Objective: To assess mortality impact of HIV in rural Uganda.
Methods: An open cohort of 19 983 adults aged 15–59 years, in Rakai district was followed at 10 month intervals for four surveys. Sociodemographic characteristics and symptomatology/disease conditions were assessed by interview. Deaths among residents and out-migrants were identified household census. Mortality rates were computed per 1000 person years (py) and the rate ratio (RR) of death in HIV-positive/HIV-negative subjects, and the population attributable fraction (PAF) of death were estimated according to sociodemographic characteristics. Mortality associated with potential AIDS defining symptoms and signs was assessed.
Results: HIV prevalence was 16.1%. Mortality was 132.6 per 1000 py in HIV-infected versus 6.7 per 1000 py in uninfected subjects, and 73.5% of adult deaths were attributable to HIV infection. Mortality increased with age, but the highest attributable risk of HIV associated deaths were observed in persons aged 20–39 years (PAF > 80%) and in women. HIV associated mortality was highest in the better educated (PAF ≥ 75%) and among government employees (PAF ≥ 82%). Of the HIV-positive subjects 40.5% reported no illness < 10 months preceding death, symptoms were poor predictors of death (sensitivity 1.6–38.8%), and only 9.1% met the World Health Organization clinical definition of AIDS. Infant mortality rates in babies of HIV-infected and uninfected mothers were 209.4 and 97.7 per 1000, respectively.
Conclusion: HIV is taking substantial toll in this population, particularly among the younger better educated adults, and infants. Symptomatology or the World Health Organization definition of AIDS are poor predictors of death.
From the Department of Medicine and Clinical Epidemiology Unit, Faculty of Medicine, Makerere University, Kampala, Uganda, the aDepartment of Population and Family Health Sciences, Johns Hopkins University, School of Hygiene and Public Health, Baltimore, Maryland, USA, the bInstitute of Public Health, Faculty of Medicine, Makerere University, Kampala, Uganda, the cRakai Project, Uganda Virus Research Institute, Entebbe, Uganda, and the dCenter for Population and Family Health, Columbia University, New York, NY, USA.
Received: 11 November 1999;
revised: 24 May 2000; accepted: 8 June 2000.
Sponsorship: Supported by grants RO1 AI34826 and RO1 AI34826S, National Institutes of Allergy and Infectious Diseases, and grant 5P30HD0626, National Institute of Child Health and Development, Fogarty International Center (D43TW00010), US National Institutes of Health; the Rockefeller Foundation; the World Bank Uganda Sexually Transmitted Infections (STI) Project, and John Snow Inc. grant 5024-30.
Requests for reprints to: N. K Sewankambo, Faculty of Medicine, Makerere University, P.O. Box 7072, Kampala, Uganda.
There is increasing evidence of the spread and magnitude of the HIV epidemic in sub-Saharan Africa  and measurement of the contribution of HIV to mortality is of fundamental importance for estimating the current and future impact of the epidemic on populations. It is more than 15 years since the existence of AIDS in Uganda was first documented , and the impact of HIV infection on mortality was reported from rural populations in Rakai and Masaka districts of southwestern Uganda, during the early 1990s [3,4]. However, as the epidemic matures, death rates and age-specific mortality may change, depending on the average duration of infection among prevalent HIV-positive individuals, so there is need to examine more recent trends in mortality.
There is also a need to assess the prognostic utility of symptoms and signs in HIV-infected individuals as a guide for clinical management. Various studies have attempted to determine the natural history of HIV infection using the World Health Organization (WHO) clinical definition of AIDS  and the WHO staging system for HIV infection . However, there are problems with the use of these classification systems in resource-poor environments or under field conditions. The WHO case definition identifies the more advanced AIDS cases and fails to detect most people with HIV disease. Lindan et al., in a study of predictors of mortality among HIV-infected women in Rwanda, questioned the utility of WHO clinical definition and showed that it lacked prognostic value. They proposed use of mortality as a more precise measure for evaluating the effect of the epidemic in Africa. The WHO staging system is difficult to implement because of the need for detailed clinical information based on laboratory diagnoses, especially for stages 3 and 4 .
The present study examines HIV associated mortality in the late 1990s using data derived from community-based studies conducted between November 1994 and December 1997. We also examine mortality in relation to symptoms and signs to assess the prognosis in HIV-infected individuals in a field setting.
