Predicting response of patients with NSCLC to immunotherapy using innate immune fitness (IIF) profiling.

Abstract

Listen

Translate article

8038 Background: Aberrant activity of AID/APOBEC deaminase enzymes can cause 'off-target' somatic mutations in cancerous cells. Mutations associated with deaminases and other mechanisms can be quantified using metrics relating to motif usage, strand bias, transitions/transversions, codon context, and amino acid changes. Collectively, these metrics form an Innate Immune Fitness (IIF) profile (US Patent 20200370124). The aim of this project was to conduct IIF profiling on a cohort of Non-Small Cell Lung Cancer (NSCLC) patients treated with immune checkpoint inhibitors (ICI), and use the IIF profiles to build and evaluate a predictive model. Methods: Whole exome and progression-free survival (PFS) data was obtained from Rizvi et al. 2015, Hellmann et al. 2018, Miao et al. 2018, Fang et al. 2019, Frigola et al. 2021, and Ravi et al. 2023 (n = 515). IIF profiles were generated using CRIS (v5.0.0; GMDx Genomics Ltd). Patients were classified as a ‘Responder’ (PFS > 12 months or Complete Response; n = 148) or ‘Non-Responder’ (PFS ≤ 12 months; n = 367). Machine learning models were generated using the H2O.ai AutoML platform and evaluated using multiple rounds of cross-validation. Patient response predictions were collated for each patient and a consensus ‘IIF Score’ was calculated. Multivariable analysis of IIF Score, TMB (10mut/Mb) and PD-L1 TPS ( < 1%, 1% - 50%, > 50%) was conducted using a Cox proportional-hazards model. Results: The predictive accuracy of IIF Scores was 76% (Sensitivity = 58%; Specificity = 83% , NPV = 83% , PPV = 58%) with a Hazard Ratio (HR) of 0.39 (0.29-0.53; p < 0.001; corrected for TMB and PD-L1). In comparison, the predictive accuracy of TMB was 67% (Sensitivity = 53%; Specificity = 79% , NPV = 79% , PPV = 44%), HR = 0.77 (0.59-1.00; p = 0.049; corrected for IIF Score and PD-L1). ‘PD-L1 > 50%’ accuracy was 53% and HR = 0.64 (0.44-0.94; p = 0.023; corrected for IIF Score and TMB). The Area Under the Curve (AUC) for IIF Score was significantly higher than TMB (0.77 vs 0.70; DeLong test: p < 0.001). Conclusions: IIF Score was the strongest predictor of patient response to ICI. Despite the inherent limitations of combining data from multiple cohorts, IIF Score outperformed TMB and PD-L1 in predictive accuracy, HR and AUC. With a Negative Predictive Value of 83%, these results support the use of IIF Score as a biomarker for identifying ICI Non-Responders in NSCLC patients. Research Sponsor: GMDx Genomics Ltd.