Truace
Subscribe

Development and External Validation of a Machine Learning-Derived Risk Stratification Score for Metastatic Castration-Resistant Prostate Cancer

Purpose Metastatic castration-resistant prostate cancer (mCRPC) shows marked clinical heterogeneity and poor long-term survival, highlighting the need for simple and reliable risk stratification tools beyond the limited usability of existing complex prognostic models. In this study, we developed and validated PROGRESS (Prostate Cancer Global Risk Evaluation and Stratification Score), a simplified prognostic score derived through machine learning‑guided feature selection distilled into a simple laboratory-based t…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-2138· cite & verify
Development and External Validation of a Machine Learning-Derived Risk Stratification Score for Metastatic Castration-Resistant Prostate Cancer
"Kent Oncology Centre" by llamnudds is licensed under CC BY-SA 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by-sa/2.0/.

In brief

Researchers trained a model on baseline data from 2,035 patients with metastatic castration-resistant prostate cancer. They used unsupervised machine learning to find subgroups with different survival, then reduced the findings to a score called PROGRESS.

The score matters because it relies on cheap, routine blood tests rather than complex prognostic models, which could make risk assessment easier in practice. The authors say it still needs prospective real-world validation before its usefulness is established.

Main points

  1. The training set contained 2,035 patients with mCRPC, and external validation used two independent data sets totaling 1,239 patients.
  2. Unsupervised modeling identified two risk subpopulations with significantly different overall survival, with a median of 27.4 versus 17.7 months in one comparison.
  3. PROGRESS is composed of three independent predictors: PSA, ALP and AST.
  4. The authors suggest uses in individual risk stratification, trial enrichment and early decision making.

The gain

A machine learning-guided process produced PROGRESS, a simple score built from PSA, ALP and AST that separated mCRPC patients into risk groups and held up in two independent data sets.

The rundown

The method embedded baseline patient data into synthetic representations, reduced them with UMAP, and clustered them with DBSCAN to discover subpopulations related to outcome. Feature contribution analysis then informed a multivariable model.

Performance was assessed with Kaplan-Meier and Cox regression for overall survival, and with ROC-AUC, AUC(t) and Harrell C-index for discrimination. The validation data sets covered different therapeutic settings.

The authors frame the tool as an alternative to existing prognostic models whose complexity limits their usability in mCRPC, a disease with marked clinical heterogeneity and poor long-term survival.

Sources

  1. Peer-reviewedJCO Clinical Cancer Informatics2026-10-20

The debate