The HIV-1 epidemic in sub-Saharan Africa is approaching its third decade. Various approaches have been used to monitor HIV trends in different countries. The group that has been most commonly studied is antenatal clinic attenders [1–4] but data from other groups such as patients with sexually transmitted diseases and sex workers have also been used for this purpose . Observations from these studies have suggested that in countries such as Uganda the epidemic is stabilizing or falling but in others, most notably in southern Africa, it is rising .
Although large quantities of data have been accumulated for specific groups, inferences drawn on HIV trends from these are subject to several limitations. Most groups are urban and are, therefore, unlikely to represent the general population nationwide, the majority of whom live in rural areas. As not all pregnant women attend antenatal clinics, and HIV-positive women experience reduced fertility, infection rates obtained from this source could be underestimated [7,8]. In contrast, rates obtained from clients at clinics for sexually transmitted diseases will tend to be overestimates because their sexual behaviour puts them at increased risk of HIV. General populations would, therefore, be more appropriate for assessing epidemic trends.
Some studies have reported HIV-1 prevalence rates for general populations using longitudinal and cross-sectional study designs [9–13], though data from the latter can be hard to interpret. Long-term cohort studies are best for monitoring HIV epidemic trends as they allow estimation of prevalence, incidence and associated mortality; they are, however, very rare because of their cost and complexity. It is also valuable to link these trends with patterns of sexual behaviour in the same population to help to interpret the observations. However, trends in changes of sexual behaviour at the general population level have also rarely been evaluated [14,15].
We describe here 7-year trends of adult HIV-1 prevalence, incidence and mortality for a rural Ugandan general population cohort and also assess changes in reported sexual behaviour during this period.
The Medical Research Council general population cohort is located in Masaka district, a rural area of southwest Uganda; it comprises around 10 000 people residing in 15 neighbouring villages, approximately half of whom are adults (aged 13 years and over). The population was first surveyed in 1989–1990 and has been re-surveyed annually since then . Briefly, an annual census enumerates all de jure residents, including joiners and newborns since the previous survey, and obtains information on those that have left or died. A medical team interviews and obtains a blood specimen from all adults within the census who provide informed consent. Since the third survey round, a monthly birth and death register maintained by village registrars has been used to record births and deaths that might otherwise have been missed by the census team. This information is used to supplement census data through regular meetings to reconcile information on births and deaths. The analyses described in the present report are based on data collected during the first eight annual survey rounds, unless stated otherwise.
A range of indicators was used to assess the extent to which sexual behaviour had changed during follow-up. Data on condom use (ever versus never and, if ever, whether with last partner) and numbers of sexual partners in the past 4 weeks were collected using standardized questions administered during the fourth and eighth medical surveys for those who reported ever having had sex. Birth rates for women aged 15–19 and 20–49 years were derived by disaggregating general fertility rates for all women aged 15–49 years reported previously for the first seven survey rounds . Data collected on marital status at annual census were used to estimate the proportion married by single year of age at each survey round.
All sera were tested for HIV-1 antibodies at the central laboratories in Entebbe by two independent enzyme-linked immunoabsorbent assay systems: Recombigen HIV-1 (Cambridge, Biotech, Worcester, MA, USA) and Wellcozyme HIV-1 Recombinant (Wellcome Diagnostics, Dartford, UK). Cambridge Biotech HIV-1 Western blot (Cambridge Biotech, Rockville, MD, USA) was used to determine HIV status on discordant results and confirm the status of newly identified seroconverters [17,18]. HIV test results are provided to study participants, on request, by trained counsellors who are stationed in the study area . Apart from the counsellors and the programme statisticians based in Entebbe, all field survey and research staff remain blind to the HIV status of study participants.
