AI-based cluster analysis enables outcomes prediction among patients with increased LVM

Abstract Background The traditional classification of left ventricular hypertrophy (LVH) solely based on left ventricular geometry disregards associated comorbidities and clinical characteristics. We aimed to identify unique clinical phenotypes of LVH using unsupervised cluster analysis and explore their association with clinical outcomes. Methods From UK Biobank (UKBB) participants with available left ventricular mass (LVM) estimations, we identified those meeting CMR-based criteria for increased LVM: LVM index ≥ 72 g/m2 for males and LVM index ≥ 55 g/m2 for females. Baseline demographic, clinical, and laboratory data were collected. Using Ward's method, patients were clustered based on 27 variables. The primary outcome was a composite of all-cause mortality with heart failure (HF) admissions, ventricular arrhythmia, and atrial fibrillation (AF). Cox proportional hazard model and Kaplan-Meier survival analysis were applied. Results Among CMR-assessed UKBB participants, 4,255 individuals exhibited increased LVM, with an average age of 64 ± 7 years; 2,447 (58%) were females. Cluster analysis identified four distinct subgroups. Over a median 5-year follow-up (IQR: 4-6), 100 patients (2%) died, 118 (2.8%) were hospitalized due to HF, 29 (0.7%) were hospitalized due to VA, and 208 (5%) were hospitalized due to AF. Univariate Cox analysis revealed that compared to the 1st cluster (n=1,578), patients in the 2nd (n=1,296), 3rd (n=824) and 4th (n=557) clusters had 60%, 104%, and 164% increased risk of a major event, respectively (95% CI 1.2-2.16, p<.001 for cluster 2; 95% CI 1.49-2.78, p<.001 for cluster 3; 95% CI 1.92-3.62, p<.001 for cluster 4). Each cluster manifested different phenotypes: cluster 2 comprised mainly overweight females, with the highest prevalence of chronic lung disease; cluster 3, predominantly composed of males, had the highest burden of comorbidities and cardiovascular risk factors; and cluster 4, mostly males, presented with the most abnormal cardiac measures, including left ventricular end-diastolic volume (LVEDV), left ventricular end-systolic volume (LVESV), and left ventricular ejection fraction (LVEF). Conclusions Unsupervised cluster analysis identified four subgroups among patients with increased LVM. These subgroups bear notable relevance to overall mortality risk and clinical outcomes. This phenotypic classification holds promise in offering valuable insights regarding clinical course and outcomes in LVH patients.