Development and external validation of multiple machine learning-based models for breast cancer-specific survival prediction in postoperative patients with invasive breast cancer: a study based on the SEER database and an external cohort

Description

ShengSheng Liu and colleagues built machine learning models to forecast long-term survival after surgery for invasive breast cancer. Published in 2026, the study used United States SEER registry data from 2010 through 2020 plus an outside patient group from a cancer hospital in Guangxi. Four models competed to predict cancer-specific survival at three, five, seven, and ten years.

A random survival forest model led the field, scoring strongest on the training data and holding the best accuracy on both internal and outside checks. Its calibration was sound and its clinical value stayed stable on decision analysis. Feature-importance diagrams showed which patient factors drove the forecasts, and the model sorted patients into distinct risk bands. The authors present it as a tool for tailoring follow-up and treatment talks. Readers get a validated predictor, with the usual caution that models need local testing before routine use.

Catalog record

Source: https://europepmc.org/articles/PMC13520318