Adult mortality in the cities of Bulawayo and Harare, Zimbabwe: 1979-2008

Background Zimbabwe has been severely affected by the HIV/AIDS and tuberculosis epidemics, with an estimated 80% of tuberculosis patients being HIV infected. We set out to use annual population-mortality records from the cities of Harare and Bulawayo to describe trends and possible causes of mortality from 1979 to 2008. The specific objectives were to document overall, sex and age-specific mortality, proportion of deaths attributed to AIDS and tuberculosis, and changes in death rates since the start of antiretroviral therapy in 2004. Methods This retrospective descriptive study used existing mortality records of the Health Services departments in Harare and Bulawayo. Data points included: estimated yearly total population; groupings by sex and age; deaths (total and by sex and age groups for each year of the study period); and most frequently reported causes of death (for age groups <15 years, 15-44 years and ≥45 years). Data on deaths were aggregated by year, and crude, sex- and age-specific death rates were calculated per 1000 population. Tuberculosis and HIV-related disease-specific death rates and proportion of deaths attributed to these conditions were computed. Results In both cities, crude death rates were lowest in the late 1980s, increased three- to five-fold by the early 2000s, and began a slow and, in the case of Bulawayo, intermittent decline from 2004. Sex-specific death rates followed a similar trend, being higher in males than in females. The death rates in the age groups <5 years, 15-44 years and ≥45 years showed significant increases, with a gradual levelling off and decline from 2002 onwards; death rates in those aged 5-14 years were relatively unaffected. Tuberculosis and HIV caused 70% of deaths in the age group of 15-44 years from the early 1990s. Conclusions This study used routinely collected population-mortality data that are rare in resource-limited settings, and it described, for the first time in Zimbabwe, the effects of the HIV/AIDS epidemic and the introduction of antiretroviral therapy on death rates in two large cities. After a substantial rise in crude mortality rates, there has been a decline associated with the introduction of ART. Such routine population data must continue to be collected, collated and analyzed.

2004 and of these, three were situated in the cities of Harare and Bulawayo.
At the end of December 2009, of the 1.2 million individuals living with HIV, 219,000 (of whom 22,000 were children) had been started on ART [2]. ART coverage was estimated at 56%, based on the Guidelines for Antiretroviral Th erapy in Zimbabwe [3] that recom mended treatment for HIV-infected individuals with CD4 counts of <200 cells/mm 3 .
Th e fi rst death from HIV infection in Zimbabwe was reported in 1988 in Harare. Since then, the country has seen a huge increase in HIV-related TB deaths, although there is almost no published data available on population mortality and its causes [4]. However, mortality statistics, collected and collated annually, have been available in bound volumes in the city councils of Harare and Bulawayo for several decades. Th ey are believed to be fairly reliable for reasons described later.
Th e aim of the current study was to use these city records to describe the trends and possible causes of adult mortality over a 30-year period in Harare and Bulawayo. Th e specifi c objectives were to document by city: (i) overall mortality, including sex-and age-specifi c (15-44 years) mortality, from 1979 to 2008; (ii) the proportion of deaths attributed to AIDS and tuberculosis; and (iii) whether there was any reversal in death rates since the start of provision of ART.

Study design
Th is was a retrospective descriptive study using yearly mortality records in Bulawayo and Harare from 1979 to 2008.

Study setting General
In Zimbabwe, municipal authorities are responsible for provision of primary healthcare services and basic diagnosis and management of common conditions. Large municipalities, such as Harare and Bulawayo, have Health Services departments to carry out these duties. Th e study was conducted using annual reports from the Health Services departments of Harare and Bulawayo. Th e population estimates of the two cities were based on the national census reports of 1982, 1992 and 2002, with interim numbers provided by the projections of the Central Statistical Offi ce.
HIV prevalence in the two cities has been high since the start of the HIV/AIDS epidemic, and in 2009, it was found to be 12% in Harare and 14% in Bulawayo among antenatal care clients aged 15 to 49 years [5]. By the end of 2008, there were fi ve public health sector ARTinitiating sites in Harare and six in Bulawayo. Th e number of individuals ever started on ART in the Bulawayo sites, including four primary healthcare clinics, was approximately 30,000 by 31 May 2010, which was estimated to represent 45% of the need. Th ese estimates were unavailable for Harare.

