The standard way of tracking the HIV epidemic usually depends on measuring HIV prevalence, which is important for planning, programming, and policy.1 Measurement of national HIV prevalence relies on sentinel surveillance and population-based surveys: in sentinel surveillance, HIV trends are inferred to the general population from the results of different studies of HIV burden in specific groups (eg, antenatal clinic attendees, blood donors, people with other sexually transmitted infections), whereas in longitudinal surveys of population-based cohorts, HIV prevalence trends are directly estimated.
Recent progress in expanding access to antiretroviral therapy (ART), with, for example, a 10-fold expansion in access to ART in low- and middle-income countries over the 5 years up to 2007,2 has implications for HIV prevalence estimates. ART has radically transformed the lives of people with HIV infection worldwide, leading to a pronounced decline in AIDS-related deaths.3 Because the prevalence of HIV depends on the incidence and duration of infection, as people with HIV infection on ART live longer, the duration of HIV infection increases, and HIV prevalence tends to increase. Therefore, interpreting HIV prevalence as a measure of epidemic severity depends on knowing the contribution to HIV prevalence of lives saved due to ART. As ART coverage increases, assessing its impact on the use of HIV prevalence to measure the epidemic becomes increasingly important, especially in sub-Saharan Africa, which bears the brunt of the epidemic.
Recent studies in Malawi4 and South Africa5 demonstrated a reduction in population mortality shortly after the introduction of ART, but the impact of ART on HIV prevalence has not been directly estimated from empirical data. To date, published estimates of the impact of ART on HIV prevalence rely on modeling to estimate prevalence trends from surveillance and survey data. The population-based cohort study in rural southwest Uganda established in 1989,6 in which ART was introduced in 2004, enables the direct assessment of the impact of ART on HIV prevalence. We estimated HIV prevalence and the number of deaths, while taking into consideration the possible influence of changes in migration and HIV incidence.7 We estimated the number of lives saved in the ART period (2004–2010) and the hypothetical HIV prevalence that would have occurred without the lives saved due to ART. We compared this hypothetical prevalence with the actual estimated HIV prevalence.
Since 1986, Uganda has been recovering from previous civil, political, and economic turmoil. The estimated 30 million population is mostly engaged in subsistence agriculture. Annual Gross National Income is $340 per capita, and mean life expectancy at birth is 51 years.8 It is one of the countries in Africa where the HIV epidemic was first reported.9 The study site is in rural southwestern Uganda, not far from Lake Victoria, with the Trans-African Highway passing nearby. The study population is mostly subsistence farmers, living in semipermanent structures built from locally available materials. There are no tarmac roads and access may be difficult during the rains. The community mostly comprises people from the Baganda tribe, with 15% of Rwandese origin, who are well assimilated. Religious affiliation is mostly Christian, with a significant Muslim minority (28%).10
Annual HIV Survey
Full details of the annual HIV serosurvey in this open cohort have been published elsewhere.11,12 In brief, starting at the end of 1989, an annual household survey has been conducted, with all study village residents eligible for inclusion. The annual household survey was initially conducted among the residents of 15 villages, increased to 25 villages in 2000. There are currently about 18,000 study village residents, of whom about half are adults (defined as aged ≥13 years). Because each annual survey begins at the end of a year and finishes at the end of the following year, the calendar year corresponding to each survey round is regarded as the year after survey initiation, for example, the first survey is regarded as taking place in 1990. Community sensitization activities precede each survey, including local council briefings and village meetings. All households are visited by, in turn, the mapping, census and survey teams (all accompanied by a village councilor). Average annual survey participation ranges from 61% to 71% of those censused and eligible for survey.
Census staff administer a standardized questionnaire to enumerate household members (including any who are temporarily absent from the household at the time of the census) and obtain information on deaths, births, and inmigration and outmigration since the previous census. The census is supplemented by village-based birth and death recorders. In the survey, consenting residents are interviewed at home in the local language by trained survey staff and provide a blood sample for HIV testing. Two enzyme immunoassays were used with set algorithms to establish HIV-1 serostatus.13 Participants in the general population-based cohort who are found to have HIV infection are offered referral to the study clinic,14 which in 2004 introduced free ART for participants eligible according to Ministry of Health criteria15 and is the main ART provider in the locality.
