Initial independent validation of a genomic heart failure survival prediction algorithm

Abstract

Background: Biological determinants of survival in advanced heart failure (AdHF) are linked to systems biological properties of disease severity including age, comorbidities, and frailty. We hypothesize that an algorithm trained to predict the survival in severely ill mechanical circulatory support (MCS) AdHF patients can be independently validated in AdHF-cohorts of varying severity undergoing etiology-specific interventions including heart transplantation (HTx), transcatheter aortic valve replacement (TAVR), and continued guidelines directed medical therapy (GDMT). Research Design & Methods: We independently validated our previously published multi-dimensional algorithm, based on 4 clinical parameters and 12 transcriptomic biomarkers, and trained in AdHF patients undergoing MCS-surgery (n = 29), in AdHF patients undergoing TAVR, HTx, MCS, and GDMT-interventions (n = 48). Results: In the independent validation cohort, our algorithm demonstrated 71% sensitivity, 90% specificity, 56% positive predictive value, and 95% negative predictive value, allowing for construction of a prototype survival prediction score. While prediction of 1-year survival using clinical parameters alone achieved an AUC = 0.69, addition of 12 differentially expressed genes to the clinical model improved the AUC = 0.90. Conclusions: Our initial validation data suggests that the proposed multi-dimensional algorithm is applicable across various AdHF-risk groups and Surgical-Interventional Therapies (S/IT), increasing survival prediction accuracy compared to clinical data alone and warranting further study in larger cohorts.