A community randomized trial was conducted of sexually transmitted disease (STD) control for AIDS prevention in Rakai District, a rural area in southwestern Uganda and the design and results of this study have been described previously [8,9]. In brief, all households in 56 communities located on secondary roads were mapped and enumerated by censuses, and consenting residents, defined as persons living in the community for 6 months or more, were enrolled. The communities were aggregated into 10 clusters which were randomly assigned to the STD intervention and control arms. The STD intervention consisted of mass antibiotic treatment administered to all consenting subjects irrespective of the presence or absence of STD symptomatology, and the control arm received anti-helminth drugs and vitamin supplements. This was an open cohort and new subjects were enrolled at each survey round conducted at intervals of 10 months. The average follow up rate was approximately 75%. All contact (interview and sample collection) occurred in the respondents’ homes.
At each home visit a household census was repeated to record, births, deaths and in- or out-migration. In this open cohort, newly in-migrating subjects were enrolled at each study visit. There were 11 571 persons enrolled at the first study round, 4171 newly enrolled at the second round and 4241 enrolled at the third study round, giving a total of 19 983 subjects. The potential duration of follow up ranged from 10 months for those enrolled in the third study round to 30 months for those enrolled in the first study round. Consenting subjects aged 15–59 years were interviewed providing information on sociodemographic, behavioral and health characteristics. The health information included symptoms suggestive of AIDS at time of interview or over the 6 months prior to interview. These symptoms included substantial weight loss, fever, diarrhea or cough persisting for more than 1 month, and generalized pruritic rash. In addition, field workers recorded a history and physical exam for Kaposi's sarcoma, herpes zoster, oral candidiasis, lymphadenopathy and tuberculosis. No physician-based diagnoses or clinical investigation was available for specific conditions such as opportunistic infections.
Subjects were asked to provide a venous blood sample for HIV-1 testing using two enzyme linked immunoassays (EIA) (Vironostika HIV-1; Organon Teknika, Charlotte, North Carolina, USA and Cambridge Biotech, Worcester, Massachusetts, USA). Persons with concordant EIA results were considered to be HIV positive, and those with discordant EIA results were tested further by Western-blot using HIV-1 WB (Bio-Merieux-Vitek, St Louis, Missouri, USA. For persons who declined to provide blood, or whose sample was insufficient for HIV testing, a urine EIA (Calypte HIV-1 urine EIA; Calypte Biomedical, Alameda, California, USA) was used with Western-blot confirmation . Over the course of the study approximately 10% of HIV results were derived from urine testing. These data provided information on HIV infections at enrollment and during follow up.
HIV prevalence was assessed at time of first enrollment (i.e. either at the first baseline survey or among newly enrolled subjects at later survey rounds). The repeat household census data were used to identify deaths that occurred within each 10 month follow up interval, and to estimate the person-years (py) exposure to risk of death. It was assumed that on average, deaths occurred in the mid-point of each follow up interval, and persons who died were accorded 5 months of exposure to risk during the interval in which the death occurred. Subjects who out-migrated were treated as censored observations at the time they were last reported to be alive or to have died by the census respondent. If no information on the vital status of out-migrants was available, and py exposure could not be estimated, these subjects were considered as lost to follow up, and were dropped from the analyses after the last interview. Death rates among out- and in-migrants and non-migrants were also examined as separate subgroups to assess the effects of selective migration on mortality. Mortality rates were estimated per 1000 py of observation for HIV-positive and HIV-negative individuals in the whole population, and for subgroups by age, sex, marital status, education, occupation and migration status . The rate ratios (RR) and 95% confidence intervals (CI) of deaths among HIV-positive relative to HIV-negative subjects were computed to estimate the relative excess of deaths associated with HIV infection. In addition, the population attributable fraction of death (PAF) associated with HIV-infection was estimated to assess the contribution of HIV to mortality in the whole population or in subgroups of interest. The population attributable fraction was based on the proportion of deaths of HIV-positive subjects to all deaths in the population or subgroup (Pd) and the RR of the risk of death in HIV-positive : HIV-negative subjects [PAF% = Pd (RR – 1/RR) × 100]. The absolute excess of mortality among HIV-infected and non-infected persons was also examined.
