Secondary Logo

Journal Logo


HIV prevalence and incidence are no longer falling in southwest Uganda: evidence from a rural population cohort 1989–2005

Shafer, Leigh Annea; Biraro, Samuela,b; Nakiyingi-Miiro, Jessicaa; Kamali, Anatolia; Ssematimba, Duncana; Ouma, Josepha; Ojwiya, Amatoa; Hughes, Petera; Van der Paal, Lievea; Whitworth, Jimmyb; Opio, Alexc; Grosskurth, Heinera,d

Author Information
doi: 10.1097/QAD.0b013e32830a7502
  • Free



Uganda was one of the first African countries to report cases of HIV [1]. By 1989, when large-scale tracking of the epidemic including the establishment of two population cohorts had begun [2,3], the HIV epidemic had already reached a generalized level. Largely due to the early response initiated by the Ugandan Government [4], Uganda was among the first countries to achieve a decline in HIV prevalence.

Throughout the 1990s, HIV prevalence and incidence fell in Uganda [5–8] and this was attributed primarily to reductions in risky sexual behaviour [6,9–11]. Neighbouring countries, as well as the international community at large, have looked to Uganda as a model of effective control of the epidemic.

Recently, some researchers noticed that HIV-1 prevalence seems to be levelling off [4]. Until now, however, this evidence has been scanty and has been based on trends in HIV prevalence seen in antenatal clinics (ANCs) and voluntary counselling and testing sites (VCTs). Much research has warned against relying solely on prevalence data to support conclusions about a changing epidemic [12–14].

We examine both prevalence and incidence trends in a rural population cohort in southwest Uganda over 16 years, from 1989 to 2005. We also describe trends in sexual behaviour to assess the possible link between changing sexual behaviour and changing epidemiological trends.


We report prevalence by survey round and incidence by calendar year from a general population cohort study. The study population has been described elsewhere [3,15,16]. Briefly, the study began in 1989 with 15 villages in a rural area in southwest Uganda. In 1999, an additional 10 villages were added to the cohort. Survey rounds began in November and ended in October of the following year. The first survey was performed in 1989/1990 and the 16th in 2004/2005. After obtaining consent, survey staff administered risk factor questionnaires to individuals, after which blood was taken for HIV-1 tests. Interviewers obtained census data through annual household visits.

Statistical methods

Individual HIV status was determined as follows: in a given round, if an individual was tested, the HIV test result was used. If the individual could not be tested and had been classified as seropositive during a previous round, it was assumed that this person was still HIV positive. Similarly, if an individual was classified as seronegative in a subsequent survey round, then it was assumed that the individual was HIV negative. This approach takes advantage of the longitudinal character of the data. We earlier showed that this method yields similar results to the method of using only the serostatus of individuals who could be tested in a given round [7].

Prevalence was the proportion of the population de-jure resident during the survey round, known to be HIV infected. Because the estimation method makes use of data from previous and subsequent rounds, data from rounds 1–18 are used to provide estimates in this paper for rounds 1–16. This results in an estimated prevalence for round 1 that is biased downward. The estimate from round 16 may be slightly biased upward, but inclusion of rounds 17 and 18 data made this bias negligible. Although we do provide an estimate of prevalence from round 17, completed in October 2006, this estimate is provisional. Because we make use of data from subsequent rounds, the round 17 estimates will be more stable once we have completed round 19, due for completion in October 2008.

Although prevalence was calculated by annual survey round, incidence was calculated by calendar year. The mid-date between the last negative and first positive HIV test was estimated as the date of seroconversion. In incidence analyses, all time up to the last HIV test if uninfected, or estimated date of seroincidence if an HIV seroconversion occurred, was included in calculating the person-years at risk (PYAR). This includes time in which the person may have temporarily shifted de-jure residence away from the catchment area, as long as the individual did move back to the area and did receive at least one HIV test after moving back.

To examine the statistical significance of trends, we employed a logistic regression to examine prevalence and a Poisson regression [17] to examine incidence. Both regressions contained two main components: a linear trend with time, and a nonlinear (quadratic) trend with time. This allowed an assessment of whether the trends changed over time. A Huber correction accounted for multiple measurements per individual across rounds.

