Introduction
More HIV-positive people live in South Africa than in any other country in Africa (5.5 million children and adults in 2005) [1]. In order to estimate the current need for prevention, and future need for antiretroviral treatment (ART), it is important to know HIV incidence [2]. To date, estimates of HIV incidence in South Africa have come from two sources: changes in prevalence over time [3] and laboratory tests to detect recent HIV infections [4]. Both sources of HIV incidence information have major limitations. Incidence models rely on assumptions about mortality among HIV-positive people, which is difficult to measure and changes quickly as ART is scaled up [5–7]. Immunoassays to measure recent HIV infections may not produce valid results, especially in settings with high HIV prevalence [8,9]. In contrast, prospective population-based cohort studies are widely recognized as the gold standard to measure HIV incidence because they directly measure individual HIV seroconversion [10]. Only a few such studies have been conducted in Africa, none of them in South Africa [11,12].
This study reports for the first time HIV incidence rates from a large prospective population-based HIV survey in rural KwaZulu-Natal, South Africa. While prospective population-based studies have the advantage over cross-sectional surveys in that they directly measure individual HIV incidence, they may be more likely to suffer from selection biases because some people are willing to test for HIV once but not repeatedly [13,14]. Multiple imputation (MI) was used to adjust for selection effects on HIV incidence estimates.
Methods
HIV serosurvey
Data were used from the first round (June 2003 to December 2004) and the second round (January to December 2005) of a prospective population-based HIV survey and a household socioeconomic survey conducted by the Africa Centre for Health and Population Studies, University of KwaZulu-Natal, South African (Africa Centre). The surveys took place in a demographic surveillance area (435 km2, resident population approximately 86 000 Zulu-speaking people) in the rural district of Umkhanyakude in northern KwaZulu-Natal. Details of the administration of the Africa Centre HIV serosurveys have been described elsewhere [15]. Ethics permission for the HIV surveys and the Africa Centre Demographic Surveillance System was obtained from the Research Ethics Committee at the College of Health Sciences, University of KwaZulu-Natal.
Sample
Inclusion criteria were age eligibility in both rounds of the HIV serosurvey (women aged 15–49 and men aged 15–54 years) and residency in the demographic surveillance area during the first round of the HIV survey. In addition, migrants (i.e., individuals who changed from being a resident to being a nonresident member of a household in the demographic surveillance area between the first and the second survey round) were included. Exclusion criteria included mental disability, severe illness, and physical handicap precluding informed consent. Of the 17 861 individuals eligible for HIV testing in both the first and second round of the HIV survey, 84% and 98% were contacted in the first and the second survey round, respectively. Amongst those successfully contacted, 59% (8882 individuals) and 39% (6872 individuals) consented to an HIV test in the first and the second round, respectively. At baseline, the average HIV prevalence was 22.3 [95% confidence interval (CI), 21.2–23.4] in women and 11.5 (95% CI, 10.5–12.6) in men.
To be included in the HIV incidence analysis, individuals had to have tested negative for HIV antibody during the first round of the HIV survey and tested either HIV negative or HIV positive during the second round. Of the 7293 individuals who tested HIV negative in the first round, 57% (4046 individuals) participated in the second round of the survey. Table 1a in the Appendix (available online) compares eligible participants and nonparticipants in each of the two survey rounds by demographic characteristics (age, sex, migration status), behavioural factors (marital status, self-assessed previous and current HIV risk, previous voluntary testing and counselling), socioeconomic characteristics (urban versus periurban/rural residence, employment status, self-assessed financial status) and geographic characteristics (distance from the individual's household to the nearest health clinic and to the nearest primary road). Table 2a in the Appendix compares the individuals who tested HIV negative in the first round and tested again in the second round with those who did not test again after a negative HIV test in the first round (i.e., were lost to follow-up).
Statistical analysis
The date of HIV seroconversion was assumed to occur midway between the date of the first and second HIV test. Survival analyses in which midpoint imputation is used yield unbiased results if the interval between the two tests is less than 2 years [16]. The median of the interval between the first and second test among seroconverters in our sample was 516 days; more than 93% of the test–retest intervals were < 2 years.
