2.7. Data Processing and Analysis

Data were entered using double data entry into EpiData 3.1 (The EpiData Association, Odense, Denmark) and exported into Stata 14 (StataCorp, College Station, TX) for analysis. Descriptive statistics were used to describe participants' characteristics, sexual behaviors, condom use, and HIV prevalence. We adjusted all proportions separately for each city to account for the RDS method. This adjustment takes into consideration the probability of each participant to be included in the study. This probability was measured through weighting based on the size of each participant's network. Network size was determined using the survey question: “how many different people do you know personally who are female sex workers or sell sex? i.e., you know them and they know you, and you could contact them if you needed to?” The mean network size was 39: the network size by city was 69 in Ouagadougou, 21 in Bobo-Dioulasso, 39 in Koudougou, 13 in Ouahigouya, and 27 in Tenkodogo. Network size ranged from 1 to 1000. We presented population estimates and 95% confidence intervals (CI) adjusted for RDS design using the RDS Analysis Tools (RDSAT) version 6.0.1 (RDS, Inc., Ithaca, NY). Bivariate and multivariate logistic regression analyses were performed using Stata to identify factors associated with HIV infection at the p < 0.05 level of significance along with their 95% confidence interval (CI). These pooled bivariate and multivariate analyses were not RDS-adjusted because data from all cities were combined. Multivariate analyses were not conducted separately for each city due to smaller sample sizes in Koudougou, Ouahigouya, and Tenkodogo. Age categories were generated according to existing HIV planning goals with adolescent FSWs categorized as age 24 and younger.

Our outcome variable was HIV status (positive or negative) as determined by blood tests. Predictor variables included sociodemographic variables including age, education level, marital status, employment, and migration to Burkina Faso. Other predictor variables included those related to sex work including experience, number of clients, and condom use. First, sociodemographic and behavioral variables associated with HIV infection at the significance level of p < 0.2 in bivariate analyses were included in a backward elimination model selection procedure, and variables independently associated with HIV infection were retained in the multivariate model to produce the final results.