Machine Learning Shows Potential in Diagnosing Myocardial Infarction in Patients With Chest Pain
The algorithm accurately predicted the diagnosis of myocardial infarction 94% of the time in the validation phase.
ORLANDO — Researchers have found that machine learning can improve the sensitivity and specificity of diagnosis of myocardial infarction (MI) in patients with chest pain who present at emergency departments (EDs). The research was presented at the American College of Cardiology 67th Annual Scientific Session, held March 10-12, 2018, in Orlando, Florida.
Although chest pain is an extremely common complaint for patients presenting to the ED, the majority of patients with chest pain do not have MI. In this study, researchers trained a gradient boosting machine (GBM) to achieve higher diagnostic sensitivity and specificity than high-sensitivity troponin T (hsTnT) alone.
Using data from past ED visits for chest pain (N=5802), researchers trained the machine to predict MI using results of laboratory testing, vital signs, and all prior diagnoses from the Swedish National Patient registry. They validated the results using a separate set of ED visit data (N=2485).
When comparing the results to those obtained by using hsTnT alone, the researchers found the GBM to demonstrate greater accuracy in identifying MI. In terms of discrimination, the GBM model had an area under the curve (AUC) of 0.993 in the training set, 0.946 in cross-validation, and 0.946 in the validation cohort. In the standard hsTnT model, AUC was 0.902 for the training cohort and 0.888 for the validation cohort.
Clinically, these results suggest that “machine learning improves prediction of MI in patients presenting with chest pain in the ED. Its potential role as a decision support should be further studied.”
Lindholm D, Holzmann M. Machine learning for improved detection of myocardial infarction in patients presenting with chest pain in the emergency department. Presented at: American College of Cardiology 67th Annual Scientific Session & Expo; March 10-12, 2018; Orlando, FL. Abstract 13883.