A machine learning algorithm dubbed “MI3” can reportedly predict a person’s risk of heart attack with more nuance than existing algorithms, prompting its developers to claim it as “one of the first effective demonstrations” of how AI can be used to inform treatment decisions in the cardiology unit.
MI3—or the myocardial ischemic injury index—was piloted by Martin P. Than, MBBS, and his team, who published their findings in Circulation August 16. Than, of Christchurch Hospital in Christchurch, New Zealand, and colleagues noted that, while contemporary risk-assessment algorithms for MI are in many ways effective, they’re often based on fixed time points for sampling, fixed troponin thresholds and don’t account for any interaction between input variables.
Just one exception, the Troponin only-Manchester Acute Coronary Syndrome rule, wraps factors including age, sex, a handful of clinical variables and one high-sensitivity troponon T measurement into a singular model, which in testing achieved an area under the curve (AUC) of 0.90. Still, Than et al. said, even that model fails to take into account the “dynamic interaction” between variables and can only classify patients in one of four risk categories.
MI3 attempts to mitigate those shortcomings by generating a risk score that takes into consideration age, sex, paired cardiac troponin I concentrations and rate of change in troponin concentration to estimate both negative (NPVs) and positive predictive values (PPVs) for individual patients. The algorithm was trained on 3,013 patients with suspected MI and tested in a similar cohort of 7,998 subjects.
Than and co-authors reported that heart attack occurred in 13.4% of patients in the training set and 10.6% of patients in the test set. MI3 seemed well-calibrated, achieving an AUC of 0.963 in the test set and with similar results in early and late presenters. Example MI3 thresholds identifying low-risk and high-risk patients in the training cohort were 1.6 and 49.7, respectively.
In the test set, the authors said MI3 values were less than 1.6 in 69.5% of patients, with an NPV of 99.7% and a sensitivity of 97.8%. Values were 49.7 or less in 10.6% of the pool with a PPV of 71.8% and a specificity of 96.7%.
According to those thresholds, Than et al. said MI3 performed better than the ESC 0/3-hour pathway (sensitivity 82.5%, specificity 92.2%) and the 99th percentile at any time point (sensitivity 89.6%, specificity 89.3%).
“Consistent with previous reports, both of these approaches gave a low diagnostic sensitivity and poor PPV despite being widely used in clinical practice,” the authors wrote in their paper. “MI3 compared well to the ESC 0/1-hour pathway, with the primary advantage of MI3 being flexibility in the timing of serial testing and the simplicity of using probabilities rather than multiple thresholds to to stratify risk in individual patients.”
Than and colleagues said their approach is unique in that it considers multiple, individual metrics for each estimate, creating a more tailored output than its predecessors. Still, they said, prospective studies will need to validate MI3.
“This represents one of the first effective demonstrations of how machine learning could be used to guide clinical decision-making in patients with suspected acute coronary syndrome,” the team wrote. “The MI3 algorithm is more versatile than existing algorithms as it is not dependent on fixed cardiac troponin thresholds, does not require serial testing to be performed at specific time points, and recognizes that different healthcare systems have different priorities and tolerances of risk.”
After graduating from Indiana University-Bloomington with a bachelor’s in journalism, Anicka joined TriMed’s Chicago team in 2017 covering cardiology. Close to her heart is long-form journalism, Pilot G-2 pens, dark chocolate and her dog Harper Lee.