Abstract 5390: Robust prediction of pan-cancer immune checkpoint blockade response using machine learning

Abstract

Abstract Immune checkpoint blockade (ICB) has revolutionized our approach to cancer treatment. However, the response rate is still low. With the accumulation of data, efforts to use these data to build machine learning predictors of ICB response are rising. However, as most datasets are still quite small, pertaining machine learning models may often ‘overfit’ the data, i.e., have a much weaker performance on independent test data than on the data they were learned upon. Here we analyzed ~2000 patient samples from multiple cohorts across 16 solid tumor types. Based on 10 genomic & clinical features, we built different machine learning models, comparing their performance in terms of the Area Under ROC Curve (AUC) and importantly, in terms of the AUC difference between training vs validation sets, using a standard cross validation procedure. As a result, the Linear LASSO Regression (LLR) model outperformed all other models by having the highest AUC on the validation set (0.74) and notably, the smallest AUC difference between training and validation sets (0.02). In contrast, these two AUC values for the FDA approved tumor mutational burden (TMB) biomarker are 0.63 and 0.03. The highest predictive features were, receiving chemotherapy before ICB treatment or not, TMB, cancer type, blood albumin level, blood neutrophil-to-lymphocyte ratio, and patient age. The LLR score also consistently predicted overall survival and progression-free survival across all individual cancer types. More importantly, ICB response probability increased near-monotonically from 0% to > 43% with the LLR score, which is an important feature for patient exclusion. In contrast, ICB response probability is ~25% when TMB = 0 and not always higher with higher TMB. In summary, the LLR model identifies key features predicting pan-cancer ICB response and survival. The use of combined genomic and clinical features holds potential to further facilitate clinical ICB patient stratification beyond TMB. Citation Format: Tiangen Chang, Saugato R. Dhruba, Yingying Cao, Luc G. Morris, Eytan Ruppin. Robust prediction of pan-cancer immune checkpoint blockade response using machine learning. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5390.