Machine learning-based prediction of acute and complicated appendicitis using readily available data in low-resource...

Area of Science:

• Medical Informatics
• Computational Biology
• Diagnostic Medicine

Background:

• Acute appendicitis diagnosis is challenging, particularly in pediatric and elderly patients, with delays leading to severe complications.
• Imaging is not universally accessible, highlighting the need for alternative diagnostic methods.
• Machine learning (ML) presents a promising avenue for improving diagnostic accuracy using readily available data.

Purpose of the Study:

• To develop a simple and accurate ML model for diagnosing acute appendicitis.
• To utilize basic demographic and laboratory data for improved diagnosis, especially in low-resource settings.
• To assess the feasibility of ML in aiding appendicitis diagnosis without advanced imaging.

Main Methods:

• A retrospective analysis of 453 patients who underwent appendectomy for suspected acute appendicitis.
• Collection and preprocessing of clinical, laboratory (including Complete Blood Count - CBC and C-reactive protein - CRP), and histopathological data.
• Training and evaluation of seven ML models using stratified five-fold cross-validation, with SHAP for interpretability.

Main Results:

• Appendicitis was confirmed in 68.87% of patients, with significant differences in age, white blood cell (WBC) count, neutrophils, CRP, and lymphocyte levels compared to controls.
• The Support Vector Classifier (SVC) model achieved the highest accuracy (75.82%) and ROC-AUC (76.39%) for appendicitis classification.
• SHAP analysis identified WBC, lymphocyte percentage, gender, age, and neutrophil percentage as key predictors. SVC showed moderate accuracy for differentiating complicated appendicitis but struggled with precision.

Conclusions:

• ML models utilizing CBC and CRP data demonstrate preliminary potential for predicting appendicitis.
• The study's findings suggest that ML can aid in appendicitis diagnosis, particularly in resource-limited environments.
• Further validation is required due to the surgical-only cohort and modest performance before widespread clinical implementation.