The Integrated Biological Behavioural Surveys (IBBS) are a series of surveys taken once every two years in Nepal, beginning in 2003. There have been 7 iterations of this survey. The surveys are intended to track the prevalence of various diseases among targeted populations, in this case men who inject drugs. In addition to assessing disease prevalence, the surveys recorded demographic, behavioural and social factors as part of the surveillance of HIV in Nepal (17). The key populations that the IBBS focus on are people who inject drugs (PWIDs), labour migrants, sex workers spouses and men who have sex with men (17).
This study uses a series of behavioural surveys taken between 2003-2011 and 2015-2017. The total number of participants across this period was 2,235 men over the age of 16 who had been injecting drugs for a period of at least three months. Participants were recruited using respondent driven sampling methods, where seeds from the specific target population of men who inject drugs were selected. A full description of the selection and sampling methods are available in the IBBS (17). The 2003, 2005, 2007 and 2009 surveys all had 300 participants, while the 2011, 2015 and 2017 surveys each had 345.
Face-to-face interviews were performed with participants’ answers being recorded on a set questionnaire, with the 2017 survey being performed on tablets. The IBBS surveys recorded a wide array of variables, a selection of which were used in the present study (a full list of which is available at the end of each IBBS survey). HIV testing was performed using the Determine HIV 1/2 test (manufactured by Abbot, Japan), which detects the presence of antibodies against HIV. If this first test produced a positive result, a second and third test were used for confirmation. These tests were Uni-Gold (Trinity Biotech, Ireland) and Stat-Pak HIV 1/2 (Chembio Diagnostics). The third test was only used if there was a disagreement between the first two tests used. If the first test was negative, then no further testing was done. The full testing algorithm is available in the IBBS reports (17), and also shown in Figure 1. In the 2017 IBBS study, the WHO HIV testing strategy was used. This involved using three consecutive reactive tests as the basis for HIV positive diagnoses. Stat-Pak HIV1/2 was the mandatory kit to confirm any HIV positive diagnoses.
The main outcome of interest in this study was HIV infection. This was defined by use of two rapid detection kits and collected from participants by testing blood samples. The tests were used in a cascading series, where if the first test returned a positive result, the second test would be used to confirm or contradict this result (17). Cases were determined to be HIV positive only if all three tests showed the blood sample to be reactive. The testing cascade differed in the earlier surveys (2003-2011) so that positives were counted either if the first and second tests were both positive. This difference was due to the different testing kits used from 2003-11.
Socio-demographic factors included age, education, and marital status, which were all defined as binary variables to maximise the size of comparator groups given the small sample size available at each time-period. Age was defined as being either “above” or “below” the median age. Similarly, education status was defined as being “primary or lower school education” and “secondary or higher school education”, while marital status was categorised as a “married” or “not married”. Living situation, which was defined as living “with” or “without” a sexual partner was also included in socio-demographic factors.
Behavioural factors included having undergone treatment for drug addiction, being tested for HIV, visiting an HIV testing and counselling centre, duration of intravenous drug use, use of a female sex worker, needle sharing behaviours, alcohol intake, condom use, use of unsterilized injecting equipment and the age at which drugs were first injected. Most of these factors were binary “yes” or “no” variables. Duration of drug use and age at first injection were defined as being “above” or “below” the median, while alcohol intake was broken into “everyday”, “sometimes” and “never”.
Overall, the average age of male PWIDs interviewed between 2003 and 2017 increased. The lowest average age occurred in the 2003 survey (average age of 23.4 years), and generally increased with each iteration of the survey. Average age decreased slightly between 2007 and 2009 (24.8 to 24.5 years), before increasing in 2011 (24.9 years). The highest average age occurred in the 2017 (28.2 years).
The education status of the respondents increased every year between 2003 and 2015, before decreasing in 2017. The highest completed grade of school on average in the surveys occurred in 2011 and 2015 (Grade 9). The average highest completed grade in 2003 was Grade 7, before increasing to Grade 8 in all other years.
Factors relating to knowledge of HIV included knowing someone with HIV/AIDS, having met and discussed HIV with a peer/community educator (PE/CE), and knowledge of where to access antiretroviral therapy. A summary of participant characteristics is provided in Supplementary Table 1.
Prevalence estimates and 95% confidence intervals for HIV were examined over the period spanning 2007-2017, stratified by each of the study variables described above.
A separate series of logistic regression models that were restricted to the period 2007 to 2017 were also conducted. Models were restricted to this period because the prevalence numbers expressed in 2003 and 2005 were likely substantially affected by measurement bias. This was due to overestimation of the positive results, which may have been due to a change in the HIV testing regime and the definition of a positive case. In these years, HIV testing was performed using two rapid tests (“Capillus” and “Determine”), with disagreements being resolved with the use of “Uni-Gold TM”.
Univariable and multivariable logistic regression models were conducted to investigate the association between the socio-demographic, behavioural and knowledge factors described above and HIV status. Models were aggregated over the period 2007-2017. Additionally, single years were investigated to evaluate the magnitude of change of relative associations between different time points. Multivariable models adjusted for the potential confounding factors of age, education level and marital status. Multivariable models adjusted for the key potential confounding factors of age, education level and marital status. Knowledge, behavioural and health service determinants were considered for inclusion in multivariable models, however given small numbers of cases within strata for some variables and the cross-sectional data used, the putative causal directions between these variables could not be clearly specified, therefore a minimal set of socio-demographic factors were adjusted for. All analyses were conducted in R Studio using the glm and confint.lm functions, which produced the estimates for odds ratios and 95% confidence intervals for each of these models.