UNSUPERVISED CLUSTER ANALYSIS OF HYPERTENSIVE INDIVIDUALS IDENTIFIES A ROBUST HIGH-RISK SUBGROUP

Abstract

Abstract Objective: Despite hypertension being a crucial risk factor for cardiovascular disease, the current classification of hypertension does not reflect the heterogeneity or the wide variation in cardiovascular outcomes of hypertensive individuals. Our objective was therefore to identify subgroups of hypertensive individuals with distinct cardiovascular outcomes using data-driven cluster analysis. Design and method: To identify phenotypically distinct subgroups of hypertensive individuals, we performed unsupervised hierarchical clustering in participants with grade 2 hypertension from two population cohorts (FINRISK 2002 and 2012; n = 1971, mean age 59.8 years, 49% women) with up to 16 years of follow-up for incident cardiovascular events (n = 260). We defined hypertension as either systolic blood pressure (SBP) over 160 mmHg, diastolic blood pressure (DBP) over 100 mmHg, or use of antihypertensive medication. We computed the clusters separately for men and women based on eight factors related to hypertension and cardiovascular outcomes: mean arterial pressure (MAP; steady state pressure), pulse pressure (PP; pulsatile pressure), non-high-density lipoprotein cholesterol (non-HDLc), glycated hemoglobin (HbA1c), body-mass index (BMI), C-reactive protein (CRP), estimated glomerular filtration rate (eGFR), and alcohol consumption (rAlcohol). In analyses with men and women combined, we used Cox regression models adjusted for age and sex to assess the risk of cardiovascular outcomes between the resulting clusters. Results: We observed two comparable clusters for both sexes (Figure). These clusters had significantly different patient characteristics. Notably, mean HbA1c (5.5 ± 0.4 vs. 7.6 ± 0.9 in men; 5.5 ± 0.3 vs. 7.7 ± 0.8 in women) and BMI (29.0 ± 4.5 vs. 31.1 ± 4.3 in men; 29.4 ± 5.4 vs. 33.3 ± 5.5 in women) were higher in Cluster 2 than in Cluster 1. Individuals in Cluster 2 had 2.1-fold greater risk (95% CI 1.4–3.1, p < 0.001) of cardiovascular disease than individuals in Cluster 1. Conclusions: Using unsupervised hierarchical clustering, we stratified patients into two subgroups with differing disease progression and risk of complications. Substratification of hypertensive patients by low vs. high blood glucose and BMI could help to tailor and target early treatment to patients who would benefit most from it.