A machine learning-derived echocardiographic algorithm identifies people at risk of heart failure with distinct...

An echocardiographic algorithm derived by machine learning (e'VM) characterizes pre-clinical individuals with different cardiac structure and function, biomarkers, and long-term risk of heart failure (HF). Our aim was the external validation of the e'VM algorithm and to explore whether it may identify subgroups who benefit from spironolactone.

The HOMAGE (Heart OMics in AGEing) trial enrolled participants at high risk of developing HF randomly assigned to spironolactone or placebo over 9 months. The e'VM algorithm was applied to 416 participants (mean age 74 ± 7 years, 25% women) with available echocardiographic variables (i.e. e' mean, left ventricular end-diastolic volume and mass indexed by body surface area [LVMi]). The effects of spironolactone on changes in echocardiographic and biomarker variables were assessed across e'VM phenotypes. A majority (>80%) had either a 'diastolic changes' (D), or 'diastolic changes with structural remodelling' (D/S) phenotype. The D/S phenotype had the highest LVMi, left atrial volume, E/e', natriuretic peptide and troponin levels (all p < 0.05). Spironolactone significantly reduced E/e' and B-type natriuretic peptide (BNP) levels in the D/S phenotype (p < 0.01), but not in other phenotypes (p > 0.10; p_interaction  <0.05 for both). These interactions were not observed when considering guideline-recommended echocardiographic structural and functional abnormalities. The magnitude of effects of spironolactone on LVMi, left atrial volume and a type I collagen marker was numerically higher in the D/S phenotype than the D phenotype but the interaction test did not reach significance.

In the HOMAGE trial, the e'VM algorithm identified echocardiographic phenotypes with distinct responses to spironolactone as assessed by changes in E/e' and BNP.