For each annual survey round, age- and sex-specific prevalence rates were derived from the numbers of adults testing HIV positive and negative at that round. All adults identified as HIV negative at a given survey round who had a further definitive HIV test result at one or more subsequent rounds were eligible for inclusion in the incidence analyses. Person-years (PY) at risk for incidence commenced at the date when an HIV-negative individual first gave a blood sample and ceased at the date of their last blood sample, or, for seroconverters, their estimated date of seroconversion. Dates of seroconversion were estimated as midway between the date of the last negative and the first positive HIV test result. For mortality, PY commenced from the date a blood sample was first given with a definitive seropositive status and ceased at the date of the round 8 census, date of death or date of leaving. For adults who were recorded as having left and died, PY ceased at the date of death. Seroconverters contributed PY for mortality as HIV-negative individuals until their estimated date of seroconversion, after which they contributed PY as HIV positive.
Approximate 95% confidence intervals (CI) for rates were estimated using the Gaussian approximation to the log-likelihood; tests of statistical significance for differences in rates were based on the chi-squared distribution . Death rates were standardized for age using the direct method with the total population as standard . Approximate one degree of freedom chi-square (χ) tests for trend across year of follow-up in age-adjusted mortality rates were based on likelihood ratio test statistics derived from Poisson regression models using four age groups (13–34, 35–44, 45–54, 55+). Statistical analyses were performed using the STATA computer package, using standard programs for the analysis of cohort studies where appropriate .
An average of 5500 adults were within the census as resident at each of the eight survey rounds. Approximately 65% gave a blood sample at each round with overall compliance ranging from 74% at round 1 to 56% at round 4. Over 90% of adults gave a blood sample at one or more rounds. Table 1 shows the age-specific HIV prevalence rates by annual survey round and sex. Overall HIV-1 prevalence in all adults declined from 8.2% (95% CI 7.3–9.0) in 1989–1990 to 6.9% (95% CI 6.1–7.7) in 1996–1997 (test for trend:P = 0.008) The respective overall decline among adult males was from 7.5 to 6.3% and this was of borderline significance (test for trend:P = 0.05) whereas that among females was from 8.7 to 7.5%, which was not significant (test for trend:P = 0.09). The age-specific decline was most evident among men aged 20–24, from 11.7 to 3.6% (test for trend, P < 0.001), and among women aged 13–19 and 20–24, from 4.4 to 1.4% (test for trend:P = 0.003) and from 20.9 to 13.8% (test for trend:P = 0.003), respectively. Prevalence however increased significantly among women aged 25–34 years from 13.1% to 16.6% (test for trend:P = 0.04). The median age among infected males was 32 years in 1989–1990 and 33 years in 1996–1997 (P = 0.08, Wilcoxon Rank test for trend). The corresponding median ages for females were 25 and 28 years, respectively (P = 0.03).
A total of 156 HIV-1 incident cases (74 males and 82 females) were identified during 23 578 PY of follow-up. The overall incidence declined from 7.7/1000 PY (95% CI 4.9–12.2) in 1990 to 4.6/1000 PY (95% CI 2.8–7.6) in 1996, but this trend was not significant (test for trend:P = 0.2). The incidence among all males declined from 8.9/1000 PY (95% CI 4.8–16.6) to 5.6/1000 PY (95% CI 2.9–10.7) and among females from 6.6/1000 PY (95% CI 3.3–13.2) to 3.6/1000 PY (95% CI 1.6–8.0). Shown in Table 2 are the age-specific HIV incidence trends by calendar year; neither these nor the overall declines were statistically significant (P > 0.2). However among women aged 13–34 years, incidence peaked at 17.6/1000 PY (95% CI 11.3–27.2) in 1993 and then fell in subsequent years (χ2 = 21.3; degrees of freedom 6;P for unequal rates 0.002). The age–sex distribution of those contributing to incidence analyses remained stable over the study period.
