Risk prediction and effect evaluation of complicated appendicitis based on XGBoost modeling

Area of Science:

• Medical diagnostics
• Machine learning in healthcare
• Surgical outcomes analysis

Background:

• Distinguishing complicated appendicitis (CAP) from uncomplicated appendicitis (UAP) presents a clinical challenge.
• Accurate and rapid differentiation is crucial for appropriate patient management and resource allocation.

Purpose of the Study:

• To develop a safe, economical, and accurate diagnostic model for differentiating CAP from UAP.
• To leverage machine learning for improved appendicitis diagnosis.

Main Methods:

• Retrospective analysis of 773 appendectomy patient data.
• Feature selection using Random Forests and data splitting (3:1 ratio).
• Comparison of Extreme Gradient Boosting (XGBoost) with Support Vector Machine (SVM), Random Forest (RF), and Decision Tree (CART) algorithms using AUC, sensitivity, specificity, and other metrics.

Main Results:

• All four models showed predictive capabilities, with XGBoost achieving the highest Area Under the Curve (AUC) of 0.914.
• XGBoost demonstrated superior accuracy (0.855), sensitivity (0.865), specificity (0.846), Negative Predictive Value (NPV) (0.848), and Positive Predictive Value (PPV) (0.897).
• XGBoost and SVM models exhibited excellent calibration, and XGBoost showed superior clinical utility via decision curve analysis.

Conclusions:

• XGBoost-based predictive models effectively predict the risk of acute appendicitis progressing to complicated appendicitis.
• These models offer a valuable tool for optimizing clinical decision-making in appendicitis diagnosis and management.