Routine, voluntary HIV testing in Durban, South Africa: correlates of HIV infection

Abstract

Background

Routine HIV testing is increasingly recommended in resource-limited settings. Our objective was to evaluate factors associated with a new diagnosis of HIV infection in a routine HIV testing programme in South Africa.

Methods

We established a routine HIV testing programme in an out-patient department in Durban, South Africa. All registered adults were offered a rapid HIV test; we surveyed a sample of tested patients.

Results

During the 12-week study, 1414 adults accepted HIV testing. Of those, 463 (32.7%) were HIV-infected. Seven hundred and twenty (50.9%) were surveyed. Compared with married women, unmarried men were at the highest risk of HIV [odds ratio (OR) 6.84; 95% confidence interval (CI) 3.45–23.55], followed by unmarried women (OR 5.90; 95% CI 3.25–10.70) and married men (OR 4.00; 95% CI 2.04–7.83). Age 30–39 years (compared with ≥50 years; OR 5.10; 95% CI 2.86–9.09), no prior HIV test (OR 1.45; 95% CI 1.07–2.27) and an imperfect HIV knowledge score (OR 2.32; 95% CI 1.24–4.35) were also associated with HIV infection.

Conclusion

In a routine HIV testing programme in South Africa, rates of previously undiagnosed HIV were highest among men, young and unmarried patients, and those with poorer HIV knowledge. Better interventions are needed to improve HIV knowledge and decrease HIV risk behaviour.

Introduction

Despite increasing access to antiretroviral therapy in sub-Saharan Africa, rates of HIV counselling and testing remain low. A 2005 US Agency for International Development (USAID) survey in 12 high-burden countries found that only 12% of men and 10% of women in the general population have both been tested for HIV and received their test results [1]. The World Health Organization and Joint United Nations Programme on HIV/AIDS issued guidelines in May 2007 promoting provider-initiated HIV testing in health facilities to promote earlier HIV diagnosis and improve access to services [2]. However, in both industrialized and resource-constrained settings, patients may have multiple contacts with the medical care system prior to undergoing HIV testing [3,4]. In generalized epidemics (defined as an HIV prevalence > 1% in pregnant women) at sites where a package of HIV care is locally available within the framework of a national health plan, healthcare providers are urged to recommend HIV counselling and testing to all adults and adolescents regardless of presenting signs and symptoms [2].

Provider-initiated testing has been successfully integrated into prenatal care programmes in multiple African countries, including Malawi, Zimbabwe, South Africa and Botswana [5-9]. Botswana began incorporating provider-initiated testing into a broad range of healthcare settings in 2004, with uptake as high as 89% and no apparent decline in healthcare facility use [10]. Demographic factors associated with self-reported HIV testing in a large population-based survey in Botswana included female gender, higher education, marital status and more frequent healthcare visits [11].

However, few data describe correlates of being HIV-infected in the setting of routine, provider-initiated HIV testing. Our objective was to evaluate factors associated with a new diagnosis of HIV infection, within a routine HIV testing intervention in an urban out-patient department in Durban, South Africa.

Methods

Study setting and participants

McCord Hospital is a semi-private hospital in Durban, South Africa. The out-patient department (OPD) serves a largely African Zulu-speaking clientele (~70%), with about 150–200 patients seen per day. Women comprise 65–70% of the OPD population. The majority of medical patients present episodically or for a single visit; antenatal and surgical patients are seated in a separate waiting area and were excluded from this study. At a separate site on the hospital grounds, patients who are HIV-infected undergo CD4 cell count testing and evaluation for entry into McCord’s antiretroviral treatment programme.

Routine HIV testing intervention

The routine testing intervention in this study has been described previously [12]. Briefly, we offered routine, voluntary HIV testing to all adult OPD patients (age ≥18 years) who spoke English or Zulu for a 12-week intervention period starting in January 2005. HIV tests were offered at no extra charge to all medical patients registered for care during daytime working hours (8 am–4 pm) in the OPD. Patients received a handout at registration in both English and Zulu informing them that they would be offered an HIV test during their visit. Trained bilingual HIV counsellors led educational sessions in the waiting room.

