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

Background Ruling out myocardial infarction (MI) in patients with an initial indeterminate (detectable to mildly elevated) troponin measure is challenging. Myocardial-Ischaemic-Injury Index (MI 3 ) is a machine-learning algorithm designed to diagnose MI, but its utility in patients with indeterminate troponins is unclear. This study seeks to evaluate its diagnostic performance in patients with an initial indeterminate troponin. Methods We conducted a secondary analysis of a cohort (Cardiovascular Magnetic Resonance-Invasive-based Strategies in Patients with Chest Pain and Detectable to Mildly Elevated Serum Troponin) of adult patients with symptoms suggestive of acute coronary syndrome and an initial clinical contemporary troponin of 0.006–1.0 ng/mL across four US hospitals. Patients with initial and 3-hour high-sensitivity cardiac troponin I (Abbott Laboratories) measures were classified by MI 3 into low-risk, intermediate-risk and high-risk groups. The primary outcome was adjudicated MI at 30 days. The sensitivity, specificity and negative likelihood ratio (−LR) of MI 3 for MI at 30 days were calculated and reported with 95% CIs. A receiver operator characteristics curve for MI at 30 days was created and area under the curve (AUC) for MI 3 was calculated. Results Among 207 patients, 34.3% (71/207) were female with a mean age of 61±11 years. MI at 30 days occurred in 43.5% (90/207). The AUC for MI 3 for the detection of MI at 30 days was 0.882 (95% CI 0.833 to 0.932). MI 3 classified 34.8% (72/207) of patients as low-risk, of which 8.3% (6/72) had MI at 30 days, yielding a sensitivity of 93.3% (95% CI 86.1 to 97.5%) and −LR of 0.12 (95% CI 0.05 to 0.26). Among the 47.3% (98/207) classified as intermediate-risk, MI at 30 days occurred in 48.0% (47/98). MI 3 classified 17.9% (37/207) as high-risk, among which 100% (37/37) had MI at 30 days, yielding a specificity of 100% (95% CI 96.9% to 100%). Conclusions Among emergency department patients with an initial indeterminate troponin measure, the MI 3 machine-learning algorithm had high AUC and specificity for 30-day MI.