Heterogeneity of the Baseline Risk within Patient Populations of Clinical Trials: A Proposed Evaluation Algorithm

In this paper, the authors present an evaluation algorithm for systematic assessment of the observed heterogeneity in disease risk within trial populations. Predictive models are used to estimate the predicted patient hazards, the odds of having an event in the upper risk quartile (ODU) and the lower risk quartile (ODL), and the odds ratio (rate ratio for time-to-event analyses) for having an event in the upper risk quartile versus the lower risk quartile (extreme quartile odds ratio (EQuOR) and extreme quartile rate ratio (EQuRR)). The ranges for these metrics depend on the extent to which predictors of the outcome of interest exist and are known and the extent to which data are collected in the trial, as well as on the eligibility criteria and the specific patients who are actually enrolled. ODU, ODL, and EQuOR values are used to systematically interpret the results for patients at different levels of risk, to evaluate generalizability, and to determine the need for subgroup analyses. Individual data for five outcomes from three trials (n = 842, 913, and 1,001, respectively) are used as examples. Observed EQuOR values ranged from 1.5 (very little predicted heterogeneity) to 59 (large heterogeneity). EQuRR values ranged from 2 to 46. ODU values ranged from 0.24 to 3.19 (generally high risk), and ODL values ranged from 0.01 (clinically negligible risk) to 0.16 (clinically meaningful risk). The algorithm may also be used for comparing diverse trials (e.g., in meta-analyses) and used prospectively for designing future trials, as shown in simulations.