Consenting patients were taken to a private counselling room, where they received brief pretest counselling and underwent a rapid HIV 1/2 antibody test (Determine HIV-1/2 Assay; Abbott Laboratories, Abbott Park, IL, USA). Positive rapid tests were immediately confirmed with a second rapid test kit (Smart Check HIV-1/2 Assay; World Diagnostics, Miami Lakes, FL, USA) as per McCord Hospital protocol. Eligible patients were free to decline testing without any effect on their OPD care.

We also administered a survey to a convenience sample of approximately every other HIV-tested person. This sampling technique was chosen so that more counsellor time could be devoted to HIV testing rather than to conducting surveys. The surveys, available in both English and Zulu, included demographic information such as age, ethnicity, education, employment and responses to four HIV knowledge statements (adapted from Family Health International Surveys [13]). These statements were answered “yes”, “no” or “don’t know”. The four statements were: (1) All pregnant women infected with HIV will have babies born with AIDS; (2) A person with HIV can look and feel healthy; (3) There is a vaccine that can stop people from getting HIV; and (4) There are medicines available to help people with HIV/AIDS live longer. Based on distribution, we dichotomized knowledge question scores to less than perfect score (≤3 questions answered correctly) or perfect score (all four questions answered correctly).

The project was approved by the McCord Hospital Ethics Committee (Durban, South Africa) and the Partners Institutional Review Board (Boston, MA).

Statistical analysis

Baseline cohort characteristics were compared between HIV-infected and HIV-uninfected patients using the χ² test for categorical data and Student’s t-test for continuous variables. Univariate analysis with odds ratios (ORs) were used to examine correlates of testing HIV positive during the intervention period. Logistic regression was used to examine multivariate models of correlates of testing HIV positive and returning for a CD4 cell count within 3 months of the HIV test. In the model controlling for confounding factors associated with HIV positivity, interaction occurred between marital status and gender; for this reason, we separated men and women by marital status for further analyses. Associations were examined at P<0.05 significance level with a two-sided test. All survey data were double-entered into a Microsoft ACCESS^® Database. Statistical analyses were performed using SAS software (version 9.1; SAS Institute, Cary, NC).

Results

During the 12-week study period, 1414 of 2912 adult patients (48.6%) accepted HIV testing in the OPD. Of those tested, 463 [32.7%; 95% confidence interval (CI) 30.3–35.2%] were HIV-infected. One hundred and fifty of the newly identified HIV-infected patients (32.4%; 95% CI 28.1–36.7%) underwent CD4 cell count testing at the hospital HIV care site.

Among patients who accepted HIV testing, 766 of 1414 (54.2%) were surveyed. Thirty-two patients reported a previous positive HIV test and were excluded from subsequent analyses. Seven hundred and twenty patients had complete survey information and were included in the analysis; 193 were HIV-infected (26.8%; 95% CI 23.6–30.0%) and 73 of those infected underwent a CD4 cell count (37.8%; CI 31.0–44.7%). Surveyed patients did not differ significantly from the rest of the HIV-tested cohort with respect to gender, age, and HIV prevalence. Women comprised 64.0% of the surveyed patients and 58.9% of the rest of the cohort (P = 0.051). The mean age for surveyed patients was 40.8 years and the mean age of the rest of the tested cohort was 40.4 years (P = 0.63).

Sociodemographic data, HIV testing history, and knowledge score for the cohort are shown in Table 1. Marital status is separated by gender because of the interaction between gender and marital status found in the logistic regression model. Men and women were similar with respect to most characteristics; however, more men than women were married (55% vs. 47%; P = 0.045) and were employed (78.0% vs. 52.5%; P <0.0001). More women scored perfectly on the four questions of HIV knowledge (15.0% vs. 8.9%; P = 0.019).

Table 1.

