Comparative Performance of Machine Learning Algorithms and Logistic Regression for Predicting In-Hospital Cardiac Arrest (Preprint)

BACKGROUND In-hospital cardiac arrest (IHCA) remains a catastrophic event with persistently low survival, even amid advances in resuscitation and critical care. Conventional early warning scores provide only limited predictive accuracy, often failing to identify patients at highest risk. To overcome these limitations, we evaluated a multimodal prediction framework that integrates traditional logistic regression with advanced machine learning (ML) approaches to optimize risk stratification. OBJECTIVE This study aimed to develop and compare the performance of multiple machine learning algorithms against traditional logistic regression for predicting IHCA using a comprehensive set of electronic health record (EHR) data. METHODS We conducted a retrospective case–control study including 800 IHCA cases and 3,464 matched controls from a large tertiary medical center. Candidate predictors comprised demographics, comorbidities, laboratory values, and vital signs. Five models—logistic regression, decision tree, random forest, XGBoost, and multivariate adaptive regression splines (MARS)—were trained and validated. Model performance was assessed using accuracy, sensitivity, specificity, F1 score, and area under the receiver operating characteristic curve. RESULTS Among all models, XGBoost demonstrated the best predictive performance (AUC 0.909; accuracy 0.883), followed by random forest (AUC 0.910; accuracy 0.876). Logistic regression achieved robust but comparatively lower performance (AUC 0.895; accuracy 0.876). Importantly, ML models highlighted clinically relevant predictors—such as blood urea nitrogen, heart rate, and presence of heart failure—providing novel insights beyond traditional regression. CONCLUSIONS Integrating ML-based approaches with conventional regression substantially improved IHCA risk prediction. While logistic regression offers transparency and interpretability, ML models capture complex, non-linear interactions that enhance accuracy. This multimodal framework holds potential to strengthen hospital early warning systems, enabling earlier detection, timely intervention, and ultimately improved patient outcomes. CLINICALTRIAL The study protocol was approved by the Institutional Review Board of National Taiwan University Hospital (IRB No. 201807063RINC). Due to the retrospective nature of this study, which involved the analysis of pre-existing data, trial registration was not required.