Clustering patterns of evidence-based therapy identifies distinct phenotypes in heart failure with reduced ejection fraction: a machine learning analysis of the Swedish heart failure registry
A Uijl, D Stolfo, L Benson, M Imazio, F W Asselbergs, G SavareseAbstract
Background
In routine clinical practice, use of guideline-directed medical therapy (GDMT) for heart failure with reduced ejection fraction (HFrEF) remains suboptimal. Patients are exposed to often unrecognized patterns of treatment implementation. Identifying the patterns of GDMT use may enable targeted interventions to address therapeutic inertia, optimise initiation, uptitration, and resource allocation.
Purpose
We aimed to derive and characterize meaningful clusters of patients with HFrEF based on patterns of foundational therapy use.
Methods
Latent class analysis (LCA) was used to derive clusters in 8443 HFrEF patients from the Swedish Heart Failure Registry (2021-2025), with a duration of disease ≥6 months to allow time for full GDMT implementation. In the cluster model we included multiple variables related to use and target dose (TD) of GDMT, and clinical characteristics (see Figure 1). The Bayesian Information Criterion determined the optimal number of clusters. Multivariable Cox proportional hazard models assessed the association between cluster membership and the outcomes cardiovascular (CV) mortality, HF hospitalisation and all-cause mortality.
Results
LCA identified seven distinct clusters matching specific clinical features and patterns of GDMT use/underuse: 1) cardio-metabolic-Atrial fibrillation phenotype with high use of RAS-inhibitors and beta-blockers, but only beta-blockers on optimal TD, and with 2/3 of the patients on MRAs and SGLT2-inhibitors 2) a less severe HF cluster (lower NYHA and NT-proBNP values) with suboptimal TD use of RAS-inhibitors and beta-blockers, low use of MRA (35%) and SGLT2-inhibitors (50%), 3) an older cluster with moderate renal impairment which was characterized by high use of all GDMT but at suboptimal TD, 4) an older cluster with severe HF and severe renal impairment mainly treated with beta-blockers and RAS-inhibitors at suboptimal TD and very rare use of MRA (5%) and SGLT2-inhibitors (17%), 5) the youngest cluster, with mild HF (lowest NT-proBNP and NYHA class), highly treated with 4 GDMT at optimal TD, 6) a cluster with high use of implantable devices (40%), characterized by likely use of all 4 GDMT, with mainly beta-blockers at optimal TD and 7) a hypertensive cluster frequently on 4 GDMT but at suboptimal TD for all therapies. Lowest risk of CV mortality and HF hospitalisation was seen in cluster 2, 5 and 7, whereas cluster 3, 4 had the highest risk. Cluster 1 and 6 had intermediate risk for CV mortality, yet differed in their risk for HF hospitalisation, with only cluster 6 at high risk.
Conclusions
Clustering may aid clinicians in the detection of patient profiles at risk of undertreatment. In our cohort advanced age, renal impairment, and greater HF severity characterized those clusters where GDMT was less implemented and risks of CV mortality and HF hospitalisation were highest. Our findings could support more targeted interventions on therapy optimisation in clinical practice.ClusterCharacteristicsFor image description, please refer to the figure legend and surrounding text.SurvivalPlotsFor image description, please refer to the figure legend and surrounding text.