Answers are grounded in this paper and cite their sources.
Abstract Background: Randomized clinical trials provide high internal validity but may not fully reflect outcomes in routine clinical practice. We propose a population-alignment and decomposition framework for interpreting how measured baseline differences contribute to trial--real-world outcome differences when individual-level real-world data are available but trial information is limited to aggregate baseline summaries. Methods: The framework was applied to human epidermal growth factor receptor 2--positive breast cancer by comparing a Helsinki University Hospital real-world cohort of 267 patients with three randomized neoadjuvant therapy trial arms. Crude pCR rates and crude differences were estimated using Bayesian binomial models. Population-aligned pCR estimates were obtained using matching-adjusted indirect comparison (MAIC) and Bayesian multilevel regression with post-stratification and raking. Shapley decomposition was used to identify which measured variables contributed most to the adjustment differences. Results: Across comparisons, baseline alignment shifted the real-world estimates downward, indicating less favorable measured baseline profiles in the trial populations with respect to pathological complete response. Adjustment differences were modest in the NeoSphere comparison and larger in the KRISTINE and HELEN-006 comparisons, where positive residual differences were observed after alignment. Conclusions: The application illustrates how crude trial--real-world comparisons can obscure the contribution of measured baseline composition. Population alignment and decomposition provide a practical framework for interpreting such comparisons under aggregate trial reporting.
No takes yet. Share an insight, caveat, or question.
Lähteenmaa et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: