Participant characteristics and exclusion from trials: a meta-analysis of individual participant-level data from phase 3/4 industry-funded trials in chronic medical conditions

ObjectivesTrials often do not represent their target populations, threatening external validity. The aim was to assess whether age, sex, comorbidity count and/or race/ethnicity are associated with likelihood of screen failure (i.e., failure to be randomised to the trial for any reason) among potential trial participants. DesignBayesian meta-analysis of individual participant-level data (IPD). SettingIndustry-funded phase 3/4 trials in chronic medical conditions. Participants were identified as "randomised" or "screen failure" using trial IPD. ParticipantsData were available for 52 trials involving 72,178 screened individuals of whom 24,733 (34%) failed screening. Main outcome measuresFor each trial, logistic regression models were constructed to assess likelihood of screen failure, regressed on age (per 10-year increment), sex (male versus female), comorbidity count (per one additional comorbidity) and race/ethnicity. Trial-level analyses were combined in Bayesian hierarchical models with pooling across condition. ResultsIn age- and sex-adjusted models, neither age nor sex was associated with increased odds of screen failure, though weak associations were detected after additionally adjusting for comorbidity (age, per 10-year increment: odds ratio [OR] 1.02; 95% credibility interval [CI] 1.01 to 1.04 and male sex: OR 0.95; 95% CI 0.91 to 1.00). Comorbidity count was weakly associated with screen failure, but in an unexpected direction (OR 0.97 per additional comorbidity, 95% CI 0.94 to 1.00, adjusted for age and sex). Those who self-reported as Black were slightly more likely to fail screening (OR 1.04; 95% CI 0.99 to 1.09); an effect which persisted after adjustment for age, sex and comorbidity count (OR 1.05; 95% CI 0.98 to 1.12). ConclusionsAge, sex, comorbidity count and Black race/ethnicity were not strongly associated with increased likelihood of screen failure. Proportionate increases in screening these underserved populations may improve representation in trials. Trial registrationRelevant trials in chronic medical conditions were identified according to pre-specified criteria (PROSPERO CRD42018048202).