Annual mortality reports
Annual reports are prepared by the Health Services departments and presented to city councillors and shared with relevant stakeholders. Th ese reports have been available for several decades in Bulawayo and Harare. Th ough the reporting styles are not identical and have changed, especially after 1980 when the country gained its independence, most reports include a section on vital events, such as deaths, that are frequently disaggregated by sex and age group. Mortality data include deaths that occur both in health facilities and homes (the latter accounting for up to 40% of all deaths [unpublished data]). Th ese data are believed to be fairly comprehensive because no burial in city cemeteries that are managed by the Health Services departments is permitted without registration of a death. Transportation of a body out of the city without necessary documentation is also impossible as it is reinforced by police offi cers who man roadblocks along the highways.
A death notifi cation is issued by a medical offi cer at a facility where the death occurred or where the deceased received care. Th is form captures the direct, underlying and contributing causes of death, and allows the acquisition of an internment order (issued by the Health Services departments) and a burial order and death certifi cate (issued by the Registrar for Births and Deaths).
Th e departmental clerks collect information from the Registrar's offi ces on the number of deaths that have occurred among residents in the city during the year. Clerks also prepare line listings that consist of age, sex and causes of death as written in death certifi cates, and exclude personal identifi ers, such as name and address of the deceased. Th ese manual lists are then examined by medical offi cers who code the causes of death, if more than one has been included in the certifi cate, into broad categories, such as tuberculosis, AIDS, pneumonia/chest infections, and accidents. Th e information is then summarized in tables and presented in the departmental annual reports.

Data collection
Th e annual reports from each city from 1979 to 2008 were retrieved and data on deaths were extracted. Th e yearly population estimates that the departments had used were also summarized. A spread-sheet template was developed to enter: total population and population by sex and age group; total number of deaths and number of deaths by sex and age groups (less than fi ve years, 5-14 years, 15-44 years and 45 years and above); and the most frequently reported causes of death (for age groups younger than 15 years, 15-44 years, and 45 years and above).

Analysis
Data on deaths, including deaths stratifi ed by sex and age, were aggregated by year and the crude, sex-and agespecifi c death rates were calculated per 1000 population based on population estimates for each city. Chi-square for trend test was performed on the yearly reduction in the crude mortality rate from 2003-2004.
For cause-specifi c mortality, we focused on tuberculosis and HIV-related disease for the age group of 15-44 years and from 1988 to 2008 only. We included all types of tuberculosis, and the proportion of deaths attributed to tuberculosis was calculated for each year. For HIV-related disease, we included death due to HIV-related disease, tuberculosis, pneumonia/Pneumocystis jirovecii (or previously, carinii) pneumonia, meningitis, malnutrition, Kaposi's sarcoma and septicaemia; the proportion of deaths attributed to HIV-related disease was calculated for each year. Deaths due to tuberculosis and to HIV/ AIDS are, therefore, not mutually exclusive in this analysis.

Ethics approval
Permission for the study was obtained from the Directors of the two Health Services departments and the Ethics Advisory Group of the International Union Against Tuberculosis and Lung Disease.

Population
Th e estimated total population in Harare increased from 653,800 in 1980 to 1,532,800 in 2008 [6 and unpublished data]. In 2008, 765,000 (50%) inhabitants were females. Th e estimated total population in Bulawayo increased from 391,100 in 1980 to 719,500 in 2008 [6 and unpublished data]. In 2008, there were 369,500 (51%) females.

Crude death rate
Th e total number of deaths during the 30-year period was 283,926 in Harare and 200,124 in Bulawayo. Th e trends in crude death rates were similar in both cities, with Bulawayo having consistently higher rates than Harare.
Th e crude death rate per 1000 population was 6. rate showed a steady decline in Harare and an intermittent decline in Bulawayo. In 2008, the rates were 9.9 and 12.6 per 1000 population in Harare and Bulawayo, respectively. Overall, there was a 19% decrease in crude death rates in both cities between 2003 and 2008: for Harare, the chi-squared test for trend was 629.8 (p <0.001) and for Bulawayo, it was 121.4 (p <0.001).

Sex-specifi c death rate
In both cities, there were some years when mortality infor mation by sex, age group and cause was unavailable ( Table 1). In Harare, sex-specifi c death rates were available for a 26-year period (1982-1993 and 1995-2008), and these rates are shown in Figure 2. During all years, mortality in males was higher than in females. Th e overall trend of the sex-specifi c death rates resembled the trend seen in crude death rates in that the lowest rates were recorded in the late 1980s (when the death rate was 2.3 and 2.5 per 1000 population in females and males, respec tively) and the highest in the early 2000s (when the death rate was 11.1 and 13.3 per 1000 population in females and males, respectively), after which the rates started to decline.
Th ere were more missing data in Bulawayo, but these (not shown) followed the same trends as were seen in Harare, although death rates were consistently higher than in Harare, especially among males, in whom a mortality rate of 17.8 per 1000 population was reported in 2003.

