Evaluating Treatment Outcomes of Acute Infectious Diseases Using Machine Learning and Ordered Logit Models

Abstract

Acute infectious diseases, especially of viral origin, remain a major global health concern. This study evaluates recovery outcomes using nine machine learning (ML) algorithms including Ordered Logit, Random Forest, Light GBM, and Naive Bayes to identify key predictors such as age, hospital stay duration, and treatment costs. Data were collected from 5,066 patients hospitalized for respiratory infections at the National Center for Communicable Diseases, Mongolia (2022–2024). Recovery was assessed at admission and discharge, categorized into four ordinal levels (0–3). Machine learning models such as Gradient Boosting and SVM achieved the highest predictive accuracy, while the Ordered Logit model offered interpretability, highlighting significant variables including age, length of stay, drug expenditures, pregnancy status, and year of hospitalization. The study demonstrates the complementary value of statistical and ML approaches in predicting clinical outcomes. Future research should explore additional variables such as genetics and mental health to improve model performance.