Abstract 5426: Multi-omic characterization and predictive features of advanced ovarian cancer patients in a large phase III cohort

Abstract

Abstract Discovery and validation of biomarkers derived from multi-dimensional clinico-genomic datasets have become critical in precision medicine and oncology drug development. The integration of multi-dimensional genomic and imaging datasets from patients in late-stage oncology clinical trials can be difficult, in part due to limited patient enrollment and sample collection, especially tumor tissue biopsies. The objective of this project was to conduct predictive biomarker discovery on integrated clinico-genomics data spanning tumor genomics, germline genetics, tumor imaging, and circulating blood-based biomarkers for a cohort of 800+ advanced ovarian cancer patients enrolled in the phase III trial of the PARP-1 inhibitor Veliparib (VELIA). Genomic (DNA & RNA) and imaging datasets were generated from 800+ patient-matched tumor biopsies, liquid biopsies and whole blood to enable various biomarker analyses for this study, notably in BRCA-deficient and HRD+ subgroups. Pairwise analysis of individual features with clinical outcomes shows that increased tumor-mutation burden (TMB), RNA-based estimates of immune activity (ICR and MIRACLE scores), CA-125 elimination constant (KELIM score), and image-based estimates tumor-infiltrating lymphocytes (TILs) were each significantly associated with longer PFS and less progressive disease (PD). Similarly, homologous recombination deficiency (HRD) and BRCA alteration were associated with better clinical outcomes, while high CA-125 was associated with worse outcomes. To understand the concerted impact of these features on clinical outcome, we developed multivariate classifiers of PD and regressors (Cox Proportional Hazards) of PFS using XGBoost; using a train-test split of 75/25, we trained the models with 500 rounds of 10-fold cross validation hyperparameter tuning, which resulted in fit models. The PD classifier achieved a validation accuracy of 0.71 and F1-score of 0.78, with high immune activation, high TMB, high KELIM, and HRD and BRCA alteration predicting better outcomes. The PFS survival model achieved a validation C-index of 0.61, with a rank importance of features similar to that of PD classification models; for both models, patients receiving Veliparib had clear benefit relative to that of control arms. Collectively, this study illustrates the value of integrating multi-dimensional datasets with predictive machine learning to identify clinically-relevant biomarkers. CR, DM, PA, BR, JD, TB, PN, JS, XH, and JFW are employees of AbbVie. RLC an employee at The University of Texas MD Anderson Cancer Center and has no funding to disclose. The design, study conduct, and financial support for this research were provided by AbbVie. AbbVie participated in the interpretation of data, review, and approval of the publication. Citation Format: Cyril Ramathal, David Masica, Peter Ansell, Bridget Riley-Gillis, Jacob Degner, Thanh Bui, Priya Narayanan, Josue Samayoa, Xin Huang, Robert F. Coleman, Jeffrey F. Waring. Multi-omic characterization and predictive features of advanced ovarian cancer patients in a large phase III cohort. [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 5426.