Statistical Analysis

First, a summary of all variables was made, including mean and standard deviation for continuous variables and a frequency count for categorical variables. To identify the association of deductibles with the outcome while correcting for other covariates, we used an ordered beta regression mixed model with a logit-link function and autoregressive covariance structure using the glmmTMB package [35] in R to model the adherence as a continuous number between 0 and 1. We investigated antihypertensive and antihyperlipidemic monotherapy separately. We used a mixed model to account for repeated measurements and within patient variation. The exponent of the β coefficients using this model were interpreted as the ratio in formula (1) and were referred to as the relative adherence ratio (RAR). This is similar to the definition of an odds ratio [36].

Adherence in the new setting could be for instance having reached the deductible limit, compared with the old setting were the deductible limit is not reached yet, keeping all the other factors constant [36]. Furthermore, we plotted the difference in expected values for binary deductible related variables in our dataset to show the direct impact on adherence, similar to an average treatment effect (ATE).