Data on HIV prevalence trends are used to monitor the epidemic. Careful interpretation is essential, as prevalence is influenced not only by recent HIV incidence but also by past HIV incidence, AIDS-related mortality and migration . Recognizing these limitations, it appears that the HIV epidemic is now stable (although at unacceptably high levels) or declining in many sub-Saharan Africa countries, including Tanzania [2,3]. However, HIV epidemic dynamics may differ between individuals with different levels of education or socioeconomic position (SEP).
In many African settings, access to school education is influenced by socioeconomic factors, and educational attainment is the most widely available marker of SEP in HIV prevalence data sets. Many studies have shown, perhaps unexpectedly, that HIV infections were more common among individuals of both sexes with higher levels of education . A similar association has also been shown between greater household asset–wealth and HIV prevalence . Wealthy, educated individuals are often the most mobile in a population and may have had wider sexual networks earlier in the epidemic. However, new HIV infections may now be concentrating among less-educated individuals over time . Evidence to support this hypothesis comes from studies of HIV prevalence [7–10] and incident HIV infections  in selected areas and population groups (including in Tanzania ) and systematic reviews of these studies [4,6]. However, nationally representative HIV prevalence data have not previously been available from any country to examine this.
Data on educational attainment and HIV prevalence have recently become available from two nationally representative, population-based HIV serosurveys conducted in Tanzania in 2003–2004 and 2007–2008. We used these data to explore the hypothesis that HIV infections are concentrating among those with the lowest levels of education in this sub-Saharan African setting.
We downloaded data from the AIDS Indicator Surveys (AIS) conducted in Tanzania in 2003–2004 and 2007–2008 from www.measuredhs.com. These surveys use a two-stage sampling procedure to identify a large, nationally representative sample of adults aged 15–49 years and collected data on sociodemographic characteristics of research participants. Both surveys achieved high response rates (RR) . A dried blood-spot sample was also collected from consenting participants in both surveys. These were tested for the presence of HIV antibodies using an ELISA. All positive and 10% of negative tests were confirmed with a second ELISA test.
We restricted our analysis to individuals for whom full data were available on a small set of following variables relevant to our purpose: sex, age, urban/rural residence, educational attainment, household asset–wealth and prevalent HIV infection. All of our analyses used the ‘survey’ commands in Stata v.10 (Stata Corp., College Station, Texas, USA) to reflect the sampling strategy and ensure that our results would be representative of the national population structure.
First, we described the characteristics of the research participants in the 2003–2004 and 2007–2008 surveys. Second, we described changes in HIV prevalence between 2003–2004 and 2007–2008 in urban and rural settings and after stratification by the other characteristics. Third, we explored the association between educational attainment and prevalent HIV infection in both surveys. We initially stratified this analysis on the basis of sex and age group (15–24 and 25–49 years). HIV risk-factor analyses are often conducted for men and women separately because epidemic dynamics often differ between the sexes. Age-group stratification is useful because HIV incidence is high after 15 years of age but is then a life-long infection. Consequently, a higher proportion of infections are likely to have been recently acquired among 15–24-year-olds than among 25–49-year-olds and, therefore, changes in HIV prevalence among this group may give a better indication of recent HIV incidence patterns. We used logistic regression to calculate odds ratios (ORs) describing the association between educational attainment and prevalent HIV infection after adjustment for the potential confounding influence of age (coded into population terciles within our larger age groups), urban/rural residence and household wealth. We adjusted for residence status because access to education and HIV prevalence differed between urban and rural areas but we did not think it plausible that residence status was on the causal pathway between educational attainment and HIV status. We adjusted for household wealth because we wanted to assess whether the effect of educational attainment was independent of any influence of background household wealth. Although education and household wealth can both be considered domains of SEP, educational attainment was our preferred exposure variable for this analysis, we return to this issue in our Discussion section. We used Wald tests to explore whether educational attainment was an important explanatory factor in our adjusted models separately for the 2003–2004 and 2007–2008 surveys. Following stratum-specific analyses, we also tested for evidence of an interaction between educational attainment and a combined age–sex variable with four categories. Because we found some evidence to support the existence of an interaction in the data, we did not conduct pooled analysis.
Finally, we pooled the data from both surveys to examine changes in HIV prevalence between 2003–2004 and 2007–2008 among participants of different levels of education. For each age–sex stratum, we specified an adjusted logistic regression model with HIV infection as the dependent outcome variable and independent terms for our other variables. We included an interaction term between year of survey and level of educational attainment. We report the results of these models and the findings of Wald tests assessing the statistical significance of the interaction terms. We judged that there was some evidence of interaction between education and year of survey but little evidence that the strength of this interaction differed by age or sex. We, therefore, conducted a final analysis including all research participants to provide an overall description of how changes in HIV prevalence from 2003–2004 to 2007–2008 varied between individuals with different levels of education in Tanzania.