To assess the need for standardization, we examined the composition of those with a confirmed HIV status by age/sex and survey round. We determined that the population composition did not vary much from round to round and for this reason crude results are presented.

Ethical issues

Survey staff encouraged participants to request their HIV-1 test results from counsellors [3]. In line with Uganda guidelines for HIV testing, results were only issued to respondents in person after pretest and posttest counselling [18]. Residents accessed free medical care from a study clinic, stocked according to the Uganda Essential Drug List. Since 2004, antiretroviral medication was available. The Science and Ethics Committee of the Uganda Virus Research Institute and the Uganda National Council for Science and Technology gave ethical approval for the study.


An average of 6279 adults (age 13 years or more) were censused each year in survey rounds 1–10. In survey rounds 11–16, after 10 villages were added to the cohort, an average of 11 139 adults were censused each year. Across all 16 rounds, 75.8% of censused adults had confirmed HIV-status results in the rounds in which they were censused (Table 1).

Table 1
Table 1:
Census by survey round, sex, and age.


During the first 10 rounds, an average of 4980 adults had a definitive HIV status per round. In rounds 11–16, the average number of adults with a definitive HIV status per round was 8086. Some of these results are imputed, as discussed in ‘Methods’ section. The average number of adults physically tested for HIV was 3722 during rounds 1–10, and 6422 during rounds 11–16.

HIV-1 prevalence declined from 8.5% in the 1990/1991 survey round to 6.2% in 1999/2000, and thereafter rose to 7.7% in 2004/2005. The prevalence estimate from the first round (1989/1990) was 7.2%, and the early provisional estimate from 2005/2006 is 8.0%. As explained above, the first round estimate is known to include a downward bias. Women had higher prevalence than men in all rounds, but the trend did not differ between men and women (see Fig. 1). Overall, prevalence fell during the 1990s, and has been increasing since 1999/2000 (P value: 0.01 for significance of change in trend). The slope of prevalence from 1990 to 2000 was −0.14 per year, whereas the slope from 2000 to 2005 was +0.06 per year. The newly added villages showed a steeper increase in prevalence (5.5% in 2000/2001 and 8.3% in 2004/2005) than the original villages (P value for trend among new villages: <0.001). The prevalence estimate among newly added villages in 1999/2000 was 4.4%. This being the first round among these newly added villages, however, the estimate of 4.4% is biased downward.

Fig. 1
Fig. 1:
HIV prevalence: original and new villages by sex and year.


Between November 1989 and October 2007, 14 449 individuals were HIV negative at first test, had two or more tests, and contributed to the analysis of incidence. The latest incidence figures calculated are from 2005. Note that we do provide an estimate of incidence from 2006, but this most recent incidence estimate is only provisional. Less-stable residents have higher incidence and they are also less likely to be tested in a given round, resulting in a delay in catching the incident case.

Similar to prevalence, incidence fell throughout the 1990s (Fig. 2). It appears that the trends in incidence began to change about 2 years before the corresponding change in prevalence. Between 1998 and 2004, incidence stabilized and showed signs of rising, but the 2005 and (early) 2006 estimates make incidence trends unclear. Although it is about 1 year too early for a stable estimate of incidence for 2006, the early estimate shows an increase over 2005– across all villages, it is 3.7 per 1000 PYAR [95% confidence interval (CI) 2.3–5.8]. Other than more noise due to wider CIs in the newly added villages, the pattern of incidence from the time that the newly added villages joined the cohort was similar in the new and original villages (Fig. 2). In both groups of villages, there was a high incidence estimate in 2002 and a low incidence estimate in 2005 (Fig. 2).

Fig. 2
Fig. 2:
HIV incidence: rural Masaka by village group and year. CI, confidence interval; PYAR, person-years at risk.