A total of 574 deaths were observed in adults with a definitive seropositive status: 296 deaths occurred during 29 150 PY for HIV-negative individuals (rate 10.2/1000 PY; 95% CI 9.1–11.4) and 279 occurred during 2075 PY for HIV-positive individuals (rate 134.0/1000 PY; 95% CI 119.1–150.7). The age-standardized death rate was 6.5/1000 PY (95% CI 3.6–9.4) for HIV-negative adults in 1990 and 8.2/1000 PY (95% CI 5.6–10.9) in 1996. Corresponding rates for HIV-positive adults were 129.7/1000 PY (95% CI 72.6–186.9) in 1990 and 102.7/1000 PY (95% CI 60.7–144.8) in 1996. There were no statistically significant trends in age-adjusted death rates when analysed by single year of follow-up (Table 3) for either HIV-negative (test for trend:P = 0.4) or HIV-positive adults (test for trend:P = 0.9). Similarly there was no significant trend in overall age-adjusted death rate for HIV-positive and HIV-negative individuals combined (test for trend:P = 0.8) nor for age-specific death rates (test for trend:P > 0.1).
Trends in sexual behaviour
At round 4, 59% (1709/2886) females and 48% (1367/2858) males within the census had information on sexual behaviour, the corresponding proportions at round 8 were 59% (1668/2816) and 57% (1541/2720). This is slightly lower than the proportions giving a blood sample at each round as some individuals preferred providing only a blood sample and not to be subjected to responding to the sexual behaviour questionnaire.
Age at first sex
Among young men aged 13–19 years, there was a trend towards an increase in reported age of first sex (Fig. 1). At round 4, the median age at first sex was 17.5 years, while at round 8 it was 18.2 years (test for trend:P = 0.002). There was no equivalent delay in young women, the median age at first sex being 16.7 years at both rounds (test for trend:P = 0.2).
Age at first marriage
There was a decrease in annual age-specific prevalence of being married for young women aged 13–19 (test for trend:P < 0.001). This was particularly evident between the ages of 16 and 19 (Fig. 1 shows data for rounds 4 and 8 as examples). At round 4, the median age at marriage was 18.5 years, while by round 8 this had increased to 19.5 years. There was no equivalent increase in age at first marriage for men (test for trend:P = 0.8). The median age at marriage at both survey rounds was 24 years for men.
The fertility rate for all women aged 15 years or over fell significantly from 231.7/1000 women-years (WY) in the first year of follow-up to 188.0 in the sixth year of follow-up (test for trend:P < 0.001), (Fig. 2). A substantial proportion of this fall can be attributed to a reduction in the fertility rate of teenage women (15–19 years). In the first year of follow-up their fertility rate was 205.2/1000 WY and by the sixth year was 138.7/1000 WY (test for trend:P = 0.001). The unmarried teenage pregnancy rate halved over this period from 119.9 to 57.9/1000 WY (test for trend:P < 0.001).
Number of recent sexual partners
The distribution of reports from males of numbers of sexual partners in the past 4 weeks altered significantly between rounds 4 and 8 (P < 0.001). The proportion of men with two or more sexual partners in this period rose from 10 to 14%. There were also significant changes for single men and married men and for males in age groups 25–34 and 45+ (Table 4). For women, there were no significant changes overall nor for any age group. There was, however, a significant change among married women, where the number reporting no sexual partner decreased from 20 to 14% (P = 0.008).
The rate of reported condom use increased very significantly between survey round 4 and survey round 8 for males of all age groups (Fig. 3). The overall prevalence of males who had ever used condoms increased from 12 to 35% in this 4-year period (test for trend:P < 0.001). Analysis by marital status showed significant increases in reported use for single, married and divorced or widowed men (data not shown). There were similar increases for females, rising overall from 4 to 12% (test for trend:P < 0.001), with significant increases in all age groups under 35 years (Fig. 3) and also for all marital groups (data not shown).
In contrast, reported condom use with the last partner did not alter substantially over this period (51% at round 4 and 43% at round 8 for men and 47% and 36%, respectively, for women). None of the differences was statistically significant, but this analysis was restricted to those who reported having ever used a condom (94 men and 43 women at round 4; 365 men and 154 women at round 8). However, condom usage rates were high at round 8 among those reporting a casual partner in the past 12 months (only 175 men and 23 women). Overall, 60% of men and 39% of women reported using a condom with their most recent casual partner. There were no equivalent data collected at round 4.
