Behavioural survey data were collected from samples of 1,404 IDUs in six cities between August and November 2007: Bandung (West Java), Jakarta, Malang (East Java), Medan (North Sumatra), Semarang (Central Java) and Surabaya (East Java). IDUs interviewed in Bandung, Jakarta, Medan and Surabaya were asked to provide biological samples and were tested for HIV and syphilis (n = 992), and chlamydia and gonorrhoea (n = 728).
In Jakarta, Malang, Semarang and Medan, IDUs were selected through two-stage, time-location sampling. Lists of venues where IDUs congregated (streets, parks, private houses and drop-in) were developed by non-governmental organizations providing services to IDUs and local health authorities. Samples of venues were chosen via systematic-random sampling with probability proportional to venue size. All IDUs present at the time of data collection were selected for participation.
In Bandung and Surabaya, IDUs were recruited through respondent-driven sampling (RDS) as access to IDUs was insufficient to use time-location sampling. Eight male IDUs “seeds” were recruited purposively in each city, ensuring that they (1) lived in the city; (2) were aged 15–49; and (3) were part of an extended network of IDUs. All seeds and subsequent recruits were each given three coupons to recruit other IDUs. Recruiters received 40,000 Indonesian Rupiah (equivalent to USD 4, depending on exchange rates) for each recruit who could be verified to be an IDU and completed the survey interview. The survey was terminated when the target sample size of 250 was reached.
Survey field teams were drawn from staff of provincial offices of the Central Statistics Bureau and Provincial Health Offices; all received specialized training. Interviews were conducted in locations that offered visual and auditory privacy.
Interviewers obtained witnessed verbal consent and gathered behavioural data using a structured, pre-coded questionnaire. In four cities, a nurse then collected blood through finger prick, and in three cities participants provided self-collected, first-void urine. Behavioural and biological data were gathered anonymously and linked by identification numbers. Per MOH surveillance guidelines, participants received a coupon for free HIV counselling and testing at a nearby Community Health Centre and were given their participant number in order to access their STI test results and receive treatment free of charge if needed.
Blood specimens were collected in EDTA tubes, stored at 4-60C and transported to the nearest government laboratory to be tested for HIV and syphilis. HIV was tested using two parallel rapid tests: SD Bioline® HIV 1/2 3.0 (Standard Diagnostics, Suwon City, South Korea) and Determine® HIV-1 (Abbott, Abbott Park, IL, U.S.A.). Discrepant results were re-tested at the national research laboratory using two ELISA assays: Murex® (Murex Biotech, Dartford, United Kingdom) and Vironostika® (Biomérieux, Marcy l’Etoile, France). Syphilis was tested using a treponemal test – Determine Syphilis-TP® (Inverness Medical, Bedford, United Kingdom). Urine samples were tested for chlamydia and gonorrhoea by Cobas Amplicor® (Roche, Basel, Switzerland).
Behavioural data were double-entered using CSPro 2.6.007 (U. S. Census Bureau). Laboratory data were entered using Microsoft Excel. Analysis was performed using Stata 9.0 (Stata Corporation, College, Station, TX). Differences in frequencies were assessed using the Wald test, and means were tested with the Wilcoxon ranksum test. All tests were double sided and p-values <0.05 were considered significant. Background characteristics and behaviours were presented in term of frequency, mean and median. Associations between HIV and individual characteristics and/or behaviours were assessed using generalized estimating equations (GEE), which control for correlations within sampling “clusters.” Histograms representing the distribution of continuous variables that were tested in GEE were examined to ensure that they had a single mode and limited skewness. Independent contributions of factors to predicting HIV infection were assessed by fitting variables associated with HIV infection in bivariable analyses significant at the p ≤ 0.20 level into multivariable GEE models. Backward stepwise elimination was used and logits significant at the p < 0.05 level on Wald tests were retained in the final model. To accommodate readers more familiar with logistic regression, the logit results were converted to odds-ratios by taking their anti-logs.
Preliminary data analyses indicated that respondents sampled via RDS differed from those sampled via TLS with regard to key background characteristics – they were somewhat better educated and more likely to hold salaried positions, have resided in their interview city their entire life, and be somewhat longer-term drug injectors (all p < 0.01). However, because such differences can be controlled in multivariable analyses, we opted to use the full set of data available irrespective of sampling method. Analyses were performed on the multi-site pooled data assuming stratified cluster sampling, with drug injecting venues and RDS recruitment chains being considered as clusters.
The study protocol was approved by the Ethics Committee of the Indonesian Centre for Biomedical and Pharmaceutical Research, as well as the Protection of Human Subject Committee of Family Health International. Oral witnessed informed consent was obtained from participants for publication of the survey results.