Uptake of HIV testing in Burkina Faso: an assessment of individual and community-level determinants

This study uses data collected during the 2010 Burkina Faso Demographic and Health Survey (DHS), the first nationally-representative survey in Burkina Faso to collect information on previous HIV testing among both women and men. The survey protocol has been published elsewhere [26]. Briefly, the survey used a stratified, two-stage cluster sampling design. The country was stratified into rural and urban areas. Among each stratum, the first stage consisted of a random sample of clusters (primary sampling unit, PSU) using sampling probabilities proportional to the number of households in the cluster. Clusters were established by a general population and housing census conducted in 1996 (the average cluster populations were 1000 and 1200 in rural and urban areas, respectively). The second stage involved the systematic sampling of households from the selected clusters. A nationally representative sample of 14,957 households was thus obtained. All women aged 15 to 49 years were eligible to be interviewed. For men, a sub-sample of half of the selected households was randomly selected, from which all men aged 15 to 59 were eligible to be interviewed. Hereafter, the term community refers to all respondents living within the same geographical environment, corresponding to the PSU of the survey.

The main outcome for this study was the participants’ self-reports of having ever been tested for HIV. Our independent variables were individual-level or community-level characteristics.

Individual variables covering socio-demographical, economical and behavioral dimensions, as well as knowledge and attitudes toward HIV/AIDS, were considered in the analyses. Socio-demographic covariates included: age, education, marital status, religion, occupational status, media exposure, location of residence (urban/rural), and an asset-based index of wealth [28]. The number of children ever born was also included for women. Sexual behavior was measured by self-reported lifetime number of sexual partners. HIV/AIDS knowledge was assessed by creating an index of correct responses to five (for men) and nine (for women) questions related to HIV transmission. Those sets of questions were summarized using principal component analysis, where the first axis explained 34% and 32% of the variance for women and men, respectively. This first axis was used to create a three-category index of HIV knowledge. Stigma towards PLWH was proxied using a set of four questions included in the DHS questionnaires: 1) willingness to share a relative’s HIV infection status, 2) willingness to care for an infected relative, 3) belief that a female teacher infected with HIV should teach, and 4) willingness to buy food from an HIV infected vendor. This choice of independent variables was guided by prior research on HIV testing in sub-Saharan Africa countries [18, 25].

The main explanatory community-level variables of interest were: HIV prevalence, knowledge of HIV testing service, HIV/AIDS knowledge, HIV-related stigma (percentage of respondents with accepting attitudes towards people living with AIDS), and community socio-demographic characteristics (proportion of respondents who are at least high-school educated in the community, proportion of households belonging to the poorest quintiles of wealth, and proportion of individuals reporting low media exposure). Consideration of these variables was based on their potential to be modified and their known or hypothesized association with HIV testing [22, 25].

Community-level variables were aggregated from individual responses pertaining to individuals of the same sex and the same PSU (except for HIV prevalence which was aggregated at the regional level due to low number of cases). To avoid endogeneity problems related to the double inclusion of variables in the model – first at the individual-level and again at the aggregate-level – community-level variables were calculated by excluding an individual’s own response from the numerator and denominator of the aggregate-level variables. Indicators from each of the community-level determinants of HIV testing uptake were dichotomized using the median as a cut-off point. The stigma question regarding willingness to care for an infected relative was not considered at the community level because preliminary analyses showed that, for both women and men, the communities had low levels of stigma with small between-community variations that were not deemed qualitatively important.

For descriptive statistics at the individual and community levels, estimates accounted for the complex survey design using the sampling weights provided by DHS [29]. Prevalence ratios (PR) were obtained using a modified Poisson regression model that used Generalized Estimating Equation to take into account clustering of observations and perform the unbiased variance estimation [30, 31]. We did not adjust for survey weighting in the regressions. Univariate and multivariable analyses of individual-level covariates were first conducted. A complete case analysis was used and missing observations for women (n = 385) and men (n = 100) were disregarded. This was followed by multilevel models integrating the community-level variables. All multilevel models were adjusted for the following individual-level covariates: age, education, number of children ever born (for women only), marital status, religion, wealth index, place of residence, media exposure, HIV knowledge, lifetime number of sexual partners, and the four personal stigma questions. Two multilevel models were fitted for both gender. The first multilevel model adjusted for individual-level variables and each of the main community-level determinants hypothesized to be related to the community variation in HIV testing separately: 2010 HIV prevalence, knowledge about VCT service, HIV/AIDS knowledge and HIV related stigma (model 1). The second multilevel model is fully adjusted for all individual-level and community-level variables enumerated above (model 2). Analyses were restricted to participants reporting having ever had sexual intercourse. All analyses were stratified by gender because attitudes toward HIV testing are likely to vary between women and men [32, 33]. Statistical analyses were performed using the R statistical software and the “geepack” package was used to fit the regression models [34].

The sociodemographic characteristics of the study population are presented in Table 1. About half of the male participants were aged 35 years or older whereas only a third of female respondents were in this age group. About two thirds of respondents had no schooling and pproximatively 90% of women and 80% of men were married or living in a union. More than half were Muslims and two thirds lived in rural areas.

Among men and women, three-quarters of participants knew where they could get an HIV test (Table 2). Men were more educated, had higher level of media exposure, and higher HIV/AIDS knowledge than women. Additionally, men were more likely to have views that stigmatized HIV/AIDS than women (21% vs. 9% expressed stigmatizing views to all four stigma-related questions) but few individuals had no stigmatizing views on all questions (7% and 6% of women and men, respectively).

The proportion of women reporting having ever been tested for HIV was 35.6% (95% Confidence Interval [CI]: 32.9–37.4%) while this proportion was 10% lower for men (25.6%; 95% CI: 24.2–26.6%). Among women having ever been tested, 90.8% have had an HIV test as part of an antenatal care visit. HIV testing differed greatly geographically and was highest for both genders in the department of Central Region, where the capital Ouagadougou is located, at 61.9% for women and 46.6% for men (Figs. 1a, b). Lowest uptake of HIV testing was found in in the Sahel for women (16.4%) and the Boucle du Mouhoun departments for men (13.9%).

For both genders, uptake of HIV testing was associated with age, education, religious affiliation, household wealth index, being employed year-round, media exposure, number of lifetime sexual partners, and level of HIV knowledge (Table 3). The probability of having been tested was highest among women aged 15–24 and men aged 25–34 years. Having non-stigmatizing views on PLWH was associated with HIV testing uptake but this was not consistent for all stigma-related questions. For women, the number of children ever born was significantly associated with HIV testing.

Table 4 presents results of the community-level determinants of HIV testing. For model 1, community-level determinants are adjusted only for selected individual-level variables. On the other hand, model 2 is fully adjusted for individual and all community-level determinants listed in Table 4. Results from this fully adjusted model (model 2) suggested that women were significantly more likely to report testing in communities where a larger percentage of respondents knew the location of a VCT service, where views towards buying food from HIV positive vendors were less stigmatizing, where the proportion of respondents in the highest wealth quintile was higher, and in communities with the highest media exposure. Furthermore, women residing in community where the proportion of respondents not working year-round was higher than the median were 10% less likely to report having ever been tested. For men, few community determinants reached statistical significance for both model 1 and 2. Men were more likely to report previous testing in communities where respondents were more educated.