Maternal mortality associated with HIV was estimated in 2997 mothers who were pregnant during the course of the investigation. Infant mortality was estimated from deaths under 1 year of age, among live born infants, ascertained by interviews with the mothers or guardians during subsequent follow up visits.
Due to the lack of clinical diagnoses the WHO Clinical Staging of HIV infection/AIDS could not be used. However, it was possible to apply the WHO clinical definition of AIDS based on the presence of two or more major symptoms (weight loss, prolonged diarrhea or fever), and one or more minor signs (cough, rash, candidiasis, herpes zoster) or the presence of Kaposi's sarcoma. To ascertain whether symptoms were predictive of death, the proportions of HIV-positive subjects who died during the 10 month follow up interval subsequent to interview was examined. Symptoms were assessed individually and by the number of reported symptoms (none, one, two, three, four or more). We evaluated the sensitivity and specificity of these symptoms with respect to death, relative to subjects reporting no symptoms. The predictive value positive of death within 10 months was estimated from the proportions of symptomatic persons who died within 10 months of interview.
Tests of statistical significance of mortality in HIV-positive versus HIV-negative subjects were based on the 95% CI of the RR, and by chi-square for trend. For comparisons with other studies direct standardization was used to estimate age-standardized mortality rates using the age composition of the current Rakai cohort as the standard.
Data were available for 19 983 subjects of whom 3210 (16.1%) were HIV-positive. These subjects provided a total of 31 432 py of observation. There were a total of 795 deaths with an overall adult mortality rate of 25.3 per 1000 py.
Table 1 shows HIV prevalence, and death rates per 1000 py for HIV-positive and HIV-negative subjects. HIV prevalence in the whole population increased with age until 30–39 years and decreased thereafter. The mortality rate among HIV-positive subjects was substantially greater than among HIV-negative subjects (132.6 and 6.7 per 1000 py, respectively; RR, 19.7; 95% CI, 16.7–23.4), and 73.5% of deaths in adults aged 15–59 years were attributable to HIV infection. The absolute excess mortality in HIV-positive versus HIV-negative persons was 125.9 per 1000. Death rates were lowest among adolescents aged 15–19 years, but mortality was increased in HIV-infected adolescent population compared to the HIV-uninfected, despite the fact that the HIV-positive adolescent presumably had more recent HIV infections. The RR of mortality in HIV-positive : HIV-negative adolescents was 11.3 and the population attributable fraction of deaths associated with HIV in this youngest age group was 24.8%. Among persons aged 20 years or older, mortality rates generally increased with age, but the highest relative and population attributable fraction of death due to HIV was observed among persons aged 20–29 years (RR, 32.3; 95% CI, 22.4–49.3; PAF, 85.6%) and 30–39 years (RR, 16.1; 95% CI, 9.90–18.63; PAF, 77.8%). The largest absolute excess of mortality was in the 40–49 year age group. Male mortality exceeded female mortality both in the HIV-infected and uninfected populations and within each age group (with the exception of HIV-negative persons aged 15–29 years). However, the attributable fraction of death associated with HIV showed different age-specific patterns in men and women. Among men, the population attributable fraction of mortality due to HIV were lower in males over the age of 40 years, whereas among women HIV the attributable fraction of mortality was high at all ages and exceeded that of males within each age group. Nevertheless, in both sexes, the absolute excess mortality associated with HIV was observed in persons over age 40 years.
Duration of infection among prevalent HIV-positive subjects is not known. However, there were 175 HIV-negative subjects who seroconverted during follow up between the first and third study rounds, among whom mortality could be determined over the four study rounds. These subjects provided 167 py of observation subsequent to seroconversion, and there were four deaths giving a mortality rate of 24.0 per 1000 py, which exceeded the death rate of 6.7 per 1000 py in HIV-negative non-seroconverting subjects (RR, 3.50; 95% CI, 1.31–9.32). Assuming that the seroconversions occurred on average at the mid-point of each 10 month follow up interval, the average duration of observation was approximately 11.5 months subsequent to seroconversion Thus, within an average period of less than 1 year's exposure following infection, there was an excess death rate in these newly infected HIV-positive persons.
