Patient Stratification for Improving Acute Chest Pain Management and Mitigate ED Crowding (Preprint)

<sec> <title>UNSTRUCTURED</title> Objectives: Artificial intelligence (AI) models were developed to support clinical decision-making in the initial management of patients with acute chest pain. The models can accurately detect ACS, as well as rapidly and reliably identify low-risk patients based on one single hsTnT test result. By integrating this AI-assisted strategy into clinical workflows, we aim to reduce ED length of stay, alleviate crowding, and ultimately improve healthcare efficiency while lowering medical expenditures. Methods: We conducted a retrospective study at a tertiary teaching hospital using data from 2015 to 2022. Models based artificial neural networks (ANN) were trained to build multi-classifiers for placing patients into three classes: critical patients with ACS, critical patients without ACS, and non-critical patients. Model performance was evaluated using AUROC, AUPRC, and subgroup-specific metrics. Results: After excluding ED visits with missing triage data, incomplete medical histories, or unsuitable dispositions, 17,935 visits were included. All ANN models demonstrated strong AUROC performance, with one showing the highest AUPRC (0.946). Regarding multi-classification, one classifier achieved excellent intended performance: it maintained high sensitivity (0.917) for identifying ACS patients, while achieving high PPV (0.901) and sensitivity (0.861) for identifying non-critical patients. Medically interpretable features which drives the predictive power were identified. Model performance supported the expansion of single-sample hs-TnT protocols. Conclusions: The ANN-based classifiers developed in this study offer effective clinical decision support for evaluating acute chest pain in the ED. By improving risk stratification and accurately identifying both high- and low-risk patients, the models can enhance diagnostic accuracy and help reduce ED overcrowding. </sec>