Development and External Validation of a Machine Learning Model for Prediction of Lymph Node Metastasis in Patients
Ali Sabbagh1, Samuel L Washington2, Derya Tilki3
1Department of Radiation Oncology, University of California-San Francisco, San Francisco, CA, USA.
European Urology Oncology
|March 3, 2023
Summary
Machine learning (ML) models, particularly XGBoost, significantly improve the prediction of lymph node involvement (LNI) in prostate cancer patients. These advanced tools outperform traditional methods, enabling better patient selection for pelvic lymph node dissection (PLND).
Area of Science:
- Oncology
- Medical Informatics
- Machine Learning
Background:
- Pelvic lymph node dissection (PLND) is standard for detecting lymph node involvement (LNI) in prostate cancer.
- Traditional tools like the Roach formula, MSKCC calculator, and Briganti nomogram estimate LNI risk but have limitations.
Purpose of the Study:
- To evaluate if machine learning (ML) can enhance patient selection for PLND.
- To compare the predictive performance of ML models against established tools for LNI risk.
Main Methods:
- Retrospective analysis of 21,589 prostate cancer patients (1990-2020) from two institutions.
- Trained logistic regression and XGBoost models using clinicopathologic variables (age, PSA, T stage, positive cores, Gleason score).
- Externally validated models and compared performance (AUC, calibration, DCA) against traditional nomograms.
Main Results:
- XGBoost demonstrated superior performance in predicting LNI on external validation.
- ML models, especially XGBoost, showed higher AUC, better calibration, and improved clinical utility compared to Roach, MSKCC, and Briganti tools.
- The study identified ML as a more accurate predictor of LNI than existing methods.
Conclusions:
- Machine learning models utilizing standard clinicopathologic variables significantly outperform traditional tools for predicting lymph node involvement in prostate cancer.
- ML offers improved accuracy and clinical utility for selecting patients who would benefit from pelvic lymph node dissection.