Mortality and socioeconomic status
Table 2 shows mortality by social characteristics. The attributable fraction of HIV associated death was modest among the never married (PAF, 54%), compared to currently married, separated or divorced (PAF, 72%). Mortality was particularly high among HIV-positive widowed subjects, as was the population attributable fraction of mortality due to HIV (PAF, 89.6%). The RR of death in the HIV-positive : HIV-negative subjects was increased among those with higher education. Thus, the attributable fraction of HIV associated death was greater among the better educated compared with the uneducated, indicating the toll taken by HIV on the more educated segments of this population. By occupation, the RR of mortality in HIV-positive : HIV-negative subjects was greatest for persons working in the military/police and for students (RR, 36.2; 95% CI, 18.81–65.93 and RR, 66.3; 95% CI, 22.02–101.40, respectively). However, the attributable fraction of HIV associated death showed a different pattern, with the lowest attributable risk (AR, 40.3%) observed among students (due to the relatively low proportion of HIV-infected persons in this occupational category), and the highest attributable fraction in government/technical employees and military/police (PAF, 82.7% and 87.0%, respectively).
Mortality was also examined by study randomization arm. In the intervention arm, mortality of HIV-infected persons was 127.0 per 1000 py (313/2467 py) compared with 6.1 per 1000 py in HIV-negative subjects (84/13 682 py), with a RR of 20.8 (95% CI, 15.04–24.07). In the control arm the death rates were 139.0 per 1000 py in HIV-positive (302/2172 py) versus 7.3 per 1000 py in the HIV-negative population (96/13 113 py) with a RR of 19.0 (95% CI, 13.49–21.07).
Maternal and infant mortality
There were seven maternal deaths in 415 HIV-positive mothers and eight maternal deaths in 2582 HIV-negative pregnant or parturient women giving maternal mortality ratios of 1686.8 and 309.8 per 100 000 births respectively, in the in the HIV-infected and uninfected mothers (RR, 5.44; 95% CI, 1.98–14.93). Among 339 infants born to HIV-positive mothers, 71 died within the first year of life (infant mortality rate 209.4 per 1000 live births) and among 2303 infants born to HIV-negative mothers, there were 225 deaths during the first year of life (infant mortality rate 97.7 per 1000 live births). The RR of infant mortality in children born to HIV-positive relative to HIV-negative mothers was 2.14 (95% CI, 1.68–2.73).
Mortality and migration
As noted in the methods section, this was an open cohort which enrolled new in-migrants at each survey round. To assess mortality within a closed cohort, analyses was conducted of those permanent residents first seen at baseline (round 1) who were observed again after 30 months at the fourth follow up round. In this latter closed cohort, the death rate in HIV-positive subjects was 134.4 per 1000 py (476/3540 py) compared with 7.2 per 1000 py in HIV-negative subjects (138/19 310). These rates are similar to those in the open cohort data shown in Table 1 (132.6 per 1000 py in the HIV-positive subjects and 6.7 per 1000 py in the HIV-negative subjects), which suggests that in-migrants have comparable risk of death to non-mobile residents.
Persons who left the community were tracked by questioning the head of household about the survival or death of absent household members. Of the 3210 HIV-positive subjects 1037 (32.3%) left the household compared with 5451 out of 16 773 HIV-negative subjects (32.5%). Thus out-mobility was similar in HIV-infected and uninfected persons. The heads of the household were able to report the survival or death (and approximate date of death) for 206 HIV-positive out-migrants and 559 HIV-negative out-migrants. Among persons who left the community, and whose vital status was ascertained by proxy reports, mortality in the HIV-positive out-migrants was 267.4 per 1000 py (77/288 py), and in the HIV-negative out-migrants mortality was 33.5 per 1000 py (30/895 py); the RR was 6.50 (95% CI, 5.11–9.56). Thus, mortality among out-migrants was higher than among non-migrants and in-migrants, irrespective of HIV status, which might indicate selective out-mobility associated with illness, but mortality of HIV-positive out-migrants exceeded that of HIV-negative out-migrants. To assess the health status of out-migrants symptomatology reported by the individual at the visit prior to out-migration was also examined. The proportion of persons reporting one or more symptoms was 19.5% among 6488 out-migrants compared with 15.5% in 13 495 non-migrants, suggesting a modest excess of illness among out-migrants in this cohort.
