Study setting

The study took place in 2009 in the Nouna Health District, north-western Burkina Faso. At the time of the study, the district had a population of approximately 311,000 distributed in 300 villages and included 31 first-line facilities, Centres de Santé et de Promotion Sociale (CSPS),and one district hospital, also located in Nouna town. A sub-portion of the district has been part of a Health and Demographic Surveillance System (HDSS) for over 15 years[24].

In line with national estimates[25], adult HIV prevalence in the area has remained stable at 3.6%[26]. At first, HIV counseling and testing based on an “opt-in” strategy was only available in the district hospital (2003–2006). Following national directives[27] starting in 2006, all CSPS were equipped to offer HIV counseling and testing to pregnant women and their partners as the entry point to PMTCT services. Explicit efforts were made to address the need for couple counseling and testing, encouraging spouses to be tested together with their pregnant wives. In addition, providers were encouraged to offer testing to anyone requesting it and/or to anyone who presented with potential symptoms of infection. Findings from a parallel study confirmed the availability of test kits even in the rural facilities[28]. The service was structured to ensure that blood samples could be drawn in the rural facilities and then regularly transported to the district hospital for laboratory testing. Both pre-test and post-test counseling were scheduled to take place at the facility where the blood sample was taken.

Study design

This population-based study adopted a triangulation mixed methods design[29]. The triangulation design was chosen as the most appropriate mixed methods design, since the research question focused on one single phenomenon, i.e. the decision to test or not to test for HIV, but aimed at capturing all its possible dimensions, in such a way that neither qualitative nor quantitative methods alone would have been able to do. The conceptual framework guiding the study recognized the decision to test or not to test for HIV as the product of the interplay between access factors (acting at the individual and at the community level) and individual opinions, beliefs, and practices related to knowledge, risk perceptions, and perceived vulnerability. The quantitative component relied on household survey data to identify socio-demographic, economic, and health system factors associated with access toHIV testing. The qualitative component relied on in-depth interviews to explore men’s motivation to test or not to test in relation to their knowledge, attitudes, and perceived vulnerability to the disease.

Study population, sampling, and data collection

Analytical approach.

Household survey data were analysed using Stata 11 (Stata Corporation, Texas, USA). The outcome variable was defined as “having tested for HIV”. Since HIV testing could be observed only among those men who had in the first place been offered an HIV test, we initially intended to rely on a Heckman estimation model to take into account the selection bias induced by the need to be offered a test before being able to undergo one[34,35]. Our definition of being offered a test included individuals who voluntarily sought a test/asked to be tested, since technically, one always needsto be offereda test before being able to undergo it.

Since it was impossible to identify an adequate instrumental variable for the efficient computation of a Heckman model, wereverted to applying a two-part model[36]. First, we computed a model to estimate factors associated with the probability of being offered an HIV test. Second, we computed a model to estimate factors associated with the probability of having undergone an HIV test. Model 1 (being offereda test) was computed using the complete sample of men having provided complete answers to the module on HIV testing (N = 937). Model 2 (HIV test behaviour) was computed on the truncated sub-sample of men reporting that they had been offered an HIV test in the first place (N = 357). This analytical approach allowed us to discern factors associated with the probability of being offered a test from those associated with the probability of actually undergoing the test, in order to depict decision-making more realistically. In both instances, we did not aim at saturating the model by exploring all possible factors associated with either level. We rather aimed at identifying a few theoretically relevant factors which could be properly identified in our dataset. Probit estimation accounting for clustering at the village level was used to compute both models. The two equations estimated separately are written as follows:

1. Model 1: P(offered_i = 1/x_1i) = ф(x_1iβ₁).
2. Model 2: P(tested_i = 1/x_2i, offered_i = 1) = ф(x_2iβ₂)

with

offered_i = 1 if being offered an HIV test and 0 otherwise.

tested_i = 1 if having undergone an HIV test having been offered one and 0 if not (having been offered one).

and ф(.) the normal cumulative distribution function.

x_1iandβ₁ are respectively the factors associated with the probability of being offered a test and the vector of coefficients associated with these factors.

x_2iandβ₂ are respectively the factors associated with the probability of actually undergoing the test and the vector of coefficients associated with these factors.

Prior to estimating both models, we carried out descriptive analysis to identify bivariate associations between the explanatory variables included in our analysis and the two distinct outcome variables.

