Introduction
Mathematical model projections of the scale of the likely demographic consequences of HIV-1 epidemics in sub-Saharan African populations vary substantially [1-3]. These variations flow principally from differences in the modelling approaches applied and the assumptions made with respect to the 'intermediate determinants' of HIV-1 epidemics and their demographic impact (i.e., the key biological and behavioural input parameters of the models). The latter result from a continuing shortage of reliable empirical data on these intermediate determinants, specific to sub-Saharan African settings. Variations also often reflect differences in the objectives of the studies. Some seek to develop predictions specific to particular countries [4] or regions [3]. Others are designed to investigate, in a qualitative sense, the intermediate determinants of demographic impact [2,5-7].
Until recently, the links between HIV, AIDS and rising mortality in sub-Saharan Africa were questioned by some commentators [8]. However, empirical data on demographic impact are now available from a small number of locations that clearly demonstrate the emergence of excess mortality associated with HIV infections [9-13]. The order and age pattern of this excess mortality has been shown to be consistent with that projected in model simulations based on similar levels of HIV-1 prevalence [14]. However, further information on the impact of the HIV-1 epidemic on mortality is needed to confirm these early findings and to improve understanding of the relationship between trends in HIV-1 infections and age-specific mortality rates. In this article, we present results from a study of the early demographic impact of the HIV-1 epidemic in two rural areas of Zimbabwe. We found that HIV-1 prevalence among women attending antenatal clinics was high and that both adult and infant mortality had begun to rise. There was indirect evidence for a causal relationship between these two phenomena.
Background and methods
Background and study approach
The HIV-1 epidemic became established in Zimbabwe during the mid-1980s and its subsequent spread has been extremely rapid. Anonymous unlinked surveillance of pregnant women in five urban and 17 rural antenatal clinics during 1992 and 1993 found that 21.2% (1205 out of 5679) were infected with HIV-1. Prevalence levels in excess of 40% were recorded in some towns and trading centres [15]. Numbers of reported AIDS and tuberculosis cases have risen sharply since 1989 (Fig. 1). In each instance, the numbers of male cases reported have consistently exceeded the numbers of female cases.
The Manicaland Study was initiated in 1993 to investigate the early demographic impact of the HIV-1 epidemic in rural areas of Zimbabwe, where 70% of the country's population are based [16]. Two exclusively rural areas in Manicaland Province, the Honde and Rusitu valleys, were chosen for the study, on the basis of their similar cultural and economic backgrounds and close proximity to the border with Mozambique, but contrasting locations with respect to the early spread of HIV-1 infection within Zimbabwe. The Honde Valley lies closer and has better access to areas where high antenatal prevalence levels were recorded in 1993 [e.g., 25 and 46% in Mutare (90 km) and Rusape (130 km), respectively] and thus it was hypothesized that the early demographic effects of the HIV-1 epidemic would be more apparent in this area.
At the time the study was initiated, HIV-1 surveillance in Zimbabwe was restricted to testing at antenatal and sexually transmitted disease clinics. The best option for assessing the impact of the epidemic on mortality appeared to be to cross-reference data on HIV-1 prevalence among women at local antenatal clinics to estimates for a range of indirect indicators for the study populations (Table 1). Although it was recognized that results from individual methods could be subject to bias, the nature of the bias would differ in each case. By focusing on the strengths of each method, a range of information on recent trends and patterns in mortality could be obtained. To the extent that a consistent picture emerged, the results could be used to test hypotheses on the impact of the HIV-1 epidemic and other factors, such as the effects of the recently introduced Economic Structural Adjustment Programme.
To this end, an hypothesis on the expected pattern of impact of the HIV-1 epidemic on mortality was developed using a previously published and validated mathematical model of the epidemiology and demographic impact of HIV-1 epidemics [14,17], qualitative observations on the predominant pattern of sexual behaviour in rural Zimbabwe, independent estimates of the key biomedical parameters of HIV-1 transmission and disease progression [18], and data on local and national levels of HIV-1 prevalence [19].
The methods used to estimate the various indicators of mortality in the study areas and surrounding districts are described below. The empirical results are presented and alternative interpretations are discussed in the light of findings from the mathematical model simulations.