Although the point estimate of incidence (per 1000 PYAR) across all villages rose from 4.1 in 1998 to 5.0 in 2004, CIs are wide and the changing trend in overall incidence is not statistically significant (P value 0.47, Poisson regression Wald test). Incidence patterns varied by age and sex, though both men and women experienced a downward trend, followed by an increase in incidence until 2004. Among women, incidence (per 1000 PYAR) fell from 5.9 (95% CI 2.8–12.5) in 1990 to 4.6 (95% CI 2.6–8.4) in 1998, and thereafter rose to 5.8 (95% CI 3.8–8.9) in 2004, with a sudden drop to 2.9 (95% CI 1.6–5.4) in 2005 (see Table 2). Among men, incidence (per 1000 PYAR) fell from 9.3 (95% CI 5.0–17.2) in 1990 to 3.6 (95% CI 1.8–7.2) in 1998, and thereafter rose to 3.9 (95% CI 2.2–6.9) in 2004, with a drop to 2.1 (95% CI 1.0–4.7) in 2005. Some age groups followed this same pattern, though in no case was the change in trends statistically significant.

Table 2
Table 2:
Incidence rates by year, age, and sex across all villages combined.

Figure 3 shows that incidence generally increased until the age of late 20s, and then declined. In the recent time period (2000–2005), there emerge new incidence patterns, albeit based on small numbers. Among men, there occurred a second incidence peak emerging in the 40–44 years age group. Incidence (per 1000 PYAR) in this age group among men was 7.5 (95% CI 2.4–23.2) in 1990–1994 and 7.3 (95% CI 2.8–19.5) in 1995–1999; in 2000–2005, however, the point estimate of incidence rose to 9.6 (95% CI 4.8–19.1). Similarly, among women in the most recent period, we note a mild second peak at age of 30–34 years. These trends and differences in incidence rates were not statistically significant.

Fig. 3
Fig. 3:
HIV incidence by age and sex over time all villages combined. PYAR, person-years at risk.

Sexual behaviour

We examined the following sexual behaviour indicators: casual partners, condom use, pregnancies, and age at sexual debut. Using logistic regression Wald tests with a Huber correction to account for multiple measurements per person across years, very few of the trends in sexual behaviour reached statistical significance (P < 0.05). We describe the data here and in Table 3.

Table 3
Table 3:
Sexual behaviour indicators in the catchment area by year of survey and selected subgroups.

The number of casual partners during the last year was asked on seven occasions. Among all age groups except age 13–19 years, the percentage claiming to have had one or more casual partners rose between 1997 and 2004. There was a dip in 2005, but the older age groups still indicated a larger percentage than in 1997. The only trend that reached statistical significance was that among the age group 35–44 years (P = 0.003). This coincides both with stabilized incidence in the whole adult population from 1998 to 2004, and with the increasing incidence in men 35 years or more. By contrast, among the youngest age group of 13–19 years, the trend towards fewer participants claiming to have had one or more casual partners in the last year was nearly significant (P = 0.069).

Youth in their 20s appear to have decreased their condom use with casual partners. Condom use with casual partners dropped from 74.1% in 1997 to 50.7% in 2005 among those aged 20–24 years (P value 0.057). By contrast, condom use with casual partners may be rising among the older age groups. Among those aged 45 years or more, condom use increased from 10.0% in 1997 to 24.4% in 2005 (P value 0.071).

The percentage of never-married adolescent women (age 15–19 years) who have ever been pregnant fell rapidly throughout the 1990s to 1998, but since 1998 this decreasing trend has stopped. The change in trend of the percentage of never-married adolescent women who have ever been pregnant is statistically significant (P value 0.034).

For both sexes, the median age at sexual debut as reported by 16–19-year olds rose steadily between 1997, the first year in which the age at first sex was asked, and 2005. In most years from 1993 onwards, information was also collected on whether young people ever had sex. Among young men, we see a pattern, which holds true for each age (15, 16, 17, and 18-year olds) separately. That is, among these young men, the percentage of individuals who have ever had sex declined between 1993 and 1997, but increased between 1998 and approximately 2000. Finally, from 2002 onward, the percentage is slightly lower than between 1993 and 1997 and generally stable. This interesting pattern among young men displays a temporary increase in the percentage who have ever had sex, which coincides with the time period during which overall incidence in the population showed a rising trend (Fig. 2) and incidence among young men also had a temporary increase (see Table 2). Among young women, a similar temporary increase in the percentage who ever had sex occurred between 1997 and 2000, but this was observed in two age groups only.



