Machine Learning for Predicting Mortality in Intensive Care Unit Patients: A Prognostic Performance Systematic Review and Meta‐Analysis

ABSTRACT Background Predicting patient outcomes upon ICU admission remains challenging; however, early mortality prediction could facilitate timely interventions. A comprehensive synthesis of the literature on the performance of machine learning models for predicting ICU mortality is currently lacking. Aim This study aimed to review the literature on machine learning models for predicting mortality in ICU patients and to conduct a meta‐analysis to summarise pooled performance estimates. Study Design This is a systematic review and meta‐analysis. PubMed, Cochrane Library, Web of Science and Embase were systematically searched for studies published between January 1, 2014, and December 10, 2024, that evaluated the performance of machine learning models in predicting ICU mortality. Two reviewers independently extracted data and assessed the risk of bias. A bivariate mixed‐effects model meta‐analysis was performed using Stata (version 14.0) to synthesise predictive performance indicators. Results Forty studies involving 317 028 patients and comprising 123 machine learning models were included. The most commonly used methods were logistic regression, random forest and XGBoost. Commonly identified predictive features included age, blood urine nitrogen, heart rate, respiratory rate, SpO 2 and Glasgow Coma Scale score. The pooled area under the receiver operating characteristic curve was 0.83 (95% CI: 0.80–0.86). The pooled sensitivity was 0.72 (95% CI: 0.68–0.76, I 2 = 99.66%, p < 0.001), and the pooled specificity was 0.81 (95% CI: 0.77–0.84, I 2 = 99.93%, p < 0.001). Conclusions Machine learning models show strong potential for predicting mortality in ICU patients. However, most studies lacked external validation, and reliance on retrospective cohorts for model development increased the risk of bias. Further multicenter prospective studies are necessary to validate and enhance model reliability. Relevance to Clinical Practice Machine learning‐based mortality prediction models, when integrated into clinical decision‐support systems, have the potential to significantly enhance nursing practice. These tools could assist in the early identification of high‐risk patients, enabling prioritisation for intensified monitoring, facilitating timely escalation of care and supporting proactive goals‐of‐care discussions with families. Future implementation should focus on improving model interpretability and seamless workflow integration to foster trust and utility among frontline physicians and nurses.