Differentiating Type 1 and Type 2 myocardial infarction using a machine learning algorithm and biomarkers

Treatment of myocardial infarction (MI) differs based on MI type, which can be difficult to determine in the emergency department. This study sought to evaluate whether a machine learning algorithm (MI³) in combination with N-Terminal pro B-type natriuretic peptide (NT-proBNP) and galectin-3 (Gal-3) can accurately differentiate MI type.

We conducted a secondary analysis of the multisite CMR-IMPACT trial, which prospectively enrolled adults with symptoms of acute coronary syndrome and an initial indeterminate troponin. Patients with an adjudicated diagnosis of MI and an initial high-sensitivity cardiac troponin I (hs-cTnI; Abbott Laboratories) measure were included. Incidence of MI and MI type were adjudicated by expert reviewers. Receiver operator characteristic curves for MI-type were created and area under the curve (AUC) calculated for MI³ and MI³ with NT-proBNP and Gal-3. AUCs were compared using DeLong's method.

Among 123 patients with adjudicated MI, the mean age was 60±12 years and 37.4% (46/123) were female. Type 1 MI occurred in 58.5% (72/123) and type 2 MI in 41.5% (51/123). MI³ based on an initial hs-cTnI yielded an AUC of 0.704 (95% CI 0.611-0.797) for MI type. When combined with NT-proBNP and Gal-3, AUC improved to 0.789 (95% CI 0.709-0.869, p=0.0165). In patients with serial hs-cTnIs (n = 86), the AUC of MI³ for MI type was 0.721 (95% CI 0.614-0.829) and increased to 0.797 (95% CI 0.700-0.894, p=0.09) with the addition of NT-proBNP and Gal-3.

Adding NT-proBNP and Gal-3 to the MI³ machine learning algorithm shows promise in differentiating type 1 from type 2 MI.