In order to bring you the best possible user experience, this site uses Javascript. If you are seeing this message, it is likely that the Javascript option in your browser is disabled. For optimal viewing of this site, please ensure that Javascript is enabled for your browser.

Late-Breaking Science: A novel machine-learning approach redefines beta-blocker response in patients with HF

30 Aug 2021
Late-Breaking Science presented at ESC Congress

Today, Doctor Andreas Karwath from the team of Professor Dipak Kotecha (University of Birmingham, UK) reports on the use of a novel artificial intelligence (AI) pipeline to identify clusters of patients with heart failure (HF) and reduced left ventricular ejection fraction (LVEF) that respond to beta-blocker therapy.

Including 15,659 patients from nine double-blind placebo-controlled randomised trials, the study used a combination of neural network-based variational autoencoders and hierarchical clustering, with objective determination of the number of clusters and dimensions. All-cause mortality was assessed during a median of 1.3 years of follow-up, and stratified by heart rhythm into sinus rhythm and atrial fibrillation (AF).

Among 12,822 patients in sinus rhythm, most of the six clusters identified demonstrated a consistent overall mortality benefit from beta-blockers, with odds ratios (ORs) ranging from 0.54 to 0.74, including in those at the highest mortality risk. However, there was no significant efficacy in one cluster of 2,537 patients who were on average older, with less impairment of LVEF and a lower baseline heart rate.

Among 2,837 patients with AF, four of five clusters were consistent with the overall neutral effects of beta-blockers vs. placebo (OR 0.92; 95% confidence interval [CI] 0.77 to 1.10; p=0.37). The fifth cluster, on average younger AF patients with lower rates of prior myocardial infarction but similar LVEF, had a statistically significant mortality reduction with beta-blockers (OR 0.57; 95% CI 0.35 to 0.93; p=0.023). The robustness and consistency of clustering was confirmed for all models (p<0.0001 vs. random) and cluster membership was externally validated across the nine independent trials.

This AI-based approach may provide a useful, unbiased tool to estimate prognostic response to beta-blocker therapy in patients with HF and reduced LVEF, and across other therapies and conditions. Further details can be seen in the paper to be published in The Lancet.


Click on the heart to add this presentation to your favourites then watch it live or on demand.