A Proposed Statistical Approach for Conducting Longitudinal Assessment of Circulating Tumor DNA
As circulating tumor DNA (ctDNA) may reflect cancer progression, understanding its temporal evolution can inform clinical decision-making in precision oncology. Temporal changes in ctDNA often exhibit complex patterns, varying significantly between and within patients. Factors such as patient characteristics and treatment regimens can further impact these changes. Traditional statistical methods may fall short in adequately characterizing patient-level ctDNA evolution over time, highlighting the need for advanced approaches. In this study we provide a framework to guide the identification of optimal models for the analysis of genetic biomarkers in liquid biopsy settings. Specifically, we focus on hierarchical mixed-effects models as they provide both cohort and patient-level insights. To illustrate the versatility of these models, we conduct an exploratory analysis of a real-word data consisting of patients with advanced colorectal cancer. In our analysis, we discovered that a hierarchical linear spline mixed-effects model was most optimal. Based on this finding, we used the model to generate cohort and patient-level response patterns, where patient-level results compare ctDNA trajectories using different combinations of patient demographics, medical history, treatments regimen, and outcomes. Finally, we discuss how results could potentially assist in constructing a patient monitoring system to help inform patient care.