Hepatitis and tuberculosis testing are much less common than HIV testing among adults in Kisumu, Kenya: results from a cross-sectional assessment

These analyses were conducted using data from the RV393 prospective observational cohort in the Kisumu region of Kenya, which was conducted to determine HIV incidence and to inform HIV vaccine development. Male and female participants were eligible for enrollment if they were at risk for HIV, were aged 18–35 years, and reported sex with at least two partners in the preceding three months. These criteria were developed based on prior studies from the region demonstrating high HIV incidence among young adults with multiple sexual partners [22‐24]. A small group of PLWH who satisfied other inclusion criteria were enrolled as a comparator group and to mask the HIV status of participants due to substantial HIV-related stigma in the community. Participants were excluded from the study if they had previously received investigational agents that could change HIV risk such as monoclonal antibodies or candidate HIV vaccines. Study participants were recruited via community outreach and mobilization activities facilitated by community-based organizations, local ministry of health offices, and community leaders. Targeted recruitment was conducted at locations known to be focal points for activities associated with HIV risk including bars, markets, and fishing villages as well as healthcare settings such as HIV testing centers and family planning clinics.

All participants provided written or witnessed informed consent in English, Kiswahili or Luo prior to enrollment. The study protocol was approved by institutional review boards at the Walter Reed Army Institute of Research, Silver Spring, Maryland (WRAIR #2290) and the Kenya Medical Research Institute, Nairobi, Kenya (SSC 3307).

Data for these cross-sectional analyses were collected over screening and enrollment visits approximately 28 days apart. A complete medical history was taken and physical examination was performed. Participants completed comprehensive questionnaires that included questions about sexual and other HIV risk behaviors as well as questions about prior testing. In three separate questions, participants were asked if they had ever been tested for HIV, hepatitis and tuberculosis. If the answer to a question was affirmative, participants were asked when the last test for that infection was performed with answer choices that included less than 6 months ago, between 6 and 12 months ago or more than 1 y ago. They were also asked if they had ever been treated or were currently being treated for HIV, viral hepatitis, and tuberculosis. A social history was taken that included documentation of any history of alcohol abuse, defined as having more than 5 alcoholic drinks in 1 d and/or causing harm while intoxicated. Participants were screened for symptoms of possible tuberculosis including cough, fever, weight loss, chest pain or breathlessness, or night sweats of greater than 2 weeks.

At study entry, all participants underwent rapid HIV testing with First Response® Rapid HIV 1–2-0 card tests (Premier Medical Corporation, Maharashtra, India). Discordant rapid tests were resolved with HIV ELISA via the GenScreen ULTRA HIV Ag-Ab Combo Kit and HIV was confirmed with the Geenius HIV-1/2 Supplemental Assay (Bio-Rad Laboratories, Hercules, CA). Hepatitis B surface antigen was detected using the GS HBsAg EIA 3.0 qualitative immunoassay (Bio-Rad Laboratories, Hercules, CA). Hepatitis C antibody testing was performed using the Ortho HCV version 3.0 ELISA Test System (Bio-Rad Laboratories, Hercules, CA). All testing was performed according to package inserts.

Enrollment characteristics of the study population were summarized using descriptive statistics including medians with interquartile ranges (IQRs) for continuous variables and frequencies with percentages for categorical variables. Age was categorized as ≤22, 23–29, or ≥ 30 years based on preliminary data exploration to create relatively balanced groups and also align roughly with life-event stages. Because HIV testing is recommended at least yearly individuals at risk for HIV, our primary outcome of interest was a dichotomized variable for prior testing for HIV based on whether or not it had been conducted within the year prior to enrollment. PLWH who were aware of their HIV seropositivity prior to enrollment were excluded from analyses evaluating this outcome, since routine HIV testing would not be indicated in such participants, but were included in analyses of hepatitis and tuberculosis testing. Chi-squared or Fisher’s exact tests were used to evaluate differences between categorical groups of interest. We employed a robust Poisson model with a log link using a general linear modeling framework to estimate prevalence ratios (PRs) and 95% confidence intervals (CIs) for factors potentially associated with HIV testing in the year prior to enrollment. Participants with missing covariate data were excluded from analyses using listwise deletion as the total missingness constituted about 1% of participants. Variables that had a significance of p < 0.2 in unadjusted models were included in the multivariable model. This p-value threshold was used for down-selection of variables in order to liberally allow potential confounders into the multivariable model, reduce bias, and minimize type II error as compared to a more stringent selection criterion such as the p < 0.05 typically used to define statistical significance. Collinearity of variables included in the multivariable model was assessed via multiple means, including calculation of variance inflation factors for all variables modeled, calculation of the condition index values for the tolerance test, and examination of the Pearson correlation matrix. Lastly, in order to assess whether prior testing was associated with aspects of differential access due to spatial location, we constructed a 2-tier hierarchical model with level 1 as the individual respondent and level 2 as the residential ward. Residential wards were aggregated into 12 groups based on expert knowledge from in-country staff. Due to the number of geographic categories necessary to consider, the model was limited in the number of covariates that could be included as independent variables. Additionally, the comparison group for self-assessed HIV risk in the hierarchical model was modified to be the ‘some risk’ category as using the ‘no risk’ group resulted in model errors. All analyses were conducted in SAS 9.4; the hierarchical model was constructed using the GLMMIX environment.

