Emulating the LEADER trial in China: a regulatory science case study on non-interventional research

Integrating real-world evidence (RWE) into regulatory decision-making requires validation against pivotal randomized controlled trials for the same estimand. We emulated the LEADER trial using Chinese claims data to evaluate liraglutide's cardiovascular safety, assessing RWE-RCT concordance and examining the methodological adaptations and operational challenges encountered when emulating RCTs with claims data.

Using the Beijing Municipal Medical Insurance Database (2020-2024), we emulated the LEADER protocol. We identified new users of liraglutide and, as an active comparator, new users of DPP-4 inhibitors (DPP4i). The primary outcome was 3-point Major Adverse Cardiovascular Events. Propensity score matching was employed to balance baseline characteristics. Hazard Ratios were estimated using Cox proportional hazards models. Negative control outcomes (fractures and head injuries) and E-values were employed to assess the potential for unmeasured confounding.

The final cohort included 17, 772 patients (8, 886 per group). Over a median follow-up of 35 months, the incidence of MACE was 7.0% in the liraglutide group versus 7.8% in the DPP4i group. In the primary analysis, liraglutide was associated with a trend toward reduced cardiovascular risk (HR 0.90; 95% CI: 0.81, 1.01; P = 0.063), consistent with the effect size observed in the LEADER trial (HR 0.87), though not reaching statistical significance. However, sensitivity analysis adjusting for baseline comorbidities and drug exposure demonstrated a significant benefit (HR 0.90; 95% CI: 0.80, 1.01). The negative control analysis showed no association (HR 1.02; 95% CI: 0.93, 1.12), supporting the absence of unmeasured confounding.

This target trial emulation study validated the cardiovascular safety profile of liraglutide observed in the LEADER trial. Liraglutide demonstrated non-inferiority to DPP-4 inhibitors for the risk of MACE. While the primary analysis for superiority showed a trend toward benefit, it did not reach statistical significance. However, sensitivity analyses suggested potential benefits and supporting the robustness of the safety findings.