Inferring causal treatment effects from real-world data is often hindered by unmeasured confounding. This study was presented at the European Society of Cardiology Congress 2025.

A novel computational framework, DISCO, was applied to patients with heart failure with reduced ejection fraction receiving angiotensin receptor–neprilysin inhibitors or angiotensin-converting enzyme inhibitors. The approach used artificial intelligence to decompose electrocardiograms into high-dimensional embeddings, enabling the creation of digital twins and phenotypic matching.

Among 4,705 patients, traditional observational analyses failed to replicate the hazard ratio observed in clinical trials. In contrast, digital-twin cohorts produced treatment effects consistent with randomized trial results and balanced prognostic factors between arms. Subgroup analyses revealed heterogeneity in response, identifying populations likely to derive greater benefit from therapy.

These findings highlight the potential of AI-enabled digital twins to support robust causal inference in real-world data and to inform precision trial designs, including synthetic control strategies in cardiovascular medicine.