HIV-1 prevalence among pregnant women
Routine antenatal serological testing for syphilis is not feasible in remote areas of Zimbabwe, so special arrangements were required to obtain surveillance data on HIV-1 prevalence. Women attending antenatal clinics at Hauna, Gatsi and Ruda in the Honde Valley, and Ngorima and Rusitu Mission in the Rusitu Valley, over the 15-month period until June 1994, were asked to participate in the study. It was explained that blood samples and background information were needed for medical research purposes and verbal consent was sought. Where this was given, four 30 µl whole blood samples were obtained from the thumb using Ancoscreen filter paper tabs (Ancos Ltd, Hojby, Denmark). A short sociodemographic questionnaire, covering age, education, marital status and current parity, was administered. The questionnaire and filter paper tabs for each woman were marked with a unique identification number, but names were not recorded, in order to ensure anonymity. To minimize repeat cases, women were asked whether they had previously participated in the research.
Blood samples were tested for the presence of HIV-1 antibodies at the Public Health Laboratory in Harare, using two enzyme-linked immunosorbent assays: Organon Teknika BV Vironostika (Boxtel, Funen, Holland) HIV MIXT (sensitivity, 96.0%; specificity, 99.4%) and Behring Enzygnost (Marburg, Hessen, Germany) Anti-HIV-1/HIV-2 (sensitivity, 97.6%; specificity, 99.4%) [20]. When contradictory results were obtained in the initial tests, the same tests were repeated. Where the results still differed, an Ancoscreen immunoblot assay was applied and the result of this test was taken as being conclusive. Identical procedures were used in the Zimbabwe National AIDS Coordination Programme's sentinel surveillance system.
Mortality indicators
Data on numbers of deaths registered were extracted from the records of the seven district registry offices in Manicaland for each year between 1982 and 1995. Annual estimates of the numbers of deaths registered per 1000 population were derived, using Zimbabwe Central Statistical Office (CSO) figures for the district populations in 1982 and 1992, the most recent census years, and by assuming constant growth rates during intervening and subsequent years. Numbers of deaths registered, by reported cause, were obtained over the same period for Mutasa District (the district that includes the Honde Valley), and the same procedure was applied to derive annual estimates of cause-specific death rates. Crude death rates estimated in the 1992 census were compared with the average death rates for the period 1982-1995, estimated from the vital registration data, to assess the level of coverage in the latter system.
Household censuses were conducted in the national census enumeration areas immediately surrounding the growth points at Hauna (Honde) and Dzingire (Rusitu), between March and June 1994. In each case, a group of enumeration areas was selected comprising approximately 600 women of childbearing age (13-49 years), using CSO records as a yardstick. Household members were enumerated on a de jure (usual residence) basis. Age at last birthday was recorded for each household member and population pyramids were constructed for the two study areas. For children aged under 15 years, data on the survival status of natural parents were sought and used to calculate estimates of maternal and paternal orphanhood prevalence. A follow-up survey of all households containing children in 1994 was conducted between March and April 1995. The identities and survival status of natural parents of children aged under 16 years were established and confirmed, where possible, by reference to birth certificates, child health cards and other official documentation. Where a parent had died, the year of death was recorded. The data obtained in 1995 were used to validate the 1994 findings and as the basis for calculating estimates of the incidence of parental death among children aged 0-7 years, by period prior to the survey date.
The survival status in 1995 of individuals enumerated in the 1994 household census was ascertained in the follow-up survey. These data were used to estimate contemporary levels of age-specific death probabilities (nqx) by applying the approximation nqx = 1 - Sxn, where Sxn is the proportion of individuals aged between x and x + n at the time of the 1994 survey who were still alive 1 year later. The patterns of estimates obtained, by location, sex and age, were compared with figures for the period immediately preceding the onset of the AIDS epidemic in Zimbabwe [21], with patterns projected in mathematical model simulations (see below), and with findings from other empirical studies in sub-Saharan Africa [9-13].
Women of childbearing age living in the study areas were identified in the 1994 census and asked to participate in an in-depth survey of knowledge, attitudes, beliefs and practices on fertility and HIV-1/AIDS [18]. The data obtained in this survey included detailed birth history information (i.e., details of the date of birth, survival status and, where appropriate, date of death of each child). These data were used to calculate estimates of infant mortality rates, by year prior to the survey date and by religion.
