First, we examined participants’ baseline characteristics, stratified by participants who discontinued MMT. For categorical variables, we used Pearson’s χ2 test, and for continuous variables we used Mann-Whitney (Wilcoxon) rank sum test. Second, we examined factors associated with MMT discontinuation in the past six months during study follow-up using Generalized Estimating Equations (GEE) with a logit link function and an exchangeable correlation structure. These methods provide standard errors adjusted by multiple follow-ups per participant using an exchangeable correlation structure for the analysis of correlated data. Therefore, data from every participant follow-up visit were considered in these analyses. As a first step, we used bivariate GEE analyses to determine factors associated with MMT discontinuation. All variables with p < 0.1 in bivariate analyses were then entered in the multivariable logistic GEE model. In sub-analyses, among those who reported having discontinued methadone at follow-up, we explored self-reported reasons for methadone discontinuation.