HIV prevalence in Uganda is no longer falling and is beginning to rise at least in some parts of the population. This is confirmed by the prevalence trends within the rural cohort described in this paper, as well as by similar trends seen in the antenatal clinic surveillance system of the Ministry of Health (MOH) [19], and in all four VCT sites of the Uganda-based AIDS Information Center (AIC) [20]. The estimate of prevalence in our cohort in 2004 was 7.2%, and in 2005 it was 7.7%. This compares to the estimate of prevalence in the Western region of Uganda from the 2004/2005 national HIV/AIDS serobehavioural survey of 6.9% [4].

Batter et al.[12] and others have argued that a cohort population may not be representative because research conducted in a cohort may change the actual and reported behaviours in the cohort population. Several factors, however, support our confidence in the fact that trends seen in this cohort are representative of the area in general. Our cohort was established as an observational cohort, without providing interventions that go beyond the efforts of the national HIV control programme.

Shortly after the inclusion of 10 villages in the cohort in 1999, we had established that the prevalences found in the newly added villages were similar to those found in the villages, which had been in the cohort for 10 years [21]. We nonetheless did notice a difference between prevalence trends in the new villages compared with the original villages. In both cases, prevalence increased between 1999 and 2005. In the newly added villages, however, the increase was much steeper. Although an attempt was made to add villages to the cohort which were similar to the original villages, this was not completely possible. One difference is that the newly added villages include the main trading centre of the subcounty. Trading activity and mobility in this area has grown in recent years, which may partly explain the difference in the slope of increasing prevalence between the original and newly added villages since 1999/2000.


HIV incidence was falling from 1990 to about 1998. The trend from 1998 is less clear. It levelled off and even showed a slight rise between 1998 and 2004, but the 2005 estimate of incidence reached an all time low. These results were reported at the XVI International Conference on AIDS in Toronto in August 2006 [19]. Since then, the eighteenth survey round has been completed. These most recent data did not change our assertions with respect to prevalence trends. However, the most recent incidence estimate (2005) became a surprising all time low. This finding should be interpreted with caution. Preliminary data with an early 2006 incidence estimate gives evidence that it is likely that the low 2005 estimate is a result of random fluctuation, similar to that observed for incidence in 2002.

In addition to an apparent levelling off, incidence rose in some subgroups. For example, when comparing three time periods, men aged 40–44 years showed an incidence peak in the most recent period (2000–2005). If these peaks represent a real change in trends of HIV incidence among older men and women, as opposed to random fluctuation, then public health messages and campaigns targeting these subgroups are urgently needed. Interestingly, similar trends among men aged 40–44 years have been observed in an observational cohort in Tanzania (Basia Zaba, personal communication). The AIC also estimated an age shift in incidence towards older ages in recent years in Uganda [20].

Sexual behaviour

Several driving factors may have influenced recent trends in prevalence and incidence. These include a shift towards more risk-taking sexual behaviour, the natural epidemiological cycle, possible changes in migration patterns, and reduced mortality due to improved healthcare. Although the relative contribution of all factors is important in understanding recent trends, sexual behaviour is the one factor which, if its influence on recent epidemiological trends is large, could be modified in a positive way.

Some parameters of sexual behaviour, such as casual partners in older age groups, seem to indicate more risk-taking behaviour in recent years. The higher percentage of young men who have ever had sex between 1998 and 2002 also points towards more risk taking behaviour in recent years. However, this evidence conflicts with other indicators such as the age at sexual debut, which seems to be increasing in recent years.

The evidence of changing sexual behaviour in recent years is ambiguous. It is for this reason that we are currently in the process of studying sexual behaviour in this cohort in detail (S. Biraro, L. Shafer, L. Van der Paal, H. Grosskurth, Trends in sexual behaviour between 1993 and 2005 in a rural Ugandan population cohort where HIV prevalence and incidence are no longer declining, unpublished observation). If sexual behaviour is becoming more risky in this population, an exploration of the reasons for this changing behaviour is paramount in order to advise public health policy. The MOH in Uganda has already compiled a document, in which possible reasons for changing sexual behaviour are listed (Uganda_AIDS_Commission, Accelerating HIV prevention in Uganda: the road towards universal access, unpublished observation). One possibility is that in the perception of the population, HIV has become a normal feature. Although 10 years ago, people may have feared AIDS and taken great measures to avoid it, today people may think of AIDS as simply a normal part of life. If this is true, public health awareness programmes directed specifically at this perception, could ameliorate the problem.

