Real-World Body Composition and Blood Pressure Changes Following Glucagon-Like Peptide 1 Receptor Agonist Initiation: A Causal Inference Study of Connected Device Data

Figures

• 

  Figure 1. Study flowchart and cohort formation. Flow diagram illustrating the selection process for the final analytic cohorts from the initial survey population. The diagram details the exclusion criteria applied sequentially. Two distinct analytic samples were derived: cohort 1 (n=2031) for the analysis of body composition outcomes and cohort 2 (n=672) for the analysis of blood pressure outcomes. aDefined as the requirement for at least 1 measurement in each of the 3 key temporal windows: the preinclusion history (6 months before t0 for adjustment), the active study period (10 months following t0), and the 1-year efficacy window (months 10-14 postinitiation). bRefers to users possessing valid measurements in all 3 periods specified above.

• 

  Figure 2. Distributional and longitudinal effects of glucagon-like peptide-1 receptor agonist (GLP-1 RA) initiation on body composition (cohort 1). (A, C, E) Plots displaying the distributions barycenters of values of percentage change in total body weight (A), device-estimated muscle mass (C), and device-estimated fat mass (E) relative to inclusion weight across the baseline period (the preinclusion window of 6 months) and the “after 10 months” 4-month period (the 1-year efficacy window, months 10-14). The left side (green) represents the barycenter distribution in the exposed group, and the right side (red) represents the barycenter distribution in the matched unexposed group (counterfactuals). The bracketed values display the associated scalar average treatment effect on the treated estimate, the estimated difference between the median of the 2 barycenter distributions, with 95% bootstrap CIs. (B, D, F) Longitudinal cross-sectional mean trajectories of percentage change in total body weight (B), device-estimated muscle mass (D), and device-estimated fat mass (F) from 6 months before 14 months postinclusion (t0). Trajectories were reconstructed using functional data analysis via…

• 

  Figure 3. Heterogeneity of treatment effect on body composition by baseline characteristics. Forest plot displaying the scalar conditional average treatment effect on the treated (sCATT) on percentage weight loss, device-estimated fat mass loss, and device-estimated muscle mass loss across stratifications of self-reported gender, age, baseline body mass index (BMI), and baseline muscle-to-fat ratio. Points represent the median difference between the barycenter distributions of the exposed and unexposed outcome distributions within each subgroup. Error bars indicate 95% CIs derived from 500 bootstrap replicates. Selection frequency indicates the proportion of bootstrap replicates in which the variable was retained as a robust independent effect modifier in the multivariable variable selection model. Adjusted β coefficients represent the median linear contribution of each baseline characteristic to the treatment effect size; unlike the stratified sCATT estimates, these coefficients isolate the independent contribution of each variable by accounting for correlations between baseline covariates.

• 

  Figure 4. Distributional and longitudinal effects of glucagon-like peptide-1 receptor agonist (GLP-1 RA) initiation on blood pressure (cohort 2). (A, C) Plots displaying the distributions barycenters of values of systolic (A) and diastolic (C) blood pressure (mm Hg) across the baseline period (the preinclusion window of 6 months) and the “after 10 months” 4-month period (the 1-year efficacy window, months 10-14). The left side (green) represents the barycenter distribution in the exposed group, and the right side (red) represents the barycenter distribution in the matched unexposed group (counterfactuals). The bracketed values display the associated scalar average treatment effect on the treated estimate, the estimated difference between the median of the 2 barycenter distributions, with 95% bootstrap CI. (B, D) Longitudinal trajectories of systolic (B) and diastolic (D) blood pressure from 6 months before 14 months postinclusion (t0), estimated via Principal component Analysis through Conditional Expectation smoothing. The green line represents the exposed group, and the red line represents the matched unexposed group (counterfactual). Shaded areas indicate 95% bootstrap pointwise CI.