Clinicogenomic Insights for Progression-Free Survival in Prostate Cancer

Prostate cancer (PrCa), the second most common cancer diagnosed in men globally, remains a critical challenge in precision oncology. While PrCa can be deadly, it is highly treatable if detected early. Identifying associative factors influencing disease progression risks can help inform preliminary steps that will further the expedition of clinical therapeutic intervention decisions, which will improve treatment outcomes. While conventional PrCa progression assessment tools rely heavily on a few clinical parameters, the importance of genomic information is increasingly recognized. In this study, we evaluate the prognostic value of patients' clinicogenomic profiles in modeling progression-free survival (PFS) of PrCa. Three survival models, namely the penalized Cox model, random survival forest, and a deep learning survival neural network, were deployed with extensive tuning applied to a dataset for a cohort of 494 patients with PrCa. This dataset, compiled from public data in The Cancer Genome Atlas (TCGA) accessed via cBioPortal, consists of relevant clinical features and single-nucleotide variant information on likely PrCa-related genes. The survival models demonstrated satisfactory discriminatory performance, with Harrell's concordance index ranging from approximately 0.80 to 0.87 on held-out test data, indicating their ability to rank patients according to their relative progression risk among patients, while exhibiting distinct dynamics, all three models consistently identified clinical variables that indicated neoadjuvant treatment history, neoplasm cancer status, and tumor recurrence as well as the gene MYH6 as important predictor variables for PrCa PFS. Our findings suggest the incorporation of genomic data into the survival modeling workflow, thereby allowing the use of integrated clinicogenomics information to gain insights into progression risks for patients with PrCa.