Association between demographic and socioeconomic factors and HIV status^†

Variable	No. in category (% of surveyed)	No. HIV positive (% of each category)	Crude OR (95% CI)	P-value	Adjusted OR (95% CI)	P-value
Age (years)				<0.0001		<0.0001
18–29	144 (20)	44 (31)	3.04 (1.75–5.30)		2.43 (1.32–4.51)
30–39	188 (26)	77 (41)	4.67 (2.79–7.82)		5.10 (2.86–9.09)
40–49	178 (25)	44 (25)	2.29 (1.33–3.97)		2.75 (1.53–4.96)
≥50	210 (29)	28 (13)	1.00		1.00
Gender and marital status				<0.0001		<0.0001
Male, unmarried	116 (16)	50 (43)	8.20 (4.40–15.29)		6.84 (3.45–13.55)
Female, unmarried	240 (34)	86 (36)	5.89 (3.34–10.36)		5.90 (3.25–10.70)
Male, married	141 (20)	36 (26)	3.55 (1.89–6.70)		4.00 (2.04–7.83)
Female, Married	213 (30)	19 (9)	1.00		1.00
Education				0.1162		0.21
<High school	578 (81)	164 (28)	1.46 (0.91–2.34)		1.39 (0.83–2.32)
≥Matriculation course	128 (18)	26 (20)	1.00		1.00
Employment status				0.0208		0.79
Employed	438 (62)	128 (29)	1.53 (1.07–2.21)		0.95 (0.62–1.44)
Unemployed	272 (38)	62 (23)			1.00
Previous HIV testing				0.0196		0.003
Never tested	461 (65)	136 (30)	1.56 (1.07–2.27)		1.87 (1.24–2.83)
Ever tested	249 (35)	52 (21)	1.00		1.00
Knowledge score				0.0362		0.008
Less than perfect	92 (13)	17 (18)	1.87 (1.04–3.36)		2.32 (1.24–4.35)
Perfect *	628 (87)	176 (28)	1.00		1.00

‘Unmarried’ includes single, widowed, separated/divorced, and living with partner.

^*Perfect score: 4 of 4 knowledge questions correct; see ‘Methods’ section.

^†Denominators vary slightly because of missing data.

CI, confidence interval; OR, odds ratio.

Correlates of testing HIV positive during the intervention period in univariate analysis were age 18–29 years (OR 3.04 compared with ≥age 50 years; 95% CI 1.75–5.30), age 30–39 years (OR 4.67; 95% CI 2.79–7.82), age 40–49 years (OR 2.29; 95% CI 1.33–3.97), being employed (OR 1.53; 95% CI 1.07–2.21), no previous HIV test (OR 1.56; 95% CI 1.07–2.27) and an imperfect score on the four knowledge questions (OR 1.87; 95% CI 1.04–3.36) (Table 1). We assessed marital status stratified by gender. Being an unmarried man (OR 8.20; 95% CI 4.40–15.3), an unmarried woman (OR 5.89; 95% CI 3.34–10.4), and a married man (OR 3.55, 95% CI 1.89–6.70) were each associated with HIV infection compared with being a married woman (reference group). In univariate analysis, we did not find a significant association with HIV infection for education level, or for knowing someone living with HIV infection or being treated for HIV.

We used logistic regression to better determine the effect of these covariates on the likelihood of being HIV infected. Multivariate analysis indicated that those at highest risk of infection were those aged 30–39 years (OR 5.10 compared with those ≥50 years; 95% CI 2.86–9.09) and unmarried men (Table 1). Compared with married women, unmarried men were at the highest risk of testing positive for HIV (OR 6.84; 95% CI 3.45–23.55), followed by unmarried women (OR 5.90; 95% CI 3.25–10.70) and married men (OR 4.00; 95% CI 2.04–7.83). No prior HIV test (OR 1.45; 95% CI 1.07–2.27) and a less than perfect score on the HIV knowledge questions (OR 2.32; 95% CI 1.24–4.35) were also independently associated with testing HIV positive (Table 1). Employment was the only correlate that was no longer significantly associated with HIV infection in the adjusted model of HIV infection.

A logistic regression model with the outcome of returning for a CD4 cell count within 3 months of the HIV test among those who tested HIV-infected considered the same covariates as in the first model. In this adjusted model with returning for CD4 cell count as the outcome, being employed was the only significant predictor of returning for a CD4 cell count (OR 2.20; 95% CI 1.03–4.70).

Discussion

During a 12-week routine HIV testing intervention in an OPD in Durban, South Africa, nearly half of patients accepted HIV testing; the prevalence of HIV was 32.7%. A detailed survey of 720 test acceptors revealed that those at highest risk for a new diagnosis of HIV were unmarried men, patients aged 30–39 years, those who had not been tested before, and those with a low HIV knowledge score.

