Identifying clinical phenotype clusters in patients with coronary artery disease

Guideline recommendations for the prevention of cardiovascular (CV) events in patients with coronary artery disease (CAD) are predominantly one-size-fits-all. Clinically identifiable phenotypes needing specific considerations might exist. The purpose of this study is to identify such clinical phenotypic clusters in patients with CAD and assess their relationship with the risk of recurrent CV events.

Unsupervised machine learning through latent class analysis was performed in patients with CAD from the Swedish Web-System for Enhancement and Development of Evidence-Based Care in Heart Disease Evaluated According to Recommended Therapies (SWEDEHEART) registry (n=88 894) and Utrecht Cardiovascular Cohort-Second Manifestations of Arterial Disease (UCC-SMART) cohort (n=5506). Characteristics for clustering were based on availability, missingness and clinical relevance. Clustering was performed in SWEDEHEART and validated in UCC-SMART. Association between clusters and the composite of myocardial infarction, stroke or CV death was assessed using Cox proportional hazard models.

Four phenotypes could be distinguished: cluster 1 (38%, n=33 777) of predominantly younger males with increased body mass index, blood pressure and C-reactive protein, cluster 2 (21%, n=18 775) of smokers with few traditional risk factors, cluster 3 (30%, n=26 501) of older patients with few comorbidities and cluster 4 (11%, n=9841) of patients with multimorbidity. Compared with cluster 1, cluster 4 was at the highest risk (HR 4.38 95% CI (4.01 to 4.78)), followed by cluster 3 (HR 1.78 (1.70 to 1.85)), and cluster 2 (HR 0.97 (0.88 to 1.07)). Validation in UCC-SMART yielded similar results.

Four distinct and reproducible phenotypes, with differences in the risk of recurrent CV events, were identified among patients with CAD. These may be relevant in practice and warrant research into specific pathophysiology and differences in treatment effects.