Notable heterogeneity exists in the clinical presentation of heart failure (HF) patients. Current subtype classifications are
based on ejection fraction may not fully capture the aetiological and prognostic heterogeneity of HF.
The use of unsupervised machine learning (ML) approaches, such as cluster analysis, on large-scale observational data from
electronic health records (EHR), can enable the discovery of novel subtypes and guide the characterization of their clinical
manifestation. Clustering methods can group HF patients based on similarities between their clinical features without making
a priori assumptions about the distribution of the data.
We sought to discover, characterize and replicate HF subtypes by applying a clustering method on a heterogeneous HF
population derived from phenotypically rich EHR. Characterization of HF subtypes using EHR derived variable may enable
more precise large-scale genomic analysis to inform better prevention, diagnostic and treatment strategies.