Long-Term Survival Heterogeneity Across PD-L1-Defined Subgroups in Pembrolizumab-Treated Non-Small-Cell Lung Cancer: A Model-Based Analysis

Abstract

Background/Objectives: Immune checkpoint inhibitors have substantially improved survival outcomes in non-small-cell lung cancer (NSCLC). However, survival heterogeneity across PD-L1-defined and other clinically relevant patient subgroups remains incompletely characterized, limiting subgroup-specific interpretation of pembrolizumab outcomes within a precision oncology framework. In this study, we quantified survival heterogeneity and estimated long-term outcomes of pembrolizumab using reconstructed clinical trial data. Methods: A systematic review identified 36 trials in advanced NSCLC. Individual time-to-event and censoring records were reconstructed from published Kaplan–Meier curves, while patient-level clinical and demographic covariates were stochastically assigned according to trial-level summary distributions. Parametric time-to-event models were developed for overall survival (OS) and progression-free survival (PFS) across programmed death-ligand 1 (PD-L1) tumor proportion score subgroups. Model performance was evaluated using bootstrap analyses and external validation, and model-based simulations were conducted to project survival for up to 10 years. Results: A total of 7165 reconstructed time-to-event records for OS and 6730 for PFS were included in the analyses. Model-predicted survival outcomes varied across clinically relevant characteristics, with higher PD-L1 expression, treatment-naïve status, non-squamous histology, and combination therapy generally associated with more favorable outcomes. Long-term projections demonstrated substantial heterogeneity across covariate-defined subgroups, with projected 10-year OS probabilities ranging from 7.2% to 18.0% among treatment-naïve subgroups. Conclusions: These findings provide a quantitative characterization of subgroup-specific survival heterogeneity and long-term outcomes in pembrolizumab-treated NSCLC, supporting a more refined interpretation of treatment outcomes in the context of precision oncology.