Adverse events (AEs) –derived biomarker in predicting clinical outcomes in patients with colorectal cancer (CRC) treated with immunotherapy (IC).

132 Background: There are insufficient studies evaluating the role of AEs in CRC patients treated with IC. This study aims to transform AEs from traditional toxicity evaluation tools to informative biomarkers for assessing treatment efficacy by utilizing modern analytical strategies. Methods: We analyzed a single arm study of 51 mismatch repair proficient, refractory CRC patients (NCT03712943) treated with regorafenib and nivolumab. Data for analysis included CTCAE version 5 AE data, RECIST treatment response, progression-free survival (PFS), and overall survival (OS). Analytic approach leveraged multiple AE parameters to develop a set of innovative AE biomarkers. One AE biomarker of interest was treatment related low-grade early AEs (TR-LG-AE) which was defined as frequency of all TR-LG-AEs occurring between the 1st day and day 30 after the 1st treatment. Results: TR-LG-AE was associated with treatment response (p<0.001 by ANOVA trend test). Patients with higher AE events tended to be a responder, compared to PD patients with a lower frequency of AE events. The TR-LG-AE also correlated with improved PFS and OS with hazard ratio of 0.83 (p=0.01) and 0.89 (p=0.03), respectively. Accordingly, patients were grouped into high versus low TR-LG-AE based on optimal cutoff of 6 events for further analysis. The dichotomized TR-LG-AE (high vs low) showed significant association with survival outcomes (median PFS: 15.3 vs 3.7 months; p<0.001; median OS: 21.9 vs 9 months; p=0.02). Further interaction analysis showed patients with high TR-LG-AE had improved PFS regardless of occurrence of high-grade AE (HG-AE) (p=0.01), suggesting its independence of HG-AE. Individual AE analysis identified patients experienced with early low-grade rash maculo-papular, lipase increased, or hypertension had improved PFS (p<0.001). In biomarker analysis, one key mutation biomarker, RAS, did not show significant association with survival outcomes (PFS: p=0.66; OS: p=0.81). In contrast, integration with TR-LG-AE led to remarkable improvement. Specifically, the TR-LG-AE was able to help improve RAS classification by identifying patients with better treatment benefit. Specifically, patients with RAS mutation and high TR-LG-AE experiences had a median PFS of 15.3 months compared to others (median PFS of 3.7 months; p=0.007), and a median OS of 21.9 months compared to others (median OS of 9.4 months; p=0.04). Conclusions: The study demonstrated utility of AE in predicting clinical outcomes in CRC, similar to our previous findings in lung cancer. Moreover, it discovered that integration of AE-derived biomarkers with molecular biomarkers could produce a more robust classifier than when considered individually. This analysis is the first of its kind to identify subgroups of CRC patients that benefit from IC treatment.