OBJECTIVES: Mediation analyses are useful to disentangle possible causal pathways from an exposure/treatment to its clinically relevant outcomes. Mediation effects are commonly presented in changes in hazard ratios obtained from Cox proportional hazard model analyses, yet the assumption of proportionality is however violated with a post-exposure mediator added into the analysis. Dynamic path analysis evolved more recently where the proportionality is not required and all time-varying mediators can be simultaneously modeled, thereby reflecting the complexity of mediators in causal pathways. We sought to apply dynamic path analyses for assessing the heterogeneity of mediation effects on liraglutide-associated major adverse cardiovascular event (MACE) outcome by racial and smoking status of patients.

METHODS: LEADER trial data was utilized and mediators of interest included glycated hemoglobin (HbA1c), urine albumin-to-creatinine ratio (UACR), body weight (BW), systolic blood pressure (SBP), and low-density lipoprotein (LDL). Mediation percentage of treatment effect was estimated as dividing mediated treatment effect by total treatment effect and corresponding bootstrap-based 95% confidence interval (CI) was generated.

RESULTS: Individual HbA1c, UACR, SBP, BW, and LDL data mediated liraglutide-associated MACE outcome by 38.2% (95% CI: 11.9-286.7%), 17.9% (6.3-100.2%), 6.8% (2.6-33.2%), 2.9% (-0.5-18.2%), 1.6% (-0.8-6.0%), respectively, in the third year following treatment initiation. Heterogeneity of HbA1c/UACR-mediated liraglutide outcomes was revealed across racial (i.e., 53.9%/24.9%, 27.4%/18.5%, 18.7%/-0.46%, and 17.4%/13.5% in white, black/African Americans, Asians, and others, respectively) and smoking (i.e., 11.4%/8.5% for non-smokers, 64.8%/22.2% for smokers) groups.

CONCLUSIONS: Significant mediators identified using dynamic path analyses are of clinical value and consistent with previous findings, thereby supporting the validity of dynamic path analyses in our illustration. Given the possibility of disparities in HbA1c/UACR-mediated treatment effects by racial and smoking status, further investigations on different causal pathways inherent to diverse patient clinical and behavioral characteristics are warranted in support of individualized medicine.