Table 1 displays the explanatory variables included in the analysis as well as the expected sign of the coefficient, indicating the a priori underlying hypothesis we held. Most variables are straightforward. The decision to dichotomise ethnicity into “Bwaba” and “other ethnicity” was informed by previous research showing that the Bwaba often display different health seeking patterns [37–39]. Standard procedures applying principal component analysis were used to estimate household wealth [40,41].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Variables and their measurement.

https://doi.org/10.1371/journal.pone.0130216.t001

Analysis of the qualitative material took place on the transcribed material, directly in French. We translated into English only the citations included in this manuscript. First, the materialwas independently coded by three of the authors, all relying on an inductive approach which let codes emerge as the reading proceeded. Second, the three analystscompared their analyses and returned to the material to resolve divergent interpretations. Third, the collective emerging interpretation was discussed among all authors, informed in the light of the quantitative findings, and finalised.

The interpretation of the findings as presented in this paper is based on the joint appraisal of the quantitative and qualitative findings. The process of bringing together in one single conceptual model findings from the two components was conducted at the end of the two distinct analytical approaches.

Quantitative findings

Complete information could be ascertained for 937 (89%) of 1058 male household heads. Selection bias at this level was excluded since no systematic differences were observed between households heads with complete vs. household heads with incomplete information (data not shown). Of the 937 men included in the final sample, 357 (38%) had been offered an HIV test. Bivariate analysis indicated the existence of important differences between the men who had been offered an HIV test before, and the men who had never been offered an HIV test, suggesting the existence of selection bias at this level (Table 2). Men who were offered testing were more likely to be younger, of Bwaba ethnicity, wealthier, live closer to a CSPS and in a HDSS village,tohave attended some form of formal schooling, and to be aware that HIV testing could be performed in a CSPS.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Descriptive statistics and t-test comparing men being offered and men not being offered HIV test (see Table 1 for variable values).

https://doi.org/10.1371/journal.pone.0130216.t002

Of 357 men who were offered HIV testing, 97 (27% of the sub-sample being offered testing and 10.3% of the total sample of men included in this study) had taken up the opportunity at any prior point in time. Bivariate analysis revealed that only two factors were significantly positively associated with the probability of having tested for HIV: having attended some form of formal schooling and living closer to a CSPS. Household wealth also showed indication of a positive association with HIV testing (Table 3). Table 4 reports the estimation of the two probit models, one estimating the probability of being offered an HIV test (n = 916) and one estimating the probability of actually having undergone an HIV test once offered it (n = 349). Due to missing values on some explanatory variables, the number of observations included in the estimation of the two models is slightly lower than the total samples indicated above. Younger age, household wealth, living in an HDSS village, and knowing that HIV testing is available at the CSPS were all factors positively associated with the probability of being offered an HIV test (model 1). Household wealth and literacy were also found to be positively associated with the probability of actually having taken an HIV test, while the distance to the closest CSPS was found to be negatively associated with this probability (model 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Descriptive statistics and t-test comparing men who tested and men who did not test for HIV (see table 1 for variable values).

https://doi.org/10.1371/journal.pone.0130216.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Probit regressions of being offered HIV test and being tested (when offered).

https://doi.org/10.1371/journal.pone.0130216.t004

Qualitative findings

One of the 38 men selected for the qualitative study component refused to be interviewed. When asked to reveal whether they had ever undergone an HIV test, only eight (and not the expected 19) confirmed the information shared during the quantitative phase of the study, meaning that twenty-eight out of thirty-seven men reported never having tested for HIV.

Knowledge on HIV transmission was generally very high among all respondents, with no striking differences between those who had tested and those who had not tested. Men were aware that HIV could be transmitted both sexually and during pregnancy and were generally also aware that treatment was available both to avoid vertical transmission and to delay progression towards the disease. Very few men indicated the sharing of eating utensils and mosquito bites as potential modes of HIV transmission.

If you see a woman and you do not know that she is ill and you have sex with her, you can get the disease. This is why, even if you love women, you need to protect yourself before having an intimate relationship with them. Otherwise, it can be severe if you just say that you love them and forget to wear the condom that retains the germs (ID 1250008).

Thirty-four out of thirty-seven respondents were aware that a test is available to detect HIV, but only one out of eight men who had tested was aware that the test could be performed at the CSPS. All other men believed that testing facilities were only available in the district hospital in Nouna or at the regional hospital in Dedougou and declared having been tested at one of these two locations. A few respondents mentioned that health workers or volunteers from local NGOs conduct outreach campaigns to promote testing, but that they do not offer testing opportunities during these campaigns.

It is just in Nouna that you can do the HIV test (ID 106 18 00).

Perceived health status was the key motivating factor in deciding whether to undergo or not to undergo the test, both among men who tested as well as among men who did not test. On the one hand, men who had not tested justified their decision not to test with the fact that they felt healthy and therefore perceived no need to undergo the test. They supported their decision not to test with the belief that the illness would have surely manifested itself if they had been infected with HIV. On the other hand, men who had tested explained that upon falling ill, they had feared being infected with HIV and therefore had decided to undergo the test. Only two out of the eight men who reported testing had done so to respond to the request of their pregnant wives, following the couples counselling delivered during antenatal care.

