Evaluation of machine learning framework for personalized prediction of benefit from beta-blockers in myocardial infarction with preserved or mildly reduced ejection fraction
Y Chen, C H Zhang, Y Qiao, R Li, S Y Dong, Y Ma, W Y Wang, Y D TangAbstract
Background
Long-term clinical benefit of β-adrenergic receptor blocker (β-blocker) in patients with acute myocardial infarction (AMI) and preserved or mildly reduced left ventricular ejection fraction (LVEF) remains controversial, with conflicting findings from large-scale randomised controlled trials. Conventional risk stratification tools cannot identify which patients in this population derive significant survival benefit from β-blocker therapy, creating an urgent unmet need for phenotype-guided personalised treatment.
Purpose
We aimed to develop and validate an interpretable multimodal machine learning (ML) framework (Acute Myocardial Infarction Multimodal Outcome Calculator, AMIMOC) to identify the patient phenotype associated with significant survival benefit from β-blocker therapy, and to elucidate the underlying physiological pathways driving this treatment effect.
Methods
We developed and internally validated the AMIMOC using pooled clinical, electrocardiographic, and echocardiographic data from 5726 AMI patients with LVEF>40% in the MIMIC-III and IV databases. External validation was performed in an independent multicentre cohort of 918 patients. Patients were stratified into low (<17.00%), intermediate (17.00–50.00%), and high (>50.00%) risk groups by AMIMOC-predicted 1-year all-cause mortality. Within each stratum, the survival benefit from β-blocker was estimated using inverse-probability-of-treatment weighted Cox proportional hazards models. Multiple mediation analyses were conducted to explore the mechanistic pathways of treatment benefit.
Results
The AMIMOC showed robust discriminative performance with an area under the receiver operating characteristic curve of 0.89 (95% confidence interval [CI] 0.86–0.91) in internal validation and 0.86 (95% CI 0.84–0.89) in external validation. The framework identified a distinct high-risk phenotype characterised by LVEF 40–49%, elevated heart rate, and prolonged QTc interval. In this high-risk stratum, β-blocker use was associated with a significant 48% relative reduction in 1-year mortality (hazard ratio 0.52, 95% CI 0.37–0.73). Mediation analyses revealed that 78.5% of this benefit was mediated through improvements in LVEF, heart rate control, and QTc interval shortening. The AMIMOC has been deployed as a free, interactive web calculator to support clinical decision-making.
Conclusions
This interpretable multimodal ML framework can reliably identify a clinically and biologically coherent phenotype of AMI patients with LVEF>40% who derive significant survival benefit from β-blocker therapy. These findings support phenotype-guided treatment to optimise β-blocker prescribing in post-AMI care. The accessible web calculator further enables translation of this framework into routine clinical practice.AMIMOC Framework Overall Study DesignBeta-Blocker Survival Analysis in Risk Strata