Knowledge on voluntary medical male circumcision in a low uptake setting in northern Uganda

The study was conducted in Gulu district in mid-northern Uganda. This region was the centre of a 20-year civil war, still has a high HIV prevalence above the national average (7.2 versus 6.2), and has a persistently low VMMC coverage [7]. Gulu district has an urban county (Gulu municipality) and two rural counties (Achwa and Omoro). VMMC services are available free-of-charge using funding from development partners, and are provided mostly through scheduled community outreaches, and also routinely at four health facilities - two in the urban county and two in the rural counties.

In this cross-sectional study, we recruited consenting adults aged 18–49 years living in a household within Gulu district for more than 6 months by December 2015, and were within a 4 km radius from a health facility with free circumcision services. Men were the primary interest group, but women with male sexual partners were also included to determine if their knowledge had any association with circumcision status of their male sexual partners. Initially, service providers were included but were interviewed and found to have universal knowledge.

Systematic sampling was used to select households in each of the three counties of Gulu district (2 rural and 1 urban) using lists from a recent mosquito net distribution campaign. Eligible residents were ranked by age in descending order, and potential participants were randomly selected using a KISH grid [11]. These were then contacted for a possible interview, with two additional attempts for respondents who were not initially reached. A semi-structured questionnaire which was developed for this study (Additional file 1), was pre-tested at a non-study site, was used to collect cross-sectional data on sociodemographic characteristics, men’s circumcision status or circumcision status of the male sexual partner for female respondents (irrespective of whether the circumcision was medical or non-medical), and knowledge on VMMC. Sociodemographic characteristics included age, location, marital status, tribe, religion, education level, and occupation. The questions for knowledge on VMMC were selected from the National communication strategy [10], administered in a respondent’s preferred language, and included priority questions on VMMC benefits (n = 4 questions), risks (n = 1), procedure (n = 3), and misconceptions (n = 6). Misconceptions ought to be rejected, while facts on benefits, risks, and procedure ought to be accepted.

Frequencies and percentages were used to describe categorical sociodemographic characteristics of the participants, and means(SD) and medians(IQR) were used for age. Correct acceptance of a fact or correct rejection of a misconception was scored 1, otherwise 0. Chi square tests or fisher exact tests were used to determine association between responses to each of the 14 selected/priority knowledge questions and circumcision status, the outcome of interest. For questions significantly associated with circumcision status, propensity score weighting (inverse probability of treatment weights) was then applied to balance those who gave correct responses and those who did not with respect to sociodemographic characteristics as potential confounders. Propensity scores were preferred over conventional regression modelling since they are much less vulnerable to model misspecification when dealing with measured confounders [12, 13] The balanced groups were then compared using modified poisson models to obtain prevalence ratios (PR), excluding subjects outside the common range of propensity scores. Prevalence Ratios (PR) were used because data were cross-sectional, they can convey strength of association between exposure and outcome, and are more conservative in magnitude than Prevalence Odds Ratios (POR) when the outcome is relatively common i.e. > 10% [14, 15], and have been used elsewhere in cross-sectional studies relating knowledge to circumcision [16, 17]. Modified poisson regression with robust variance can be used to estimate PR for cross-sectional data when log-binomial models fail to converge, which occurs quite often [15, 18]. To check potential influence of missing data on results, participants with missing responses to the selected knowledge questions were compared to those with complete responses using chi square tests or fisher exact tests as appropriate. Male and female respondents were analysed separately but results are presented side by side for comparison. STATA 14.2 was used for analysis [19].

The median age of male respondents was 25 years (IQR 22–29), and most were of the Acholi tribe (86.6%), 68.9% were catholic, and 72.7% had a secondary school education or higher. The Acholi tribe are a traditionally non-circumcising tribe. Location (rural or urban), tribe, education level, and occupation were associated with circumcision status of male respondents. In contrast, the median age of female respondents was 26.5 years (IQR 23–32), most were also of the Acholi tribe (84%), 64% were catholic, but less than half (46%) had a secondary school education or higher. Female respondents’ sociodemographic characteristics had no association with circumcision status of their male sexual partners, Table 1.

Bivariate analyses (Table 2) showed that majority of men did not know that VMMC reduces cervical cancer risk for female partner (92.7% failed to accept), and most erroneously thought VMMC significantly reduces risk of HIV transmission to female partner (75.5% failed to reject). However, these were not associated with circumcision status (p > 0.05). Instead, the misconception that VMMC reduces sexual performance was strongly associated with circumcision status (p = 0.001). Also associated with circumcision status was knowledge whether men normally bleed after VMMC (p = 0.006), the misconception that men usually desire more sexual partners after VMMC i.e. that VMMC increases promiscuity (p = 0.016), and knowledge whether VMMC differs from traditional circumcision (p = 0.042).

After applying propensity weights in poisson models to balance the comparison groups (Additional file 3: Table S2), male respondents who failed to reject the misconception that VMMC reduces sexual performance had 42% lower prevalence of circumcision (PR = 0.58, 95% CI 0.38–0.89, p = 0.012). Also, men who failed to reject the misconception that VMMC increases a man’s desire for more sexual partners had 35% lower prevalence of circumcision (PR = 0.65, 95% CI = 0.46–0.92, p = 0.014), Table 3.

Bivariate analyses (Table 2) for female respondents showed that majority of female respondents (91.2%) did not know that VMMC reduces their risk for cervical cancer, and most (88.2%) also erroneously thought VMMC reduces their risk of HIV acquisition from male sexual partners. However, these were not associated with circumcision status (p > 0.05). Instead, the misconception that VMMC reduces a man’s sexual performance was significantly associated with circumcision status of their male sexual partners (p = 0.032). After applying propensity weights in poisson models, male sexual partners of female respondents who failed to reject misconception that VMMC reduces sexual performance had significantly lower prevalence of circumcision, although the estimate was imprecise (PR = 0.22, 95% CI = 0.07–0.76, p = 0.016).

We could not establish whether the knowledge preceded circumcision given the cross-sectional design i.e. temporality between exposure and outcome, but the observed associations could plausibly explain the reluctance of men in this setting to get circumcised, and the reluctance of women to encourage their male sexual partners. Unmeasured confounders may explain away the observed association, but sensitivity analyses for these would also involve more untestable assumptions. Subsequent studies could establish temporality between messaging and subsequent circumcision; and also explore exposure to specific VMMC messages especially those addressing misconceptions and relationship with actual knowledge or attitudes to check whether messages are truly effective or not. A qualitative study in the same setting could complement our findings by further exploring cultural influences and other attitudes.