Diagnostic performance of the Myocardial-Ischaemic-Injury index machine-learning algorithm in patients with an initial indeterminate troponin

Comprehensive Summary

Snavely et al. retrospectively evaluated the diagnostic performance of the Myocardial-Ischaemic-Injury index (MI³) in adult patients with symptoms of acute coronary syndrome (ACS) and detectable to mildly elevated troponin. Across four emergency departments (EDs), 207 patients were classified by MI³ into low, intermediate and high-risk based on initial and 3-hour high-sensitivity cardiac troponin I. The primary endpoint was the proportion of MIs at 30 days. MI³ was assessed by sensitivity, specificity, negative likelihood ratio (−LR), and area under the curve (AUC) for MI at 30 days. The study found that MI³ detected MI at 30 days with an AUC of 0.882 (95% CI: 0.833-0.932). Of 72 patients (34.8%) classified as low-risk, 6 patients (8.3%) had MI at 30 days, giving a sensitivity of 93.3% (95% CI: 86.1-97.5%) and −LR of 0.12 (95% CI: 0.05-0.26). There were 98 patients (47.3%) classified as intermediate-risk, of which 47 (48%) had an MI at 30 days. The algorithm determined 37 patients (17.9%) to be high-risk, among which all (100%) had MI at 30 days, resulting in a specificity of 100% (95% CI: 96.9%-100%).

Outcomes and Implications

Clinically, it remains challenging to effectively detect MI in patients with detectable to mildly elevated troponin. This study demonstrated that the MI³ had a strong performance and specificity for 30-day MI in ED patients with ACS symptoms and indeterminate troponin. Therefore, the algorithm may guide clinicians in diagnosing MI in this patient population. However, the algorithm lacked sensitivity to rule out MI and major cardiovascular events (MACE) and adding natriuretic peptide measures did not significantly enhance its diagnostic performance. In addition, the study was limited by a small, non-representative sample, the use of retrospective stored samples and its secondary analysis design. Larger prospective studies are warranted to generalize the MI³ algorithm’s performance across broader clinical settings.