AI-guided refinement of coronary revascularization need in patients suspected of acute coronary syndrome

TL;DR

This study uses machine learning to better predict the need for coronary revascularization in patients suspected of heart attacks, reducing unnecessary procedures.

Contribution

A novel machine learning approach integrating prehospital and in-hospital data to improve revascularization decision-making and reduce overdiagnosis.

Findings

Machine learning models improved prediction accuracy as more diagnostic data became available.

Model 4, including sequential troponin testing, achieved the highest AUROC of 0.87.

Overdiagnosed patients often had ST-elevations and wall motion abnormalities but no significant troponin elevation.

Abstract

Overdiagnosis in patients suspected of acute coronary syndrome (ACS) leads to unnecessary coronary angiographies, particularly in cases with non-specifically elevated troponin (Trop) levels. We established machine learning (ML) models integrating sequentially available prehospital and in-hospital variables to improve early prediction of the need for coronary re while minimizing overdiagnosis.

Retrospective cohort study analysing patients with suspected ACS from 2016 to 2020. Machine learning models were trained using data available at different diagnostic time points, including prehospital assessment, arterial blood gas analysis, full laboratory results, and sequential Trop measurements. A total of 2756 patients were included, identified through emergency physician protocols for ACS-related complaints. Patients with incomplete data or prehospital mortality were excluded.

Model performance improved with additional diagnostic data. Model 1 (prehospital data only) achieved an area under the receiver operating characteristic (AUROC) of 0.76 (95% confidence interval [CI] 0.72–0.79), while Model 4 (including sequential Trop testing) reached 0.87 (95% CI 0.83–0.91). Adding early hospital diagnostics (Model 2) significantly improved accuracy compared with Model 1 (0.65 vs. 0.78). Sequential Trop testing in Model 4 did not substantially enhance performance compared with single Trop testing in Model 3 (AUROC 0.87, 95% CI 0.83–0.91 vs. 0.86, 95% CI 0.82–0.91). Misclassification analysis revealed that underdiagnosed patients were typically older females with dyspnoea and known coronary artery disease but no ST-elevations. Overdiagnosed patients had higher body mass index, ST-elevations, regional wall motion abnormalities, and impaired left ventricular ejection fraction but lacked significant sequential Trop elevation.

Prehospital assessments combined with early in-hospital diagnostics provide reliable stratification of coronary intervention need, potentially optimizing clinical decision-making and resource utilization.

Graphical Abstract

Figures