Explainable machine learning can outperform Cox regression predictions and provide insights in breast cancer survival
Problem
ML survival models are often disregarded in oncology due to lack of transparency, leaving Cox regression as the standard.
Why it matters
More accurate, explainable survival predictions could improve treatment decisions for breast cancer patients.
What the AI did
Compared Cox Proportional Hazards with Random Survival Forests, Survival SVMs, and Extreme Gradient Boosting on registry data, showing XGB outperformed CPH while providing explainability.
Results
ML performed at least as well as CPH; XGB performed better; c-index values omitted from abstract as formula placeholders
Impact
Showed explainable ML can match or exceed Cox regression for survival prediction on one large dataset.
Record
| Venue | Scientific Reports |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Arturo Moncada‐Torres, Marissa Corine van Maaren, Mathijs P. Hendriks, Sabine Siesling, Gijs Geleijnse |
| Citations | 335 (OpenAlex, at import time) |
| References | 72 |
| Model or method | Random Survival Forests, Survival Support Vector Machines, Extreme Gradient Boosting (XGB) |
| Organization | Netherlands Cancer Registry (data source) |
| Primary source | https://doi.org/10.1038/s41598-021-86327-7 |
| Open access | gold |
| Keywords | machine learning, breast cancer, survival analysis, XGBoost, explainability |
Importance rating
real but narrower advance
Explainable ML matching/exceeding Cox regression could improve breast cancer survival prediction and clinical adoption; single-dataset study, narrower advance.