Age-specifi c death rate
In Bulawayo, there were several years when information on age-specifi c death rates was unavailable (Table 1). However, in Harare, these data were available for a 29-year period from 1980 to 2008 for the age groups of 5-14 and 15-44 years, and for a 30-year period for the other age groups (Figure 3).
In all age groups, the lowest death rates during the study period were recorded from the mid-1980s to 1988 and thereafter, the mortality rose, peaked at 9.9 per 1000 population in the age group of 15-44 years from 2002 to 2004, and subsequently declined. Th ree age groups were principally aff ected during these 30 years (younger than fi ve years, 15-44 years, and 45 years and above), with all three showing an increase in mortality from 1989 to 2002, and thereafter, a decline reaching 7.6 per 1000 popu lation (15-44 years) in 2008. Th e age group of 5-14 years was relatively unaff ected.
In Bulawayo, all-cause mortality in the age group of 15-44 years followed a similar trend and reached its highest point in 2003 at 15.7 per 1000 population and then declined to 12.3 per 1000 population in 2008 (not shown). Similar to Harare, the three age groups predominantly aff ected were younger than fi ve years, 15-44 years, and 45 years and above, while the age group of 5-14 years was relatively unaff ected.

Cause-specifi c mortality in the 15-44 year age group
Cause-specifi c mortality due to tuberculosis and HIVrelated disease in the age group of 15-44 years from 1988 to 2008 is shown in Figures 4a and 4b. Data were unavailable in Bulawayo in 1990, 1992 and 1998. TB-specifi c mortality increased 18-fold in Bulawayo and 25-fold in Harare, and was 2.3 per 1000 population in both cities in 2008. Mortality due to HIV-related disease increased 23to 28-fold, and the death rate was 5.7 and 9.5 per 1000 population in Harare and Bulawayo, respec tively, in 2008.

Proportionate mortality ratio in the 15-44 year age group
Th e proportion of all deaths in this age group that were recorded as due to HIV/AIDS and related conditions, as stated in the Methods section, is shown in Figure 5. Data were unavailable in Bulawayo in 1990, 1992 and 1998. Initially low, the proportion increased to more than 70% in 2000 and has remained at that level ever since.

Discussion
Th is is the fi rst study in Zimbabwe to use routinely collected mortality data to assess the impact of the HIV/ AIDS epidemic and introduction of antiretroviral therapy on death rates in Harare and Bulawayo. Crude death rates    in both cities rose staggeringly between the late 1980s and early 2000s and began a slow and, in the case of Bulawayo, intermittent decline from 2003-2004, coincident with the advent of antiretroviral therapy. Interestingly, during the entire study period, overall mortality was consistently higher in Bulawayo than in Harare, and the reason for this diff erence is not known. Sex-specifi c death rates followed a similar trend, being higher in males than in females.