Mortality and symptomatology
To determine prognostic indicators, reports of symptomatology for 3210 HIV-positive subjects were examined to assess the symptoms predictive of death during the succeeding 10 month following interview (Table 3). Among HIV-positive persons with no symptoms at interview, 12.2% died within the subsequent 10 months, and these asymptomatic subjects were used as the reference group for estimating the relative risk of death associated with symptomatic illness, as well as the sensitivity and specificity of symptoms with respect to death or survival. It is noteworthy that 40.5% (249/615) of all HIV-positive subjects who died reported no symptomatology prior to their demise. The proportions dying (i.e. positive predictive value), increased with increasing number of reported symptoms from 25.1% with one symptom to 73.1% among those reporting four or more symptoms (P = 0.0001, χ2 for trend). Specific symptoms were associated with an increased risk of death, and the range of positive predictive values was 36.7–72.2%, but the sensitivity of these symptoms was relatively poor (range, 1.6–38.8%). The proportion dying was highest among persons meeting the WHO criteria for AIDS (80.0%), but the sensitivity of the WHO criteria was low (18.4%). Moreover, only 56 out 615 deaths (9.1%) met the WHO criteria. Although the number or nature of symptomatic illnesses were insensitive, the specificity (i.e. absence of symptoms) as predictors of survival was high (range 87.7–99.8%).
Fig. 1 and Fig. 2 show mortality associated with symptomatology in HIV-positive and HIV-negative individuals. Mortality rates increased with the number of reported symptoms or conditions in both HIV-positive and HIV-negative subjects, but irrespective of the number of symptoms/conditions present, the mortality was consistently higher among the HIV-infected persons. Fig. 2 shows mortality associated with specific symptoms. Death rates were highest among HIV-positive persons with weight loss, chronic diarrhea and chronic cough, Kaposi's sarcoma, thrush and tuberculosis. Death rates in HIV-positive symptomatic persons consistently exceeded those in HIV-negative symptomatic persons, and differentials were most marked (RR > 20) for prolonged fever, diarrhea, rash and lymphadenopathy. In contrast, the RR of mortality in HIV-positive : HIV-negative subjects were lowest for diagnoses of Kaposi's sarcoma and tuberculosis. (There were no deaths in HIV-negative subjects with thrush or herpes zoster infection, so RR could not be estimated).
Specific symptoms associated with mortality in HIV-infected persons were also assessed. Relative to asymptomatic subjects, mortality was increased among those reporting any symptom. Mortality exceeded 50% among persons reporting weight loss (57.9%), prolonged diarrhea (72.2%) or prolonged cough (54.6%), thrush (60.9%), Kaposi's sarcoma (57.1%) and tuberculosis (69.2%). However, the prevalence of these symptoms was relatively low (< 10%) in the overall HIV-positive population, and symptoms were reported by 0.7–22.0% of those who subsequently died. Among persons meeting the WHO definition of AIDS (two major and at least one minor symptom/sign or Kaposi's sarcoma), 80% died within the subsequent 10 months, but only 2.2% of the population and 9.1% of those who died met the WHO definition. Thus, in this representative rural population, in which individuals are not self-selected by health care seeking, the majority of HIV-positive subjects and HIV associated deaths have no antecedent symptoms suggestive of terminal HIV 0–10 months prior to death.