Possible biases

With the possible exception of HIV prevalence in 2005, it is unlikely that the widespread availability of ART has impacted the trends discussed in this paper. MRC-Uganda did not start administering ART to eligible persons in this cohort until the beginning of 2004, well after the epidemiological trends began to change. In 2004, 7.5% of HIV-positive residents in the cohort had started on ART, and by 2005, 15.6% had started on ART. It is therefore likely that the round 16 (2005) estimate of prevalence was partially influenced by reduced mortality due to ART, but unlikely that any of the earlier rounds were influenced. There has been no change in the proportion of de-jure resident individuals who agreed to an HIV test since 2004, so it is unlikely that sick people were more likely to agree to testing as a result of the availability of ART.

Other possible biases to be considered are selective enrolment, and the occurrence of very rapid HIV progression to death. Age standardized mortality among adults who were censused but did not have blood samples taken was higher than the combined mortality rate among HIV-positive and HIV-negative participants (data not shown). This could have been due to a higher HIV-positive rate among those who did not participate in the serosurvey. This bias may have influenced estimated trends, as the percentage of people who agreed to an HIV test has decreased in recent years. If anything, however, the bias would make our estimate of a rising trend in HIV prevalence conservative. Similarly, cohort participants were tested for HIV annually. Incidence estimates are likely to be slightly biased downward because some rapid HIV progressors may have become infected and died before their infection could have been detected.

In summary, we documented that HIV-1 prevalence and incidence decreased in this rural surveillance cohort throughout the 1990s. In recent years, prevalence has levelled off and showed a rising trend, an observation that is consistent with other sources in Uganda. For incidence, the picture is less clear. Incidence declined during the 1990s. Other than the 2005 data point, incidence has appeared to level off since 1998. Factors influencing the recent trends of the epidemic are not yet clear, but there are indications that the observed changes in trend may be partly explained by increased sexual risk behaviour. Uganda has been highly successful in controlling the HIV epidemic in the country in the past. To solidify this success, the ongoing efforts in HIV prevention need to be re-strengthened, using all strategies known to reduce HIV transmission.


This study was funded by the Medical Research Council (MRC) and the Department for International Development (DFID) of the UK. We thank the study population, field staff, and administrative support. We also thank the Alpha network, funded by the Wellcome Trust, for their workshops during which some ideas were contributed for analyses. We thank the Uganda Ministry of Health for their support.

L.A.S. and J.O. conducted statistical analysis. J.O. and D.S. provided data management for the longitudinal data from the cohort from which data for this paper come. S.B. is the project leader of the cohort and manages day-to-day activities of data collection for the cohort. J.N.-M. was the head of the statistical unit during the time that data were collected and analysed for this study; she contributed to analysis. A.K. was the project leader of the cohort during the 1990s, when the cohort was still in its early stages of development. P.H. and A.Ojwiya managed the laboratory and helped to assess HIV status when the ELISA and/or western blot test results were ambiguous. From 2001 to 2007, L.VdP. contributed to study design and analysis as the epidemiologist in charge of the observational cohorts between 2001 and 2007. J.W. was the director of the MRC Unit in Uganda between 1994 and 2002, and H.G. has been director since 2002. Both directors contributed greatly towards questionnaire development in the cohort and data collection practice. A.Opio provided an assessment of HIV results from Ministry of Health antenatal clinics throughout Uganda, which ultimately led to the development of this paper on prevalence and incidence trends in the MRC/UVRI cohort. L.A.S. conducted the literature review and wrote the first draft of this paper. All authors have contributed to subsequent drafts of the paper until the final version.


