01 · The Question
What If the People Who Most Need the Evidence Are Missing From the Study?
Imagine a treatment intended for a condition common among older adults, but the trial recruits mostly younger participants. A digital health system is designed for broad public use, but people with disabilities cannot use the research interface. A study of healthcare access excludes people who cannot complete materials in the researchers' preferred language.
The studies may still produce statistically valid results for the people actually enrolled.
But can systematic exclusion itself become an ethical problem?
Yes, in some circumstances. When a study repeatedly or structurally excludes people central to the research question, the concern extends beyond statistical representativeness. The design may distribute research opportunities unfairly, generate evidence poorly suited to the population expected to use it, perpetuate disparities, or weaken the social value used to justify participant burdens.
03 · What You Need to Know
Underrepresentation Can Become More Than a Generalizability Problem
Representativeness and justice are not the same thing
A study does not become unethical merely because its sample is not statistically representative of a population.
Qualitative studies may intentionally use small purposive samples. Early-phase trials may study narrowly defined populations. Mechanistic experiments may deliberately control participant characteristics. Research can be scientifically valuable without estimating population parameters.
Representativeness question
How well does the sample reflect a target population for the inference researchers want to make?
Justice question
Are people and groups being included or excluded for scientifically and ethically defensible reasons, with fair attention to research burdens and opportunities?
The two can overlap. A systematically unrepresentative sample becomes ethically important when the pattern of exclusion itself is unjust or when it prevents research from serving populations central to its stated purpose.
CIOMS explicitly connects exclusion with disparities
CIOMS states that categorical exclusion from research can result in or exacerbate health disparities and therefore requires justification when groups needing special protection are excluded. It also states that underrepresented groups should receive appropriate access to participation and that inclusion and exclusion criteria should not rest on potentially discriminatory characteristics without sound scientific or ethical reasons.
This reverses a familiar assumption in research ethics.
Ethical scrutiny is not only about whether researchers have included a population that needs protection. Sometimes the ethical question is why the population is absent.
The 2024 Declaration of Helsinki also treats underrepresentation as an ethical issue
The current Declaration of Helsinki states that groups underrepresented in medical research should be provided appropriate access to participation. It situates medical research within structural inequities and asks researchers to consider how benefits, risks, and burdens are distributed.
That does not create a numerical quota for every study. It does establish that access to research participation can have ethical significance rather than being purely a sampling decision.
Systematic exclusion can make evidence less applicable to the people expected to use it
Suppose an intervention is intended for a population in which 40% of users are older adults, but researchers repeatedly exclude people above 65 because comorbidities complicate analysis.
The study may answer a valid question about younger adults. The problem arises if the resulting evidence is then treated as though it adequately answers the clinical question for the entire intended population.
NIH's Inclusion Across the Lifespan policy was developed specifically to improve the applicability of research knowledge to populations affected by the conditions under investigation. NIH requires inclusion across ages unless scientific or ethical reasons justify exclusion.
The issue is therefore partly epistemic: whose evidence is missing from what we later call evidence-based practice?
Systematic exclusion can move uncertainty into ordinary practice
Researchers sometimes reduce uncertainty inside a study by excluding complicated participants.
Clinicians and policymakers may later inherit that uncertainty.
If people with multiple chronic conditions are consistently excluded from trials, healthcare professionals still need to treat people with multiple chronic conditions. If pregnant people are excluded from drug research, those drugs may still be used during pregnancy. If disabled people are excluded from technology evaluations, they may still encounter the technology once deployed.
Exclusion can therefore make the research internally cleaner while leaving important real-world questions unanswered.
This is especially visible when considering automatic exclusion during pregnancy, where avoiding research exposure may preserve uncertainty about treatments pregnant patients nevertheless need.
Exclusion can arise without anyone explicitly deciding to exclude a group
Some of the most important exclusions are produced indirectly.
A study requires weekday appointments between 9 a.m. and 4 p.m. People unable to leave work disappear from the sample. Recruitment is entirely online, reducing participation among people with limited digital access. A research site has physical barriers. Materials exist in one language. Transportation is not supported. An app is incompatible with assistive technologies.
None of these procedures may be labeled an exclusion criterion.
Yet collectively they determine who can participate.
Watch Out
Do not audit inclusion only by reading the eligibility section of the protocol. Recruitment channels, site location, scheduling, technology, consent procedures, language, transportation, compensation, and accessibility can create de facto exclusion even when the formal eligibility criteria appear neutral.
Systematic exclusion can weaken the social value of research
Human research asks participants to contribute something: time, information, inconvenience, privacy, procedures, or risk.
One important justification for those burdens is that the research has social or scientific value.
If a study claims to address a major problem but is designed in a way that prevents it from producing useful evidence for a substantial portion of the affected population, the social value supporting those burdens may be weaker than researchers assume.
This does not mean every limitation in external validity becomes an ethical violation. All studies have boundaries. The concern becomes stronger when there is a substantial mismatch between the population used to justify the importance of the research and the population actually represented in the evidence.
Exclusion can also distribute opportunities unfairly
Participation is not always a benefit. Research can involve burdens and risks.
Yet some research offers access to potentially beneficial interventions, specialist monitoring, new technologies, or opportunities to contribute to knowledge relevant to one's community.
Systematically excluding a population can therefore mean shielding it from research burdens while simultaneously denying access to potential research benefits and opportunities.
CIOMS recognizes both sides of this problem: no group should bear an unfair share of research burdens, but categorical exclusion can also exacerbate disparities.
Researchers should not solve underrepresentation through token inclusion
Once exclusion is recognized, the tempting response is to recruit a few members of the missing population.
That may change the demographic table without changing the science.
If a trial enrolls five adults over 80 but cannot meaningfully examine whether the intervention behaves differently in older populations, researchers should not imply that the study has resolved the evidence gap. Similarly, translating a consent form without translating the measurement instrument does not produce meaningful linguistic inclusion.
Inclusion should be capable of contributing relevant evidence, not merely improving the appearance of the sample.
Some exclusions remain entirely appropriate
A study can legitimately focus on a narrower population when the research question requires it. A separate study may be preferable for a population requiring substantially different methods. Risk may make inclusion unacceptable. The disease may not occur in a particular group. Relevant knowledge may already exist.
NIH explicitly recognizes several such justifications in its lifespan policy.
The ethical problem is therefore not “somebody is missing.” It is unjustified systematic exclusion of people whose absence matters to the question, benefits, burdens, or eventual application of the research.
The strongest warning sign is a mismatch between the target of the claim and the source of the evidence
Ask two questions:
Who is this research supposed to help us understand?
Who actually had a realistic opportunity to become part of the evidence?
If the answers differ substantially, researchers should either improve inclusion or narrow their claims.
This is where the justification for individual exclusion criteria connects to the ethics of the sample as a whole.
07 · A Quick Checklist
Look for the People Your Study Design Quietly Removes
Before recruitment begins, check:
Who is most affected by the condition, intervention, policy, technology, or question being studied?
Which of those populations are formally excluded, and why?
Which populations may be excluded indirectly by recruitment channels, location, scheduling, language, technology, accessibility, or cost of participation?
Are those barriers scientifically necessary or primarily consequences of how the study was organized?
Will the resulting sample generate evidence applicable to the populations for whom the research claims relevance?
Could exclusion perpetuate an existing evidence gap or disparity?
If meaningful inclusion is impossible, have we narrowed our claims accordingly?
Are we measuring meaningful inclusion rather than merely demographic presence?