Statistical analysis
HIV-1 prevalence levels were computed by age-group and for all women of childbearing age. The STATA package was used to carry out logistic regressions to examine sociodemographic determinants of HIV-1 infection [22]. χ2 tests for the significance of differences between proportions, using the Normal approximation (with continuity correction) to the binomial distribution, were applied to assess the significance of recent changes in mortality and of differentials between the two study sites and in population subgroups. Exact binomial 95% confidence intervals (CI) were calculated for the 1994-1995 estimates of age-specific death probabilities for the principal sexually-active age-intervals (25-44 years for men and 15-44 years for women) and for each 10-year segment within these intervals. 95% CI for each value of nqx were calculated and the approximation nqx = 1 - Sxn was applied to these to estimate the limits for the broader age intervals. The latter were then compared with CSO lifetable estimates for 1987 [23] to test for recent mortality increases.
Development of mathematical model simulation using qualitative data on sexual behaviour
Qualitative data were collected using focus-group discussions, in-depth interviews with key informants, pocket-chart voting and personal observation methods [24,25]. The pocket-chart voting method is based on a community participation technique [26] and entails secret voting by a small group of participants (n = 6-15) on a short set of questions. After each question, the numbers of votes are counted for the different possible responses and reasons for the results are explored with the group. In total, 16 focus-group discussions and four pocket-chart voting sessions were held in 1994 and 1995. Topics covered included traditional and contemporary beliefs, attitudes and practices in respect of marriage and sexual behaviour.
Data from these qualitative studies were used in validation of the statistical results and in development of a mathematical model simulation designed to show the kinds of demographic changes which might be expected at the current stage of the epidemic in Zimbabwe. Detailed descriptions of the reliability and validity tests conducted and of the construction and validation of the mathematical model are beyond the scope of this article but are available in earlier publications [14,17,18,27].
Results
Population size and coverage
The total populations enumerated in Manicaland Province and in the two districts containing the study areas, Mutasa (Honde) and Chimanimani (Rusitu), in the 1992 census, were 1.5 million, 166 000 and 110 000, respectively. On the basis of census data, the Zimbabwe CSO estimated that the crude death rates in Mutasa and Chimanimani in 1992 were 10.9 and 10.6 per 1000, respectively [28]. Comparison of these levels with the average per capita registered death rates between 1982 and 1995 (1.9 and 1.6 per 1000, respectively) suggests that coverage in the death registration system in each of the two districts is of the order of 15-20%. Only 3% of deaths registered in each area were infants', compared with an expected value of 24%, based on 1992 census estimates [28], which indicates that registration coverage is higher for adults, as reported elsewhere [29]. Among adults, 73% of deaths registered in Chimanimani in 1993 were among men.
Within the study areas, 431 households were enumerated in the Honde Valley and 498 in Rusitu Valley in 1994. These numbers represent 96 and 98% of all households in the respective study areas, on the basis of reconciliations with 1992 census lists. Of the women of childbearing age enumerated in this household census, 593 (95% coverage) and 644 (97%) were interviewed in depth on knowledge, attitudes, beliefs and practices regarding fertility and HIV/AIDS. Coverage in the 1995 follow-up survey was estimated at 95%, after adjusting for 14 households that had moved away from the study areas since 1994. In total, 745 households containing 2320 children aged under 16 years participated in this survey. Data on survival status of father and mother was obtained for 2188 and 2217 of these children, respectively.
A total of 487 pregnant women aged between 15 and 49 years were tested for HIV-1 infection at the antenatal clinics in Honde Valley (n = 220) and Rusitu Valley (n = 267). Only one woman (at Hauna) is known to have declined to participate. The HIV-1 testing procedure yielded results for 470 women (96.5%). The remaining tests were indeterminate.
Qualitative studies on sexual behaviour
Participants in pocket-chart voting sessions and focus group discussions indicated that although extramarital relationships are common among men in rural areas, irrespective of marital status, female partners in such relationships are primarily divorced and, to a lesser extent, single and widowed. For example, in (anonymous) pocket-chart voting exercises in Rusitu, nine out of 14 men but no married women (out of 14) reported casual partners in the past month. This impression conforms to the view in traditional culture that serious misfortunes afflict married women who are unfaithful to their husbands, and this was endorsed by information obtained in visits to local beer halls and in key informant interviews with commercial sex workers. Men cited absence from the home area for work purposes and post-partum abstinence by their wives as the main reasons for having casual partners.
