A Machine Learning-Derived Score to Effectively Identify Heart Failure With Preserved Ejection Fraction

Background: The diagnosis of heart failure with preserved ejection fraction (HFpEF) in the clinical setting remains challenging, especially in patients with obesity.

Objectives: This study aimed to identify novel predictors of HFpEF well suited for patients with obesity.

Methods: We performed a retrospective analysis of a well-characterized cohort of patients with obesity with HFpEF (n = 404; mean body mass index [BMI] 36.6 kg/m²) and controls (n = 67). We used the machine learning algorithm Gradient Boosting Machine to analyze the association of various parameters with the diagnosis of HFpEF and subsequently created a multivariate logistic model for the diagnosis.

Results: Gradient Boosting Machine identified BMI, estimated glomerular filtration rate, left ventricular mass index, and left atrial to left ventricular volume ratio as the strongest predictors of HFpEF. These variables were used to build a model that identified HFpEF with a sensitivity of 0.83, a specificity of 0.82, and an area under the curve (AUC) of 0.88. Internal validation of the model with optimism-adjusted AUC showed an AUC of 0.87. Within the studied cohort, the novel score outperformed the H2FPEF score (AUC: 0.88 vs 0.74; P < 0.001).

Conclusions: In a HFpEF cohort with obesity, BMI, estimated glomerular filtration rate, left ventricular mass index, and left atrial to left ventricular volume ratio most correlated with the identification of HFpEF, and a score based on these variables (HFpEF-JH score) outperformed the currently used H2PEF score. Further validation of this novel score is warranted, as it may facilitate improved diagnostic accuracy of HFpEF, particularly in patients with obesity.