OBJECTIVES: Population-adjusted indirect comparisons (PAICs) support the comparative efficacy of treatments in health technology assessments (HTAs). Typically, researchers have access to the individual patient data (IPD) from one trial and the aggregate data (AgD) for comparators. In a PAIC, IPD data is adjusted such that it reflects the AgD data, providing a relatively unbiased relative treatment effect (RTE) estimated in AgD population. However, HTA discussions focus on the IPD population, not the AgD population. Assuming shared-effect modification (SEM), the RTE can be projected into any target population, but violation of SEM can lead to misinterpretation of an indirect treatment comparison (ITC) result and is demonstrated here.

METHODS: Patient-level data were simulated (n=1000) for two hypothetical clinical study populations (i.e., IPD; AgD) using the ‘simsurv’ package in R. Effect modification of the baseline characteristic genomic risk (high-risk [HR], low-risk [LR]) was set to present in the IPD and absent in the AgD. The distribution of genomic HR between-study was set as significantly different, 50% (IPD) and 30% (AgD). Equivalent PAIC models were applied to each dataset providing ITC results, which were interpreted in the context of the IPD and AgD populations assuming SEM.

RESULTS: The RTE in the IPD was 0.60 (intention-to-treat [ITT]), 0.50 (HR) and 0.70 (LR) and 0.70 (ITT), 0.90 (HR) and 0.50 (LR) in the AgD. After adjustment, the estimated RTE for an ITC of IPD vs AgD was 0.82 (p<0.05) when interpreted in the AgD population context and 0.92 (p<0.05) in the IPD population context. The estimated RTE from the standard NMA was 0.85 (p<0.05).

CONCLUSIONS: Violation in the SEM assumption led to an incorrect and more optimistic generalization of the RTE in the IPD population. This research demonstrates that SEM is critical for interpretation of the results in the IPD population.