Rakai district, Uganda is facing a mature, generalized HIV epidemic with an HIV prevalence of 16% among adults aged 15–59 years. An overall mortality among HIV-infected persons of 132.6 per 1000 py was observed in a population of 3210 HIV-infected subjects during the period 1994–1998. To the best of our knowledge, these data represent the largest cohort of HIV-positive subjects observed in eastern Africa, and the only published information on mortality from a mature HIV epidemic during the mid- to late 1990s. The present mortality rates in HIV-infected subjects are higher than estimates from our earlier studies in Rakai conducted during 1990–1991, in which the mortality rate among HIV-infected persons was 114.7 per 1000 py . The current estimates are also higher than in a rural population in Masaka district to the north of Rakai with an HIV prevalence of 8%, in which the death rate was 115.9 per 1000 py during the period 1989–1992 (age range 15–54 years) . The present estimates are higher, although compatible with a more recent analysis of Masaka data from a closed cohort followed between 1989 and 1995, which reported an adult death rate of 124 per 1000 py . Direct comparisons between these cohort studies are constrained by differences in population sampling, the age range of enrolled subjects and the age groups reported in the published literature. The age-standardized mortality rate for HIV-positive subjects aged 13–54 years in the Masaka cohort observed in 1989–1992, was 112.4 per 1000 py. In our previous Rakai cohort study conducted in 1990–1991, the age-standardized mortality rate was 119.3 per 1000 py. Both of these estimates are less than the lower 95% CI of the age-standardized Rakai mortality from 1994 to 1998 (95% CI 122.8–142.4 per 1000 py), and indicate that age-standardized mortality among prevalent HIV-positive persons may have increased during the 1990s in these rural areas of southwestern Uganda. This suggests that as the epidemic matures over time the average duration of prevalent HIV infections may be longer and HIV-positive subjects may, on average, be at a later stage of infection and thus have a higher probability of dying. This is consistent with demographic projections of mortality during the evolution of the HIV epidemic .
The mortality in these Ugandan cohorts is, higher than that reported among HIV-positive subjects in two northern Tanzanian studies. In the Mwanza region of Tanzania, HIV prevalence was 4%, and mortality between 1992 and 1994 was 93.5 per 1000 py (age range, 15–54 years) . In Kisesa, HIV prevalence was 8%, and mortality in 1995–1996 was 72.8 per 1000 py (age range, 15–44 years) . This suggests that the HIV epidemic in northern Tanzania may be at an earlier stage than that in Uganda, and that the mean duration of HIV infection among HIV-prevalent cases may be shorter in these former populations. The mortality rates observed in representative community-based studies from east Africa are substantially higher than those reported in follow up of antenatal clinic attendees (range, 35.0–50.1 per 1000 py), [7,17] and selected factory populations (19.1 per 1000 py) . These latter populations are likely to be unrepresentative with respect to age and health status, and appear to underestimate the magnitude of HIV associated mortality.
In all of the African cohorts, mortality in the HIV-positive subjects exceeded that in the HIV-negative subjects substantially, but the RR of mortality in HIV-infected : uninfected persons were higher in the present Rakai study (RR, 19.8) than in the lower prevalence populations (RR, 11.7 in Masaka ; RR, 15.6 in Mwanza ; RR, 9.5 in Kisesa ). The population attributable fraction of death due to HIV among adults was also high in the present study and did not differ from our previous Rakai study  (PAF, 73.5% and 74%, respectively), but exceeded the population attributable fractions reported from other investigations in lower HIV prevalence settings (69% in Masaka ; 35% in Mwanza ; 47% in Kisesa ). Thus, as the epidemic progresses, it is likely that HIV will contribute an increasing proportion of total mortality and lead to major increases in death rates in affected regions. Such increases in adult deaths and reductions in life expectancy have been observed in demographic studies from eastern and southern Africa [19,20]. The African experience stands in stark contrast to that of the USA which has reported a decline in AIDS mortality since 1996  attributed to the use of combination antiretroviral therapies and treatment of opportunistic infections.
In the present study, male mortality exceeded female mortality in both HIV-positive and negative persons (Table 1), but the age-specific attributable risk of death due to HIV among women consistently exceeded the attributable risk in males. Our finding of a generally increasing mortality with age when stratified by sex is in agreement with other studies suggesting that older age at infection may be associated with more rapidly progressive disease [13,22]. With regard to socio-demographic characteristics, mortality in the HIV-infected was highest among divorced, separated or widowed persons. As indicated by the attributable risk estimates in Table 2, HIV is taking a heavy toll among the better educated and among persons employed in government service. The loss of the most highly skilled members of the society will have a substantial impact on future economic development. However, these socio-demographic differentials in mortality reflect the epidemiology of HIV infection over the past decade, and the differentials are likely to diminish over time as the HIV epidemic becomes more generalized throughout the population.
