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

## TL;DR

This paper explores using machine learning to predict appendicitis using basic patient data, aiming to help in areas with limited medical resources.

## Contribution

The study introduces a machine learning model for predicting appendicitis using accessible clinical data in low-resource settings.

## Key findings

- The Support Vector Classifier achieved 75.82% accuracy in predicting appendicitis.
- White blood cell count and age were key predictors of appendicitis.
- The model showed moderate success in distinguishing complicated from uncomplicated cases.

## Abstract

Acute appendicitis often presents diagnostic challenges, especially in pediatric and elderly patients. Delays can cause serious complications. While imaging aids diagnosis, it is not always accessible. Machine learning offers a promising solution. This study aimed to develop a simple, accurate model using basic demographic and laboratory data to improve diagnosis in low-resource settings.

This retrospective, single-center study analyzed 453 patients undergoing appendectomy for suspected acute appendicitis. Clinical, laboratory, and histopathological data were collected and classified into normal, uncomplicated, or complicated appendicitis. Data preprocessing included feature encoding, scaling, and balancing. Seven machine learning models were trained and evaluated using stratified five-fold cross-validation, and interpretability was assessed using SHapley Additive exPlanations (SHAP).

Among 453 patients, appendicitis was confirmed in 68.87%, predominantly in males (p < 0.001). Patients with appendicitis were significantly older (p = 0.0012), exhibiting elevated white blood cells(WBC) count, neutrophils, and C-reactive protein (CRP) levels, and lower lymphocytes (all p < 0.001). The Support Vector Classifier (SVC) performed best in classifying appendicitis (accuracy = 75.82%, ROC-AUC = 76.39%). SHAP analysis identified WBC, lymphocyte percentage, gender, age, and neutrophil percentage as influential predictors. For differentiating complicated versus uncomplicated appendicitis, SVC achieved moderate accuracy (70.19%) and ROC-AUC (76.33%), but low precision (14.85%) indicated challenges in minimizing false positives.

Machine learning models based on CBC and CRP show preliminary potential for predicting appendicitis, but given the surgical-only cohort and modest performance, further validation is needed before clinical use, particularly in low-resource settings.