Dynamic Predictions in Non-Small Cell Lung Cancer Using Joint Modeling of Longitudinal and Time-To-Event Outcomes...

Joint modeling (JM) of longitudinal and time-to-event (TTE) data is a powerful statistical technique that elucidates how temporal changes in a biomarker relate to TTE outcomes while accounting for study confounders. The growing use of next-generation sequencing (NGS) in precision oncology, particularly in capturing circulating tumor DNA (ctDNA) methylation via liquid biopsy, offers new opportunities to apply JM for patient prognostication. In this study, we used JM to analyze a prospective cohort of 251 non-small cell lung cancer patients receiving standard-of-care immune checkpoint inhibitor treatment regimens. The JM is comprised of two sub-models-a hierarchical cubic spline random effects model for the longitudinal biomarker component and a Cox proportional hazards model for the TTE outcome component. Each sub-model incorporated baseline factors of age, gender, smoking status, and cancer stage. The main objective was to establish whether the evolution of a methylation-derived tumor fraction (TF) is associated with patient outcomes. Our investigation revealed that TF is strongly associated with both real-world overall survival and real-world progression-free survival. These results enable the generation of various patient-level predictions that capture the dynamic interplay between TF and patient outcomes, offering a path toward real-time, personalized prognostication. Although validation in larger, multi-center cohorts is needed before routine adoption, this work represents an important step toward integrating dynamic predictions into serial liquid biopsy monitoring. This approach demonstrates the potential of JM to provide more nuanced and adaptive disease evolution monitoring, offering better clinical decision guidance compared to static models or imaging alone.