Construction of a clinical prediction model for complicated appendicitis based on machine learning techniques

Area of Science:

• Geriatric Surgery
• Medical Informatics
• Machine Learning in Medicine

Background:

• Acute appendicitis is a common surgical emergency in the elderly, posing diagnostic challenges.
• Accurate diagnosis is crucial for effective treatment planning and improved patient outcomes in this demographic.

Purpose of the Study:

• To develop and validate a simple, fast, and accurate machine learning model for early diagnosis of acute appendicitis in elderly patients.
• To compare the performance of various machine learning algorithms for appendicitis diagnosis.

Main Methods:

• Analysis of clinical data from 322 elderly patients (2012-2022) with acute appendicitis.
• Implementation and comparison of nine machine learning techniques: LR, CART, RF, SVM, Bayes, KNN, NN, FDA, and GBM.
• Interpretation of the optimal model (GBM) using SHAP and analysis of clinical utility via calibration and decision curves.

Main Results:

• The Gradient Boosting Machine (GBM) algorithm achieved optimal diagnostic performance.
• Key performance metrics for GBM included: sensitivity 0.9167, specificity 0.9739, precision 0.9429, and F1-score 0.9296.
• The developed Shiny application demonstrated clinical and economic benefits for diagnosing complicated appendicitis.

Conclusions:

• Machine learning, particularly the GBM algorithm, offers a highly accurate method for diagnosing acute appendicitis in the elderly.
• The developed tool can assist clinicians in making prompt and informed treatment decisions for acute appendicitis.
• The study highlights the potential of AI in improving healthcare delivery for surgical emergencies in geriatric populations.