Identifying Phenogroups in patients with subclinical diastolic dysfunction using unsupervised statistical learning

Area of Science:

• Cardiology
• Medical Informatics
• Biostatistics

Background:

• Subclinical diastolic dysfunction precedes heart failure with preserved ejection fraction (HFpEF).
• Not all individuals with diastolic dysfunction progress to HFpEF.
• Identifying predictors of HFpEF progression is crucial for patient management.

Purpose of the Study:

• To identify clinical and echocardiographic variables that predict the development of HFpEF from subclinical diastolic dysfunction.
• To stratify patients with diastolic dysfunction into distinct subgroups based on progression to HFpEF.

Main Methods:

• Retrospective data collection from 162 patients (81 HFpEF, 81 controls).
• Utilized unsupervised clustering (density-based and hierarchical) on 65 variables.
• Logistic regression analysis performed on the overall cohort and individual clusters.

Main Results:

• Three distinct patient subgroups were identified through clustering.
• Subgroups differed in gender, cardiac hypertrophy, aortic stenosis severity, NT-proBNP levels, and HFpEF progression timing.
• Predictors of HFpEF varied across subgroups, with common factors including diabetes, chronic kidney disease, atrial fibrillation, and diuretic use.

Conclusions:

• Cluster analysis effectively identifies phenotypically distinct subgroups of patients with diastolic dysfunction.
• These subgroups exhibit differential HFpEF and mortality outcomes.
• Predictive variables for HFpEF outcomes are cluster-specific, underscoring the heterogeneity of the disease.