Artificial Intelligence-Based Risk Prediction Models for Complications After Tongue Cancer Surgery

Area of Science:

• Oncology
• Surgical Oncology
• Data Science in Medicine

Background:

• Glossectomy for tongue tumors presents significant postoperative risks.
• Current risk stratification tools lack individualized precision for surgical planning.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting major 30-day postoperative complications after glossectomy.
• To compare the performance of ML models against the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) risk calculator.

Main Methods:

• Retrospective cohort study using ACS-NSQIP data (2008-2024) from over 700 US hospitals.
• Trained logistic regression and five ML models (neural network, support vector classifier, LightGBM, XGBoost, stacked generalization) on 85:15 train-validation split (2008-2023 data).
• Tested models on 2024 data, assessing prediction performance using risk stratification, discrimination (AUC-ROC, AUC-PR), and calibration (Brier score).

Main Results:

• Developed PRO-TONGUE, an outcome-specific risk prediction tool for tongue cancer surgery, using data from 8266 adult patients.
• ML models demonstrated performance comparable to the ACS-NSQIP calculator across outcomes.
• XGBoost and LightGBM were optimal for specific complications, with superior prediction for bleeding requiring transfusion (AUC-ROC 0.88-0.90).

Conclusions:

• ML models show excellent performance in predicting major 30-day complications following glossectomy for tongue cancer.
• The developed PRO-TONGUE tool offers individualized, interpretable risk estimates to enhance preoperative planning.
• Further external validation of these ML-driven risk prediction models is warranted.