Presentation (CTI)

OBJECTIVES: Recent studies indicate the presence of an inverse relationship between methylation-based tumor fraction (TF) and overall survival in patients with non-small cell lung cancer (NSCLC). Similarly, KRAS mutations, common oncogenic drivers in NSCLC, evolve under therapeutic pressure and reflect clonal adaptation. While temporal TF measurements capture disease progression, KRAS mutation dynamics reflect genomic evolution. We hypothesize that leveraging joint modeling (JM) to interrogate TF and KRAS trajectories will offer a more complete understanding of how TF and the KRAS mutation work in tandem to modify outcome.
METHODS: We analyzed a cohort of 251 patients with NSCLC extracted from RADIOHEAD, a prospective study of pan-cancer immunotherapy naive patients receiving immune checkpoint inhibitor (ICI) regimens. Patients with baseline and at least two on-treatment samples were included. TF was quantified using Guardant Reveal, a methylation-based assay, and KRAS alterations were assessed via Guardant 360 Liquid. For each subject, the mutation within the KRAS gene with the highest variant allele frequency was selected. Patient longitudinal trajectories were characterized using hierarchical cubic spline mixed-effects sub-model, linked to a Cox regression sub-model for real world overall survival (rwOS). Covariates included age, smoking status, stage (III vs IV), and gender.
RESULTS: The JM demonstrated that current measures of TF and KRAS burden were each significantly associated with rwOS (p < 0.05). Rising TF and KRAS trajectories corresponded to inferior survival outcomes while declining trends showed improved prognosis. The JM captured interactions between TF and KRAS, highlighting co-evolving biomarker patterns. Dynamic survival predictions reflecting the change in the biomarker trajectories over time were generated to provide patient-level interpretation.
CONCLUSIONS: Joint modeling of TF and KRAS dynamics integrates disease progression and clonal adaptation, yielding continuously updated predictions. This approach holds promise in enhancing real-time disease monitoring, improving prognostic performance, and informing clinical decision-making along the patient journey.