Interpretable machine learning model predicts 1-year inguinal hernia risk after robot-assisted radical prostatectomy

Area of Science:

• Urology
• Surgical Oncology
• Artificial Intelligence in Medicine

Background:

• Inguinal hernia is an underreported complication following robot-assisted radical prostatectomy (RARP).
• Limited predictive tools exist for post-RARP inguinal hernia, impacting patient quality of life.
• There is a need for accurate prediction models to identify patients at high risk.

Purpose of the Study:

• To develop and validate the first machine learning (ML) model for predicting inguinal hernia within one year of RARP.
• To utilize explainable AI (XAI) for clinical interpretability of the predictive model.
• To identify key clinical predictors of post-RARP inguinal hernia.

Main Methods:

• Retrospective analysis of 652 patients undergoing RARP for localized prostate cancer.
• Development and evaluation of five ML algorithms using a 70:30 training-test split.
• Utilized LASSO regression for feature selection and SHAP for model interpretability.

Main Results:

• The XGBoost model achieved an AUC of 0.833 (validation) and 0.791 (test) for predicting inguinal hernia.
• Key predictors identified: age, BMI, preoperative albumin, T stage, and history of abdominal surgery.
• The model provides interpretable insights into hernia risk factors.

Conclusions:

• The first ML-based predictive model for post-RARP inguinal hernia was established.
• The XGBoost model demonstrates robust performance in identifying at-risk patients.
• Personalized interventions can be offered to high-risk individuals identified by the model.