Stratifying cardiovascular benefits from GLP-1RA: a multisource analysis of patient-level CVOT and real-world data using AI-driven methods

Cardiovasc Diabetol. 2025 Oct 17;24(1):401. doi: 10.1186/s12933-025-02952-w.

Abstract

Background: It remains unclear whether certain individuals with type 2 diabetes (T2D) derive greater cardiovascular benefit from GLP-1 receptor agonists (GLP-1RAs). Here, we integrate individual-level data from cardiovascular outcome trials (CVOTs) and electronic health records (EHRs), applying machine learning methods to confirm the cardiovascular benefits of GLP-1RAs in real-world populations and to identify subgroups with enhanced treatment response.

Methods: Data from two CVOTs (LEADER and SUSTAIN-6) and a large real-world study (DARWIN-T2D) were analyzed. We first transposed the hazard ratio (HR) for 3-point major adverse cardiovascular event (3P-MACE) from CVOTs to the real-world population. Then, we used PRISM (Patient Response Identifiers for Stratified Medicine) against 3P-MACE reduction by GLP-1RA in a training/test setting. Findings were validated with external cohorts of new-users of GLP-1RA or comparators (DPP-4 inhibitors or basal insulin).

Results: Despite notable differences in clinical characteristics between CVOT and real-world patients, the real-world-transposed HRs for 3P-MACE closely paralleled those from CVOTs. PRISM identified subgroups with differential treatment responses, based on history of myocardial infarction (MI) or stroke and age. Participants aged over 71 years without MI/stroke (41% of the real-world population) had the greatest relative benefit (HR 0.46; 95% CI 0.24-0.89 in the test set) and a greater absolute risk reduction (ARR 4.5%, 95% CI 1.2-7.7) than other subgroups (Gail-Simon p = 0.02). The external validation cohort confirmed these results (HR 0.67; 95% CI 0.51-0.89 and ARR 3.8%, 95% CI 1.5-6.1) showing significant differences in absolute risk reduction (p < 0.05).

Conclusions: This study supports the integration of individual data from CVOT with those from EHR to confirm the transposition of results from CVOT to real-world populations, and enables the identification and validation of subgroups with greater cardiovascular benefits from cardioprotective treatment such as GLP-1RA treatment. This precision medicine approach represents a new framework for deploying cardiovascular prevention strategies in T2D.

Keywords: Cardiovascular prevention; GLP-1RA; Precision medicine; Randomized clinical trial; Real-world evidence.

Publication types

  • Multicenter Study

MeSH terms

  • Aged
  • Cardiovascular Diseases* / diagnosis
  • Cardiovascular Diseases* / epidemiology
  • Cardiovascular Diseases* / mortality
  • Cardiovascular Diseases* / prevention & control
  • Decision Support Techniques*
  • Diabetes Mellitus, Type 2* / blood
  • Diabetes Mellitus, Type 2* / diagnosis
  • Diabetes Mellitus, Type 2* / drug therapy
  • Diabetes Mellitus, Type 2* / epidemiology
  • Diabetes Mellitus, Type 2* / mortality
  • Electronic Health Records
  • Female
  • Glucagon-Like Peptide-1 Receptor
  • Glucagon-Like Peptide-1 Receptor Agonists*
  • Heart Disease Risk Factors
  • Humans
  • Incretins* / adverse effects
  • Incretins* / therapeutic use
  • Machine Learning*
  • Male
  • Middle Aged
  • Risk Assessment
  • Risk Factors
  • Time Factors
  • Treatment Outcome

Substances

  • Glucagon-Like Peptide-1 Receptor Agonists
  • Incretins
  • GLP1R protein, human
  • Glucagon-Like Peptide-1 Receptor