Introduction
Zimbabwe has experienced one of the largest generalized HIV epidemics, with prevalence in the adult population estimated at 20% in 2005 [1]. HIV prevalence has, however, started to decline since 2000 [2,3]. This results from a combination of reduced HIV incidence, a trend associated with reductions in sexual risk behaviour [4], and rising adult mortality caused by HIV/AIDS [2]. These national trends in prevalence, sexual risk and adult mortality have also been observed in the population-based cohort followed by the Manicaland HIV/STD Prevention Project in the eastern province of Zimbabwe [4].
Trends in mortality are informative per se, but they also provide insights into the history of the epidemic as they reflect incidence rates some years previously. The median survival time after seroconversion has previously been estimated to be approximately 10 years in African populations [5]. Data presented in this special issue demonstrate that the median survival may be somewhat longer overall but substantially shorter for those infected at older ages [6]. Therefore, although incidence may have peaked already, mortality rates would be expected to rise for approximately a decade after this, as a result of the long latent period of infection.
We reported previously that from 1998 to 2003, the male adult mortality rate in Manicaland remained high and fairly stable. Female mortality increased during the same period, bringing the rates of men and women level at approximately 23 deaths per 1000 person-years. Sixty-one per cent of adult male deaths were attributable to HIV compared with 70% of adult female deaths [7]. This paper uses recently available data from a second follow-up period to update mortality trends, allowing longitudinal analysis from 1998 to 2005. We explore mortality patterns across different demographic strata including age and socioeconomic location, represented by small towns, commercial estates, roadside business centres (RBC) and subsistence farming areas (SFA).
Methods
The Manicaland HIV/STD Prevention Project is a population-based open cohort study conducted in 12 sites in the Manicaland province in eastern Zimbabwe. Study sites differ in terms of their socioeconomic location, comprising two small towns, four commercial forestry, tea and coffee estates, two RBC, and four SFA. Baseline surveys were carried out in a phased manner between July 1998 and February 2000, with two subsequent follow-up rounds conducted approximately 3 and 5 years later.
All members of households (8211 households at baseline; 7102 at first follow-up; 11 183 at second follow-up) were initially enumerated in a census. Then, adults (men aged 17-54 years and women aged 15-44 years; expanded to include all men and women aged 15-54 years at second follow-up) were invited to participate in the individual cohort study, from which all mortality rates are calculated. Only one person within each cohabiting marital union was eligible for the cohort study, and individuals reaching the upper age boundary during the study period were still included in subsequent follow-up rounds. Participation in the cohort required being interviewed on a detailed questionnaire and giving a dried blood sample, which was anonymously tested for HIV using a highly sensitive and specific antibody detection test (ICL Dipstick; ImmunoChemical Laboratory, Thailand) [8].
Deaths of cohort members occurring between rounds were identified at each follow-up. Using lists of households and their inhabitants, enumerators questioned household representatives about the survival status of individuals who participated in the previous round of the cohort. If no household representative was available at follow-up or if the household had dissolved or outmigrated, then a neighbour or community informant was questioned on the survival status of the cohort member. Dates of seroconversion were assumed to be halfway between the last negative and the first positive test. Further information about the study design is detailed elsewhere [9].
As reported previously 54% (2242/4142) of the men and 66% (3265/4922) of the women interviewed at baseline, and not known to have died subsequently, were re-interviewed at follow-up [9]. Similar rates were achieved between the first and second follow-up for men at 56% (1816/3246) and women at 65% (3035/4645). Outmigrants and those lost to follow-up were censored at the last interview. Outmigrants had similar HIV prevalence to non-migrants [10] so their exclusion is unlikely to have introduced major bias.
In order to analyse trends in mortality rates, follow-up was split by time period (1998-2000; 2001-2002; 2003-2005) rather than follow-up round; this was facilitated by the availability of dates of death. In cases in which year of death information was missing (first follow-up 12%; second follow-up 10%), a date of death was randomly assigned from a uniform distribution spanning the time between the two rounds of the study in which the individual died. When the year of death was known but the month was missing (6% of all deaths), a month of death was randomly assigned; this did not affect stratification by year so these are not included in the missing date of death statistics. Mortality rates were directly standardized to the population age distribution at baseline (taken to be the 1998-2000 period) and stratified by sex in order to compare trends within the two sexes. To examine patterns of mortality within different areas of residence, rates were directly standardized to the age distribution at baseline of each sex and site type separately. Improved approximate 95% confidence intervals (CI) for a Poisson distribution were calculated using the method proposed by Dobson et al. [11].
