Abstract 2481: Employing joint modeling to support clinical interpretation of ctDNA dynamics during the treatment of advanced colorectal cancer

Abstract Background: Recent evidence suggests dynamic changes in cell-free circulating tumor DNA (ctDNA), defined as the maximum variant allele fraction (maxVAF), is predictive of overall survival (OS) in colorectal cancer (CRC) patents. As such, we leverage an advanced statistical approach, joint modeling of longitudinal and time-to-event data (JM), to associate ctDNA dynamics and OS. Contrasting with traditional methods, a unique advantage of JM is results assume the form of patient-specific adaptive dynamic predictions, where utilizing these predictions can enhance clinical decision-making in longitudinal monitoring. Methods: We applied JM to a retrospective cohort of patients with advanced CRC who received chemotherapy (chemo). The cohort was extracted from GuardantINFORM, a real-world clinico-genomics database. Patients analyzed had baseline Guardant360 (G360) ctDNA tests within 14 days before chemo initiation and >2 G360 tests on treatment. JMs are comprised of two sub-models: a hierarchical cubic spline random effects sub-model evaluated the serial ctDNA measures and a Cox-regression sub-model examined OS. Covariates incorporated into each sub-model included: age, gender, smoking status, and the Charlson Comorbidity Index. JMs, based on different association structures (AS) designed to connect information between sub-models, were investigated. ASs included the biomarker’s estimated value, instantaneous rate of change, and cumulative effect. Results: In total, 137 patients with 759 ctDNA time points were analyzed using three different JMs, one for each AS. After controlling for confounders, each AS demonstrated a significant relationship between ctDNA progression and OS (respective p-values < 0.005). As results are best interpreted graphically, various dynamic predictions are provided, each accentuating how the AS modifies the relationship between ctDNA and OS. Conclusions: To our knowledge, using JM to link dynamic ctDNA changes and OS in CRC patients is novel. In this study we demonstrate that a biomarker’s updated estimated value, instantaneous rate of change, or cumulative effect can forecast OS. As such, JM may afford healthcare providers different options to monitor patient progression and update prognosis as new ctDNA information becomes available on clinical reports. In sum, these results establish a foundation for integrating serial ctDNA measures for prognostic clinical interpretation. Citation Format: Christopher R. Pretz, Jiemin Liao, Aaron Hardin, Amar Das. Employing joint modeling to support clinical interpretation of ctDNA dynamics during the treatment of advanced colorectal cancer. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2481.