Estimation of Mortality Rates
Mortality rates per 1000 persons were estimated in each annual survey, between 2000 and 2010, for HIV-negative and HIV-positive participants. Mortality rates among HIV-positive survey participants were standardized for age and sex to the HIV-negative population of the 2000–2010 surveys. We calculated a single annual age-standardized mortality rather than age-specific mortality by age group because we did not readily have the ages of those on ART. Aligning names, age, and gender to ART data would require us to go through the clinic forms, and this would breach confidentiality of these data.
We compared standardized mortality rates among HIV-positive participants before and after ART introduction in 2004. The number of deaths averted was calculated by multiplying the mortality rate in HIV-positives from the pre-ART era, by the number of HIV-positives not on ART at the beginning of the year.
Estimation of HIV Prevalence
Methods of estimating adult HIV prevalence in this study population have been described previously.12,16 Missing HIV serostatus data were imputed as follows: in a given survey, those found to be HIV-positive were assumed to be HIV-positive in all subsequent surveys that they participated in and HIV-negative in all surveys before the estimated date of seroconversion, and those found to be HIV-negative were assumed to be HIV-negative in all previous surveys that they participated in.12,16 HIV status was regarded as unknown if an individual tested HIV-negative in a given survey but was not tested during subsequent surveys. Mbulaiteye et al12 showed that the imputed HIV status yielded similar HIV prevalence trends to the method where only the tested serostatus of individuals in a given year was used; but noted that estimates from imputed HIV status tend to be biased downward in the initial year and upward in the final year. In this study, this is not a problem because we do not report estimates in the first and final rounds. Annual HIV prevalence was calculated from 2000 to 2010.
Estimation of HIV Incidence
Incident cases were determined based on whether the date of seroconversion could be established. Date of seroconversion is an estimated date half way between the date of the last negative test before the first positive test and the date of the first positive test. HIV-positive individuals were considered to have been at risk of HIV acquisition before the first HIV-positive test only for periods that they were resident/present in the study area. HIV-negative individuals were considered to be at risk of HIV infection when they were resident in the study area. Person-years at risk were calculated and then HIV incidence was calculated based on the number of cases in a given year and the total number of person-years at risk in that year.
Estimation of Contribution to HIV Prevalence of Lives Saved Due to ART
We calculated the average annual mortality rate among HIV-positive participants in the 4 years before ART introduction (2000–2003). We then calculated the number of deaths expected among HIV-positive participants each year during 2004–2010 if ART had not been available, assuming an unchanged average annual mortality rate. We calculated the expected number of lives of HIV-positive participants saved due to ART during 2004–2010 as the difference between the number of deaths expected if ART had not been available and the number of observed deaths. Finally, we calculated the hypothetical HIV prevalence as the prevalence that would have occurred if HIV mortality had remained at pre-2004 levels and lives had not been saved due to ART. We compared the hypothetical and estimated HIV prevalence.
The difference in mortality rates in the periods before ART (2000–2003) and during ART (2004–2010) was compared with its estimated standard error, using standard methods for comparing incidence rates.17 The χ2 test of trend was used to test for the increased hypothetical HIV prevalence between 2004 and 2010. The analysis was done using STATA 10 (Stata Corp, College Station, TX).
The study was part of a research program approved by the Science and Ethics Committee of the Uganda Virus Research Institute and by the Uganda National Council for Science and Technology.
Characteristics of Cohort Participants
The mean age of census and survey participants during 2000–2010 was generally significantly higher in females (range 32–35 years) than in males (range 32–33 years; Appendix, Table 1). The proportion of male participants was significantly greater in the census (range 47%–48%) than in the survey (range 42%–47%).
During 2000–2010, the unstandardized mortality rate (per 1000) among survey participants was much higher in those who were HIV-positive (range 26.3–115.3) than in those who were HIV-negative (range 1.9–4.0, Table 1).
The average annual standardized mortality rate (per 1000) among HIV-positive people significantly fell from 107.3 in the period before ART (2000–2003) to 35.3 in the period during ART (2004–2010; P < 0.01). The average annual mortality rate (per 1000) among HIV-negative participants was similar in the 2 periods (2.8 and 2.7, respectively).