Analysis files in Stata including the procedures we used to link AIS datasets, recode variables and conduct statistical analysis of the data are available on request.
The final sample sizes on which data on all variables of interest were available were 10 934 in 2003–2004 (82% of all eligible adults) and 15 542 in 2007–2008 (88%). In both years, the surveys included more women than men (Table 1). Just over 40% of participants were aged 15–24 years. Similar proportions of participants in both surveys had received no formal schooling. However, the proportion of individuals reporting having attended at least secondary education was slightly higher in 2007–2008 as compared with 2003–2004 (11.8 vs. 9.5%, χ 2 P = 0.11). In both years, the sample was predominantly rural with a slightly lower proportion of urban participants in 2007–2008 as compared with 2003–2004 (24.6 vs. 30.5%, χ 2 P = 0.11). Overall, HIV prevalence had declined between 2003–2004 and 2007–2008 (7.0 vs. 5.7%, χ 2 P = 0.02).
HIV prevalence was highest among urban residents, women and those aged 25–49 years (Table 2). Greater educational attainment and household wealth were more common among urban than rural residents. HIV prevalence was lower in 2007–2008 than in 2003–2004 in all but three of the population strata we examined, these being individuals from rural areas with no formal education and from the bottom two household wealth quintiles. Declines in HIV prevalence were statistically significant among urban individuals aged 15–24 years, and rural individuals who had attended secondary education and were from the fourth highest household wealth quintile. The point estimate of the OR describing the size of the HIV prevalence decline from 2003–2004 to 2007–2008 was largest among individuals who had attended secondary school.
Among men, there was little evidence of an association between educational attainment and prevalent HIV infection in either survey (Table 3). Among the younger men in 2003–2004, HIV prevalence was higher in the more educated groups but confidence intervals (CIs) were wide because the number of infections was low. Among women aged 15–24 years, educational attainment was significantly associated with HIV prevalence in both surveys. Specifically, there was good evidence for the lowest HIV prevalence being among young women who had attended secondary school (adjusted odds ratio, aOR, comparing secondary with primary educated individuals in 2003–2004, 0.31; 95% confidence interval, CI, 0.12–0.76 and, in 2007–2008, 0.33; 95% CI, 0.13–0.83). Among women aged 25–49 years, there was little evidence of an association between educational attainment and HIV infection in either year, although HIV prevalence was highest in the most educated group in both surveys. In all analyses, additional adjustment for urban/rural status and household wealth increased the OR for no education as compared with primary education, and decreased the OR for secondary education, suggesting that it is important to account for these factors in future analyses, as not to do so risks masking genuine negative associations between level of educational attainment and HIV risk.
There was some statistical evidence for an interaction between educational attainment and age/sex in explaining HIV prevalence in both 2003–2004 (P = 0.07) and 2007–2008 (P = 0.05). We, therefore, did not go on to present coefficients for the education parameters from this pooled analysis.
Finally, we pooled the data from both years, specified adjusted models with HIV as the dependent variable and included an interaction term between educational attainment and year of survey (Table 4). In each of the age/sex stratum-specific analyses, there was weak evidence that this term improved the fit (Wald test P values ranged from 0.18 to 0.61), perhaps because this interaction test had relatively low power in each stratum. However, multiplying out the terms in these stratum-specific models gave point estimates, suggesting that after adjustment for potential confounding factors, HIV prevalence was constant between 2003–2004 and 2007–2008 among those with no formal education in all strata (ORs ranging from 0.83 to 1.50). Among those with primary education, there was good evidence for a decline in HIV prevalence during this period among men aged 15–24 years (aOR 0.41, 95% CI 0.23–0.74), with weaker evidence of this in the other three strata (ORs 0.74–0.91). Among those who had attended secondary school, there was good evidence for a decline in HIV prevalence among men aged 15–24 years (aOR 0.18, 95% CI 0.03–1.02) and women aged 25–49 years (aOR 0.50, 95% CI 0.28–0.92), and again weaker evidence of this between the other two strata (ORs 0.63–0.85).
From these stratified analyses, we judged that there was some evidence that HIV prevalence declined differently among individuals with different levels of education and that the magnitude of these declines was similar in each age/sex stratum. Therefore, in order to maximize the statistical power of our model, we analysed men and women aged 15–49 years together, instead adjusting for age and sex in our analysis. In this analysis, we found good evidence that the interaction term between educational attainment and year of survey improved the fit of the model (Wald test P = 0.07). Parameter estimates suggested that there had been stable HIV prevalence from 2003–2004 to 2007–2008 among those with no education (aOR 2007–2008 vs. 2003–2004, 1.03; 95% CI, 0.72–1.47), a small but borderline-significant decline in HIV prevalence among those with primary education (aOR 0.85, 95% CI 0.69–1.03) and a larger statistically significant decline in HIV prevalence among those with secondary education (aOR 0.53, 95% CI 0.34–0.84).
