A joint model for integrating serial methylation-based tumor fraction and ESR1 assessment to forecast overall survival in patients with ER+/HER2- metastatic breast cancer

Background: Methylation-based tumor fraction (TF) dynamics are highly associated with outcomes in metastatic ER+/HER2- breast cancer. Gain-of-function ESR1 mutations frequently arise under therapeutic pressure and serve as actionable biomarkers. Because TF reflects tumor burden and ESR1 mutations reflect clonal evolution, jointly modeling their trajectories may improve understanding of biomarker co-evolution and strengthen outcome prediction. We developed a statistical framework that simultaneously characterizes serial ESR1 mutation burden and TF, capturing both genetic adaptation and epigenetic quantification of tumor fraction. Methods: A joint model (JM) was applied to serial liquid biopsy data from ER+/HER2- mBC patients enrolled in a prospective plasma-collection study while receiving endocrine therapy (ET) and CDK4/6 inhibitors (CDK4/6i). Patients contributed baseline and ≥ 2 on-treatment samples. TF was measured using Guardant Reveal and ESR1 alterations were assessed using Guardant360 Liquid. TF and ESR1 variant allele frequency (VAF) using the ESR1 alteration with the highest VAF were logit-transformed for modeling and back-transformed for interpretation. A hierarchical cubic spline mixed-effects sub-model captured longitudinal TF and ESR1 patterns, paired with a Cox regression sub-model for overall survival (OS). Baseline covariates, incorporated in both sub-modules, included age, CDK4/6i agent, line of therapy, and prior treatment. Results: Forty-nine patients (279 ctDNA timepoints) met inclusion criteria. After covariate adjustment, current TF and ESR1 values were significantly associated with OS (p < 0.05). Rising TF or ESR1 trajectories corresponded to poorer survival, whereas decreasing trajectories delineate improved outcomes. The model also captures interactions between TF and ESR1 , showing how molecular signals evolve together and impact outcome. Individualized survival predictions, informed by the dynamic biomarker trends, are visualized through patient-level dynamic prediction plots. Conclusions: Joint modeling of TF and ESR1 mutation dynamics provides an integrated view of tumor evolution, capturing both clonal adaptation and tumor burden over time. This approach yields continuously updated, patient-specific survival predictions that may support real-time clinical decision-making. Integrating genomic and epigenomic biomarkers within a unified dynamic model represents a meaningful advance in disease monitoring. Future work should validate this framework in larger, more diverse cohorts and assess feasibility in routine clinical use.