MI was used to adjust for selection effects using the variables described above (Appendix, Table 1a) [17]. MI adjusts for bias if selection is on observed variables. However, in many realistic cases it provides robust results even if selection on unobservable variables has occurred [18–23]. Imputation by chained equations was carried out in STATA 10.0 (Stata Corporation, College Station, Texas, USA) with five imputations and 10 iterations [24,25]. (See the Appendix for further details on the MI implemented here.) All individuals who had a negative HIV result (measured or imputed) in the first round of the survey and either a positive or a negative HIV result (measured or imputed) in the second round were eligible for the analyses using the imputed datasets. Weibull hazard regressions were carried out on each of the five imputed datasets in order to investigate risk factors for HIV incidence (Table 3a in the Appendix). Coefficient and standard errors were calculated following Rubin [17].
Results
Among eligible individuals who consented to an HIV test in both survey rounds, 170 individuals seroconverted during 5253 person-years at risk. The crude HIV incidence rates per 100 person-years were 3.8 (95% CI, 3.2–4.6) in women aged 15–49 years and 2.3 (95% CI, 1.8–3.1) in men aged 15–54 years. In the five imputed datasets, 14 583–14 729 individuals were eligible for HIV incidence analysis. In comparison with the people eligible for HIV incidence analysis before MI, average age increased in the younger five-year age groups (while the prevalence of some factors positively associated with HIV incidence declined) and the prevalence of a range of factors positively associated with HIV incidence increased in the older age groups.
Table 1 shows crude HIV incidence rates and MI-adjusted rates by sex and 5-year age group as well as by the other variables used in MI. The imputed HIV incidence rate was 2.08 and 2.22 times higher than the nonimputed rate in women and men, respectively. In both men and women, the absolute increase in HIV incidence was larger in the 5-year age groups above 29 years of age than in younger age groups. While the highest HIV incidence in women occurred in the age group 25–29 years, both without and with MI, in men MI shifted the highest HIV incidence from the age group 25–29 to the age group 30–34 years (Fig. 1).
Table 1: Crude and multiple imputation-adjusted HIV incidence rates by demographic, behavioural, socioeconomic and geographic characteristics.
Fig. 1: HIV incidence rates by 5-year age groups [crude rates and adjusted for multiple imputation (MI)] in the Africa Centre HIV survey, 2003–2005. (a) Rates for women; (b) rates for men.
Weibull multiple regression analysis showed that the hazard of seroconversion was 53% higher among women than among men (P < 0.001), approximately twice as high among people who were currently unmarried but had a partner than for individuals who were married (P < 0.001) and 24% lower among people who stated that they were ‘just getting by’ than in people who stated that they were ‘extremely poor’ (P = 0.022). The hazard of seroconversion increased with the logarithmically transformed distance between the individual's household and the nearest government health clinic (adjusted hazard ratio, 1.174; P = 0.051) and decreased with the logarithmically transformed distance to the nearest primary road (adjusted hazard ratio, 0.856, P = 0.002) (Table 3a in the Appendix).
Discussion
We report for the first time directly measured HIV incidence rates from a prospective population-based HIV survey in rural KwaZulu-Natal, South Africa. The crude HIV incidence rate in the survey community was high both in women and men. Consent rates to repeated HIV testing were comparatively low among the eligible population. We used MI to adjust for selection biases owing to selection on observable variables into the sample for HIV incidence analysis. MI significantly increased the already high HIV incidence rates both in women (to 7.9/100 person-years; 95% CI, 7.4–8.4) and men (to 5.1/100 person-years; 95% CI, 4.1–6.2) because population groups with higher risk of HIV incidence were less likely to have tested for HIV in the two survey rounds than groups with lower risk of HIV incidence.
The very high HIV incidence in the province of South Africa that currently has the highest HIV prevalence in the country emphasizes the acute need for HIV prevention. Examples of interventions that might be effective in diminishing the spread of HIV in KwaZulu-Natal include school-based HIV prevention education, condom promotion or treatment of sexually transmitted diseases [26–28].