A desirable approach for monitoring HIV-1 epidemic trends is to determine prevalence and incidence levels in a longitudinal cohort, of which incidence is the most appropriate. However, available data on trends have been generally limited to HIV prevalence from sentinel surveillance, which on their own are subject to limitations. We have been able to use information from this longitudinal cohort study in a rural area of Uganda to investigate the trends over 7 years of epidemiological parameters of HIV incidence, prevalence and mortality with linked information on aspects of sexual behaviour. This cohort study has shown a significant reduction in overall adult HIV seroprevalence over the study period, from 8.2 to 6.9%, which is an approximate 15% reduction. This has been mainly a result of larger falls in seroprevalence in young men and women, where the declines have been strongly statistically significant. However older women aged 25–34 years do not appear to be benefiting from the declining rates seen in younger women. This is largely a cohort effect of women aged 20–24 years moving into this age group. There have been no similar reductions in either HIV seroincidence or mortality in this population. However, this may be because even with a cohort of this size we are not yet able to detect any trend. We have found considerable heterogeneity in the HIV seroincidence and mortality rates over the years, such that even the documented 40% fall in seroincidence between the first and the sixth year of follow-up does not produce a statistically significant trend.
There is also some evidence of sexual behaviour change over this period. The reported rates of ever using condoms have increased substantially, but we do not know if this reflects consistent condom use. We have found a trend towards marriage at a later age for girls and towards a later age at sexual debut for boys. Fertility rates have fallen generally in the study area, and this is particularly evident for teenagers, especially unmarried teenagers. In contrast, we found no evidence of a reduction in numbers of recent sexual partners; indeed, there may even have been a small increase. Taken together, these are generally favourable trends in reported sexual behaviour and are consistent with the epidemiological information. Secular trends such as contraception and polygamy do not seem to explain these trends. For example, we have found that only 3% of women were reported to use modern contraceptives at round 8 survey and that the rate of married adults in polygamous marriages did not greatly change: 12% at round 1 and 14% at round 8.
These findings support and extend previous reports from Uganda. We have previously shown a significant reduction in HIV seroprevalence in young men and women in this same rural cohort  and a significant fall in overall adult seroprevalence has been reported from the neighbouring Rakai district over a 2-year period . Surveillance data based on repeated cross-sectional surveys from urban antenatal clinics have reported a reduction in HIV seroprevalence in young pregnant women and the same research group has also found evidence suggestive of change in sexual behaviour among young urban adults . Similar prevalence trends and corresponding reductions in risk behaviour have also been observed with other antenatal sentinel surveillance data in western Uganda, particularly in the 15–19 year age group .
The issue of whether falling HIV seroprevalence in young adults can confidently be taken to indicate that seroincidence is also falling remains unresolved. Wawer et al.  and Nunn et al.  have both shown empirical data where HIV seroprevalence has fallen in young adults but seroincidence has remained stable or even risen. It is clear that the population seroprevalence depends upon the balance of incidence, mortality, migration and coverage; however, the power of these studies to detect any trend in seroincidence, especially for age- and sex-specific rates, is marginal. A significant reduction in HIV seroincidence would provide the clearest indication of the epidemic trends.
It is important that the information about sexual behaviour is not overinterpreted. The complexity and sensitivity of sexual behaviour means that this topic does not lend itself well to relatively superficial epidemiological surveys, which should be complemented by more in-depth qualitative methods in order to test the reliability of the data. In addition, for some of the reported trends, the numbers of subjects involved were small and, therefore, our estimates have wide margins of confidence. We also lack important pieces of behavioural information, for example the relationship with the last sexual partner and the consistency of condom use. It is also possible that respondents’ perceptions of acceptable and desirable answers to standardized questions have changed over the study period as societal norms have adjusted to the realities of the era of the AIDS epidemic.
