Munim Technologies

Research/Medicine

One in five trials excludes disabled people, and the share is rising

Under rules published before the data were seen, 20.1% of 450,052 interventional trials on ClinicalTrials.gov exclude people in at least one of six disability categories. The yearly share has roughly doubled since the early 2000s.

Oct 14, 2026 · 7 min read

We read the eligibility criteria of every interventional study on ClinicalTrials.gov and counted how many exclude people by disability. Under rules we published before any registry result was seen, 20.1% of the 450,052 trials with an exclusion section shut out people in at least one of six categories: cognitive, consent capacity, visual, hearing, communication or mobility. The share has grown. Among trials that started between 2000 and 2008, 11.5% to 14.1% had such an exclusion each year; among 2025 starts it was 24.9%. Trials are where the evidence for treatments comes from. When a group is routinely left out, the evidence says less about them.

20.1%

of 450,052 trials exclude people in at least one disability category

17.9%

when exclusions that match the trial's own condition are left out

24.9%

of trials that started in 2025, against 11.5% to 14.1% for 2000 to 2008 starts

38.3%

of trials led by other US federal agencies, mostly the VA

The question

Earlier studies coded trial protocols by hand. DeCormier Plosky and colleagues read 97 protocols in four disease areas; McDonald, Schwartz and Sabatello read NIH-funded trials for exclusions of adults with intellectual disability. Both found exclusion was common. Neither could say how common it is across the whole registry, or whether it is changing. We wanted an open, repeatable count that anyone can re-run and argue with.

What we did

The census pulls every interventional study through the ClinicalTrials.gov API, splits out the exclusion criteria and flags items that exclude people by disability, from dementia and MMSE cut-offs to "unable to provide informed consent", blindness, deafness, aphasia and wheelchair use. It also marks whether a flagged exclusion matches the trial's own condition, since excluding vision loss from a glaucoma trial is not the same as excluding it from a diabetes trial. The v0.1.1 rules were written and published before the data, fetched on 2026-10-14, were seen.

To check the rules, 720 random flagged items were read one by one, 60 per category. Precision was 95% for cognitive, 100% for consent capacity, 95% for hearing, 97% for communication, 90% for mobility and 75% for visual, where the errors were eye findings in ophthalmology trials. A separate random sample suggests the rules miss some exclusions, so the registry figures are more likely too low than too high.

What we found

Cognitive exclusions are the most common, followed by consent capacity. Sensory and physical exclusions are each rarer, but together they appear in 7.0% of trials.

Share of trials excluding people, by disability category

450,052 interventional trials with an exclusion section, pre-registered v0.1.1 rules. A trial can fall in several categories. Source: results/v0.1.1/summary.json

The yearly share climbs steadily from about 2008 onward. Not every exclusion grew: substance-use exclusions fell over the same years, from 18.7% of 2000 starts to 15.8% of 2025 starts. Trials starting in 2026 read 27.5%, but they are still being registered, so the chart stops at 2025.

Share of trials with a disability exclusion, by start year

Any disability exclusionSubstance-use exclusion
01020200020052010201520202025Start yearShare of trials (%)
Pre-registered v0.1.1 rules. 2026 starts (27.5%, still being registered) are left out. Source: results/v0.1.1/by_year.csv

A rising line could simply mean criteria got longer, wording changed, or the mix of trials shifted toward behavioural and device studies, which exclude more. We tested this with the registry's own search, counting disability phrases against non-disability control phrases by start year. Under a pre-registered test, in trials with no phase (behavioural, device and procedure studies) the cognitive, hearing and visual families rose more than all nine kept controls (p = 0.0045). In phased trials, mostly drug trials, they sat in the middle of the controls (p = 0.45). So the rise looks specific to disability in no-phase trials and is not shown in the others.

Sponsor matters. Trials led by industry exclude least; trials led by other US federal agencies exclude most, at 38.3% of 4,174 trials. Two-thirds of those are led by the Department of Veterans Affairs. Matched on condition in a pre-registered test, VA-led no-phase trials name a cognitive wording 2.2 times as often as other trials of the same conditions.

Share of trials with a disability exclusion, by lead sponsor class

Sponsor classes as recorded by ClinicalTrials.gov; classes with fewer than 1,000 trials are left out. Source: results/v0.1.1/by_sponsor.csv

The telephone

Reading VA hearing exclusions one by one, we noticed several shut people out because the study runs by phone. We pre-registered a test across all sponsors: do no-phase trials delivered by telephone name a hearing wording more often than other trials of the same condition? They do, 2.45 times as often, while naming every non-disability control less often. Visual and cognitive wordings rise too, so the finding is that telephone trials name sensory and cognitive disability much more often, with hearing first. Leaving out VA trials gives 2.41.

Telephone-delivered trials: times as often as other trials of the same condition

Mantel-Haenszel ratios over six condition strata; 7,620 telephone trials against 202,729 other no-phase trials. Selected controls shown; all eight controls range from 0.24 to 0.61. Source: results/phone-check/PHONE-MATCHED-RESULT.md

The known fixes are cheap: captioned telephone, relay services, text or chat, and video with captions. None of the VA hearing-exclusion trials we read offered one in its registry entry.

What it means

A count does not show that any single exclusion is unjustified. Some trials have good reasons. But a registry-wide figure that rises year on year, concentrated in some sponsors and in remote delivery, points to where review effort would pay off. The repository includes a browser check that flags each disability exclusion in a study's criteria with an inclusive alternative to consider, and a draft guide for study teams and review boards.

More research