Improving Classification of Myocardial Infarction with Machine Learning in a Diverse Population

Researchers at Harvard Medical School have made a significant breakthrough in the classification of myocardial infarction (heart attack) in diverse populations using machine learning algorithms. The study, published in the American Journal of Epidemiology, aimed to investigate the performance of machine learning (ML) phenotyping pipelines in classifying myocardial infarction in a general and self-reported Black population. The findings demonstrate that a machine learning pipeline, PheCAP, can outperform a traditional ICD-based algorithm in classifying heart attacks, particularly in the Black population.

Key Takeaways:

• The study used data from the Veterans Health Administration (VHA) electronic health records (EHRs) from 2002 to 2019, involving 11,523,175 Veterans, with a mean age of 67.5 years and 93.8% male.
• The machine learning pipeline, PheCAP, incorporated natural language processing and was trained and validated against a gold standard of 403 Veterans who were chart-reviewed for MI.
• PheCAP achieved high precision and sensitivity in classifying MI, especially in the Black population, outperforming the traditional ICD-based algorithm.
• The authors conclude that applying PheCAP to the entire VHA population would provide increased power to replicate findings from published MI risk factor studies compared to the ICD algorithm.

Statistics:

• 11,523,175 Veterans were examined in the study.
• 93.8% of the Veterans were male.
• 14.3% of the Veterans were Black.
• 79.1% of the Veterans were White.
• The machine learning pipeline, PheCAP, achieved a precision of 0.90 and sensitivity of 0.66 in classifying MI in the entire VHA population.
• In the Black population, PheCAP's precision was 0.81, and sensitivity was 0.79.

Sources:

• American Journal of Epidemiology, "Improving classification of myocardial infarction with machine learning in a diverse population."
• Oxford University Press Inc, Journals Dept, 2001 Evans Rd, Cary, NC 27513, USA.
• Harvard Medical School, Department of Biomedical Informatics, Chuan Hong, et al.