The average coverage rate of the annual surveys was 65%. This could potentially lead to bias if HIV-infected individuals systematically increased or decreased their involvement in the surveys over time. In fact there are very few long-term resident non-compliers, and we find that the majority of non-compliers at any one survey round are likely have participated in either the previous or subsequent surveys. In addition, our data collection methods have generally improved over the study period and this may also have affected our reported rates. For example, census information about births and deaths has improved over the years as we have added questions about whether emigrants are known to have died or not, and by the addition from the third annual survey onwards of a supplementary system of local birth and death recorders. Migration is a well-recognized factor in the HIV epidemic and is likely to introduce bias into results from studies such as these. We have looked at the effect of both out and in migration on HIV prevalence in the cohort and found excess infection rate among adults who join (16.3%) compared with those who leave (11.5%) the study area . Therefore, any migration effect would not explain the observed decline in prevalence.
Despite these limitations, this study provides the strongest evidence to-date from Uganda of an overall long-term reduction in seroprevalence of HIV in adults and parallel evidence of sexual behaviour change. It is important to appreciate that no particular HIV interventions are carried out in this study area. Condoms were distributed through the counselling offices and field survey teams and later during the study period through social marketing of condoms. HIV testing and counselling facilities are available and the population is exposed to national health education messages in addition through village-based meetings and visits to schools. While these activities may have been at a somewhat higher intensity in the study area, this is fairly typical of most rural Ugandan populations and these encouraging trends are, therefore, likely to be generalized to the wider population. We plan to continue to monitor the population in order to detect any emerging trends in HIV seroincidence and mortality. These findings should provide other countries in sub-Saharan Africa and elsewhere with hope that HIV epidemic trends can be reversed. It is encouraging that policy-makers from other developing countries are visiting Uganda to view what can be achieved with modest resources through a national AIDS control programme. However, the burden of HIV-associated morbidity and mortality remains unacceptably high in this population, as in many similar populations in Uganda and elsewhere. Monitoring epidemic trends is not in itself sufficient; these findings should spur the scientific community to renewed efforts to mitigate the cataclysmic impact of HIV in sub-Saharan Africa.
1. Sokal DC, Buzingo T, Nitunga N, Kadenda P, Standaert B. Geographic and temporal stability of HIV sero-prevalence among pregnant women in Bujumbura, Burundi.
AIDS 1993, 7: 1481 –1484.
2. Asiimwe-Okiror G, Opio A, Musinguzi J, Madraa E, Tembo G, Carael M. Change in sexual behaviour and decline in HIV infection among young pregnant women in urban Uganda.
AIDS 1997, 11: 1757 –1763.
3. Kigadye R-M, Klokke, Klokke, Nicoll A. et al. Sentinel surveillance for HIV-1 among pregnant women in a developing country: 3 years’ experience and comparison with a population serosurvey.
AIDS 1993, 7: 849 –855.
4. Magazani K, Laleman G, Perriens JH. et al. Low and stable HIV sero-prevalence in pregnant women in Shaba Province, Zaire.
J Acquir Immune Defic Syndr 1993, 6: 419 –423.
5. Laga M, Alary M, Nzila N. et al. Condom promotion, sexually transmitted diseases treatment, and declining incidence of HIV-1 infection in female Zairian sex workers.
Lancet 1994, 344: 246 –248.
6. Tarantola D, Schwartlander B. HIV/AIDS epidemics in sub-Saharan Africa: dynamism, diversity and discrete declines? AIDS 1997, 11 (Suppl B): S5 –S21.
7. Carpenter LM, Nakiyingi JN, Ruberantwari A, Malamba SS, Kamali A, Whitworth JAG. Estimates of the impact of HIV infection on fertility in a rural Ugandan population cohort.
Health Transit Rev 1997, 7 (Suppl 2): 113 –126.