Previous analysis of the AIS datasets from Tanzania concluded that national HIV prevalence declined significantly from 2003–2004 to 2007–2008 . Our analysis of the same data adds an important nuance to following conclusion: HIV prevalence remained constant among Tanzanians with no formal education from 2003–2004 to 2007–2008, whereas it declined most steeply among those who had attended secondary school.
The sample size of both surveys was large, allowing us to examine interactions between education and survey year in our analysis, the most appropriate analytic strategy with which to address this hypothesis. The data come from the AIS that use standardized sampling and data collection protocols to maximize the comparability of results across time and country. The surveys achieved a high RR and cover a 4-year time span. In addition, we controlled for key potential confounding factors in our analysis. Nevertheless, our study had limitations. One plausible explanation for the pattern we report here is a cohort effect, as more educated individuals may have been more likely to be infected early in the epidemic; by 2007–2008, these individuals would have been infected longer than those of lower education and, therefore, have been more likely to die from AIDS. However, our analysis suggested little evidence that this pattern differed between age groups. As AIDS-related mortality among those aged 15–24 years was likely to be low over this period, our analysis provides some argument against this being the main explanation. Indeed, AIDS-related mortality may have been higher among the least educated because of their lower SEP and consequent lower access to antiretroviral therapy, lower nutritional status or both, leading to less physiological resilience to disease progression as has been seen in some high-income settings . In this case, any bias might have been expected to weaken our conclusion. Overall, our findings are most consistent with the hypothesis that the different pattern of changes in HIV prevalence among groups of different levels of education seen between 2003–2004 and 2007–2008 reflect lower incidence in the most educated group over this period in Tanzania.
Recent studies have explored the association between HIV prevalence and household wealth as measured by a summed, weighted index of a selected group of fixed and durable assets. As for the association between education level and HIV, these results have been presented as an unusual association between SEP and health [5,13]. To our knowledge, no analyses have explored whether this association has changed over time. Such data were available to us from Tanzania, and the unadjusted data we present in Table 2 provide some weak evidence that HIV prevalence was stable or rising amongst those from lower wealth quintiles and fell among those from wealthier quintiles. However, in adjusted analyses, we retained our focus on educational attainment as the key exposure variable of interest, as we believe that this has some advantages. First, educational attainment is an individual-level characteristic that develops during youth and then remains constant over the life course. In contrast, household asset–wealth is a transient characteristic of households so that an individual's status may change over time because of fluctuations in a given household's wealth or if an individual changes household residence. Second, asset indices may not always be an ideal tool to differentiate levels of relative household welfare. Asset indices generally have poor agreement with the more traditional economic measure of consumption expenditure. They may be a marker of population density and general community development more than of household economic status, as they incorporate a number of assets that are driven by community-level availability of services such as those that require electricity to function (e.g. ownership of a fridge or television) [14,15]. It is well documented that HIV has been more common among urban than rural populations, probably because of the increased opportunities for sexual contact these settings provide . Therefore, it is important that our conclusion with regard to the association between educational attainment and HIV persisted after adjustment for both household asset–wealth and a marker of urban/rural setting. Further studies exploring the relative importance of different dimensions of SEP and their association with HIV prevalence are needed. In the absence of new data, however, we believe that trends with respect to educational attainment and HIV are suited to this purpose in many African settings.
Our analysis sheds little light on the social and behavioural processes that might explain our findings. It is plausible that highly educated individuals have undertaken the most rapid sexual behaviour change in recent years, perhaps because of their greater engagement with HIV prevention interventions, access to condoms and levels of self-efficacy and empowerment to change behaviour in ways contrary to established community norms [6,17]. Further analysis of the Tanzania data may provide clues in this regard. In Zambia, changes in HIV prevalence in differently educated groups have been associated with differential declines in risk behaviours . Among the younger group in our analysis, school attendance might directly help reduce infection risk by fostering group norms that are positive towards protective behaviours and putting young people in regular contact with potential sexual partners with low HIV prevalence . It is also plausible that there is some homogeneity of mixing in sexual partnerships with respect to education level . Although there is little available historical data to fully investigate these hypotheses, future studies should collect these data. Dynamic mathematical models of infectious diseases also provide a useful tool to further explore these hypotheses. Finally, further research will also be necessary to explore whether our findings from Tanzania are reproduced in other settings. If confirmed in future studies, our findings suggest the need to re-evaluate whether current HIV prevention efforts meet the needs of the least educated.