Cohort studies present a unique opportunity to assess the effects of HIV on mortality and open cohorts allow estimation of the net effects among stable residents as well as in- or out-migrants. The Rakai study was an open cohort which enrolled consenting in-migrants at each survey round and tracked out-migrants by questioning the head of households about the survival of household members who had left on a temporary or permanent basis. This allowed evaluation of how mobility affected mortality estimates by measuring differential migration of selectively healthy or sick individuals, differential mobility of HIV-positive and HIV-negative persons, or differential mortality among migrants and non-migrants. No excess of mortality was observed among new in-migrants which suggests that in these secondary road trading center communities of Rakai, selective return of seriously ill persons is not a major issue. In contrast the Masaka study observed higher mortality, particularly among males, newly entering rural villages . Other studies have not, to our knowledge, assessed mortality of out-migrants. We observed an excess mortality in both HIV-infected and uninfected people who out-migrated from the Rakai cohort, which suggests that sick persons may leave these trading villages in search of health care, or return to their natal communities for terminal care. This is compatible with the findings from the Masaka study which primarily included rural villages and suggest that ill persons may be more likely to return to their families located in such rural communities. Although dissaggregated analyses by migration status are informative, in balance, we believe that analyses of open cohorts provide more useful information of the net effects of HIV on mortality in representative populations.
It was not possible to assess the effects of WHO clinical staging on mortality in this field study, due to the lack of specific clinical diagnoses needed to define WHO stage 4 (e.g. cryptosporidial diarrhea, extra-pulmonary tuberculosis, cryptoccocal meningitis and HIV encephalopathy etc.) . However, it was possible to ascertain the presence of symptoms or signs suggestive of AIDS at the interview prior to risk of death, and it was found that the proportions dying (i.e. positive predictive value) increased with the number of reported symptoms, and was particularly high among those persons meeting the WHO criteria for AIDS (Table 3). However, it is noteworthy that 40.5% of deaths in HIV-positive persons occurred among individuals who were asymptomatic and free of identifiable opportunistic infection at time of interview within 0–10 months preceding death. Other investigations in Uganda also report that a high proportion of HIV-infected subjects are asymptomatic within 1 year prior to death . This suggests a rapid progression of illness, compatible with findings from a natural history study which found a 9 month survival time between onset of AIDS defining conditions and death . Lack of access to adequate health care services, the unavailability of antiretroviral therapy or treatment of opportunistic infections and the prevalence of other, endemic, acute or chronic diseases may contribute to the short survival times in these rural African populations. As indicated in Fig. 1, mortality rates among the HIV-positive subjects increased significantly among persons reporting one or more symptoms or conditions at interview. Nevertheless, only a minority of HIV-positive persons reported multiple symptoms/conditions, and these were associated with approximately one-third of the deaths. Mortality associated with specific symptoms was particularly high in persons reporting chronic diarrhea, severe weight loss, oral candidiasis and tuberculosis and in those persons who met the WHO criteria for clinical AIDS  (Fig. 2, Table 3). However, no particular symptoms or signs, were highly predictive of mortality in the HIV-positive as compared with the HIV-negative populations, which probably reflects the high background levels of these conditions in the HIV-negative population of rural areas [24,25].
In summary, mortality due to HIV is substantial in this rural African population with a mature generalized HIV epidemic. HIV associated mortality is the major cause of deaths among adults aged 15–59 years, particularly among women, and is differentially depleting these communities of the most educated and skilled members.
The authors thank the Rakai Project Staff for exceptional work, S. Sempala, Director, Uganda Virus Research Institute, for his support and encouragement and the people of Rakai for their generous cooperation.
1. UNAIDS/WHO. Report on the Global HIV/AIDS Epidemic
. Geneva: World Health Organization; 1998.
2. Serwadda D, Mugerwa RD, Sewankambo NK. et al. Slim disease, a new disease in Uganda and its association with HTLV III infection.
Lancet 1985, ii: 849 –852.
3. Mulder DW, Nunn J, Kamali A, Nakiyingi J, Wagner H, Kengeya-Kayondo JF. Two-year HIV-1-associated mortality in a Ugandan rural population.
Lancet 1994, 343: 1021 –1023.
4. Sewankambo NK, Wawer MJ, Gray RH. et al. Demographic impact of HIV infection in rural Rakai District, Uganda: results of a population-based cohort study.
AIDS 1994, 8: 1707 –1713.
5. World Health Organization. Acquired immunodeficiency syndrome (AIDS).
:Workshop on AIDS in Central Africa, Bangui, 22–25 October 1985.
