INTERPRETABLE PREDICTION OF GALLSTONE STATUS USING CLINICAL AND BIOCHEMICAL INDICATORS: A COMPARATIVE ANALYTICAL APPROACH

Abstract

Numerous metabolic, inflammatory, and biochemical factors influence gallstone disease, although much predictive research focuses on performance with models, but does not sufficiently consider interpretability and variable redundancy. This study examined the association between clinical and biochemical indicators and gallstone status and compared interpretable and non-linear predictive models within a structured analytical framework. A quantitative cross-sectional design based on secondary data analysis was employed. The dataset included 319 observations and 39 variables. Descriptive statistics, independent sample t-tests, correlation analysis, and multicollinearity assessment using the Variance Inflation Factor were conducted. Significant and non-collinear variables were retained for model development. Logistic Regression and Random Forest models were trained using an 80:20 train-test split and evaluated using accuracy, precision, recall, F1-score, and ROC-AUC. Eighteen variables showed significant differences between groups. C-reactive protein and Vitamin D emerged as the most influential predictors across both models. Logistic Regression achieved an accuracy of 0.781 and ROC-AUC of 0.829, while Random Forest achieved the same accuracy with a higher ROC-AUC of 0.860. These findings indicate that inflammatory, metabolic, and biochemical indicators jointly contribute to gallstone classification. A compact set of routinely measurable variables can reasonably differentiate gallstone status, and the integration of statistical analysis with machine learning supports both predictive performance and interpretability.