Phenotyping heart failure using model-based analysis and physiology-informed machine learning
To determine the underlying mechanistic differences between diagnoses of Heart Failure (HF) and specifically heart failure with reduced and preserved ejection fraction (HFrEF & HFpEF), a closed loop model of the cardiovascular system coupled with patient specific transthoracic echocardiography (TTE) and right heart catheterization (RHC) measures was used to identify key parameters representing cardiovascular hemodynamics. Thirty-one patient records (10 HFrEF, 21 HFpEF) were obtained from the Cardiovascular Health Improvement Project (CHIP) database at the University of Michigan. Model simulations were tuned to match RHC and TTE pressure, volume and cardiac output measures in each patient with average error between data and model of 4.87 {+/-} 2%. The underlying physiological model parameters were then plotted against model-based norms and compared between the HFrEF and HFpEF group. Our results confirm that the main mechanistic parameter driving HFrEF is reduced left ventricular contractility, while for HFpEF a much wider underlying phenotype is presented. Conducting principal component analysis (PCA), k-means, and hierarchical clustering on the optimized model parameters, but not on clinical measures, shows a distinct group of HFpEF patients sharing characteristics with the HFrEF cohort, a second group that is distinct as HFpEF and a group that exhibits characteristics of both. Significant differences are observed (p-value<.001) in left ventricular active contractility and left ventricular relaxation, when comparing HFpEF patients to those grouped as similar to HFrEF. These results suggest that cardiovascular system modeling of standard clinical data is able to phenotype and group HFpEF as different subdiagnoses, possibly elucidating patient-specific treatment strategies.