Emergency departments (ED) manage over 140 million visits annually, ranging from life-threatening to non-urgent conditions. [1] Chest pain is the second leading cause for ED visits and a common indication for cardiac evaluation. [1], [2] Rapidly diagnosing or ruling out acute myocardial infarction (AMI) while assessing near-term risk places considerable demand on clinical decision-making. However, cardiac troponin testing is performed not only for suspected AMI but also for diverse presentations including suspected pulmonary embolism, heart failure, and undifferentiated critical illness.

Clinical pathways for diagnosing acute myocardial infarction (AMI) in the emergency department are well established and rely on integration of electrocardiography and serial cardiac troponin measurements. The Fourth Universal Definition of Myocardial Infarction defines AMI by evidence of myocardial ischemia in conjunction with a rise and/or fall in cardiac troponin concentrations, with at least one value exceeding the 99th percentile upper reference limit. [3] With the widespread adoption of high-sensitivity cardiac troponin (hs-cTn), diagnostic evaluation for AMI is often completed within hours of ED arrival, allowing rapid identification of patients requiring hospitalization and advanced therapy. [4], [5] While hs-cTn has substantially improved the efficiency of AMI diagnosis, it provides limited granularity for individualized short-term risk prediction among patients who do not meet criteria for myocardial infarction and/or have undergone troponin testing for different clinical reasons.

For patients not meeting criteria for immediate hospitalization, emergency physicians must determine who can be safely discharged versus who requires further evaluation. These decisions are guided by estimated 30-day major adverse cardiac event (MACE) risk. Numerous clinical tools including the HEART score, EDACS, and ESC 0/2-h algorithm stratify suspected acute coronary syndrome patients into risk groups informing disposition. [6], [7], [8], [9], [10], [11] However, these pathways were derived in selected chest pain populations and rely on subjective inputs, limiting applicability to broader populations undergoing troponin testing for diverse indications. Many patients are categorized as intermediate risk, creating uncertainty in disposition. This is evident among patients with low but detectable troponin concentrations who may be low risk but are managed conservatively, representing an opportunity to optimize resource utilization. [12]

Our study derives a novel machine learning algorithm designed to estimate the likelihood of 30-day MACE at an individual patient level, relying exclusively on objective data recorded in the electronic health record (EHR) prior to ED disposition. Notably, the algorithm does not rely on inputs that are subject to variable human interpretation, such as the clinical reading of ECG waveforms or subjective assessments of medical history. We hypothesized that our algorithm could identify a large cohort of patients as low risk for 30-day MACE while maintaining a negative predictive value (NPV) greater than 99%. This approach represents a significant step toward more precise risk stratification in emergency care settings, offering the potential to improve both patient outcomes and the efficiency of ED operations.