Years when available in annual Years when available in annual Type of information reports in Bulawayo reports in Harare
Not surprisingly, the death rates in the age groups younger than fi ve years, 15-44 years, and 45 years and above showed signifi cant increases during the same period and a gradual levelling off and decline from 2002-2003 onwards. Th e age group relatively unaff ected during this time was that of persons aged 5-14 years, who epidemiologically would have been protected against the ravages of the HIV/AIDS epidemic. Finally, there was an enormous rise in tuberculosis-and HIV-specifi c Th e strengths of this study relate to the collection and compilation of death information in both cities by the Health Services departments, a system that has continued uninterrupted over the years despite major political and economic challenges faced by the country during some of this time. We believe that the numerical data on deaths in urban areas are fairly reliable because no burial in a city cemetery or transportation of a body out of the city could and can be done without documentation. Th is diff ers signifi cantly from death registration coverage in rural areas, where many deaths may go unreported [1]. Th e data also fi t with what is known about the epidemiology of the HIV infection, especially when disaggregated by age group.
However, this was an operational study based on data collected within routine systems for which there are obvious limitations. We do not know the accuracy of the annual population estimates for the cities based on the 10-year national census data with projected numbers in between. Th ese data may be aff ected by urban-rural and rural-urban migration, as well as by emigration during recent years of insecurity and economic turmoil. Th e population denominator data may, therefore, have inaccura cies which will aff ect the results.
While we believe the number of annual deaths to be fairly accurate, this may not be the case. Th e Registrar's offi ce should record all deaths that occur in the city for that year, whether the person is resident or not in the city, but deaths in infants and children in particular may not always have been captured. Over the 30-year period, there may have been changes in health information clerks, who may have used diff erent criteria for inclusion and exclusion of deaths. Th ere were also no standard operating procedures for recording deaths, so practices between medical offi cers in diff erent cities and within a city over time might have diff ered, creating inaccurate attributions regarding causes of death. Th e format of the death notifi cation requires determination of direct, underlying and contributing causes of death, although frequently, in practice, the cause of death is limited to just one line, for example, "retroviral infection". Although the International Classifi cation of Diseases and Deaths (ICD) is recognized in Zimbabwe, most causes of death are not accompanied by an ICD-recommended code.
Autopsy is the only way to truly ascertain cause of death, but in Zimbabwe, as is the case in most of sub-Saharan Africa, they are rarely performed. Many deaths also occur at home, especially in the case of chronic illness due to HIV/AIDS and TB, where medical certification might have been cursory, and this may have lead to an over-estimation of deaths attributed to HIV/AIDS and TB. Finally, it would have been interesting to assess the eff ect of the HIV/AIDS epidemic on non-HIV related deaths and broader health outcomes, but the ways in which the annual reports present the data precluded such analysis.
Our results are supported by previous studies in a Zimbabwean province, Manicaland, which also described increased mortality among young and middle aged adults, with 61% and 70% of mortality in males and females, respectively, being attributed to HIV/AIDS [1,7]. Modelling studies also strongly suggest that the increased mortality rates were due to HIV/AIDS [8]. Th e Zimbabwe Demographic and Health Surveys [9] found higher death rates for males than females in 1994, 1999 and 2005, but mortality rates were not shown by province and therefore, no data were available for Harare and Bulawayo.
Th ere have been a few population-based studies in other African countries assessing the impact of HIV/ AIDS and the advent of antiretroviral therapy on population mortality. Studies have been conducted in Ethiopia [10,11], Botswana [12], South Africa [13,14,15] and Malawi [16,17] assessing the impact of antiretroviral therapy on HIV-related deaths at the population level, and showing that the numbers of HIV/AIDS-related deaths have declined substantially with ART. Th e reductions in death rates in Africa have generally been less dramatic than those reported from New York [18] and São Paulo [19].
In our study, the decline in the crude death rate was only 19%, which may be due to several factors. First, the HIV incidence in Harare and Bulawayo appeared to have peaked early in relation to ART: back-calculation from mortality data indicated that an annual incidence peak of 4-5% could have been reached in Harare between 1988 and 1990. In Bulawayo, a higher annual incidence peak of 5-6% was estimated to have occurred between1992-1994 [20]. Second, the number of HIV and AIDS cases was and continues to be large, with consequent poor coverage of ART as a result of the immense work load and relatively weak health infrastructure. Th ird, socio-cultural and stigma-related factors frequently impede demand for HIV testing and make persons present late for HIV diagnosis and treatment, which compromises survival [17].
Interestingly, studies from Manicaland, before availability of ART, reported stabilization of mortality rates because of decreasing HIV prevalence partly due to the natural consequence of the HIV epidemic and partly due to recent behavioural changes as a result of HIV prevention programmes in the country [21][22][23][24][25]. Our data from urban Zimbabwe add to this growing literature from the African continent, support the general impression of a decline in population death rates consequent upon expansion of antiretroviral treatment, and provide a 30-year glimpse at mortality trends in two large African cities that span the pre-AIDS era, the HIV/AIDS epidemic and the advent of ART.
More importantly, this study also demonstrates the importance and value of well-kept records. Routine systems that can measure population numbers and population deaths on an annual basis are rare in resourcepoor countries [26], but where they do exist, as in Harare and Bulawayo, they need to be used, supported and developed so that the data can become more reliable and useful as a measure of what is happening at the population level. Routine usage and regular supervision will help improve the reliability of the data, and the value of such data needs to be demonstrated to policy makers so that commitment towards collection, collation and analysis is attained.

Conclusions
It is important to collect evidence to show that eff orts to provide HIV diagnosis and care, including ART, and funds invested by both national governments and various funding agencies have an impact in averting premature deaths [27,28]. Furthermore, two Millennium Development Goals have mortality-based targets, and evidence is also needed to assess whether these targets are being reached.
Th is study used routinely collected mortality data that are rare in resource-limited settings, and described, for the fi rst time in Zimbabwe, the eff ects of the HIV/AIDS epidemic and introduction of ART on death rates in two large cities. It shows the value of these data in describing the eff ect of the HIV/AIDS epidemic and the impact of HIV care and ART on population-level mortality in spite of certain limitations, such as unknown accuracy of population estimates in the cities and lack of autopsyconfi rmed causes of death. Th e data can also be used in the future to assess population-level impact of better access to care through expansion and decentralization of HIV treatment services.