A mid-chemoradiation dynamic risk model integrating tumor features and ctDNA analysis for lung cancer outcome prediction.

Abstract

9046 Background: Circulating tumor DNA (ctDNA) molecular residual disease after curative intent therapy predicts disease progression in localized lung cancer. We hypothesized that integrating pre-CRT features and ctDNA levels during chemoradiation therapy (CRT) can predict patient outcomes earlier to enable response-adapted therapy. Methods: We identified pre-CRT features prognostic of disease progression after CRT for Stage II-III non-small cell lung cancer (NSCLC) in a historical “pre-CRT” training cohort of 109 patients. In addition, we applied CAPP-Seq ctDNA analysis pre-CRT and a median of 21 days into CRT (mid-CRT) to a “ctDNA” training cohort of 42 patients treated at MD Anderson and an independent validation cohort of 21 patients treated at Stanford. Prognostic pre-CRT features and mid-CRT ctDNA concentration were integrated using a Bayesian proportional hazards approach to generate a Continuous Individualized Risk Index (Kurtz et al. Cell 2019) for NSCLC (CIRI-NSCLC) to predict freedom from progression (FFP). Results: Adenocarcinoma histology (HR 2.6, P = 0.0005) and KEAP1 mutation (HR 2.7, P = 0.002) but not stage (P = 0.16), age (P = 0.60), or gender (P = 0.98) were significantly associated with FFP in the pre-CRT training cohort. Mid-CRT ctDNA concentration as a continuous variable was significantly associated with FFP in the ctDNA training cohort (HR 1.6, P = 0.04), and applying an optimal threshold identified in the training cohort (3.2 hGE/ml) significantly stratified FFP in the independent ctDNA validation cohort (HR 4.8, P = 0.02). CIRI-NSCLC enabled individualized real-time updating of the probability of FFP as model features became available over the course of CRT. CIRI-NSCLC outperformed individual model features in the independent validation cohort when compared by C-statistic (CIRI-NSCLC: 0.85; mid-CRT ctDNA: 0.76; histology: 0.66; KEAP1: 0.60). Across the whole cohort, patients with a greater than 66% risk of progression predicted by CIRI-NSCLC (n = 10) had an FFP of 10.0% at 12 months while patients with a less than 33% risk of progression predicted by CIRI-NSCLC (n = 22) had an FFP of 79.7% at 12 months (HR 15.0, P < 0.001). Conclusions: Our results suggest that CIRI-NSCLC can identify patients at very high and low risk of progression. Prospective evaluation will be necessary to test the potential utility of adapting treatment based on CIRI-NSCLC.