Exploratory mediation analysis of the effect of semaglutide on cardiovascular outcomes in people with overweight or obesity in the SELECT randomised trial

Abstract Background Glucagon-like peptide 1 (GLP-1) receptor agonists have been shown to reduce major adverse cardiovascular (CV) events (MACE) in people with diabetes. The SELECT trial investigated weekly subcutaneous semaglutide 2.4 mg versus placebo in people with pre-existing CV disease and body mass index ≥27kg/m², but without diabetes at randomisation. Over a mean duration of follow-up of 39.8 months, risk of MACE was lower in patients treated with semaglutide vs placebo (6.5% vs 8.0%, hazard ratio, 0.80; 95% CI, 0.72, 0.90; P<0.001). There were improvements in several CV risk factors; however, mechanisms underlying the MACE reduction are unclear. Purpose This prespecified analysis explored whether the effects on measured CV risk factors may mediate the 20% reduction in MACE in SELECT. Methods CV risk factors measured during the trial and included as potential mediators in the analysis were: body weight, waist circumference, high-sensitivity C-reactive protein (hsCRP), glycated haemoglobin (HbA1c), lipids, blood pressure, estimated glomerular filtration rate (eGFR) and urinary albumin-creatinine ratio (UACR). Mediation analyses were performed using the Vansteelandt repeated regression approach. First, the observed difference in MACE-free survival between semaglutide and placebo at 36 months was estimated as the total effect. The direct effect of semaglutide was then estimated as the difference in MACE-free survival, had the semaglutide arm experienced the values of the mediator observed in the placebo arm during follow-up. The % mediation was then calculated as 100 × (1 – [direct effect] / [total effect]). A reversed counterfactual approach estimating the direct effect as the difference between arms in MACE-free survival, had the placebo arm experienced the mediator values observed in the semaglutide arm, was also conducted. A large disparity between the % mediation from the two approaches can indicate potential treatment–mediator interaction or hidden confounding. Results Statistically significant (P<0.05) improvements in all potential mediators were observed with semaglutide. Table 1 shows the estimated mediation of each potential mediator when evaluated separately. Point estimates for the % mediation were largest for waist circumference (64.0%), hsCRP (42.1%), HbA1c (29.0%) and body weight (19.5%), but with wide 95% CIs. For waist circumference, the estimated mediation was markedly lower using the reversed counterfactual approach, suggesting potential hidden confounding or treatment–mediator interaction (Figure 1). In a multivariable analysis, the estimated joint mediation was 31.4% (−30.1%, 143.6%). Conclusions The results suggest that neither change in body weight nor other measured CV risk factors fully explain the effect of semaglutide on MACE in SELECT. Substantial unmeasured pleiotropic effects of semaglutide on MACE not mediated through these risk factors remain possible.