We would like to acknowledge the excellent work of MEASURE DHS, ICF Macro, 11785 Beltsville Drive, Suite 300, Calverton, MD 20705, USA, in collecting the data that we analysed for this study.
J.H. conceived the study and led the statistical analysis of data and writing of the paper. L.H. contributed to both the data analysis and the drafting of the paper.
There are no conflicts of interest.
1. Hallett TB, Zaba B, Todd J, Lopman B, Mwita W, Biraro S, et al
. Estimating incidence from prevalence in generalised HIV epidemics: methods and validation. PLoS Med 2008; 5:e80.
2. UNAIDS. Report on the global HIV/AIDS epidemic.
Geneva, Switzerland: UNAIDS; 2008.
3. National Bureau of Statistics (NBS) [Tanzania] and Macro International Inc. Tanzania HIV/AIDS and Malaria Indicator Survey: key findings.
Calverton, Maryland, USA: NBS and Macro International Inc.; 2009.
4. Hargreaves JR, Glynn JR. Educational attainment and HIV-1 infection in developing countries: a systematic review. Trop Med Int Health 2002; 7:489–498.
5. Mishra V, Assche SB, Greener R, Vaessen M, Hong R, Ghys PD, et al
. HIV infection does not disproportionately affect the poorer in sub-Saharan Africa. AIDS 2007; 21(Suppl 7):S17–S28.
6. Hargreaves JR, Bonell CP, Boler T, Birdthistle I, Fletcher A, Boccia D, et al
. Systematic review exploring time-trends in the association between educational attainment and risk of HIV infection in sub-Saharan Africa. AIDS 2008; 22:403–414.
7. de Walque D, Nakiyingi-Miiro JS, Busingye J, Whitworth JA. Changing association between schooling levels and HIV-1 infection over 11 years in a rural population cohort in south-west Uganda. Trop Med Int Health 2005; 10:993–1001.
8. Michelo C, Sandoy IF, Fylkesnes K. Marked HIV prevalence declines in higher educated young people: evidence from population-based surveys (1995–2003) in Zambia. AIDS 2006; 20:1031–1038.
9. Mmbaga EJ, Leyna GH, Mnyika KS, Hussain A, Klepp KI. Education attainment and the risk of HIV-1 infections in rural Kilimanjaro region of Tanzania, 1991–2005. A reversed association
. Sex Transm Dis
10. Kilian AH, Gregson S, Ndyanabangi B, Walusaga K, Kipp W, Sahlmuller G, et al
. Reductions in risk behaviour provide the most consistent explanation for declining HIV-1 prevalence in Uganda. AIDS 1999; 13:391–398.
11. Todd J, Grosskurth H, Changalucha J, Obasi A, Mosha F, Balira R, et al
. Risk factors influencing HIV infection incidence in a rural African population: a nested case-control study. J Infect Dis 2006; 193:458–466.
12. Regidor E, Sanchez E, de la Fuente L, Luquero FJ, de Mateo S, Dominguez V. Major reduction in AIDS-mortality inequalities after HAART: the importance of absolute differences in evaluating interventions. Soc Sci Med 2009; 68:419–426.
13. Msisha WM, Kapiga SH, Earls F, Subramanian SV. Socioeconomic status and HIV seroprevalence in Tanzania: a counterintuitive relationship. Int J Epidemiol 2008; 37:1297–1303.
14. Howe LD, Hargreaves JR, Gabrysch S, Huttly SR. Is the wealth index a proxy for consumption expenditure? A systematic review
. J Epidemiol Community Health
15. Vyas S, Kumaranayake L. Constructing socio-economic status indices: how to use principal components analysis. Health Policy Plan 2006; 21:459–468.
16. Barongo LR, Borgdorff MW, Mosha FF, Nicoll A, Grosskurth H, Senkoro KP, et al
. The epidemiology of HIV-1 infection in urban areas, roadside settlements and rural villages in Mwanza Region, Tanzania. AIDS 1992; 6:1521–1528.
17. Hallman K. Socioeconomic disadvantage and unsafe sexual behaviours among young women and men in South Africa. New York, USA: Population Council; 2004.
18. Sandoy IF, Michelo C, Siziya S, Fylkesnes K. Associations between sexual behaviour change in young people and decline in HIV prevalence in Zambia. BMC Public Health 2007; 7:60.
19. Hargreaves JR, Morison LA, Kim JC, Bonell C, Phetla G, Porter JDH, et al
. The association between school attendance, HIV infection and sexual behaviour among young people in rural South Africa. J Epidemiol Community Health 2008; 62:113–119.
20. Lopman B, Lewis J, Nyamukapa C, Mushati P, Chandiwana S, Gregson S. HIV incidence and poverty in Manicaland, Zimbabwe: is HIV becoming a disease of the poor? AIDS 2007; 21(Suppl 7):S57–S66.