If I was ill, I would accept doing the test to know exactly what I have. But I feel well, so I do not accept doing the test (ID 150 00 06).

I did the test because I wanted to know what I was ill. I did a first test at the CSPS, a second one at the hospital in Nouna, and then one more at the hospital in Dédougou, but they all turned out negative. This is what I did not understand. I am still looking for the cause of my illness (ID 129 11 00).

Men who had not tested generally perceived themselves to be at very low risk of contracting the disease. This low risk perception was motivated by the general low prevalence in the community and by the health status and behaviour of their sexual partners. Similarly to what they argued for their own health status, men also believed that had their sexual partners being infected with HIV, the illness would have manifested itself. Likewise, most men did not expect their wives to be at any risk of having contracted HIV since they trusted them to be faithful.

It is because of the disease is not very common here. Lately, we have not seen many cases. Some years ago, there were many cases. But lately, we have the feeling that the diseases had regressed. (ID 106 01 00)

If you know that you do not engage in risky behaviours and your wife is serious, then there is no need to test. (ID 153 06 00)

Views on health workers differed substantially between men who had tested and men who had not tested. Men who had tested reported proximity to the health facility and a general wish to comply with health workers’ sensitisation activities as important factors in motivating their decision to test. On the contrary, men who had not tested often expressed subtle mistrust in the health workers’ ability to keep the test result anonymous. They feared stigma should the result of the test or even the fact that they had tested ever be leaked to the community. In addition, men who had not tested did not trust that the test was offered free of charge and were convinced that they would have had to pay something for it.

In our village, if it is known that you have AIDS, you die within ten days. (ID 134 01 00)

Nobody told me that the test needs to be paid for, but I know that at the level of the health facilities, one always needs to pay something. I have never heard of any service being delivered for free. (ID 153 01 00)

Methodological considerations

Notwithstanding the strengths of this study, rooted primarily in its mixed-methods population-based approach, we need to acknowledge a few limitations. First of all, the fact of having interviewed only household heads might have limited the generalizability of the findings. Household heads represent a specific group of men, normally enjoying better access to health care services than the general population [51][52]. Thus, it is not to be excluded that they are more likely than other men to be offered testing, given their more frequent contact with health care services. However, household heads tend to be older than the general male population (average age in our quantitative sample was 46 years) and as such they may perceive themselves at lower risk of contracting HIV than their younger counterparts[5]. It is plausible to assume that the HIV testing rate would have been higher than what we observed in our sample had we been able to conduct the study on a sample of men not restricted to household heads. Given the absence of comparable population-based data from prior studies in Burkina Faso, however, we cannot assess if and to what extent the choice to interview exclusively household heads biased the direction of our estimates.

Only eight out of sixteen men who had originally been identified as having tested during the quantitative survey confirmed having done so during the qualitative interviews. This could be an indication of two alternative circumstances: either men provided socially acceptable answers during the household survey (reporting to have tested although this was not the case), ormen denied being tested during the qualitative interview, out of fear of stigma. We have no way to ascertain with certainty whether the former or the latter case applies to our data. We are, however, more inclined to believe that the latter case applies, since the survey question on testing was accompanied by a number of follow-up questions (location of testing, occasion for testing, etc.), which should have allowed the interviewer to recognize if a respondent was reporting non-realistic information. Qualitative interviews are inevitably more intimate and somewhat more threatening than quantitative surveys, which collect information in an anonymous manner. This imbalance in the qualitative sample represents an obvious weakness in our study design, since we ended up interviewing mostly men who had not(or declared not to have) tested for HIV, while our initial wish was to conduct in-depth interviews with an equal number of men having tested and not having tested. This limited our ability to fully explore what can motivate men to test and to build relevant policy guidance for future sensitization campaigns on the basis of these results.

We also need to acknowledge the limitation of estimating factors associated with the offer to test and with the decision to test using a two-part model rather than a Heckman selection model[34]. Our choice was dictated by the contingency of the dataset. Still, estimating the second equation on the truncated sample of men being offered the test is methodologically inferior to estimating the two equations simultaneously since it does not accurately account for the potential for self-selection into the sample. Lastly, the small pseudo-R2 on both models suggests that the explanatory variables included in the models were only able to explain a limited portion of the variation in the outcome variable. However, the variables included represent the complete range of theoretically relevant variables available in the dataset. We purposely chose not to improve the fit of the model by simply including non-theoretically relevant variables.