Artificial intelligence for the early prediction of clinical deterioration in emergency department patients: A systematic review and meta-analysis

Background: Emergency department (ED) visits have increased exponentially in many countries, leading to overcrowding. This overcrowding results in delays in care across most EDs and has been associated with poorer outcomes, including increased morbidity and mortality. Triage algorithms were developed and widely implemented in EDs worldwide to identify patients requiring resources and to allocate those resources appropriately. However, despite the use of algorithms, the care delivery remains highly variable. Machine learning (ML) models have been developed to improve clinicians’ predictive ability to identify high-risk patients and to improve resource allocation in EDs globally. This systematic review and meta-analysis, therefore, outlines the current evidence on the efficacy of artificial intelligence (AI) in predicting patients’ clinical deterioration in the ED. Methodology: A comprehensive literature search was conducted across five electronic databases—ScienceDirect, PubMed, Google Scholar, CENTRAL, and EMBASE—to identify all relevant articles published up to February 2025. Data were extracted from the studies that met the inclusion criteria, and the reported outcomes were pooled using the MedCalc statistical software. Results: A total of 743 articles were identified through the literature search; however, only nine met the inclusion criteria and were thus included in the review. We pooled the area under the receiver operating characteristic curve (AUROC) of ML models for predicting the occurrence of in-hospital cardiac arrest (IHCA) and observed a pooled Area Under the Curve (AUC) of 0.909 (95% CI: 0.889–0.929; p < 0.001). The models showed high predictive ability for the other clinical deterioration outcomes in the ED. However, we were unable to pool the results because only a limited number of studies reported these different outcomes. Conclusion: Our systematic review and meta-analysis found that ML models have a high predictive ability in identifying patients at risk of clinical deterioration in the ED. However, limited data exist on the additional benefits clinicians gain from incorporating AI into routine ED patient care. Moreover, although many models have been developed, further studies are needed to evaluate their clinical utility.