Mortality rate ratios (MRR), adjusted for the effects of age and stratified by sex, were calculated to compare mortality in the HIV-positive and negative populations. The population attributable fraction (PAF), an estimate of the proportion of deaths that would be avoided in the absence of HIV [12], is also age adjusted. This assumes equal follow-up time per capita in HIV-positive and HIV-negative individuals and therefore gives a slightly conservative estimate of the true value.
Ethical approval for the study was granted by the Research Council of Zimbabwe (no. 02187) and the Applied and Qualitative Research Ethics Committee in Oxford, UK (N97.039), and written informed consent was obtained from each participant as a condition of entry into the cohort. All statistical analyses were performed using STATA version 8 (STATA Corp., College Station, Texas, USA).
Results
In total, 247 deaths (93 men, 154 women) were observed among 15-59 year olds in the 2-year second follow-up period compared with 396 in the initial 3-year period (174 men, 222 women).
Change in adult mortality by sex
Mortality rates were higher for men (31 per 1000 person-years, 2003-2005) than for women (26 per 1000 person-years) across all the study periods. Mortality, when standardized to the baseline age structure, showed a general increase for both sexes. This is most striking in women, whose mortality rates stabilized recently after a significant increase during the initial years of follow-up (MRR 1.7, 95% CI 1.3-2.1; P < 0.001) [7]. The increase among men is less pronounced, with wide confidence intervals (Fig. 1).
Adult mortality rates in the HIV-negative population were stable for both sexes over the study period. There was, however, a marked increase in the mortality of the HIV-positive population, particularly for HIV-positive women who experienced an increase from 35 to 88 deaths per 1000 person-years over the study period (Table 1). This contrasts with an increase from 62 to 105 per 1000 person-years in crude mortality rate for HIV-positive men in the same period. Men were subject to higher absolute mortality rates in both HIV-negative and positive populations throughout the study period. Although the male adult MRR increased gradually over time, the female MRR showed a sharp incline over the earlier years, from 5.3 (95% CI 3.4-8.3) in 1998-2000 to 10.5 (95% CI 7.1-15.4) in 2001-2002. This trend is reflected in the age-adjusted PAF, the percentage of adult mortality that is attributable to HIV/AIDS. A greater proportional increase is seen for women than for men, partly because fewer deaths are HIV related in the early study period (55% for women compared with 61% for men). Both stabilized in the most recent years, with women overtaking men in their proportion of deaths attributable to HIV/AIDS: 69 and 74% for men and women, respectively, in 2003-2005.
Equation (Uncited)Image Tools
Age-specific mortality rates
Figure 2 illustrates the shift in the age of peak mortality in men and women over the study period. In 1998-2000, male mortality increased with age, peaking at 58 per 1000 person-years (95% CI 40-83) in the 45-54 year age group. In the later study periods, the peak shifted down an age group, to the 35-44 year range. The mortality rate increased dramatically in this group over the 8-year study period from 23 per 1000 person-years (95% CI 14-36) in 1998-2000 to 62 per 1000 person-years (95% CI 42-91) in 2003-2005 (MRR 2.7, 95% CI 1.5-5.0; P = 0.001); whereas mortality rates in the 15-24 and 25-34 year age groups remained relatively stable.
In 1998-2000, the female mortality rate increased with age up to the 35-44 year age group at 21 per 1000 person-years (95% CI 15-29). Mortality rose in all age groups from 1998-2000 to 2001-2002, with the greatest increase occurring in the 25-34 year age group. In 2003-2005, mortality was highest in the 25-34 year age group at 35 per 1000 person-years (95% CI 26-48), a significant increase from the 16 per 1000 person-years (95% CI 11-23) recorded in 1998-2000 (MRR 2.2, 95% CI 1.4-3.7; P = 0.001).
