Stefano Garzon

Research · Oct 23, 2026

Study population: eligibility criteria, enrichment, and generalization

Concept

Eligibility criteria define the study population, a subset of people with the condition of interest. Enrolled participants are a nonrandom sample of that population. Describing a trial requires both elements: the criteria and who was actually enrolled.

Internal validity concerns whether the result holds for those who meet the protocol criteria. External validity concerns whether it holds for patients outside those criteria in a comparable clinical setting.

Eligibility criteria

Friedman groups criteria into five categories:

  1. Potential for benefit. Patients in whom the mechanism of action suggests a response.
  2. High likelihood of showing benefit. Patients at enough risk to accumulate events within the trial period.
  3. Avoiding adverse effects. Excluding those with contraindications, such as pregnant women in trials of drugs with possible teratogenicity.
  4. Competing risk. Excluding those likely to die or leave the study from another cause before the outcome, such as advanced cancer in cardiovascular trials.
  5. Expected poor adherence. Excluding those unlikely to follow the protocol.

Criteria central to the question need detailed definitions. In an antihypertensive trial, the protocol must state how, when, how often, and with what instrument blood pressure is measured. Criteria apply only at entry. A participant who develops a contraindication later may stop the intervention but stays in the trial and in the analysis.

Homogeneous or heterogeneous population

A homogeneous population increases the chance of detecting an effect when the mechanism is known. When the mechanism is uncertain, a heterogeneous population allows discovery of who responds. In the Diabetic Retinopathy Study, photocoagulation showed early benefit in the four most severe of ten subgroups. With longer follow-up, the other six also benefited. A population limited to severe cases would not have shown this.

Large simple trials use broad criteria and resemble practice. The cost is effect dilution and larger sample sizes.

Enrichment

Enrichment means selecting participants more likely to show an effect. There are three main forms.

By risk. ROCKET-AF required additional stroke risk factors and randomized 14,264 patients with atrial fibrillation. The higher event rate reduced the required sample size. The drug was also approved for lower-risk patients who were not enrolled.

By biomarker. JUPITER enrolled 17,802 people with LDL below 130 mg/dL and C-reactive protein of 2.0 mg/L or higher. Rosuvastatin reduced the primary outcome (HR 0.56; 95% CI 0.46–0.69). The result applies to that population, not to everyone with low LDL.

By adherence and tolerance (run-in). In PARADIGM-HF, 10,521 patients entered run-in phases with enalapril and then sacubitril-valsartan. Of these, 2,079 were discontinued before randomization, and 8,442 were randomized. The final population tolerates both treatments better than the population seen in practice.

A severity sequence is another strategy. CONSENSUS randomized 253 patients in functional class IV and showed a 40% reduction in 6-month mortality with enalapril. Later trials extended the finding to less severe patients.

Generalization

Volunteers differ from nonvolunteers. They tend to be healthier and more adherent. Referral centers see selected patients. In a Finnish review of 400 patients hospitalized with gastric ulcer, only 29% met typical trial criteria, and nearly all deaths and serious complications occurred among the ineligible.

Sex is one example of the limit. The Physicians’ Health Study, with 22,071 men, showed a 44% reduction in myocardial infarction with aspirin. In the Women’s Health Study, with 39,876 women, the reduction in major cardiovascular events was not significant (RR 0.91; 95% CI 0.80–1.03). Aspirin reduced ischemic stroke in women (RR 0.76; 95% CI 0.63–0.93), while men in the Physicians’ Health Study had a nonsignificant increase in stroke.

A log of screened but not enrolled patients, with reasons, helps estimate representativeness.

Common errors when reading a paper

What to check in a trial

  1. The full inclusion and exclusion criteria, in the protocol or supplement.
  2. How many were screened, how many were eligible, and how many were randomized.
  3. Whether there was a run-in and how many left during it.
  4. The control group event rate compared with that of the population you treat.
  5. Representation of older adults, women, and patients with comorbidities.
  6. Type of center: referral or community.

Related posts: subgroups, absolute and relative effect.

References