Unsupervised cluster analysis of patients with recovered left ventricular ejection fraction identifies unique clinical phenotypes

Abstract

BACKGROUND: Patients with heart failure (HF) with recovered ejection fraction (HFrecEF) are a recently identified cohort that are phenotypically and biologically different from HFrEF and HFpEF patients. Whether there are unique phenotypes among HFrecEF patients is not known. METHODS: We studied all patients at a large medical center, who had an improvement in LVEF from ≤ 35% to ≥ 50% (LVrecEF) between January 1, 2005 and December 31, 2013. We identified a set of 11 clinical variables and then performed unsupervised clustering analyses to identify unique clinical phenotypes among patients with LVrecEF, followed by a Kaplan-Meier analysis to identify differences in survival and the proportion of LVrecEF patients who maintained an LVEF ≥ 50% during the study period. RESULTS: We identified 889 patients with LVrecEF who clustered into 7 unique phenotypes ranging in size from 37 to 420 patients. Kaplan-Meier analysis demonstrated significant differences in mortality across clusters (logrank p<0.0001), with survival ranging from 14% to 87% at 1000 days, as well as significant differences in the proportion of LVrecEF patients who maintained an LVEF ≥ 50%. CONCLUSION: There is significant clinical heterogeneity among patients with LVrecEF. Clinical outcomes are distinct across phenotype clusters as defined by clinical cardiac characteristics and co-morbidities. Clustering algorithms may identify patients who are at high risk for recurrent HF, and thus be useful for guiding treatment strategies for patients with LVrecEF.