Age-standardized mortality rates by socioeconomic location
Equation (Uncited)Image Tools
Figure 3 shows very different mortality patterns between the two sexes when stratified by area of residence. For men, the large overlapping confidence intervals suggest no clear differences between the study locations or over time. In contrast, women have a much higher age-standardized mortality rate in towns than in any other location from 2001 to 2005 (e.g. for 2003-2005: 46 per 1000 person-years, compared with 21, 26 and 23 for estates, SFA and RBC, respectively). Although generally stable rates are observed over time in estates and RBC, significant increases in mortality were recorded for women in both towns and SFA between 1998 and 2000. In the former, an age-adjusted MRR of 4.5 (95% CI 2.0-10.1; P < 0.001) was seen comparing the second with the first time interval; in the latter, the MRR was 2.0 (1.3-3.0; P = 0.003). Mortality subsequently levelled off in both town and SFA locations between 2001-2002 and 2003-2005.
Equation (Uncited)Image Tools
Discussion
Equation (Uncited)Image Tools
Generally stabilizing mortality rates in Manicaland, Zimbabwe, may mask more subtle dynamics occurring over the past 8 years in population subgroups. Mortality rates in the HIV-positive population increased for both sexes and an increasing fraction of deaths were attributable to HIV, especially for women. Peak mortality may have shifted towards younger age groups in both men and women.
There are three main features observed that are characteristic of a mature generalized epidemic, as predicted by mathematical modelling studies [13,14]. First, the mortality rate ratio increased sharply, driven by more and more infected people dying. At the same time, mortality rates in the HIV-negative population remained relatively constant. Similar patterns have been observed in a cohort of South African miners, in which mortality in HIV-positive individuals increased tenfold in the early 1990s [15]. Second, the age of peak mortality appears to have decreased in men and to a lesser extent in women. In the early stages of an epidemic, when HIV prevalence in the population is low, most individuals are susceptible so incident infections may occur over the whole spectrum of ages. Therefore, the resulting pattern of mortality, represented by the earlier 1998-2000 period, peaks in the older age groups as these people gradually age and die, which may occur at a particularly high rate given lower survival for those infected at older ages [6]. As infection spreads through the population, prevalence in the older age groups may saturate as the older individuals most at risk become infected. After this point, incident infections shift towards younger age groups as the younger cohorts in the population become sexually active. Although this shift is seen in both sexes, the younger age at peak mortality for women (25-34 years compared with 35-44 years for men in 2003-2005) reflects the higher incidence in young girls [4]. Similarly, in South Africa, one of the few nations with fairly comprehensive vital registration, mortality rates more than doubled from 1997 to 2005 in adults, with peak age at death occurring in 30-39-year-old men and in slightly younger (25-34-year-old) women [16]. Third, the possible levelling off of mortality rates since 2001-2002 may be characteristic of a late-stage epidemic [17,18]. Mortality is expected to peak approximately a decade after incidence. The exact timing will depend on the age distribution of infections (discussed elsewhere in this issue [6]), and will be more spread out than the peak in incidence.
The present mortality rates may also reflect historical changes in incidence, thereby providing insights into the history of the epidemic. For example, observed trends in female adult mortality could reflect an increase in incidence in the late 1980s followed by a peak in the early 1990s and a subsequent stabilization. More recently, female adult mortality reached a similar level to that seen in men. This is consistent with the hypothesis that the spread of HIV infection into the female population occurred after it had become widely disseminated among men [19].
The sharp increase in mortality rates for women in towns from 1998-2000 to 2001-2002 reflects the fact that this group had a particularly high HIV prevalence (at over 30%) and incidence [4,20]. The dramatic increase in mortality suggests that there was a period of rapid transmission at some point in the history of the epidemic in towns, and, therefore, that the HIV epidemic is in a different phase in these areas to the RBC and estates. Increasing mortality, albeit at a lower level and ratio, was also observed in SFA. It is possible that migration of HIV-infected individuals may have contributed to this. Previous studies have reported that 11 and 19% of all deaths in the rural Morogoro and Hai districts in Tanzania, respectively, were made up of homecoming sick [21]; and 17% of adults moved into the household where they died during the year before death in rural Kisesa ward, Tanzania [22]. Recent analysis of Manicaland data shows that town households that experienced an AIDS death were more likely to outmigrate than to dissolve, and outmigrating households relocated primarily to rural villages [23]. Therefore, it is possible that levels of mortality are disproportionately high compared with HIV prevalence in SFA as a result of HIV-positive individuals migrating to their rural areas before death.