1. Serwadda D, Mugerwa RD, Sewankambo NK, Lwegaba A, Carswell JW, Kirya GB, et al. Slim disease: a new disease in Uganda and its associations with HTLV III infection. Lancet 1985; 2:849–852.
2. Serwadda D, Wawer MJ, Musgrave SO, Sewankambo NK, Kaplan JE, Gray RH. HIV risk factors in three geographic strata of rural Rakai District, Uganda. AIDS 1992; 6:983–989.
3. Seeley J, Wagner U, Mulemwa J, Kengeya-Kayondo J, Mulder D. The development of a community-based HIV/AIDS counselling service in a rural area in Uganda. AIDS Care 1991; 3:207–217.
4. MOH-Uganda. Uganda HIV/AIDS sero-behavioural survey 2004–2005. Calverton, Maryland, USA: Ministry of Health (Uganda) and ORC Macro; 2006.
5. Mulder D, Nunn A, Kamali A, Kengeya-Kayondo J. Decreasing HIV-1 seroprevalence in young adults in a rural Ugandan cohort. Br Med J 1995; 311:833–836.
6. Kamali A, Carpenter L, Whitworth J, Pool R, Ruberantwari A, Ojwiya A. Seven-year trends in HIV-1 infection rates, and changes in sexual behaviour, among adults in rural Uganda. AIDS 2000; 14:427–434.
7. Mbulaiteye S, Mahe C, Whitworth J, Ruberantwari A, Nakiyingi JS, Ojwiya A, 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.
8. Nunn AJ, Mulder DW, Kamali A, Kengeye-Kayondo JF, Whitworth JAG. HIV-1 incidence in sub-Saharan Africa. Lancet 1997; 348:833–834.
9. Asiimwe-Okiror G, Opio AA, Musinguzi J, Madraa E, Tembo G, Carael M. Changes in sexual behaviour and decline in HIV infection among young pregnant women in urban Uganda. AIDS 1997; 11:1757–1763.
10. Kirungi WL, Musinguzi J, Madraa E, Mulumba N, Callejja T, Ghys P, et al. Trends in antenatal HIV prevalence in urban Uganda associated with uptake of preventive sexual behaviour. Sex Transm Infect 2006; 82(Suppl 1):i36–i41.
11. Kilian AHD, Gregson S, Ndyanabangi B, Walusaga K, Kipp W, Sahlmüller G, et al. Reductions in risk behaviour provide the most consistent explanation for declining HIV-1 prevalence in Uganda. AIDS 1999; 13:391–398.
12. Batter V, Matela B, Nsuami M, Manzila T, Kamenga M, Behets F, et al. High HIV-1 incidence in young women masked by stable overall seroprevalence among childbearing women in Kinshasa, Zaire: estimating incidence from serial seroprevalence data. AIDS 1994; 8:811–817.
13. Zaba B, Boerma T, White R. Monitoring the AIDS epidemic using HIV prevalence data among young women attending antenatal clinics: prospects and problems. AIDS 2000; 14:1633–1645.
14. Wawer MJ, Serwadda D, Gray RH, Sewankambo NK, Li C, Nalugoda F, 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.
15. Mulder D, Nunn AJ, Kamali A, Nakiyingi J, Wagner H-U, Kengeya-Kayondo JF. Two-year HIV-1-associated mortality in a Ugandan rural population. Lancet 1994; 343:1021–1023.
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. Kengeya-Kayondo J-F, Kamali A, Nunn AJ, Ruberantwari A, Wagner H-UH, Mulder DW. Incidence of HIV-1 infection in adults and socio-demographic characteristics of seroconverters in a rural population in Uganda: 1990–1994. Int J Epidemiol 1996; 25:1077–1082.
18. MOH-Uganda. Uganda national policy on HIV counselling and testing. Kampala, Uganda: Ministry of Health (Uganda); 2005.
19. Shafer L, Biraro S, Kamali A, Grosskurth H, Kirungi W, Madraa E, et al. HIV epidemic trends in Uganda 1989–2005. In: Oral presentation at XVI International AIDS Conference Toronto, 2006 [abstract THLB0108].
20. Baryarama F, Bunnell R, McFarland W, Hudes ES, Neilands TB, Ransom RL, et al. Estimating HIV incidence in voluntary counseling and testing clients in Uganda. J Acquir Immune Defic Syndr 2007; 44:99–105.
21. Mbulaiteye S, Mahe C, Ruberantwari A, Whitworth J. Generalizability of population-based studies on AIDS: a comparison of newly and continuously surveyed villages in rural southwest Uganda. Int J Epidemiol 2002; 31:961–967.

Africa; AIDS; HIV; incidence; prevalence; sexual behaviour; trends; Uganda

© 2008 Lippincott Williams & Wilkins, Inc.