Holding other factors constant in multiple regression, women faced a higher hazard of HIV seroconversion than men. However, in the population as a whole, men, especially after 25 years of age, are at very high risk of acquiring HIV as well. Therefore, a one-sided focus of HIV prevention efforts on women does not seem reasonable in rural KwaZulu-Natal. We further find that, while HIV incidence peaks in the age groups 25–29 years in women and 30–34 years in men, HIV incidence remains high in older ages, suggesting that HIV prevention efforts need to include the middle-aged and old in addition to youth. Holding other factors equal, individuals who were not married but had a partner faced more than double the hazard of HIV acquisition than married people or people without partners. Finally, our results suggest that HIV prevention interventions may be relatively more effective if they are geographically targeted at people who live close to primary roads, on the one hand, and far from government health clinics, on the other hand.
The 2005 South African National HIV Prevalence, HIV Incidence, Behaviour and Communication Survey reported an HIV incidence estimate of 3.8/100 person years (95% CI, 0.1–7.5) in the population aged 2 years and above, based on an application of the BED capture assay in a cross-sectional sample of the population of KwaZulu-Natal [4]. While our overall HIV incidence estimate cannot be directly compared with the above estimate for methodological reasons [8] and differences in the eligible populations, it is of interest to note that the national survey finds that HIV incidence peaked in the age group 20–29 in women and in the age group 30–39 in men, and that it identified single and widowed individuals as at higher risk of acquiring HIV than married people [29], supporting the similar findings in this study.
In summary, we provide evidence from the first prospective population-based HIV survey in South Africa that HIV incidence in a rural community with high HIV prevalence is currently very high in both men and women. Without a renewed emphasis on HIV prevention, it seems unlikely that the HIV epidemic in rural South Africa will be conquered.
Acknowledgements
We thank the field staff at the Africa Centre for Health and Population Studies at the University of KwaZulu-Natal, South Africa, for their work in collecting the data used in this study and the communities in the Africa Centre demographic surveillance area for their support and participation in this study. Methods used in this analysis were devised and tested at training workshops run by the Wellcome Trust-funded ALPHA network.
Sponsorship: This study was supported by grants from the Wellcome Trust through grants to the Africa Centre for Health and Population Studies, University of KwaZulu-Natal, South Africa (GR065377/Z/01/B), and the Africa Centre's Demographic Surveillance Information System (GR065377/Z/01/H), which funds household and HIV surveillance. The funding organization had no role in the design and conduct of the study, in the collection, analysis, and interpretation of the data, or in the preparation, review or approval of the manuscript.
References
1. UNAIDS.
2006 Report on the Global AIDS Epidemic. Geneva: Joint United Nations Programme on HIV/AIDS; 2006.
2. Hallett TB, White PJ, Garnett GP. The appropriate evaluation of HIV prevention interventions: from experiment to full scale implementation. Sex Transm Infect 2007; 83(Suppl I):55–60.
3. Williams B, Gouws E, Wilkinson D, Karim SA. Estimating HIV incidence rates from age prevalence data in epidemic situations. Stat Med 2001; 20:2003–2016.
4. Shisana O, Rehle T, Simbayi LC, Parker W, Zuma K, Bhana A,
et al. South African National HIV prevalence, HIV incidence, behaviour and communication survey. Cape Town: HSRC Press; 2005.
5. Ghys PD, Kufa E, George MV. Measuring trends in prevalence and incidence of HIV infection in countries with generalised epidemics. Sex Transm Infect 2006; 82(Suppl 1):i52–56.
6. Lohse N, Hansen AB, Pedersen G, Kronborg G, Gerstoft J, Sorensen HT,
et al. Survival of persons with and without HIV infection in Denmark, 1995–2005. Ann Intern Med 2007; 146:87–95.
7. Salomon JA, Hogan DR, Stover J, Stanecki KA, Walker N, Ghys PD, Schwartlander B. Integrating HIV prevention and treatment: from slogans to impact. PLoS Med 2005; 2:e16.
8. Center for Disease Control and Prevention.
Interim recommendations for the use of the BED capture enzyme immunoassay for incidence estimation and surveillance. Atlanta, GA: Center for Disease Control and Prevention, Office of the Global AIDS Coordinator; 2006.