The qualitative study results also revealed that members of a prominent minority Apostolic church in the Honde Valley follow a more restrictive behaviour code, which includes abstinence from drinking alcohol, extramarital relationships, medicines and modern methods of family planning [18].
HIV-1 prevalence among pregnant women in the Honde and Rusitu valleys
The results of the antenatal clinic based HIV-1 prevalence studies conducted in the Honde and Rusitu valleys are shown in Table 2. The overall prevalence levels among pregnant women were 24.3% (95% CI, 19.6-29.0) in Honde and 14.0% (95% CI, 11.7-16.4) in Rusitu. In the logistic regression model, HIV-1 prevalence was closely associated with age. Prevalence was high among women in their teens and early 20s and peaked in the 25-29-year age-group in both areas. Non-married women were at higher risk of infection than currently married women. Women in the Honde Valley were more likely to be infected (odds ratio, 1.4), but the difference was no longer statistically significant [30].
Recent mortality trends in Mutasa and Chimanimani districts
Death registration rates for Mutasa and Chimanimani districts, other rural districts, and Mutare, the provincial capital of Manicaland, are shown in Fig. 2 by year from 1982. In each case, the level of death registration increased in the last 5 years. In Mutasa, per capita registration in 1991-1995 was significantly greater than in the equivalent period 5 years earlier (P < 0.001). Trends in registration in Mutasa District, by cause of death, are shown in Fig. 2e and f. Small numbers of AIDS deaths have been registered since 1991. However, total numbers of deaths from causes commonly associated with HIV-1 infection (Fig. 2e) increased sharply over the same period (P < 0.001), while those from causes rarely associated with HIV-1 infection remained at 1980s levels.
Recent adult mortality trends and patterns in the Honde and Rusitu valleys
Prevalence of orphanhood among children aged under 15 years of age at the time of the 1994 survey was estimated from data obtained in both the original and follow-up household visits. The 1994 data indicated that 7.5% (n = 941) and 10.3% (n = 1247) of children had lost their fathers in Honde and Rusitu, respectively, whereas 2.5% (n = 950) and 2.4% (n = 1267) had lost their mothers. In 1995, the equivalent figures were 6.1 and 10.7%, and 3.3 and 2.6%. Inconsistencies between the orphanhood status of children recorded in the two surveys were found in 3.8% (n = 83) of cases for fathers and 0.8% (n = 18) of cases for mothers. These may have resulted from misreporting or enumeration errors in either of the two surveys, so the error rate in each survey may be smaller.
Recent trends in orphanhood and adult mortality are indicated by the data on incidence of parental death among 0-7-year-olds taken from the 1995 household survey. These results are shown in Fig. 3. In both study areas, incidence of parental death (father or mother) increased significantly between 1988-1991 and 1992-1995 (P < 0.01). The increase in paternal deaths was greater in Honde but the rise in maternal deaths was smaller. In each area, the increase in incidence of paternal death was greater in absolute terms than that in maternal deaths.
In total, 29 (n = 2245) male deaths and 20 (n = 2479) female deaths were recorded between the 1994 and 1995 surveys. One further death was recorded for an individual whose sex was not noted. The observed patterns of mortality are shown by study site, sex and 10-year age-interval in Fig. 4. The 95% CIs were wide, particularly at adult ages, because numbers of people in each 10-year interval were small. For men, there was evidence of excess mortality, compared with 1987 levels [23], in the principal sexually-active age-interval (25-44 years) in both areas (P < 0.05) and in the age interval of 25-34 years in the Honde Valley (P < 0.05). For women, the survey results in the Honde Valley were higher than the 1987 national estimates overall (15-44 years) and in each 10-year age interval within this range, although none of these differences were statistically significant. There was no evidence for excess (or reduced) mortality among women in Rusitu or among children and older adults in either area. Results from the mathematical model projections of the age pattern of mortality 5 and 10 years after the rapid onset of an HIV-1 epidemic are also shown in Fig. 4. (A summary of the principal parameter settings used in the simulation is given in the legend to Fig. 4.) These results indicate rising excess mortality over time, concentrated in the 25-44 and 15-44-year age ranges for men and women, respectively, with the early increase most pronounced among men.