Under-ascertainment of deaths can be a problem when mortality is measured by demographic surveillance [24]; for example, when households dissolve or relocate after a death [23]. In this study, however, a neighbour or community representative was interviewed instead to ascertain the vital status and whereabouts of these individuals; and we used data from the cohort component (with HIV testing), which gives more reliable estimates of mortality [7]. Our levels of loss to follow-up are comparable with similar studies [9,25]. Outmigrants may have suffered greater mortality although we found no evidence that they had greater HIV incidence [10]. Nonetheless, losses to follow-up of households and individuals may have resulted in some under-ascertainment of mortality rates.
Equation (Uncited)Image Tools
The random assignment of dates of death that were missing is not appropriate for all applications. In this analysis, however, these errors are unlikely to create biased results because the assigned dates of death were uniformly distributed over quite a short time period: a maximum of 3 years. In addition, only approximately 11% of dates of death were assigned in this way. Assigning the midpoint between two interview dates as the date of death would have resulted in a heaping of deaths around the middle of the follow-up periods; iterating the random assignment of data of death did not qualitatively change any results (not shown).
Equation (Uncited)Image Tools
We assumed that all deaths in HIV-negative individuals were not HIV/AIDS related, although this may result in a slight underestimation of HIV-related mortality. It is possible that a very small number of individuals seroconverted and died within a single 2 or 3-year follow-up period (approximately 2% of HIV-positive deaths according to Todd et al. this issue [6]).
Understanding the mortality patterns observed over the past 8 years in Manicaland can further our understanding of the epidemic. The change observed in the age, sex and burden of HIV/AIDS as a cause of death suggests that although incidence has peaked in this population, this may have occurred only a decade ago, especially among women. Combining these observations on mortality with HIV incidence will be needed to explain future changes in HIV prevalence. The Manicaland HIV/STD Prevention Project is ongoing and future results will enable these trends to be tracked further.
The possible stabilizing of mortality rates in the latter years of the study period suggests that the AIDS mortality may be reaching its peak in Manicaland. We may therefore anticipate a decline in adult HIV-related mortality over the coming years because of the declining prevalence of HIV. Such a decline may partly be a natural consequence of the epidemic; although recent behaviour change could contribute to a long-term reduction in HIV prevalence [4]. In order to have had an impact on current mortality rates, behaviour change would have had to have occurred more than 10 years ago. There is little evidence of behaviour extending back this far, although the data available on this are limited. Antiretroviral drugs were not widely available in Manicaland during the study period so will not have had a major effect on mortality or the course of the epidemic, but these data provide a baseline against which the impact of current attempts to scale up antiretroviral therapy in Zimbabwe can be measured.
Conflicts of interest: None.
References
1. UNAIDS. Report on the global AIDS epidemic. Geneva: Joint United Nations Programme on HIV/AIDS; 2006.
2. UNAIDS. Evidence for HIV decline in Zimbabwe:a comprehensive review of the epidemiological data. Geneva: Joint United Nations Programme on HIV/AIDS; 2005.
3. Mahomva A, Greby S, Dube S, Mugurungi O, Hargrove J, Rosen D, et al. HIV prevalence and trends from data in Zimbabwe, 1997-2004. Sex Transm Infect 2006; 18(suppl 1):i42-i47.
4. Gregson S, Garnett GP, Nyamukapa CA, Hallett TB, Lewis JJ, Mason PR, et al. HIV decline associated with behavior change in eastern Zimbabwe. Science 2006; 311:664-666.
5. Morgan D, Mahe C, Mayanja B, Okongo JM, Lubega R, Whitworth JA. HIV-1 infection in rural Africa: is there a difference in median time to AIDS and survival compared with that in industrialized countries? AIDS 2002; 16:597-603.
6. Todd J, Glynn JR, Marston M, Lutalo T, Biraro S, Mwita W, et al. Time from HIV seroconversion to death: a collaborative analysis of eight studies in six low and middle-income countries before highly active antiretroviral therapy. AIDS 2007; 21(Suppl. 6):S55-S63.