Weekly Epidemiol Rec 1985, 60: 342. 342.
6. The WHO International Collaborating Group for the Study of the WHO Staging System. Proposed World Health Organization Staging System for HIV Infection and Disease: preliminary testing by an international collaborative cross-sectional study.
AIDS, 1993, 7: 711 –718.
7. Lindan CP, Allen S, Serufilira A. et al. Predictors of mortality among HIV-infected women in Kigali, Rwanda.
Ann Intern Med. 1992, 116: 320 –328.
8. Wawer MJ, Gray RH, Sewankambo NK. et al. A randomized, community-based trial of intense sexually transmitted disease control for AIDS prevention, Rakai, Uganda.
AIDS 1998, 12: 1211 –1225.
9. Wawer MJ, Sewankambo NK, Serwadda D. et al. Control of sexually transmitted diseases for AIDS prevention in Uganda: a randomised community trial.
Lancet 1999, 353: 525 –535.
10. Meehan MP, Sewankambo NK, Wawer MJ. et al. Sensitivity and specificty of HIV-1 testing of urine compared with serum specimens: Rakai, Uganda.
Sex Transm Dis 1999, 26: 590 –592.
11. Cox PR. Demography, 4th edn
. Cambridge: Cambridge University Press; 1970.
12. Rockhill B, Newman B, Weinberg C. Use and misuse of population attributable fractions.
Am J Public Health. 1998, 88: 15 –19.
13. Nunn AJ, Mulder DW, Kamali A, Ruberantwari A, Kengeya-Kayondo JF, Whitworth J. Mortality associated with HIV-1 infection over five years in a rural Ugandan population: cohort study.
BMJ 1997, 315: 767 –771.
14. Stover J, Way P. Projecting the impact of AIDS on mortality.
AIDS 1998, 12 (suppl 1): S29 –S39.
15. Todd J, Balira R, Grosskurth H. et al. HIV-associated adult mortality in a rural Tanzanian population.
AIDS 1997, 11: 801 –807.
16. Boerma JT, Ngalula J, Isingo R. et al. Levels and causes of adult mortality in rural Tanzania with special reference to HIV/AIDS.
Health Trans Rev 1997, 7 (suppl 2): 63 –74.
17. Leroy V, Msellati P, Lepage P. et al. Four years of natural history of HIV-1 infection in African women: a prospective cohort study in Kigali (Rwanda).
J Accquir Immune Defic Syndr 1993, 9: 415 –421.
18. Senkoro KP, Boerma JT, Klokke AH, et al. Incidence and mortality of HIV in a factory cohort study, 1991–96, Tanzania.TANESA Working Paper
No. 20. Mwanza: TANESA/National Institute for Medical Research; 1998.
19. Timaeus IM. Impact of the HIV epidemic on mortality in sub-Saharan Africa: evidence from national surveys and censuses.
AIDS 1998, 12 (suppl 1): S15 –S27.
20. Boerma JT, Nunn AJ, Whitworth JAG. Mortality impact of the AIDS epidemic: evidence from community studies in less developed countries.
AIDS 1998, 12 (suppl): S3 –S14.
21. Centers for Disease Control and Prevention. Update: Trends in AIDS incidence, deaths and prevalence. United States, 1996.
MMWR 1997, 46: 165 –173
22. Martin JN, Colford JM, Ngo L, Tager IB. Effect of older age on survival in human immunodeficiency virus (HIV).
Am J Epidemiol 1995, 142: 1221 –1230.
23. Morgan D, Maude GH, Malamba SS. et al. HIV-1 disease progression and AIDS-defining disorders in rural Uganda.
Lancet 1997, 350: 245 –250.
24. Lifson AR, Allen S, Wolf W. et al. Classification of HIV infection and disease in women from Rwanda.
Ann Intern Med. 1995, 122: 262 –270.
25. Morgan D, Ross A, Mayanja B, Malamba S, Whitworth J. Early manifestations (pre-AIDS) of HIV infection in Uganda.
AIDS 1998, 12: 591 –596.
Keywords:© 2000 Lippincott Williams & Wilkins, Inc.
HIV; Uganda; mortality; prognosis; mature epidemic