Recent infant mortality trends and determinants in the Honde and Rusitu valleys
Fig. 5 shows the recent trends in infant mortality by study site and, for the Honde Valley, by religion. The periods 1975-1979 and 1980-1984 were aggregated, because the numbers of births in these years to women interviewed in the survey were small (approximately 100 per year) and to limit the effects of date misreporting in the earlier periods. Estimated infant mortality increased significantly 1985-1989 and 1990-1994 in Rusitu (P < 0.05) but not in Honde. Infant mortality has been closely associated with religious affiliation in Honde Valley in recent years [18]. Children of women belonging to Apostolic religions tended to experience higher mortality (odds ratio, 2.6; P < 0.001). Fig. 5b shows that this comparative disadvantage has reduced in the last decade (P < 0.01) and that infant mortality among the children of non-Apostolics may have increased over the most recent 5-year period (P = 0.055).
Demographic structure of the Honde and Rusitu valley populations
The population pyramids for the study populations (Fig. 6) indicate modest shortfalls of older men and young children, compared with the population distribution recorded in rural areas in the 1987 Zimbabwe Intercensal Demographic Survey [21]. The presence of temporary workers from Mozambique inflated the numbers of young male teenagers in 1994.
Discussion
The results indicate high levels of HIV-1 prevalence for rural areas in a sub-Saharan African country. The Rusitu Valley is one of the more remote areas of Zimbabwe. The local growth point (Dzingire) had seen little development prior to 1994 and lies over 30 min-drive from the nearest tarred road. The Honde Valley is slightly more developed, but is still 90 km from the nearest significant town (Mutare). However, the HIV-1 epidemic appears to have spread extensively into these locations. Mortality has begun to rise in both areas. The data presented here indicate increases in registered deaths and upturns in both adult and early childhood mortality. Older childhood mortality and older adult mortality appear to have changed little. The increase in mortality among younger and middle-aged adults has led to an escalation in the incidence of orphanhood. There are signs that the age and sex structure of these rural populations may have begun to be affected.
Could these results reflect biases in the data collected? Possible selection effects in the HIV-1 prevalence data from antenatal clinics have been considered elsewhere [30]. In summary, these data may slightly overstate the true levels among women of childbearing ages, due to (net) participation bias by age and religion outweighing the effect of reduced fertility among women with HIV-1 [12]. The Honde Valley results, in particular, may be on the high side. Nevertheless, it is plausible that HIV-1 prevalence is greater in this area, given its higher level of socioeconomic development and closer proximity to other areas of high prevalence. On balance, we believe that HIV-1 prevalence among women of childbearing age in the general population exceeds 15% in Honde and 10% in Rusitu.
Coverage of deaths in the vital registration system is low in rural areas. This is not a problem in itself, if interest lies primarily in assessment of trends rather than levels of mortality and the overall level of registration is, as we believe, reasonably consistent over time. However, participation bias again may be present. The higher levels of coverage among adults and particularly men suggest that the data are most representative of trends in male adult mortality. Deaths occurring outside hospital are less likely to be registered and certain causes of mortality could be selectively under-recorded. Given the emphasis on home-based care for AIDS patients in Zimbabwe, these categories could include some HIV-1-associated deaths.
Rapid increases in orphanhood and thus in parental death, initially concentrated among fathers, have also been reported in other studies in Manicaland [31]. However, under-reporting is a common problem in orphan enumerations [32]. The maternal orphan prevalence results from the current survey are slightly lower than those projected for the early stages of an HIV-1 epidemic of similar proportions in mathematical model simulations, based on comparable levels of underlying mortality and fertility [7]. Despite the preventative measures taken, a degree of under-enumeration may be present in the orphan data. Alternatively, the discrepancy could be accounted for by reduced fertility in women with HIV-1, a possibility that is not reflected in simulation results published to date. With respect to trends, vertically transmitted HIV-1 infections will have increased orphan mortality in recent years, so that trends in orphan incidence will tend to understate the impact of an epidemic on adult mortality. To some degree this effect will be offset by disproportionate omission of parental deaths occurring further into the past, resulting from excess mortality among orphans or confusion of the identities of adopting and natural parents. Deaths of parents formerly resident in urban areas, where HIV-1 initially spread more rapidly, may also distort the picture of rural adult mortality obtained from study of orphan data.
The data on age-specific mortality for 1994-1995 may be affected by selection effects among families that moved away from the study areas between the two survey dates. However, the total number of households which had moved away was small (14 which represents 1.8%). Death of the household head or his/her spouse is a common cause of household breakdown, so that the estimated death probabilities at adult ages are most likely to understate the true levels.
