Length of Stay Prediction Models for Oral Cancer Surgery: Machine Learning, Statistical and ACS-NSQIP

Area of Science:

• Oncology
• Surgical Outcomes Research
• Health Informatics

Background:

• Accurate prediction of hospital length of stay (LOS) after oral cavity cancer (OCC) surgery aids patient counseling and resource management.
• Existing prediction tools may lack sufficient accuracy for complex reconstructive procedures.

Purpose of the Study:

• To compare the predictive performance of statistical models, a machine learning (ML) model, and the ACS-NSQIP calculator for LOS in OCC patients.
• To identify novel predictors of LOS in this patient population.

Main Methods:

• Retrospective multicenter study of 837 patients undergoing free flap reconstruction for OCC.
• Development and validation of statistical and ML models using training and validation datasets.
• Performance evaluation based on correlation coefficients and percent accuracy of predicted vs. actual LOS.

Main Results:

• The ML model achieved the highest accuracy (validation correlation 0.48, 70% 4-day accuracy).
• Statistical models showed moderate performance (multivariate analysis: 0.45, 67%; LASSO: 0.42, 70%).
• The ACS-NSQIP calculator demonstrated the lowest accuracy (0.23, 59%).

Conclusions:

• Developed statistical and ML models accurately predict LOS after OCC free flap reconstruction.
• ML and statistical models significantly outperformed the ACS-NSQIP calculator.
• The ML model identified new predictors of LOS, requiring further external validation for clinical use.