What 77% median exclusion means for trial evidence

A trial's eligibility criteria exist for good reasons: participant safety, a clean enough population to detect a treatment effect, manageable heterogeneity in the data. A systematic review covering 305 trials across 31 physical conditions, drawn from 50 studies, put an actual number on how much those criteria exclude in practice, and the figure is larger than most conversations about trial generalisability tend to assume.

The median trial excludes more than three-quarters of eligible patients

The review found a median exclusion rate of 77.1%, with an interquartile range of 55.5% to 89.0%. Read plainly: a typical trial in this review excluded more than three out of every four people who actually had the condition being studied. Nearly a quarter of trials excluded over 90% of eligible patients. More than half excluded at least three-quarters. Four out of five trials excluded at least half.

These aren't figures from unusually narrow or exotic trial designs. They're drawn from studies of common chronic conditions, and the condition-specific breakdown makes the scale concrete: asthma trials excluded a median of 96.0% of eligible patients, COPD trials 84.3%, hypertension trials 83.0%, and type 2 diabetes trials 81.7%. For conditions this prevalent and this central to routine clinical practice, the population actually studied in a typical trial represents a small, specifically filtered slice of who lives with the condition in the real world.

The reported figure is likely a floor, not a ceiling

The review's authors added an important caveat that pushes the real picture further in the same direction: most of the included studies applied only a subset of a trial's full eligibility criteria when calculating exclusion rates, meaning the true exclusion rate is likely higher than what's reported. If a trial's actual protocol lists fifteen exclusion criteria and a given study only had data to assess exclusion against six of them, the calculated exclusion rate reflects a partial picture, and the real figure, accounting for every criterion actually applied, would almost certainly be higher still.

That's a meaningful methodological point in its own right: even a systematic review specifically designed to quantify this problem couldn't fully capture its true scale, because the underlying data needed to do so, complete exclusion tracking against every actual criterion, generally isn't available or reported.

What was actually driving the exclusions

The review identified age, comorbidity, and co-prescribing as the most commonly cited exclusion criteria. That's a specific and important finding, because it points at exactly the population most likely to be affected in practice: older patients, people managing more than one condition simultaneously, and people already on other medications, precisely the population that shows up most often in routine clinical care for chronic conditions like the ones studied here.

Meanwhile, factors like life expectancy or functional status, which plausibly matter enormously for whether a treatment's benefits and risks actually apply to a given patient, were largely unexamined as exclusion criteria in the reviewed trials. That's a gap between what trials are excluding on and what might actually be the more clinically meaningful basis for exclusion.

Why this connects to a broader argument about what trials are actually for

A separate, more conceptual paper addresses the underlying question this data raises: why are restrictive eligibility criteria used so widely in the first place, and is that appropriate. Its argument is that restrictive selection borrows methods that belong to a different research purpose entirely. In a laboratory setting, homogeneous selection isolates a causal signal by removing confounding variation. In epidemiology, careful sampling ensures a population is representative. Applying either logic wholesale to a clinical trial, the argument goes, misses the point: a trial's purpose is to find out which patients actually benefit from a treatment, not to isolate a signal in a laboratory sense or represent a population in an epidemiological sense. Restrictive criteria, applied with that borrowed logic, systematically exclude exactly the patients, older, multi-morbid, on other medications, who most need to know whether a treatment will help them.

The paper's conclusion, that most clinical trials should be pragmatic and as inclusive as possible, is a direct response to the exclusion data: if 77% median exclusion, concentrated on age and comorbidity, is the norm, then a large share of the people a treatment will eventually be prescribed to were never actually represented in the evidence supporting it.

What this means for how a study designs its eligibility criteria

A few practical implications follow from combining these two pieces of evidence:

• Treat every exclusion criterion as something that needs its own justification, not a default inherited from a template. Given that age, comorbidity, and co-prescribing drove most exclusions in the reviewed trials, and that these are exactly the characteristics of the population most likely to receive the treatment afterwards, each such criterion deserves scrutiny for whether it's protecting participant safety or simply narrowing the study to a more convenient population.
• Track and report exclusion rates transparently, against the full set of criteria actually applied, given the review's finding that partial tracking likely understates the true scale of exclusion.
• Favour pragmatic, inclusive designs where participant safety allows, consistent with the conceptual argument that a trial's purpose is to determine who actually benefits, not to isolate a clean signal at the cost of real-world relevance.
• Consider whether functional status or other clinically meaningful factors would be more appropriate exclusion criteria than blanket age or comorbidity thresholds, given the review's observation that these more nuanced factors were largely unexamined in current practice.

The broader point is one about the credibility of the evidence base a study eventually contributes to. A trial that excludes 77% of the eligible population, disproportionately on the basis of age and comorbidity, is answering a narrower question than "does this treatment work" for the condition it's studying. It's answering that question for a specific, filtered subset, and the growing evidence on how large that filter actually is should inform how confidently that answer gets generalised to everyone else.