An assessment of fishing communities around Lake Victoria, Uganda, as potential populations for future HIV vaccine efficacy studies: an observational cohort study

Between September 2011 and March 2013, we conducted a community-based cohort study among adults in 8 fishing communities (1 lakeshore and 7 islands) in 3 districts bordering Lake Victoria in central Uganda, a region that is mainly inhabited by people of Baganda tribe/ethnicity. Detailed study procedures have been previously described [35] but briefly, a random sample of 2200 participants aged 18–49 years resident in fishing communities for at least 6 months prior to baseline enrolment were selected from a community wide census database using Stata® 12 (StataCorp, College Station, TX) software. Of the 2200 selected, 2191 provided written informed consent and were enrolled at baseline. Interviewer- administered semi-structured questionnaires were used to collect data on socio-demographic characteristics, HIV risk behaviours, and willingness to participate (WTP) in future HIV vaccine studies. Over 99% of the questionnaires were administered in Luganda (the most commonly spoken language in the study communities) while others were in English. Two weeks to one month prior to the scheduled follow up visit date, community-based health education meetings were conducted, attended by both participants and non-participants. In addition to education and information sharing, participants were reminded of the impending follow up visit but no individualized contact, mailings, phone calls or written invitations were provided. At each visit, venous blood was collected for HIV-1 serological testing and participants were given free voluntary counseling and testing by certified HIV counselors. HIV infected participants were referred to HIV/AIDS care centres for further management and encouraged to seek care. Individuals in study communities (participants and non-participants) were given free access to HIV prevention services including health education, counseling, treatment of sexually transmitted infections (STIs) and voluntary medical male circumcision. After completion of the study procedures each study participant was reimbursed 5000 Uganda shillings (2 USD) for time and travel. Institutional Review Board approvals were obtained from the Uganda Virus Research Institute’s Science and Ethics Committee and the Uganda National Council for Science and Technology.

HIV-1 status was determined by rapid HIV tests performed in the community by certified laboratory technologists and EIA confirmation in the laboratory at Uganda Virus Research Institute. In the rapid HIV testing algorithm blood samples were first tested with Determine® HIV assay (Alere Medical Co., Ltd., Chiba, Japan), and if negative, results were reported as negative. Determine® positive samples were then tested with HIV 1/2 Stat-Pak® assay (Chembio Diagnostic Systems, Inc. Medford, NY, USA), if positive, results were reported as positive. But if negative on Stat-Pak®, Uni-Gold™ HIV test (Trinity Biotech plc, Bray, Ireland) was used as a tie-breaker. All positive rapid results were confirmed using 2 parallel Enzyme Linked Immuno Sorbent Assay (EIA) tests: Vironostika (HIV Uni-Form II plus 0 microelisa system, Biomerieux, SA, Marcy l’Etoile, France); and Murex HIV-1.2.O (Diasorin S.P.A, Dartford, United Kingdom). Concordant EIA positives were taken as positive but discordant EIA results were comfirmed using HIV RNA PCR (COBAS® AmpliPrep/COBAS® TaqMan® HIV-1 Test, v2.0 from Roche Molecular Diagnostics, Pleasanton, CA, USA).

To evaluate the suitability of fishing communities as potential populations for future HIV prevention efficacy studies, we assessed absolute HIV incidence, retention and WTP. HIV incidence rates per 100 person years at-risk (pyar) were estimated among initially seronegative participants as the number of seroconversions divided by pyar multiplied by 100. HIV infection was estimated to have occurred at the midpoint between the last negative and first positive serologic tests. Retention was calculated as the number of participants traced and interviewed at the 12 months visit divided by the number enrolled at baseline. Given the high proportion of retention/follow up (76.9%) we used log-binomial regression models to estimate unadjusted and adjusted prevalence proportion ratios (PPRs) and corresponding 95% CIs of factors associated with retention. Since odds ratios give biased estimates of prevalence ratios when the proportion of the outcome is greater than 10%, we used log binomial regression instead of logistic regression [36, 37]. The final model on retention included gender, age, tribe, occupation, marital status, and duration of stay in community. WTP was assessed hypothetically given that the study did not involve recruitment and receipt of a vaccine (experimental or licensed). This was done only at the baseline visit and the specific question on WTP was: If you were requested to participate in a research study on an experimental HIV preventive vaccine, would you be willing to join the study as a participant? The responses were willing, not willing, not sure, partner would decide and parent would decide. Only the first 3 got responses and the covariate was dichotomized into willing and not willing; not willing and not sure combined because the frequency of the latter was too small for meaningful analyses. Since we were interested in determining characteristics of fisherfolk that one would address in preparation for efficacy trials of HIV prevention, we modeled unwillingness to participate in HIV vaccine trials as the outcome. With a prevalence of 10.7%, ordinary multivariable logistic regression models were used to determine factors associated with unwillingness to participate in vaccine trials. The final model included sex, occupation, duration of stay in community, and alcohol consumption. In all multivariable models we used empirical variance estimator to estimate robust standard errors accounting for any potential correlation at household level (where more than 1 participant from a given household were selected) [38]. Inclusion of variables in multivariable models was based on biological plausibility and a bivariate statistical significance at an alpha (α) of <0.15 but statistical significance was determined by α <0.05. All regression models were constructed using stepwise logical model building method [39] and statistical analyses were performed using Stata® 12 (StataCorp, College Station, TX) software. Analyses involved participants that were HIV negative at baseline and hence at risk of infection.