Estimated HIV Prevalence
HIV prevalence was the highest among females aged 24–34 years followed by males in the same age group. Males in the youngest age group had the lowest HIV prevalence (Appendix, Table 2). Overall, HIV prevalence did not change significantly (range 6.1%–6.5%, P = 0.28) during the period before ART from 2000 to 2003, but did increase significantly from 6.9% in 2004 to 8.9% in 2010 (test for trend: P < 0.01) during the ART period and was significantly higher during the ART period 2004–2010 (range 6.9%–8.9%) than during the pre-ART period 2000–2003 (range 6.1%–6.5%; test for trend: P < 0.01).
Estimated HIV Incidence
Estimated HIV incidence (cases per 1000 person-years) was generally stable between 2000 and 2010 (P = 0.63, χ2 test for trend). Estimated HIV incidence was 5.6 during the pre-ART period (2000–2003) reducing slightly to 5.0 during the ART period (2004–2010; P = 0.28, χ2 test for trend), with little change during this last period (range 3.3–5.8; Table 2). Shafer et al16 reported an overall stable HIV incidence trend in the period 2002–2004 with a drop in 2005.
Contribution to HIV Prevalence of Lives Saved Due to ART
Assuming mortality rates remained unchanged, the number of hypothetical deaths among HIV-positive participants if ART had not been available since 2004 was estimated at 63 in 2004 and 59 in 2010, whereas the actual total number of deaths for those on and not on ART fell from 40 in 2004 to 23 in 2010. The estimated number of lives saved due to ART (ie, expected minus observed deaths) in each calendar year increased from 23 in 2004 to 36 in 2010 with a total of 228 cumulative expected lives saved during 2004–2010 among HIV-positives who had not yet accessed ART (Table 3). We note that this is likely to be an underestimate because the observed deaths may have included those on ART.
In the ART period (2004–2010), the hypothetical HIV prevalence was generally stable at 6.6% in 2004 and 6.4% in 2010 (test for trend: P = 0.99; range 6.0%–6.6%; Fig. 1) contrary to the estimated prevalence, which increased significantly from 6.9% to 8.9% (test for trend: P < 0.01) in the same period. The increase in the actual number of HIV-positive participants by 182 (from 592 in 2004 to 774 in 2010) accounted for the 29% increase in estimated prevalence from 6.9 to 8.9% during that period (Appendix, Table 2). Expected lives saved due to ART accounted for an increasing proportion of the estimated prevalence from 4.0% in 2004 to 29.4% in 2010.
Analysis of data from original villages of the general population cohort by Shafer et al18 indicated that every year since the start of the survey there are on average 40 HIV-positive in-migrants (range 28–54), and 29 HIV-positive outmigrants (range 17–41), giving a net contribution each year of 11 HIV-positive participants to the HIV prevalence numerator and an overall contribution of on average +0.2% to the HIV prevalence (range −0.3 to +0.5%). Overall migration in and out of the cohort has been variable, with an average outmigration during 2004–2007 of 75 people per year. This reduction in the HIV prevalence denominator corresponds to a further 0.25% contribution to HIV prevalence. Inmigration and outmigration data, from Shafer et al,18 among HIV-positives are presented in Table 3.
Expected lives saved due to ART largely accounted for the increase in estimated HIV prevalence from 2004 to 2010, emphasizing the importance of careful interpretation of HIV prevalence as a measure of the burden of HIV, including the increasing impact of ART. The 29% increase in estimated HIV prevalence in this study from 6.9% in 2004 to 8.9% in 2010 was not explained by increased HIV incidence or by the effect of migration. Inmigration of people with HIV (accounting in any given annual survey for >50% of the new HIV-positive respondents)7 was largely off-set by outmigration of people with HIV infection.
The 29% increase in the estimated prevalence from 6.9% in 2004 to 8.9% in 2010 arises from both the increase by 182 (from 592 to 774) in the actual number of HIV-positive participants and the 228 expected lives saved due to ART during 2004–2010. The increased estimated HIV prevalence during the ART period and the stable hypothetical HIV prevalence (range 6.0%–6.6%), that is, if ART had not been available, reflects the impact of ART on HIV prevalence. Lives saved due to ART accounted for an increasing proportion of the estimated HIV prevalence from 4% in 2004 to 29% in 2010. After ART introduction, the expected number of lives saved and therefore the impact of ART on HIV prevalence is expected initially to be relatively small, and to increase with the cumulatively increasing number of people rescued from life-threatening immunodeficiency and with the improved survival of people recruited into the program.
