Machine Learning-Based Analysis of Diagnostic Markers for Malnutrition.

Abstract

Introduction: Malnutrition leads to negative health outcomes such as delayed recovery, increased hospital stays, and higher costs. The study aimed to develop a predictive tool to diagnose malnutrition using patient data, including anthropometric, phenotypic, and laboratory information.

Methods: A cohort of 252 adult patients was assessed at a tertiary hospital. Logistic regression and decision tree analysis were applied to evaluate the role of the data in predicting the risk of malnutrition. The performances of the models were tested with Akaike information criterion, null deviance, residual deviance, accuracy, and kappa metrics, and the statistical significance of the variables was evaluated with Wald-Z test. Attribute importance ranking was obtained by Bootstrap Optimized Random Univariate Tree Analysis algorithm.

Results: A total of 252 patients, 125 female (49.6%) and 127 male (50.4%), were included in the study. The mean age was 72.57 ± 13.6 years. Malnutrition was diagnosed in 174 patients (69%). According to the equations, the most important characteristics were determined as hand grip strength, weight loss, BMI, and gender. The model achieved high accuracy (89.8%) and a kappa value of 0.79, demonstrating its potential for clinical application.

Conclusion: Machine learning model offers a faster, more efficient alternative to traditional diagnostic tools. The study concludes that machine learning-based models provide superior prediction performance and could significantly enhance the efficiency and accuracy of malnutrition diagnosis in healthcare.