Abstract 15924: Identification of Gender- and Age-specific Top Predictors of Hospitalization Due to Heart Failure Using Machine Learning in TOPCAT

Abstract

Background: NHLBI supported Treatment of Preserved Cardiac Function Heart Failure with an Aldosterone Antagonist trial (TOPCAT) (NCT00094302) investigated whether treatment with spironolactone reduces hospitalization due to heart failure (hHF) in 3,445 adults with prior heart failure and a left ventricular ejection fraction over 45%. We reused publicly available individual patient-level data from NHLBI Data Repository (BioLINCC) to perform hypothesis-generating secondary analyses by machine learning (ML) on the American TOPCAT cohort (n=1767) to identify gender and age groups specific baseline (bl) predictors of hHF. Methods: The subjects were stratified into subgroups based on gender (male and female) and age (50-59, 60-69, 70-79, 80-90). Random Survival Forest (RSF), a non-parametric ML approach, evaluated 172 bl variables as predictors of hHF. Top 10 predictors were subsequently included in a multivariate Cox proportional hazards model. Results: The top 10 predictors of hHF are shown in Figure 1. Overall, renal and hematological biomarkers appeared prominently in the top 10 predictors for these patients. While liver markers were among top predictors for hHF in males, diabetes treatment and diabetic complications were top predictors for females. Also, diabetic treatment was a top predictor among age group 50-59, diabetic complications were top predictors among age groups 50-59 and 70-79, liver markers were top predictors among age groups 70-79, and race and years of smoking were top predictors among age group 60-69. Importantly, the use of potassium sparing diuretic at bl was the top predictor among age group 80-90. Conclusion: Using ML, we uncovered in an unbiased fashion, otherwise overlooked bl predictors of hHF in a large, international, multi-center trial like TOPCAT. Thus, ML can help to identify similarities and differences of disease and treatment outcomes among gender, race, ethnicity, and age specific subgroups and advance precision medicine.