The overall picture of adult mortality in this area has to be contextualised in terms of local circumstances and effects. There was no very clear trend in mortality over the entire 18-year period, but very large and specific effects were noted during the course of the study. In the first few years, the very considerable excess male mortality in the urban area was likely associated with the final years of the Ethiopian civil war. Although the Butajira area was not heavily militarised, there were high rates of somewhat random conscription of adult males to conflict areas, which was easier to implement from higher-density urban areas. There were 50% more women than men in the 25–34 year age group resident in Butajira at that time, which gives some indication of the extent of the conscription effect and makes it plausible that the observed excess in male mortality at that time could be largely due to the war. This in turn makes it difficult to interpret longer term gender effects on mortality.
The second major effect on mortality, this time affecting primarily rural areas, was a period of unusual rainfall in 1998–2000 which led to widespread food shortages and consequent epidemics. This has been described in detail elsewhere . This also complicates the overall interpretation of mortality rates, particularly in respect of the figures at the end of the period. Subsequent fall in rates may partly be due to a shadow effect of premature mortality during the epidemic period, although this was at the same time as a new hospital coming on-stream in Butajira.
Notwithstanding these important specific mortality phenomena, the effects of urban and rural lifestyles, gender and relative poverty can be considered in relation to mortality. As has consistently been the case in previous analyses from Butajira [6, 9] there is a mortality disadvantage associated with rural living which overrides age and gender effects. The precise determinants of this effect remain to some extent unclear, since distance to the town, access to health facilities, lack of a cash economy and differential availability of food are all highly confounded within the urban-rural paradigm. The fact that the 1998–2000 rainfall and food supply phenomena discriminated against those having a rural lifestyle, even in the case of villages situated very close to the town, emphasises the urban-rural divide.
In terms of gender, adult mortality differences were minimal, particularly if the apparent war-time effects on male mortality are discounted. Butajira district still experiences relatively low levels of HIV prevalence compared with other locations in sub-Saharan Africa, and so HIV-related mortality is probably still too small as a proportion of total mortality to have any appreciable overall effect. The relatively higher rural female mortality where there was no joint household decision making pinpoints the effect of lower status for rural women . Decision making is a key indicator of women's status and interventions to empower women have shown great impact on women's quality of life, autonomy and authority, and improved infant and maternal survival [7, 18].
Poverty – whether in terms of assets, education or social capital – seems to be an important determinant of mortality. Education has been previously associated with improved survival in the area [9, 10]. Absence of a literate person in a household appears to particularly affect mortality among males and in rural areas. It is possible that presence of a literate person is more advantageous to men than to women in terms of improving awareness from which men, who are generally the decision makers, may be benefiting to a greater extent. This issue needs to be explored further.
Adjusted for other factors, house ownership affected rural residents rather than urban residents, although the proportion of residents in rural areas who do not own the house they live in is much smaller (about 5%). Rural houses in Butajira are generally small huts . Thus residents in rural areas who do not own houses could be a particularly disadvantaged group, socially and economically. By contrast, the urban lifestyle in Butajira town for many people amounts to renting a basic room or part of a house, implying that urban non-owners are part of a cash economy and may be very different from rural non-owners.
A limitation of the surveillance data is that not all important predictors of mortality were included in regular data collection and some parameters, such as cause of death, were imprecise. Thus classification between communicable and non-communicable diseases was based on the responses to simplified questions and probably introduced some misclassification bias. In addition, some missing data on literacy and marital status may have also introduced some bias.
A limitation of the case referent component is the potential for recall bias and possible gaps in information on the deceased by respondents, although the study was conducted within a reasonable period of time . Factors that were used to assess social capital, for example, did not show differences between the cases and referents and this may be related to respondents' lack of accurate information about the deceased concerning these issues. In addition, the sample size did not allow comparisons of mortality effects in sub-groups.
By comparison with a previous study , there appears a higher occurrence of non communicable disease, indicating the increasing double burden of communicable and non communicable diseases. Urban-rural differences in non-communicable and communicable diseases may have been obscured by opposing influences of increases in non-communicable diseases and HIV/AIDS . Thus, while deaths due to communicable diseases remain high, indicating a relatively early stage of epidemiological transition, increases in NCDs and HIV/AIDS would indicate a need for preparedness to deal with this triple burden.
Being unmarried was strongly associated with mortality among urban males. This was not true for other females and males in rural areas. That mortality rates are lower for married individuals has long been documented , which could be due to selection or protection. Excess mortality seen among the unmarried in rural Bangladesh supports the idea that mortality differentials may be partly attributed to selection for marriage .