What question did this study set out to answer?

The research seeks to develop and validate machine learning models for predicting inpatient deterioration using electronic health record data.

February 14, 2026Open Access

Advanced Machine Learning Techniques for Predicting Inpatient Deterioration in General Medicine

Key Points

The research seeks to develop and validate machine learning models for predicting inpatient deterioration using electronic health record data.
Conducted a retrospective cohort study on 524 hospitalized patients.
Utilized the Synthetic Minority Over-sampling Technique (SMOTE) to balance training data.
Employed feature selection with SelectKBest to refine input variables.
Trained various models including Logistic Regression, Random Forest, and Neural Networks, evaluated using AUC and accuracy.
Performed 5-fold cross-validation for model stability assessment.
Random Forest model achieved a mean AUC of 0.980 during validation.
Final optimized Random Forest model achieved an AUC of 0.837 and accuracy of 85.4% on an independent test set.
Identified functional status, oxygen requirements, and urea levels as critical predictors.

Abstract

Inpatient deterioration, marked by ICU transfer or mortality, remains a critical challenge in hospital settings. While traditional early warning systems (EWS) have limitations, machine learning (ML) offers a promising approach for the early identification of at-risk patients. This study aimed to develop and validate multiple ML models for predicting inpatient deterioration among general medical patients using electronic health record (EHR) data. A retrospective cohort study was conducted on 524 patients admitted between January 2022 and December 2023. The dataset included demographic, clinical, and laboratory variables, with time-stamped measurements treated as distinct features. After excluding variables with >15% missing data, standard imputation was performed. The training data was balanced using the Synthetic Minority Over-sampling Technique (SMOTE), and feature selection was performed using SelectKBest. A range of models—including Logistic Regression, Random Forest, Gradient Boosting, Support Vector Machines (SVMs), and Neural Networks—were trained and evaluated using AUC, accuracy, precision, recall, and F1-score. During 5-fold cross-validation, the models demonstrated high stability, with the Random Forest achieving a mean AUC of 0.980. On the final independent test set, the optimized Random Forest model yielded the highest performance with an AUC of 0.837 and an accuracy of 85.4%. Functional status, oxygen requirements, and urea levels were identified as key predictors. ML models, particularly Random Forest, can significantly enhance the early detection of inpatient deterioration. The contribution of this work lies in its systematic comparison of multiple algorithms and its robust methodology. Future research should focus on external validation, the integration of temporal data using recurrent neural network architectures, and the application of Explainable AI (XAI) to foster clinical trust and facilitate implementation.

Bookmark

View Full Paper