Towards Precision Oncology: A New Predictive Machine Learning Model for Early Progression to Castration Resistant Prostate Cancer

ABSTRACT Background and Objective Metastatic castration‐resistant prostate cancer (mCRPC) is an aggressive, lethal state of prostate cancer, for which early progression is an indicator of poor prognosis. The ability to predict this progression is of paramount clinical importance for guiding personalized treatment strategies. We aimed to develop and validate a novel machine learning (ML) model to predict early progression (≤ 12 months) to mCRPC and compare its performance against standard ML algorithms. Methods This was a retrospective analysis of 172 patients with mHSPC from the publicly available MSK‐IMPACT cohort. Inclusion criteria specified patients with mHSPC who had undergone genomic profiling and progressed to mCRPC during follow‐up. Patients with incomplete data were excluded. We collected 11 clinical, pathological, and genomic variables. The primary outcome was early progression (≤ 12 months) to mCRPC. Model performance was evaluated using a stratified fivefold cross‐validation, with AUC as the primary metric. Key Findings and Limitations A novel Rivality Index (RINH)‐based model, adapted from chemoinformatics, demonstrated significantly superior predictive performance (AUC: 0.86) compared to a panel of standard ML algorithms, none of which exceeded an AUC of 0.67. The model achieved an accuracy of 0.74, a sensitivity of 0.70, and a specificity of 0.77. Key limitations include the retrospective design and use of a single‐institution data set. Conclusions and Clinical Implications This novel RINH model offers a robust tool for risk stratification in mHSPC patients, capable of personalizing therapeutic strategies. However, external validation in multi‐center, prospective cohorts is an essential next step before its consideration as a clinical decision support tool.