INTRODUCTION:

Chronic kidney disease (CKD) is a critical comorbidity in heart failure with preserved ejection fraction (HFpEF), often linked to congestion due to elevated filling pressures. However, the role of subtle systolic dysfunction in CKD remains to be elucidated. This study aimed to investigate the predictive power of left ventricular systolic function for CKD in patients with HFpEF, using advanced machine learning (ML) techniques.

METHODS:

We analyzed data from 497 patients referred to a specialized HFpEF clinic. Using machine learning models (Random Forest, XGBoost, Logistic Regression, Decision Tree), we assessed traditional risk factors (sex, age, hypertension, obesity, diabetes, and smoking) and renal function markers (eGFR and UACR) to estimate the risk of HFpEF based on the HFA-PEFF scoring system. Echocardiography-derived variables were used to predict CKD, defined as an eGFR <60 ml/min/1.73 m2. SHapley Additive exPlanations (SHAP) provided insights into model predictions.

RESULTS:

The study population consisted primarily of women (67%) with an average age of 62 ± 12 years, increased BMI (33 ± 7 kg/m2), and prevalent hypertension (79%), dyslipidemia (71%), and type 2 diabetes (38%). The logistic regression model that incorporated traditional risk factors and renal markers achieved an AUROC of 0.80 ± 0.07, indicating significant impacts of lower eGFR and higher UACR on the risk of HFpEF (Panel A). The Random Forest model, which included variables derived from echocardiography, achieved the highest accuracy (AUROC 0.96 ± 0.02). Key systolic predictors of lower eGFR included reductions in tissue Doppler S wave and stroke volume index, as well as impaired left ventricular diastolic function (increases in E/e’ ratio) and elevated estimated pulmonary artery pressure (TRV - tricuspid regurgitation peak velocity) (Panel B).

CONCLUSION:

This study identifies subtle systolic dysfunction, along with impaired diastolic function and increased pulmonary arterial pressure, as significant predictors of CKD in HFpEF. Our findings highlight the potential of advanced ML techniques to better understand the disease mechanisms. Early detection of these impairments can improve risk stratification and guide targeted interventions in the management of HFpEF.