Abstract Body (Do not enter title and authors here): Background: GLP-1 receptor agonists (GLP1RA) agonists have been shown to reduce cardiovascular events in patients with type 2 diabetes (T2D) and atherosclerotic cardiovascular disease (ASCVD). However, it is unclear whether early subclinical therapeutic changes can be detected from routine 12-lead electrocardiograms (ECGs).
Methods: We finetuned a neural network (NN) from Pheiron’s phenotyping platform to predict subclinical risk markers from 12-lead ECGs using data from Dandelion Health, comprised of longitudinal, multimodal real-world clinical data for 106,289 patients receiving care from Sharp Healthcare in the US. Based on the finetuned NN, we predicted risk markers within an independent cohort of 48,376 patients with a history of T2D, ASCVD, or obesity and emulated a target trial to evaluate the subclinical effect of GLP1RA. The primary analysis compared percentile rank transformed ECG markers for GLP1RA and non-GLP1RA patients using inverse probability weighted (IPW) linear regression models, adjusted for key variables including demographics, comorbidities, medications, HbA1c, and BMI at baseline. Sensitivity analyses included unadjusted models, models with only covariate adjustments, and propensity score-matched models.
Results: The emulated trial study cohort was, on average, 59.6 [SD 15.9] years old, predominantly female (56%), and racially diverse: 46.6% White, 25.3% Hispanic, 9.6% Asian, and 4.6% Black. Compared to the control group, the GLP1RA cohort had a higher baseline risk profile (BMI: 35.1 [SD 7.6] vs. 31.0 [SD 6.0], HbA1c: 7.1 % [SD 1.8] vs. 5.8 % [SD 0.9]) but cardiovascular risk progressed slower during the observation time: Each year of GLP1RA treatment was linked to percentile risk reductions of -1.65 (p<0.001) for all-cause mortality, -1.74 (p<0.001) for myocardial infarction, and -1.74 (p<0.001) for stroke. These findings were directionally consistent across all sensitivity analyses.
Conclusions: Our findings suggest treatment with GLP1RA slows the progression of subclinical cardiovascular risk, as detected from routine 12-lead ECGs. This illustrates the potential of ECG-based monitoring to guide therapeutic interventions and establish treatment efficacy.