9. UNAIDS Reference Group on Estimates, Modelling and Projections.
Statement on the use of the BED assay for the estimation of HIV-1 incidence for surveillance or epidemic monitoring.
Wkly Epidemiol Rec 2006;
81:40.
10. Sakarovitch C, Alioum A, Ekouevi DK, Msellati P, Leroy V, Dabis F. Estimating incidence of HIV infection in childbearing age African women using serial prevalence data from antenatal clinics. Stat Med 2007; 26:320–335.
11. Gray RH, Li X, Kigozi G, Serwadda D, Brahmbhatt H, Wabwire-Mangen F,
et al. Increased risk of incident HIV during pregnancy in Rakai, Uganda: a prospective study. Lancet 2005; 366:1182–1188.
12. Kapiga SH, Lyamuya EF, Lwihula GK, Hunter DJ. The incidence of HIV infection among women using family planning methods in Dar es Salaam, Tanzania. Aids 1998; 12:75–84.
13. McDougal JS, Parekh BS, Peterson ML, Branson BM, Dobbs T, Ackers M, Gurwith M. Comparison of HIV type 1 incidence observed during longitudinal follow-up with incidence estimated by cross-sectional analysis using the BED capture enzyme immunoassay. AIDS Res Hum Retroviruses 2006; 22:945–952.
14. McDougal JS, Pilcher CD, Parekh BS, Gershy-Damet G, Branson BM, Marsh K, Wiktor SZ. Surveillance for HIV-1 incidence using tests for recent infection in resource-constrained countries. AIDS 2005; 19(Suppl 2):S25–S30.
15. Welz T, Hosegood V, Jaffar S, Batzing-Feigenbaum J, Herbst K, Newell ML. Continued very high prevalence of HIV infection in rural KwaZulu-Natal, South Africa: a population-based longitudinal study. AIDS 2007; 21:1467–1472.
16. Law CG, Brookmeyer R. Effects of mid-point imputation on the analysis of doubly censored data. Stat Med 1992; 11:1569–1578.
17. Rubin DB. Multiple imputation for nonresponse in surveys. New York: Wiley; 1987.
18. Zhou XH, Eckert GJ, Tierney WM. Multiple imputation in public health research. Stat Med 2001; 20:1541–1549.
19. Kessler RC, Little RJ, Groves RM. Advances in strategies for minimizing and adjusting for survey nonresponse. Epidemiol Rev 1995; 17:192–204.
20. Crawford SL, Tennstedt SL, McKinlay JB. A comparison of analytic methods for nonrandom missingness of outcome data. J Clin Epidemiol 1995; 48:209–219.
21. Greenland S, Finkle WD. A critical look at methods for handling missing covariates in epidemiologic regression analyses. Am J Epidemiol 1995; 142:1255–1264.
22. Heitjan DF. Annotation: what can be done about missing data? Approaches to imputation. Am J Public Health 1997; 87:548–550.
23. Collins L, Schafer JL, Kam CM. A comparison of inclusive and restrictive strategies in modern missing data procedures. Psychol Meth 2001; 6:330–351.
24. van Buuren S, Boshuizen HC, Knook DL. Multiple imputation of missing blood pressure covariates in survival analysis. Stat Med 1999; 18:681–694.
25. Royston P. Multiple imputation of missing values: update of ICE. STATA J 2005; 5:527–536.
26. Nyindo M. Complementary factors contributing to the rapid spread of HIV-I in sub-Saharan Africa: a review. East Afr Med J 2005; 82:40–46.
27. Wegbreit J, Bertozzi S, DeMaria LM, Padian NS. Effectiveness of HIV prevention strategies in resource-poor countries: tailoring the intervention to the context. AIDS 2006; 20:1217–1235.
28. Auerbach JD, Hayes RJ, Kandathil SM. Overview of effective and promising interventions to prevent HIV infection. World Health Organ Tech Rep Ser 2006; 938:43–78, discussion 317–341.
29. Rehle T, Shisana O, Pillay V, Zuma K, Puren A, Parker W. National HIV incidence measures: new insights into the South African epidemic. S Afr Med J 2007; 97:194–199.