Cardio-renal-metabolic risk profile strongly predicts heart failure in patients with aortic stenosis

Background Heart failure is highly prevalent and a leading cause of hospitalisations and mortality in patients with aortic stenosis. Current clinical focus in aortic stenosis is largely limited to valvular factors. Risk factors and mechanisms associated with heart failure in aortic stenosis are incompletely understood. Objective To identify risk factors for heart failure in patients with aortic stenosis. Methods Patients with aortic stenosis were identified in the UK Biobank using ICD10 (I35.0, I35.2), ICD9 (4241) and OPCS4 (K26.1, K26.2, K26.3, K26.4) codes, excluding rheumatic disease (I06) and aortic insufficiency (I35.1). Heart failure was identified using ICD10 codes (I50, I11.0, I11.3, I13.2, Z94.1, T86.2). Patients with heart failure prior to aortic stenosis were excluded. Relevant diagnoses at aortic stenosis diagnosis and lab values at recruitment were obtained. An XGBoost Cox proportional hazards model was developed to predict heart failure development, with discrimination assessed by C-index and time-dependent AUROC at 2 and 5 years. SHAP values were used to assess feature importance. Patients were stratified into tertiles for risk based on the predicted risk scores and heart failure-free survival was compared between risk tertiles using Kaplan-Meier curves. Results A total of 6385 patients (median age: 72 [66;76], 37.1% female) with aortic stenosis were included, of which 1693 (26.5%) developed heart failure over a median follow-up of 3.93 years after aortic stenosis diagnosis. Patients who developed heart failure were more often male (p=0.003), obese (p<0.001), smokers (p<0.001), more commonly had atrial fibrillation (p<0.001), hyperlipidaemia (p<0.001), diabetes (p<0.001), chronic kidney disease (p=0.004), COPD (p<0.001), coronary artery disease (p<0.001), and more often underwent aortic valve replacement (52.4% vs. 43.1%; p<0.001). Of the patients who underwent valve replacement, 12.4% had already developed heart failure at the time of replacement and 18.1% developed heart failure after the procedure. The top predictors of heart failure development based on the machine learning model were age, markers of renal function, markers of obesity, atrial fibrillation, C-reactive protein, coronary artery disease, and type 2 diabetes (Figure 1). This model demonstrated good discrimination with a C-index of 0.81 and AUROCs of 0.839 and 0.899 at two and five years. Stratification into risk tertiles showed significantly divergent heart failure-free survival across groups (p<0.001) (Figure 2). Conclusions Renal, metabolic, and inflammatory risk factors strongly predict heart failure development in aortic stenosis. Better risk factor identification and management beyond valvular features may lead to improved risk stratification and personalized treatment.