Women may have had a tendency to omit children who died from their reported birth histories, which would result in understatement of infant mortality rates. Again this is most likely to have occurred for children born longer ago, so that the greatest effect would be on the estimates for 1975-1984. Any effect in the past 10 years can only have been small. Overstatement of a recent increase in infant mortality could result from the exclusion of children born to mothers aged over 50 years at the time of the survey or who are no longer alive. However, to the extent that female adult mortality has already increased due to HIV-1 infections, study estimates of infant mortality would again be expected to understate the true impact of the epidemic, because infants of women who died of AIDS are themselves at greater risk of infection and death. A similar problem is likely to affect the equivalent results in the 1994 Zimbabwe Demographic and Health Survey and the 1992 Zimbabwe Census, for Zimbabwe, Manicaland Province and for Mutasa and Chimanimani districts [28,33]. Despite this, modest reversals of earlier declines in infant mortality were recorded in each case. The census data also indicate downturns in life expectancy of the order of 1-2 years, for men and women between 1988 and 1990.
Finally, very young children and younger adults are sometimes under-enumerated in censuses and house-hold surveys. Thus, the differences between the age/sex structures of the study populations and the results of the 1987 Zimbabwe Intercensal Demographic Survey could reflect greater age and sex-selective omissions in the Manicaland Study. This seems unlikely given the precautions that were taken [18]. More plausibly, the differences could be accounted for by a combination of more extensive labour migration from the study areas in recent years, fertility decline and, perhaps, increased male adult mortality and early childhood mortality.
Clearly, the data presented are subject to data collection and selection biases. The extent of distortion is difficult to quantify but, on the basis of the information available, there seems little reason to believe that there is a substantial and systematic upward bias underpinning the observed mortality trends. Furthermore, the levels of internal consistency between the different indicators used and the degree of consistency of the observed trends with those found in external sources are all high [18]. Thus, it seems reasonable to suppose that the overall picture of a recent increase in mortality is a correct one. If this is the case, the next question that arises is whether the high level of HIV-1 prevalence and the rise in mortality are linked. There are a number of other possible explanations for an increase in mortality. Principal amongst these are an increase in death registration due to higher levels of education and more widespread use of life insurance policies, the severe drought in 1992 and the recently introduced World Bank-sponsored Economic Structural Adjustment Programme (ESAP).
The first of these influences might be expected to exert a long-term upward pressure on the underlying level of death registration, but it is difficult to see this as a cause of the relatively sharp upturn in registration recorded in rural areas since 1990. The drought appears to have had a modest effect in 1992. A small excess above trend is detectable in the Chimanimani data, but at least a part of this reflects a temporary rise in coverage due to the use of mobile units to increase birth registration in that year. The upward trends in death registration in Mutasa and other rural districts appear to have started before the drought and have been sustained afterwards. The ESAP began to be introduced in 1990 and can only have had a minor effect so early in its implementation. Fees for health treatment were enforced first in 1991 but were not raised again until mid-1994. They are waived for people earning less than Z$ 150 (US$ 20) per month, a category that includes most people living in rural areas. It is difficult to see any of these factors resulting in the age and sex pattern of excess mortality recorded since 1987. In each case, increases in early/late age mortality would be most plausible. Young and middle-aged adults would not be expected to be particularly affected.
The observed patterns are consistent with mathematical model predictions of what would be expected 5-10 years into a rapid HIV-1 epidemic, in which men generally acquire infection first, frequently during spells of labour migration in towns or commercial areas, and then pass on the infection to their regular female partners based in rural areas ('the HIV-1 hypothesis'). This pattern of behaviour and infection appears to predominate in the study populations and may also be common in other parts of Zimbabwe. In a study of discordant couples at Parirenyatwa Hospital, Harare, only two cases were identified over an 18-month period during 1986 and 1987 where the husband was HIV-1-negative and the wife was HIV-1-positive. In each case the wife had received a blood transfusion and one wife's first husband had died of an AIDS-like illness. Over 70% of husbands reported having visited a commercial sex worker [34]. Similar age patterns of excess mortality concentrated among adults, to those observed in Manicaland, have been reported in Côte d'Ivoire and Uganda [11,12,35]. The Ugandan studies have demonstrated directly that this pattern can result from the presence of HIV-1 infection.