From January 2017 to May 2018, 671 participants were enrolled in the study, including 52 (7.7%) PLWH and 619 (92.3%) without HIV. Their median age was 24 (IQR 21–28) years, 308 (45.9%) were female, and 508 (75.7%) were unmarried. The main occupations reported were sex worker (n = 150, 22.4%), fisherman (n = 119, 17.7%), business trader (n = 99, 14.7%), and construction (n = 41, 6.1%). PLWH tended to be older than participants without HIV (median 29 [IQR 27–30] vs 24 [IQR 21–28] years, p < 0.0001). There was a marginally higher proportion of females among PLWH as compared to participants without HIV (57.7% vs 44.9%, p = 0.08) but a smaller proportion of PLWH reported being married (42.3% vs 22.8%, p = 0.0016). The distribution of occupations did not differ significantly between participants with and without HIV (p = 0.36).

Any lifetime history of testing for HIV was reported by 651 (97.0%) participants, of whom 400 (61.2%) had been tested less than 6 months prior to enrollment, 129 (19.7%) had been tested 6–12 months prior to enrollment, and 125 (19.1%) had been tested more than 1 y prior to enrollment. Twenty-seven of these participants had a previous positive test result and were known to be living with HIV. An additional 24 participants were newly diagnosed with HIV upon enrollment into the cohort. Among participants without previously-diagnosed HIV, testing for HIV within the last year was more common among those who were female and had a secondary or higher level of education (Table 1). HIV testing was more common among participants in the “bar/pub/waitress” category than among other occupation categories.

In contrast to HIV testing, prior testing for viral hepatitis and tuberculosis was uncommon. Any lifetime history of testing for viral hepatitis was reported by only 8 (1.2%) participants, of whom 1 (12.5%) had been tested less than 6 months prior to enrollment, 3 (37.5%) had been tested 6–12 months prior to enrollment, and 4 (50.0%) had been tested more than 1 y prior to enrollment. All participants with a history of testing for viral hepatitis were in the subgroup without HIV; none of the 52 PLWH enrolled in the study had ever been tested for viral hepatitis. At study entry, 28 (4.5%) participants without HIV had a positive hepatitis B surface antigen test and 21 (3.4%) had a reactive hepatitis C antibody, while 2 (3.9%) PLWH had a positive hepatitis B surface antigen test and none had a reactive hepatitis C antibody. No participants with a positive hepatitis test at study entry had any lifetime history of prior hepatitis testing.

For tuberculosis, 51 (7.6%) participants reported any lifetime history of testing, of whom 8 (15.7%) had been tested less than 6 months prior to enrollment, 9 (17.6%) had been tested 6–12 months prior to enrollment, and 34 (66.7%) had been tested more than 1 y prior to enrollment. A significantly higher proportion of PLWH participants reported any lifetime history of tuberculosis testing than was observed among participants without HIV (21.2% vs. 6.5%, Fisher p < 0.001). Participants with higher education were more likely to have been tested for viral hepatitis in the year prior to enrollment. There were no significant differences in characteristics between participants with and without a history of testing for tuberculosis in the year prior to enrollment (Table 1).

Among participants without previously-diagnosed HIV, factors associated with prior testing for HIV within the year prior to enrollment were evaluated using robust Poisson regression models. Because prior testing for viral hepatitis and tuberculosis were each rare, factors associated with prior testing could not be evaluated using regression models. In unadjusted models, HIV testing in the past year was more common among females (PR 1.08 [95% CI 1.01, 1.17]) and participants with secondary education or higher (PR 1.10 [95% CI 1.02, 1.19]; Table 2). In the multivariable model, only secondary education or higher was associated with recent HIV testing (adjusted PR 1.10 [95% CI 1.02, 1.20]) after controlling for sex, alcohol use, and occupation. Occupations often associated with HIV risk in the region due to engagement in transactional sex were not associated with prior HIV testing in the multivariable model. No collinearity was observed among variables included in the multivariable model. Variance inflation factors for all variables modeled were below 1.2. Condition index values for the tolerance test were we well below the accepted cutoff of 30, with a maximum observed condition index of 7. Examination of the Pearson correlation matrix identified that no values were problematically correlated.

In the spatial location hierarchical models, Kisumu county was divided based on 12 separate sub-counties; participant location assignment was determined by self-reported household location (Fig. 1). Due to the number of distinct geographic groups, the model was limited in the number of independent variables that could be included in the model before convergence failure; for most combinations the model was limited to 2 independent variables. Numerous combinations of variables from the unadjusted and adjusted analysis were examined (data not shown), but there were no meaningful differences in the location effect on the models. None of the sub-county locations were significantly associated with prior testing uptake for HIV. As in the Poisson model, increased education level was associated with increased likelihood of recent HIV testing in the hierarchical model (PR 1.10 [95% CI 1.02, 1.20]; Table 3).