We found an interaction between gender and marriage as correlates of a new HIV diagnosis. Unmarried men were at the highest risk, nearly seven times that of married women, with unmarried women having nearly six times the risk of that of married women. In the current study, married people seem both more willing to be tested and more likely to be HIV negative. For provider-initiated testing in Botswana and traditional voluntary counselling and testing (VCT) in rural Uganda, HIV test acceptance was higher among married individuals [11,14]. In a country-wide report from the Ministry of Health of Botswana, uptake of provider-initiated testing differed by gender; for example, uptake for women aged 15–19 years was 89.4% vs. 75.6% for men [10]. We found that unmarried men were most likely to be HIV-infected based on multivariate analysis, yet in other studies this is the group that were least likely to accept HIV testing [10,11,14].

We also found that a significantly lower proportion of men had a perfect HIV knowledge score compared with women (8.9% vs. 15.0%; P =0.019). The role of knowledge about HIV as predictive of accepting HIV testing has been described previously in Africa [6,11,15,16]. In rural Zimbabwe, knowledge of HIV was associated with VCT uptake in an adult cohort, and basic knowledge about prevention of mother-to-child transmission predicted uptake of routine HIV testing in antenatal clinics [6,16]. Similar findings were noted in studies of HIV testing uptake in both Botswana and Nigeria [11,15]. In our study of HIV test acceptors, patients with a high knowledge score were less likely to be HIV-infected, suggesting that HIV knowledge not only predicts willingness to accept testing, but also may be protective against acquiring HIV infection. In the African studies noted above, those with higher educational status were more likely to accept testing; in the current study, education level did not correlate with HIV test results [6,11,15,16]. Thus, educational interventions targeting the general population to increase knowledge about HIV transmission and infection may prevent new infections, regardless of formal educational status. In the USA, low health literacy has been associated with HIV test acceptance of provider-initiated testing, but also with lower CD4 and higher viral loads among people living with HIV [17,18].

We found that younger patients, particularly aged 30–39 years, were at high risk of having a new diagnosis of HIV during the routine testing intervention. This finding is consistent with the South African national antenatal survey which has documented an increasing prevalence of HIV among women in this age group over the last several years [19]. In Botswana, the number of women tested in their 30s was lower than in the 20–29 year age group; however, this probably reflects an increased offer of testing for childbearing women attending antenatal clinics [10]. Age in the 30s does not appear to be a strong predictor of declining HIV testing in Botswana, South Africa or Zimbabwe [10,11,16,20], but this is a high-prevalence group which deserves particular attention when designing testing interventions.

Fewer than one-third of those found to be HIV-infected through routine HIV testing in this study underwent CD4 cell count testing at the HIV care site of the hospital within 3 months, suggesting a failure to follow-up, link to care and receive therapy. The only significant predictor of returning for a CD4 cell count was being employed. The low rate of undergoing a CD4 cell count, and the fact that patients with a salary were more likely to obtain a CD4 cell count, may both reflect the fact that patients were required to pay a fee for CD4 testing (80 Rand; ~US$11.00 in 2005). This hypothesis is supported by studies on several large cohorts in resource-limited settings that have shown that access to free HIV care correlates with obtaining a baseline CD4 cell count [21]as well as retention in care at 6 months [21,22]. Promoting linkage to HIV care through the offer of free baseline CD4 testing may improve access of HIV services in our setting. After the results of this study became available, McCord Hospital made CD4 cell counts available free of charge.

This analysis has several limitations. The surveyed population represents only half of the patients who accepted HIV testing during the study; this sample may not be fully representative despite efforts to survey every second patient to minimize bias. However, the patients surveyed did not differ significantly from tested patients overall with respect to age and gender. The decision not to survey all participants was made to allow the limited staff to focus their energies on counselling, education and HIV testing itself. The knowledge questions in the survey represent only a subset of a longer validated instrument [13]; the subset was chosen based on its relevance to the study setting, but it has not been independently validated. In addition, the McCord patients pay a fee to receive care in the OPD, which may affect the generalizability of the results to fully government-subsidized hospitals in the South African public sector, where care is free of charge.

Nearly half of patients offered routine HIV testing in a high-prevalence OPD in South Africa accepted testing. Among surveyed patients, those with the highest rates of previously undiagnosed HIV included men, young and unmarried patients, those with no prior history of HIV testing, and those with poorer HIV knowledge. Unmarried men may be least likely to accept testing, but in our study this was the group with the highest risk of HIV infection; further studies to increase testing uptake should be directed at this group. Better interventions are needed to improve HIV knowledge and decrease HIV risk behaviour, particularly among men. Provision of CD4 cell counts free of charge to patients may improve linkage to care among patients newly diagnosed with HIV.