Unsupervised Learning for Automated Detection of Coronary Artery Disease Subgroups.

Area of Science:

• Cardiovascular Disease Research
• Precision Medicine
• Machine Learning in Healthcare

Background:

• Precision population health aims to tailor prevention and care using patient data.
• Advanced analytics can identify clinically informative subgroups considering variability.
• This study explored unsupervised machine learning for coronary artery disease (CAD) subgroup discovery.

Purpose of the Study:

• To evaluate unsupervised machine learning for interpreting heterogeneous clinical data.
• To discover clinically important coronary artery disease subgroups.
• To assess the prognostic value of identified subgroups for cardiovascular events and mortality.

Main Methods:

• Applied generalized low rank modeling and K-means clustering.
• Utilized 155 phenotypic and genetic variables from 1329 participants in the Genetic Determinants of Peripheral Arterial Disease study.
• Employed Cox proportional hazard models for outcome analysis and compared with ACC/AHA pooled cohort equations.

Main Results:

• Identified 4 phenotypically and prognostically distinct CAD subgroups.
• Cluster 1 had the highest all-cause mortality (26%); Cluster 2 had the highest major adverse cardiovascular and cerebrovascular events (41%).
• Cluster membership provided more informative risk assessment for myocardial infarction, stroke, and mortality than pooled cohort equations.

Conclusions:

• Unsupervised clustering successfully identified unique CAD subgroups with distinct clinical trajectories.
• Machine learning algorithms can effectively process heterogeneous patient data for disease characterization.
• This approach enhances insights into CAD subgroup characterization and risk assessment.