PI-027 - ASSESSING BASELINE AND LONGITUDINAL CIRCULATING TUMOR DNA WITH PROGRESSION FREE SURVIVAL (PFS) IN BRAF V600E MUTANT (BRAFM) METASTATIC COLORECTAL CANCER (MCRC).

S. Sun¹, X. Zhang², M. Saul³, C. Davis³, E. Hahn⁴; ¹UCSD, San Diego, CA, USA, ²Pfizer Inc., San Francisco, CA, USA, ³Pfizer Inc., La Jolla, CA, USA, ⁴Pfizer Inc., Boulder, CO, USA.

Background: Circulating tumor DNA (ctDNA) is a promising biomarker for treatment response in oncology. Here we assessed ctDNA (summarized as mean VAF) across 5 time points in BRAFm mCRC patients from the safety lead-in of the randomized phase 3 BREAKWATER study (NCT02928224) and used a nonlinear mixed effects model (NLME) and a Cox proportional hazards (CPH) model to sequentially model ctDNA with PFS by blinded independent central review.

Methods: Longitudinal ctDNA data were fit with a biexponential model with three parameters, the model estimated ctDNA at baseline, growth (Kg), and decay (Kd) rates. The estimated log-transformed Kg and Kd parameters, along with other baseline characteristics were then assessed as covariates in a CPH model to evaluate their association with PFS. NLME modeling was done in NONMEM v7.5.0 and CPH modeling was done in R v4.1.1.

Results: Fifty-six participants with both PFS data and ctDNA data were included in the CPH analysis, and 52 of these participants contributed longitudinal ctDNA measurements for NLME model development. The biexponential model provided typical values of baseline mean VAF (5.41%), KG (0.000305 year-1) and KD (0.598 year-1). In the univariate screen, line of therapy (p = 0.038), logKD (p = 0.036), and mean VAF at baseline (p = 0.012) were significant predictors for PFS. In the multivariate CPH analysis, mean VAF at baseline and line of therapy are significant predictors (p = 0.00284 and p = 0.00679, respectively), while logKD fails to be significant. The hazard ratio for mean VAF at baseline was 1.0741 (95% CI: 1.0248 – 1.126), indicating that each unit increase in mean VAF at baseline increases the hazard of progressive disease or death by 7%. The hazard ratio for line of therapy was 4.8422 (95% CI: 1.5455 – 15.171), see Table 1.

Conclusion: This analysis demonstrates the utility of ctDNA as a predictor of PFS and the potential utility of a biexponential ctDNA-based NLME longitudinal model. When incorporated into the univariate Cox proportional hazards model, the association between logKD and PFS was significant, however in the multivariate analysis baseline ctDNA was the most significant association. The integration of baseline ctDNA and longitudinal ctDNA data into CPH models offers a novel approach for the prediction of PFS.