Abstract CT024: Utility of longitudinal circulating tumor DNA (ctDNA) modeling to predict RECIST-defined progression in first-line patients with epidermal growth factor receptor mutation-positive (EGFRm) advanced non-small cell lung cancer (NSCLC)

Abstract

Abstract Circulating tumor DNA is released into the bloodstream from tumors and can reflect both intra-tumor heterogeneity and clonal evolution.1 As ctDNA levels are thought to reflect tumor burden, a decrease in ctDNA while on therapy may suggest treatment efficacy.2 We assessed whether longitudinal changes in ctDNA could supplement or improve RECIST-based measures for decision making during drug development. Plasma samples collected at regular intervals were analyzed to predict the risk of disease progression in patients with EGFRm (Ex19del or L858R) advanced NSCLC following initiation of first-line EGFR tyrosine kinase inhibitor (TKI) treatment in the FLAURA study (NCT02296125). Of 556 patients who received treatment (osimertinib 80 mg daily [QD] or comparator EGFR-TKI [gefitinib 250 mg QD or erlotinib 150 mg QD]), 353 (63%) had an available sample with detectable levels of ctDNA at baseline. Patients with both imaging and ctDNA samples at baseline and 3 additional timepoints (n=320, the analysis set) were divided (as implemented in caret R package)3 into training (n=259) and validation (n=61) cohorts. Clinical data from these patients (data cut-off: June 12 2017) were used to fit joint models of either ctDNA dynamics or sum of the longest tumor diameters (SLD) and progression-free survival (PFS). Longitudinal ctDNA levels were measured using droplet digital PCR (ddPCR; Biodesix®) for EGFR-TKI sensitizing mutations (Ex19del or L858R) in all available samples. Model covariates were selected based on their statistical significance for the longitudinal and hazard sub-models and were analyzed using JMbayes package as implemented in R software (3.5.1). Our analysis indicates that the joint model can independently characterize changes in ctDNA and SLD and can predict PFS in the validation cohort, using only ctDNA changes occurring within 6 weeks from the initiation of therapy. Similarly, in the training cohort, joint modelling indicates that the relationship between SLD and PFS using RECIST is consistent with the longitudinal ctDNA analysis. In the analysis set, our model estimated a PFS of 16.6 months (95% confidence interval [CI] 12.1, NC) for osimertinib (n=153) and 9.4 months (95% CI 8.0, 13.5) for comparator EGFR-TKI (n=167). Model-predicted PFS probability using either ctDNA or SLD data was similar, suggesting ctDNA dynamics may provide similar information to RECIST for clinical decision making. Based on this analysis, ctDNA dynamics could be utilized to predict RECIST-defined progression for decision making during drug development of treatments for patients with NSCLC.