The HIV-1 hypothesis is given added credence by the differentials in mortality increase seen between the Honde (Mutasa) and Rusitu (Chimanimani) valley populations. HIV-1 prevalence is higher in Honde where the adult mortality increase is most visible. The hypothesis fits the pattern of increase in AIDS and tuberculosis cases, by sex, found at national level, as well as that in registered deaths, by cause, seen in Mutasa District. It would also explain the stronger recent increase in paternal as opposed to maternal orphanhood - the higher underlying level of paternal orphanhood reflects large age gaps between spouses (father's age exceeded mother's age by more than 10 years for one-third of the children enumerated) and higher male adult mortality from non-HIV-1-associated causes [21]. A slower initial spread of infection among women, together with lower fertility among HIV-infected women, would account for the relatively modest increases in infant and early childhood mortality seen in the results. This and the stronger presence of the stricter Apostolic churches in the Honde Valley could explain the larger early rise in infant mortality apparent in the Rusitu Valley data. The decline in Apostolic infant mortality in Honde Valley is believed to reflect vigorous compulsory enforcement of WHO Expanded Programme of Immunization vaccinations following the mid-1980s measles epidemic in Manicaland.
The impression that the epidemic has spread recently and rapidly into these rural areas is consistent with the young age distribution of HIV-1 prevalence observed among pregnant women and the absence of evidence for a substantial shift in population structure. Significant shortfalls in middle-aged adults have been reported as resulting from AIDS epidemics in other parts of Africa [36] and are projected in mathematical model simulations 10-20 years after a major outbreak [6,18]. However, the current population structure can largely be explained by falling birth rates and, in Rusitu, more extensive male labour out-migration.
The hypothesis of early effects of a relatively recent spread of HIV-1 infection with wives generally infected after their husbands appears to explain the observed increase in mortality better than alternative theories. Given the high levels of HIV-1 prevalence recorded among women at the antenatal clinics, further increases in mortality are to be expected, particularly among women and young children. While the demographic impact of HIV-1 and AIDS epidemics may be relatively modest at the regional level [3], the effects are likely to be severe in many urban and rural localities within southern Africa. Our own estimates, based on qualitative data on the predominant behaviour patterns and fitted to past trends in HIV-1 prevalence, indicate that life expectancy in Zimbabwe could decline from 55 and 59 years, in 1986, to 30 and 32 years, for men and women, respectively, by the year 2006 [18].
Acknowledgement
We thank the Secretary of State for Health and Child Welfare in Zimbabwe for permission to publish the results of this study and T. Dyson, G. Garnett, S. Ray, T. Muzenda and colleagues at the Blair Research Institute, who provided invaluable help in the preparation of this article.
References
1. United Nations: The AIDS Epidemic and its Demographic Consequences. New York: United Nations; 1991.
2. Anderson RM, May RM, McLean AR: Possible demographic consequences of AIDS in developing countries. Nature 1988, 332:228-234.
3. Bongaarts J: Global trends in AIDS mortality. Popul Dev Rev 1996, 22:21-45.
4. United Nations: World Population Prospects. New York: United Nations; 1993.
5. Anderson RM, May RM, Boily M-C, Garnett GP, Rowley, JT: The spread of HIV-1 in Africa: sexual contact patterns and the predicted demographic impact of AIDS. Nature 1991, 352:581-589.
6. Gregson S, Garnett GP, Anderson RM: Assessing the potential impact of the HIV-1 epidemic on orphanhood and the demographic structure of populations in sub-Saharan Africa. Popul Stud 1994, 48:435-458.
7. Gregson S, Garnett GP, Shakespeare R, et al.: Determinants of the demographic impact of HIV-1 in sub-Saharan Africa: the effect of a shorter mean adult incubation period on trends in orphanhood. Health Trans Rev 1994, 4:65-92.
8. Dondero TJ, Curran JW: Excess deaths in Africa from HIV: confirmed and quantified. Lancet 1994, 343:989-990.
9. De Cock KM, Barrere B, Diaby L, et al.: AIDS: the leading cause of adult death in the West African city of Abidjan, Ivory Coast. Science 1990, 249:793-796.
10. Mulder DW, Nunn AJ, Wagner H-U, et al.: HIV-1 incidence and HIV-1-associated mortality in a rural Ugandan population cohort. AIDS 1994, 8:87-92.
