Safety after the Average Patient: Detecting Heterogeneous Drug Harm across Age, Sex, Comorbidity, Genetics, and Concomitant Treatment

Population-average safety estimates can obscure clinically important differences in drug harm because patients differ in baseline susceptibility, realized exposure, biological response, concomitant treatment, and the probability that an adverse outcome will be observed or reported. This Original Subgroup-Safety Framework Article proposes a multidimensional architecture for detecting and evaluating heterogeneous drug harm across age, sex, pregnancy, physiological state, comorbidity, frailty, organ impairment, genetics, ancestry, metabolic phenotype, and polypharmacy. The framework separates five interacting explanatory layers: baseline susceptibility, exposure modification, biological response, concomitant-treatment context, and observation or data-generation processes. It then connects subgroup-signal detection to cross-source triangulation, causal assessment, external validation, bounded communication, and decision-specific evaluation. Polypharmacy is represented as a time-varying interaction structure rather than a medication count, allowing drug–drug, drug–gene, and drug–disease relationships to be considered jointly. To limit false subgroup conclusions, the framework requires explicit subgroup definitions, interaction assessment, multiplicity control, partial pooling where appropriate, evaluation of sparse intersections, and independent replication. Validation is treated as a staged requirement encompassing construct validity, data validity, analytical validity, external transportability, causal coherence, decision utility, and continuing performance under clinical and dataset drift. The original contribution is an integrated conceptual synthesis that distinguishes observed subgroup differences from drug-attributable heterogeneous harm and separates signal detection from causal, clinical, and regulatory conclusions. The framework remains unvalidated and does not prescribe treatment, establish subgroup-specific safety, or define regulatory action. Its value must be tested through multisource retrospective studies, prospective evaluation, fairness assessment, and monitoring of subgroup-specific calibration over time.