Medications for specific phenotypes of heart failure with preserved ejection fraction classified by a machine learning-based clustering model

ObjectiveOur previously established machine learning-based clustering model classified heart failure with preserved ejection fraction (HFpEF) into four distinct phenotypes. Given the heterogeneous pathophysiology of HFpEF, specific medications may have favourable effects in specific phenotypes of HFpEF. We aimed to assess effectiveness of medications on clinical outcomes of the four phenotypes using a real-world HFpEF registry dataset. MethodsThis study is a posthoc analysis of the PURSUIT-HFpEF registry, a prospective, multicentre, observational study. We evaluated the clinical effectiveness of the following four types of postdischarge medication in the four different phenotypes: angiotensin-converting enzyme inhibitors (ACEi) or angiotensin-receptor blockers (ARB), beta blockers, mineralocorticoid-receptor antagonists (MRA) and statins. The primary endpoint of this study was a composite of all-cause death and heart failure hospitalisation. ResultsOf 1231 patients, 1100 (83 (IQR 77, 87) years, 604 females) were eligible for analysis. Median follow-up duration was 734 (398, 1108) days. The primary endpoint occurred in 528 patients (48.0%). Cox proportional hazard models with inverse-probability-of-treatment weighting showed the following significant effectiveness of medication on the primary endpoint: MRA for phenotype 2 (weighted HR (wHR) 0.40, 95% CI 0.21 to 0.75, p=0.005); ACEi or ARB for phenotype 3 (wHR 0.66 0.48 to 0.92, p=0.014) and statin therapy for phenotype 3 (wHR 0.43 (0.21 to 0.88), p=0.020). No other medications had significant treatment effects in the four phenotypes. ConclusionsMachine learning-based clustering may have the potential to identify populations in which specific medications may be effective. This study suggests the effectiveness of MRA, ACEi or ARB and statin for specific phenotypes of HFpEF.