Statistical models versus machine learning approach for competing risks in proctological surgery.

Area of Science:

• Surgical risk prediction
• Machine learning in medicine
• Proctological surgery outcomes

Background:

• Clinical risk prediction models are essential in surgery.
• Traditional models use regression analysis.
• Machine learning offers advanced predictive capabilities.

Purpose of the Study:

• Compare machine learning (ML) vs. logistic regression (LR) for preoperative risk assessment.
• Evaluate ML and LR models in proctological surgery.
• Assess prediction of complications in hemorrhoid disease surgery.

Main Methods:

• Utilized nationwide audit data of 1510 patients with Goligher's grade III hemorrhoids.
• Collected anthropometric, clinical, and surgical data for ten predictors.
• Compared LR with Decision Tree, Support Vector Machine, and Extreme Gradient Boosting ML techniques.

Main Results:

• ML and LR models demonstrated equivalent predictive performance.
• All models identified the same most important predictive factor.
• Performance metrics included AUC, balanced accuracy, sensitivity, and specificity.

Conclusions:

• ML and LR are comparable for preoperative risk assessment in this context.
• Interdisciplinary cooperation between statistical analysis and ML is encouraged.
• Focus on improving clinical decision-making through combined approaches.