Building and validating machine learning models to predict appendiceal perforation during conservative treatment of...

Area of Science:

• Medical Informatics
• Surgical Oncology
• Machine Learning in Medicine

Background:

• Fecalith-associated appendicitis poses a high risk of perforation during conservative management.
• Accurate prediction of perforation risk is crucial for timely clinical decision-making and optimizing patient care.
• Existing methods lack sufficient precision in identifying high-risk patients early in conservative treatment.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting appendiceal perforation risk in patients with fecalith-associated appendicitis undergoing conservative treatment.
• To identify key clinical, laboratory, and imaging features that predict perforation risk.
• To create a risk stratification tool for guiding clinical management.

Main Methods:

• Retrospective cohort study of 1247 patients with fecalith-associated appendicitis treated conservatively across four centers.
• Development and validation of 20 ML algorithms using clinical, laboratory, and imaging data, with LASSO regularization for feature selection.
• External validation in an independent cohort of 225 patients; performance assessed using AUC, sensitivity, specificity, PPV, and NPV.

Main Results:

• An ensemble Gradient Boosting model demonstrated high predictive performance (AUC 0.892) for appendiceal perforation within 72 hours.
• External validation confirmed generalizability (AUC 0.909), with key predictors including fecalith size, CRP, WBC count, and appendiceal wall thickness.
• Risk stratification identified distinct patient groups: low-risk (3.8% perforation), moderate-risk (24.6%), and high-risk (71.3%), showing significant clinical utility.

Conclusions:

• Validated ML models, particularly Gradient Boosting, accurately predict appendiceal perforation risk in fecalith-associated appendicitis during conservative treatment.
• These models offer clinically actionable risk stratification, aiding in treatment decisions, patient monitoring, and potentially preventing unnecessary surgeries.
• The findings support the use of ML-driven tools to enhance patient management and improve outcomes in this specific appendicitis subtype.