Using interpretable machine learning model to predict lymph node metastasis in patients with localized prostate...

Area of Science:

• Urology
• Oncology
• Data Science

Background:

• Developing accurate prediction models for lymph node metastasis (LNM) in localized prostate cancer (PCa) is crucial for treatment planning.
• Routinely available preoperative variables offer a potential data source for such predictions.

Purpose of the Study:

• To develop and externally validate machine learning (ML) models for predicting LNM in localized PCa patients.
• To compare the performance of ML models against established clinical nomograms.

Main Methods:

• Retrospective recruitment of PCa patients from two institutions for primary and external validation cohorts.
• Development of five ML models using a training set and internal validation.
• Evaluation of model performance using Area Under the Receiver Operating Characteristic Curve (AUC) and SHapley Additive exPlanations (SHAP) for interpretability.

Main Results:

• The LightGBM model achieved an AUC of 0.8613 in internal validation and 0.8148 in external validation.
• LightGBM demonstrated superior performance compared to other ML algorithms.
• SHAP analysis identified key predictive features including PSA, BMI, Gleason grade group, and various ratios (NLR, PAR, PLR).

Conclusions:

• The LightGBM model, incorporating eight key variables, offers competitive predictive performance for LNM in PCa.
• This ML model shows potential as a preoperative decision-support tool, comparable to Briganti and MSKCC nomograms.
• Further prospective validation in diverse populations is recommended before routine clinical implementation.