Statistical analyses

Behavioral and health characteristics were assessed with frequencies and percentages for categorical variables and medians and interquartile ranges for continuous variables. Comparisons between groups were performed with the Chi-square test for categorical items and the Kruskal-Wallis test for continuous items.

Three logistic regression models were estimated: crude, adjusted, and a weighted marginal structural model (MSM). The marginal structural model was used to estimate the joint effects of AUD and SUD on self-reported STI in the past 3 months. We accounted for the impact of potential informative censoring occurring when participants were lost-to-follow-up due to reasons related to the study using inverse probability of censoring weighting methods.⁴⁹ Censoring and probability weights were constructed using both baseline fixed characteristics (age, years of education, and sex work venue) and time-varying covariates (AUD, SUD, condom use with paying partners, positive PSA test result, psychological distress, income, required drinking alcohol in the workplace, ATS use with sex partners, number of sex partners, food insecurity, housing instability, and violence victimization). All visits were included and current and lagged values of each factor were used to construct weights. Censoring and probability weights were constructed separately for each outcome (AUD and SUD), and then multiplied together to create a joint weight using a two-step process.⁵⁰ Weights were stabilized to reduce the range of values and were centered around 1.3.

All models accounted for clustering by province and adjusted for intervention participation. The crude model estimated the association between AUD and SUD and self-reported STI in the past 3 months. The adjusted model accounted for baseline condom use with paying partners, psychological distress, age, sex work venue, income, food insecurity, housing instability, number of sex partners, drug use with sex partners, work place drinking, PSA test results, recent sexual or physical violence victimization, and months followed in the study. Multivariable models excluded observations without complete case information. Missing data were relatively rare (<1%) on STI reports, baseline condom use with paying partners, housing instability, drug use with sex partners, and PSA test results. Complete case data were available for AUD, SUD, psychological distress, age, sex work venue, income, food insecurity, number of sex partners, and recent sexual or physical violence victimization. All analyses were conducted in Stata version 15, College Station, TX.