In this study, all the participants eligible for ART (as per the Uganda Ministry of Health criteria) are invited to initiate ART and by 2008, 92% of those eligible had initiated ART. The proportion of HIV-positives on ART in 2004 was 7.4% increasing to 33.5% by 2010. Although ART saves the lives of a proportion of HIV-positive people with immunodeficiency, the more advanced the immunodeficiency, the lower the proportion. In a new ART program, the first-year survival is likely to be low because most patients start on ART very late, with low CD4 counts, and the number of lives saved will be limited by the ART coverage. As ART coverage increases and people with HIV infection may be identified earlier as being in need of ART, they would then start ART at higher CD4 counts, and more lives would be saved.
A strong point of this study is that the longitudinal cohort enabled the estimation of the impact of ART on HIV prevalence by empirical estimation of HIV prevalence and by direct measurement of the number of deaths, whereas other published estimates of the impact of ART on HIV prevalence rely on modeling to estimate prevalence trends from surveillance and survey data. For example, a study of the national ART program in Botswana, which was initiated in 2000 and had reached high coverage (83% of need) by 2007, found that ART provision averted an estimated 53,000 deaths from 2000 to 2007.19 If these deaths had not been averted by ART, then AIDS deaths would have exceeded new infections and adult HIV prevalence would have been 22.7% in 2007, 3% points lower than the actual estimate (25.7%).19 Our population-based cohort study has several other strengths: The data on HIV-related mortality are more representative of the general population than data from clinic-based studies, which provide information on the mortality only of individuals who have accessed health care: the accuracy of mortality data is enhanced by the system of village-based recorders, providing a crosscheck of the information on deaths obtained in the census; and the demographic data collected in the annual census enable accurate calculation of the denominators of mortality rates.
Regarding study limitations, nonresponse is a potential source of bias because survey participation is entirely voluntary. It is possible that those residents who perceive themselves to be at increased risk of HIV are less likely to participate in the survey, leading to the underestimation of HIV prevalence. Although the proportion of male participants is significantly higher in the census than in the survey (because for social and work reasons men are more likely than women to be temporarily absent from the household when the survey team visits), this is not a potential source of bias because the mortality rates calculated using the census data are similar in men and women. The significantly higher mean age of females than males in both the census and surveys is accounted for by outmigration of young women for work and marriage, leading to the possible underestimation of HIV prevalence (which is higher in younger females than in younger males). Coverage of death recording is likely to be higher than survey coverage, with deaths being recorded both by the census and by village-based recorders. Therefore, regarding the impact on HIV prevalence of decreased mortality due to ART, the study results would tend to be an overestimation of the extent of contribution of ART impact to the estimated prevalence. Finally, the calculation of expected lives saved due to ART assumes that if ART had not been available during 2004–2010 the HIV-related mortality would have remained unchanged at pre-2004 levels. The main factor, which could potentially have contributed to HIV-related mortality reduction, was the provision of cotrimoxazole as part of clinical care of HIV-positive participants from 2004 onward, although its known impact on HIV-related mortality is much less than that of ART.20
Because measuring HIV prevalence is important for planning, programming, and policy, interpreting the results of HIV prevalence surveys requires careful attention to the contributory factors, including the increasing impact of expanded access to ART in decreasing mortality.4,5 Consideration is necessary for measurement of HIV incidence, which is much more sensitive to the changing dynamics of HIV transmission and provides a more valid measure of the state of the epidemic and the impact of intervention programs than HIV prevalence. HIV incidence can be estimated indirectly from prevalence data using mathematical models,21,22 but there is as yet no reliable means of assessing it other than direct measurement by repeated population surveys23,24 (of which there have been very few in sub-Saharan Africa).25 Increased ART coverage is initially likely to increase HIV prevalence through increased survival,26 but in the medium and long term may result in decreased HIV prevalence through decreased HIV transmission and decreased HIV incidence.27 The extent to which increasing ART coverage may modify the HIV epidemic through decreased incidence is currently hotly debated.28 Especially now in the ART era, directly measuring HIV incidence in longitudinal population-based cohort studies enables accurate tracking of the HIV epidemic in Africa and elsewhere.
1. UNAIDS. Global AIDS Response Progress Reporting. Geneva, Switzerland: UNAIDS; 2012.
2. WHO/UNAIDS/UNICEF. Toward Universal Access: Scaling Up Priority HIV/AIDS Interventions in the Health Sector: Progress Report. 2010.