8. Gray RH, Wawer MJ, Serwadda D. et al. Population-based study of fertility in women with HIV-infection in Uganda.
Lancet 1998, 351: 98 –103.
9. Wawer M, Serwadda D, Gray R. et al. Trends in HIV-1 prevalence may not reflect trends in incidence in mature epidemics: data from the Rakai population-based cohort, Uganda.
AIDS 1997, 11: 1023 –1030.
10. Delaporte E, Janssens W, Peeters M. et al. Epidemiological and molecular characteristics of HIV infection in Gabon, 1986–1994.
AIDS 1996, 10: 903 –910.
11. Mulder D, Nunn A, Kamali A, Kengeya-Kayondo J. Decreasing HIV-1 sero-prevalence in young adults in a rural Ugandan cohort.
BMJ 1995, 311: 833 –836.
12. Fylkesnes K, Ndhlovu Z, Kasumba K, Musonda MR, Sichome M. Studying dynamics of the HIV epidemic: population-based data compared with sentinel surveillance in Zambia.
AIDS 1998, 12: 1227 –1234.
13. Fontanet AL, Messele T, Dejene A. et al. Age- and sex-specific HIV-1 prevalence in the urban community setting of Addis Ababa, Ethiopia.
AIDS 1998, 12: 315 –322.
14. UNAIDS Case Study. A Measure of Success in Uganda. The Value of Monitoring both HIV Prevalence and Sexual Behaviour
. Geneva: UNAIDS; May 1998. UNAIDS Best Practice Collection–UNAIDS/98.8. Geneva, Joint United Nations Programme on HIV/AIDS.
15. Ng'weshemi JZL, Burma JT, Pool R. et al. Changes in male sexual behaviour in response to the AIDS epidemic: evidence from a cohort study in urban Tanzania.
AIDS 1996, 10: 1415 –1420.
16. Mulder DW, Nunn AJ, Wagner H-U, Kamali A, Kengeya-Kayondo JF. HIV-1 incidence and HIV-1 associated mortality in a rural Ugandan population cohort.
AIDS 1994, 8: 87 –92.
17. Nunn AJ, Biryahwaho B, Downing RG, van der Groen G, Ojwiya A, Mulder DW. Algorithm for detecting antibodies to HIV-1: results from a rural Ugandan cohort.
AIDS 1993, 7: 1057 –1061.
18. Nunn AJ, Biryahwaho B, Downing RG, Ojwiya A, Mulder DW. Computer-assisted quality assurance in an HIV serology laboratory.
Meth. Inform. Med. 1994, 33:2: 170 –173.
19. Seeley JA, Wagner H-U, Mulemwa J, Kengeya-Kayondo JF, Mulder DW. The development of a community based HIV/AIDS counselling service in a rural area in Uganda.
AIDS Care 1991, 3: 207 –217.
20. Clayton D, Hills M. Analysis of follow-up studies
Stata Tech Bull 1995, 27: 19 –26.
21. Clayton D, Hills M. Statistical Models in Epidemiology
. New York: Oxford University Press; 1995.
22. Kilian AHD, Gregson S, Ndyanabangi B. et al. Reductions in risk behaviour provide the most consistent explanation for declining HIV-1 prevalence in Uganda.
AIDS 1999, 13: 391 –398.
23. Wawer MJ, Serwadda D, Gray RH. et al. Trends in HIV-1 prevalence may not reflect trends in incidence in mature epidemics: data from the Rakai population-based cohort, Uganda.
AIDS 1997, 11: 1023 –1030.
24. Nunn AJ, Mulder DW, Kamali A, Kengeya-Kayondo JF, Whitworth JAG. HIV-1 incidence in sub-Saharan Africa.
Lancet 1996, 348: 965 –966.
25. Nunn AJ, Wagner HU, Kamali A, Kengeya-Kayondo JF, Mulder DW. Migration and HIV-1 seroprevalence in a rural Ugandan population.
AIDS 1995, 9: 503 –506.