11. Mulder DW, Nunn AJ, Kamali A, et al.: Two year HIV-1-associated mortality in a Ugandan rural population. Lancet 1994, 343:1021-1023.
12. 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.
13. Garenne ML, Madison M, Tarantola D, et al.: Mortality impact of AIDS in Abidjan, 1986-1992. AIDS 1996, 10:1279-1286.
14. 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 1994, 7:839-852.
15. Zimbabwe Ministry of Health and Child Welfare: Report of an HIV/STD and AIDS Surveillance Workshop. Harare: Health Information Unit, Zimbabwe Ministry of Health and Child Welfare; 1994.
16. Parirenyatwa CN: Census 1992: Zimbabwe National Report. Harare: Zimbabwe Central Statistical Office; 1994.
17. Garnett GP, Anderson RM: Factors controlling the spread of HIV in heterosexual communities in developing countries: patterns of mixing between different age and sexual activity classes. Philos Trans R Soc Lond B Biol Sci 1993, 342:137-159.
18. Blair Research Institute/Oxford University: The Early Socio-Demographic Impact of the HIV-1 Epidemic in Rural Zimbabwe. Harare: Blair Research Institute; 1996.
19. National AIDS Co-ordination Programme: HIV and AIDS Surveillance Annual Report. Harare: Zimbabwe Ministry of Health; 1994.
20. McAlpine L, Parry JV, Mortimer PP: An Evaluation of Sixteen Combined Anti-HIV 1/2 Screening Assays. London: Medical Services Directorate; 1993.
21. Mandishona GM: Zimbabwe Intercensal Demographic Survey, 1987. Harare: Zimbabwe Central Statistical Office; 1991.
22. Hamilton LC: Statistics with STATA 3. Belmont: Duxbury Press; 1993.
23. Mandishona GM: Zimbabwe: Combined Demographic Analysis. Harare: Zimbabwe Central Statistical Office; 1992.
24. Folch-Lyon E, Trast J: Conducting focus group discussions. Stud Fam Plann 1981, 12:443-449.
25. Morgan DL: Focus Groups as Quantitative Research. Berkeley: Sage Publications; 1988.
26. Srinivasan L: Tools for Community Participation: A Manual for Training Trainers in Participatory Techniques. Washington, DC: PROWWESS/UNDP; 1990.
27. Garnett GP, Anderson RM: Balancing sexual partnerships in an age and activity stratified model of HIV transmission in heterosexual populations. IMA J Math Appl Med Biol 1994, 11:161-192.
28. Tendere R: Census 1992: Manicaland Provincial Profile. Harare: Zimbabwe Central Statistical Office; 1994.
29. Faijer DJ, Orellana H: Evaluation of Vital Statistics for the Study of Causes of Death. Edited by Timæus IM, Chackiel J, Ruzicka L. Oxford: Clarendon Press; 1996:45-68.
30. Gregson S, Zhuwau T, Anderson RM, et al.: Age and religion selection biases in HIV-1 prevalence data from antenatal clinics in Manicaland, Zimbabwe. Centr Afr J Med 1995, 41:339-345.
31. Foster G, Makufa C, Drew R, et al.: Supporting children in need through a community-based orphan visiting programme. AIDS Care 1996, 8:389-403.
32. Brass W: Indirect methods of estimating mortality: illustrated by application to Middle East and North African data. In The Population Framework: Data Collection, Demographic Analysis, Population and Development. Beirut: United Nations Economic Commission for Western Asia; 1978:121-165.
33. Parirenyatwa CN: Zimbabwe Demographic and Health Survey 1994: Preliminary Report. Harare: Zimbabwe Central Statistical Office; 1995.
34. Latif AS, Katzenstein DA, Bassett MT, Houston S, Emmanuel JC, Marowa E: Genital ulcers and transmission of HIV among couples in Zimbabwe. AIDS 1989, 3:519-523.
35. De Cock KM, Barrere B, Lafontaine M-F, et al.: Mortality trends in Abidjan, Côte d'Ivoire, 1983-1988. AIDS 1991, 5:393-398.
36. Barnett T, Blaikie P: AIDS in Africa. London: Belhaven Press; 1992.
37. Gomo E, Chandiwana SK, Chibatamoto P, et al.: Preliminary results of a HIV seroconversion study in the Lowveld. Lowveld AIDS Bull 1995, 2:12-16.
© Lippincott-Raven Publishers.