3. Mocroft A, Brettle R, Kirk O, et al.. Changes in the cause of death among HIV-positive subjects across Europe: results from the EuroSIDA study. AIDS. 2002;16:1663–1671.
4. Floyd S, Molesworth A, Dube A, et al.. Population-level reduction in adult mortality after extension of free anti-retroviral therapy provision into rural areas in Northern Malawi. PLoS One. 2010;5:e13499.
5. Herbst AJ, Cooke GS, Bärnighausen T, et al.. Adult mortality and antiretroviral treatment roll-out in rural Kwa-Zulu Natal, South Africa. Bull World Health Organ. 2009;87:754–762.
6. Mulder D, Nunn A, Kamali A, et al.. Decreasing HIV-1 seroprevalence in young adults in a rural Ugandan cohort. BMJ. 1995;311:833–836.
7. Maher D, Todd J, Shafer LA, et al.. Impact of migration on calculated HIV prevalence in a longitudinal population-based cohort in rural Uganda. Presented at: International Union for the Scientific Study of Population Conference; 2009; Marrakech, Morocco.
9. Serwadda D, Mugerwa RD, Sewankambo NK, et al.. Slim disease: a new disease in Uganda and its association with HTLV-III infection. Lancet. 1985;2:849–852.
10. Nakibinge S, Maher D, Katende J, et al.. Community engagement in health research: two decades of experience from a research project on HIV in rural Uganda. Trop Med Int Health. 2009;14:190–195.
11. Nunn AJ, Mulder DW, Kamali A, et al.. HIV-1 infection over five years in a rural Ugandan population: cohort study. Br Med J. 1997;315:767–771.
12. Mbulaiteye S, Mahe C, Whitworth J, et al.. Declining HIV-1 incidence and associated prevalence over 10 years in a rural population in south-west Uganda: a cohort study. Lancet. 2002;360:41–46.
13. Nunn AJ, Biryahwaho B, Downing R, et al.. Algorithms for detecting antibodies to HIV-1: results from a rural Ugandan cohort. AIDS. 1993;7:1057–1061.
14. Morgan D, Malamba SS, Maude GH, et al.. An HIV-1 natural history cohort and survival times in rural Uganda. AIDS. 1997;11:633–650.
15. MOH-Uganda. National Antiretroviral Treatment and Care Guidelines for Adults and Children. Kampala, Uganda: Ministry of Health; 2003.
16. Shafer LA, Biraro S, Nakiyingi-Miiro J, et al.. HIV prevalence and incidence are no longer falling in Southwest Uganda: evidence from a rural population cohort 1989–2005. AIDS. 2008;22:1641–1649.
17. Rothman K. Modern Epidemiology. Boston, MA: Little, Brown; 1986.
18. Shafer LA, Maher D, Weiss HA, et al.. The contribution of population factors to estimation of HIV prevalence trends: a cohort study in rural Uganda, 1989–2007. Am J Epidemiol. 2011;174:1175–1182.
19. Stover J, Fidzani B, Molomo BC, et al.. Estimated HIV trends and program effects in Botswana. PLoS One. 2008;3:e3729.
21. Brown T, Salomon JA, Alkema L, et al.. Progress and challenges in modelling country-level HIV/AIDS epidemics: the UNAIDS Estimation and Projection Package 2007. Sex Transm Infect. 2008;84(suppl 1):i5–i10.
22. Stover J, Johnson P, Zaba B, et al.. The spectrum projection package: improvements in estimating mortality, ART needs, PMTCT impact and uncertainty bounds. Sex Transm Infect. 2008;84(suppl 1):i24–i30.
23. Todd J, Lutalo T, Kaleebu P. Estimating incidence of HIV infection in Uganda. JAMA. 2009;301:159–160.
25. Maher D, Biraro S, Hosegood V, et al.. Translating global health research aims into action: the example of the ALPHA* network. *analysing longitudinal population-based HIV/AIDS data on Africa. Trop Med Int Health. 2010;15:321–328.
27. Lima VD, Johnston K, Hogg RS, et al.. Expanded access to highly active antiretroviral therapy: a potentially powerful strategy to curb the growth of the HIV epidemic. J Infect Dis. 2008;198:59–67.
28. Maggiolo F, Leone S. Is HAART modifying the HIV epidemic? Lancet. 2010;376:492–493.