world comparators early, when a randomised design may not be practical.

Flexibility without lower standards

Adapting a design for a rare disease does not reduce the scientific bar, and regulators agree that conventional designs may not always fit these populations. The literature supports adaptive designs, crossover or open-label extensions in some settings, external controls drawn from registries or natural-history cohorts, and early conversations with regulators about novel endpoints (Mellerio, 2022; Wu, 2020; Khazen, 2025).

Real trials also have to survive ethics committees. EPISTOP tested preventive vigabatrin in infants with tuberous sclerosis: committees at four centres did not approve randomisation. The trial ran open-label at those four and randomised at the other six. A re-analysis adjusting for selection bias gave a hazard ratio of 2.89 against the original 2.91.

Vyvoda says a chosen design has to be workable in practice. That means running the protocol consistently across a small number of specialist sites, collecting natural-history data to interpret treatment effects, choosing endpoints built for the question, rather than a template, and talking to regulators early. "The real question is not which clever design to choose. It is whether the chosen design can actually be executed with rigour once it leaves the strategy deck. This is where approaches usually succeed or fail."

The label is still important, however…

"Rare disease" can remain a regulatory label for prevalence. That is what it was built to do. But it is not a patient reality, and a trial designed as though it were one will keep missing where patients are, what they can give to a study, and which outcomes matter to them.

Clinical trials have to be built around the disease, the care pathway, and the patient burden that actually exists.

About the author

Deborah Sabinus-Chinonso, MPH, is a medical writer translating complex clinical research and health innovation into clear, evidence-based content for healthcare, healthtech, and the life sciences.