7. Lopman BA, Barnabas R, Hallett TB, Nyamukapa C, Mundandi C, Mushati P, et al. Assessing adult mortality in HIV-1-afflicted Zimbabwe (1998-2003). Bull WHO 2006; 84:189-197.
8. Ray CS, Mason PR, Smith H, Rogers L, Tobaiwa O, Katzenstein DA. An evaluation of dipstick-dot immunoassay in the detection of antibodies to HIV-1 and 2 in Zimbabwe. Trop Med Int Health 1997; 2:83-88.
9. Gregson S, Garnett GP, Nyamukapa CA, Hallett TB, Lewis JJ, Mason PR, et al. Supporting online material: HIV decline associated with behavior change in eastern Zimbabwe. Science 2006; 311:620-621.
10. Mundandi C, Vissers D, Voeten H, Habbema D, Gregson S. No difference in HIV incidence and sexual behaviour between out-migrants and residents in rural Manicaland, Zimbabwe. Trop Med Int Health 2006; 11:705-711.
11. Dobson AJ, Kuulasmaa K, Eberle E, Scherer J. Confidence-intervals for weighted sums of Poisson parameters. Stat Med 1991; 457-462.
12. Chatterjee N, Hartge P. Apportioning causes, targeting populations and predicting risks: population attributable fractions. Eur J Epidemiol 2003; 18:933-935.
13. Gregson S, Garnett GP, Anderson RM. Is HIV-1 likely to become a leading cause of adult mortality in sub-Saharan Africa. J Acquir Immune Defic Syndr Hum Retrovirol 1994; 7:839-852.
14. Anderson RM, May RM. Infectious diseases in humans: dynamics and control. Oxford, UK: Oxford University Press; 1991.
15. Glynn JR, Sonnenberg P, Nelson G, Bester A, Shearer S, Murray J. Survival from HIV-1 seroconversion in Southern Africa: a retrospective cohort study in nearly 2000 gold-miners over 10 years of follow-up. AIDS 2007; 21:625-632.
16. Statistics South Africa. Mortality and causes of death in South Africa, 1997-2003: findings from death notification. Pretoria, South Africa: Statistics South Africa; 2005.
17. Grassly NC, Garnett GP, Schwartlander B, Gregson S, Anderson RM. The effectiveness of HIV prevention and the epidemiological context. Bull WHO 2001; 79:1121-1132.
18. Wasserheit JN, Aral SO. The dynamic topology of sexually transmitted disease epidemics: implications for prevention strategies. J Infect Dis 1996; 174(suppl 2):S201-S213.
19. Gregson S, Garnett GP. Contrasting gender differentials in HIV-1 prevalence and associated mortality increase in eastern and southern Africa: artefact of data or natural course of epidemics. AIDS 2000; 14(Suppl. 3):S85-S99.
20. Lewis JJ. Behavioural, demographic and social risk factors for HIV infection in rural Zimbabwe. London: Imperial College London; 2006.
21. Kitange HM, Machibya H, Black J, Mtasiwa DM, Masuki G, Whiting D, et al. Outlook for survivors of childhood in sub-Saharan Africa: adult mortality in Tanzania. BMJ 1996; 312:216-220.
22. Urassa M, Boerma JT, Isingo R, Ngalula J, Ng'weshemi J, Mwaluko G, et al. The impact of HIV/AIDS on mortality and household mobility in rural Tanzania. AIDS 2001; 15:2017-2023.
23. Gregson S, Mushati P, Nyamukapa C. Adult mortality and erosion of household viability in AIDS-afflicted towns, estates, and villages in eastern Zimbabwe. J Acquir Immune Defic Syndr 2007; 21:188-195.
24. Preston SH, Heuveline P, Guillot M. Demography: measuring and modelling population processes, 1st ed. Oxford: Blackwell Publishers Ltd. 2001. p. 283.
25. Hosegood V, McGrath N, Herbst K, Timaeus IM. The impact of adult mortality on household dissolution and migration in rural South Africa. AIDS 2004; 18:1585-1590.
© 2007 Lippincott Williams & Wilkins, Inc.