Table 1 shows absolute retention rates, and unadjusted and adjusted prevalence ratios (PRs) of factors associated with retention in fishing communities around Lake Victoria, Uganda. Of the 2191 participants enrolled at baseline, 1685 were interviewed at the 12 months visit giving an overall retention rate of 76.9%. The highest retention rate of 89% was seen in participants who had spent 5 or more years in fishing communities. Retention rates of 80% or more were observed in farming occupation (85.4%), Baganda tribe (80.9%), married monogamous (80.5%), HIV negative at baseline (80.2%), and being aged 30 or more years (80.6% for 30–39 years and 83.7% among 40–49 years). The lowest retention rate of 54.1% was observed among participants who at baseline had spent less than a year in fishing communities. At bivariate analysis, retention was significantly lower among participants aged 18–24 years, non-Baganda, bar/hotel/restaurant workers, single never married, baseline HIV positive, and those who had spent less than a year in the fishing communities. Retention did not statistically differ by sex, age, education status, or religion but adjusted differences were observed according to ethnicity, marital status, occupation, duration of stay in the community, and baseline HIV sero-status (Table 1). Retention was significantly higher among the Baganda ethnic group (adj.PPR = 1.07; 95% CI, 1.03 - 1.12), HIV negatives (adj.PRR = 1.19; 95% CI, 1.13 - 1.27), those not currently married but previously married (adj.PRR = 1.10; 95% CI, 1.01 - 1.18), and among those who had spent more than 1 year in the community at the time of enrolment. Compared to participants with less than 1 year of stay in the community, adjusted PRRs of retention were 1.34 (95% CI 1.21 - 1.48) for 1 to 4 years, 1.57 (95% CI, 1.43 - 1.73) for 5 to 10 years, and 1.59 (95% CI, 1.44 - 1.76) for more than 10 years of stay in fishing communities.

The overall HIV incidence rate was 3.39/100 pyar (95% CI; 2.55 - 4.49). We estimated retention rates and absolute HIV incidence rates by a combination of selected risk factors so as to assess potential characteristics that would maximize both retention and incidence in case of targeted enrollment in efficacy studies in this population (Table 2). HIV rates presented in Table 2 are absolute risks that apply to specific groups of single of combined risk factors and are not comparative. Participants aged 30 or more years tended to have high retention rates ~80% regardless of whether they were engaged in actual fishing and or alcohol drinking. In this age-group however, the HIV incidence rate ranges between 2.45 and 3.57 per 100 pyar. For every combination of risk factors, HIV incidence was highest in age group of 25–29 years, ranging between 4.61 and 7.67/100 pyar. In this same age group, retention rates between 78.5 - 83.1% were observed. For every combination of risk factors, the lowest retention rate was among 18–24 year old participants.

Overall WTP was 89.1% (1953/2191) and was higher in men than women (91.2% vs 87.3%, p = 0.004) and among island communities relative to lakeshore ones (90.4% vs 85.8%, p = 0.004) (data not shown). We assessed the factors associated with lack of WTP (unwillingness to participate in future vaccine trials). As shown in Table 3, adjusted predictors of lack of WTP were sex, duration of stay in community and alcohol consumption. Lack of WTP was significantly higher among women than men [adj.OR = 1.51 (95% CI, 1.14- 2.00)] and among participants who had stayed in fishing communities for 10 or more years relative to those with less than one year [adj.OR = 1.78 (95% CI, 1.11 - 2.88)]. However, particpants who drink alcohol were less likely to lack WTP than non-drinkers [adj.OR = 0.74 (95% CI, 0.56 - 0.98)] i.e., alcohol drinkers were more willing to participate.