All-cause crude mortality was 17.2 per 1,000 person-years in the four provinces studied from 2009 to 2010; nearly half (49%) of deaths occurred at home. Overall leading causes of death were HIV/AIDS, malaria, and circulatory diseases, generally reflecting the order of sex-specific causes. Sex-specific differences in the order included injuries, the third leading cause of disease for men, and malnutrition, the fourth leading cause of death for women.
These leading causes of death were similar to those previously reported in the region [18
]. In a worldwide summary of mortality published in 2009, leading causes of adult mortality in the African region were derived largely from population-based studies, and included HIV/AIDS (35%), other infectious causes (including malaria), and injuries (for men) [22
]. Separate SAVVY-based studies in Tanzania and Kenya showed that the majority of non-infant deaths were attributable to HIV/AIDS, tuberculosis, and malaria, and approximately 6% were attributable to cardiovascular disease (CVD).
With modification based on lessons learned from this pilot, new technologies being developed, and a gradually increasing governmental commitment to fund collection of vital statistics data, it is feasible that vital registration data may be collected using SAVVY methodology in the future.
Zambia was the first country in Africa to use WHO-recommended SAVVY methodology to collect vital events data. WHO recommends conducting a dedicated census just prior to conducting verbal autopsy interviews. These activities require donor funding and the undivided attention of professional government staff. While costly on monetary, opportunity cost, and staffing bases, this rigorous methodology allows for collection of standardized census data specific for vital events registration. Adaptations of WHO-recommended SAVVY methodology that rely on national censuses, like the one conducted in Mozambique in 2007 [17
], likely realize cost and time efficiencies that may improve sustainability. However, these efficiencies must be weighed against the adverse impact that longer recall periods may have on data quality [6
]. Using a dedicated census allows for a shorter recall period for retrospective identification of deaths. Additionally, the recall period can be determined by the study implementers, rather than by scheduled national censuses. Using standardized approaches to verbal autopsy and SAVVY-based vital registration could also improve the ability to compare results across countries and regions [16
Zambia employed physicians to code interviews into ICD-10 causes of death (physician- certified verbal autopsy [PCVA]). Two physicians coded each interview and, if their codes differed, they discussed the case to agree on a final code. Physicians were unable to consider medical records in coding deaths as recommended by WHO because most families of those who had died did not keep these medical records. Despite this limitation, we viewed PCVA and duplicate coding as strengths. However, recent publications have suggested a low (30%) concordance between PCVA and a gold standard, in this case known cause of death [26
]. Recent reports have also cast doubt that duplicate coding improves data quality [27
Computer-based algorithms such as InterVA and the Symptom-Pattern and the newly developed Tariff and Random Forest methods are promising alternatives to PCVA [29
]. Algorithmic methods have been shown to perform as well as or better than PCVA in cause of death assignment without the personnel cost [29
]. For particular individual causes of death, some algorithms have been shown to perform better than PCVA. But algorithms lack the ability to identify and prioritize causes of death that are of public health importance in specific settings, to adapt to changing disease patterns, and to accurately identify less common causes of death [28
]. Overall, in recent comparisons, despite statistical differences in results generated by PCVA and algorithm-coded methods, leading causes of disease and groups most burdened have been similar and have had the same policy implications [19
During data collection for this pilot phase, we did not use algorithms to code causes of death because they have only recently been developed and validated. Once they are refined and made available for tailoring and testing in Zambia, they could be used here.
Although this was a pilot and included just four of the nine provinces in Zambia, our crude all-cause mortality rate, 17.2 per 1,000 person-years, was somewhat similar to the 13.3 per 1,000 person years estimated for 2009 and 2010 by the Central Statistics Office's (CSO) "Population Projections Report" based on projections from 2000 census data [33
]. Our estimate, which included a substantial proportion of residents of Lusaka Province, was essentially equal to CSO's projections for Lusaka Province (17.1 per 1000) and similar to results from a recently conducted analysis reporting 14.1 to 14.5 deaths per 1,000 in Lusaka Province [34
]. Although samples were not drawn to be representative of the country, our crude maternal mortality rate (1.6 per 100,000 women aged 15-49) was somewhat similar to the 1.2 per 100,000 rate reported by the Demographic Health Survey (DHS) for 2002-2003 [35
]. However, our under-5 mortality was substantially lower (80 per 1,000) than the DHS reported with regard to 2003-2004 data (119 per 1,000) [35
]. This difference could suggest an ascertainment gap in our data that should be investigated. Under-5 mortality could also be underreported because of stigma associated with discussing early childhood deaths. Otherwise, few sources of mortality data exist in Zambia. To our knowledge, there is no other source of representative data on the distribution and causes of death in Zambia.
More than 80% of people who subsequently died were reported to have sought treatment at a government clinic at some time prior to death. Many delayed too long. These findings suggest that educational material should be posted in waiting areas of government health clinics to alert people to early signs and symptoms of illnesses that should prompt them to seek medical care. Based on the demographic profile of those who died, messaging should be simple, pictorial, and in large print.
Despite rapid scale-up of national programs to provide free highly active antiretroviral treatment (HAART) and prevention of mother-to-child transmission, HIV is still the leading cause of death in these four provinces. This finding is similar to other countries in sub-Saharan Africa [24
] and the entire African region [23
]. Research suggests that HIV-related death is most common in the three months following treatment initiation and is associated with advanced HIV disease at presentation [36
], thought to indicate delays in seeking care. Long distances from homes to health care centers providing HAART have been linked to delayed treatment, particularly in rural areas [37
]. Zambia is currently incorporating new, more aggressive treatment guidelines that may improve survival [38
]. It is hoped that implementation of these guidelines will lead to reductions in HIV-related mortality, although they do not address the distance barrier. Zambia should be able to evaluate trends in HIV-related mortality before and after implementation of the new guidelines with the continued and ongoing collection of vital events data using SAVVY.
Other leading causes of death reported here were also reported by others in the region, including malaria [24
], circulatory diseases [23
], and injury [23
Challenges existed with data collection and analysis that should be taken into account when interpreting the data. First, verbal autopsy interviews were conducted by nurses who could have introduced their own professional judgment and biases into the coding process by selecting keywords associated with the illnesses that they inadvertently "diagnosed" during the interview. Secondly, the census did not collect any information about ages under 1, so it was not possible to calculate neonatal mortality rates. Thirdly, when field staff assessed health care sought, they meant in the three months prior to death, but we understand that this time frame was not consistently explained. Because of this oversight, in addition to health care sought for the illness leading to death, we may have also captured health care sought for conditions that the person had previously but from which they didn't die. Fourth, this was the pilot phase of SAVVY in Zambia and the sampled areas do not necessarily represent the country. The next phase is designed to complement the pilot and, together, provide nationally-representative estimates. Fifth, most of the households interviewed were unable to provide clinical records such as laboratory results. Our reliance on a lay description of the family member's symptoms likely resulted in misclassification of cause of death in some cases. Sixth, as a cultural practice, stillbirths and neonatal deaths are not generally acknowledged by families as deaths, and so are likely greatly undercounted in this study. Seventh and finally, sample randomization for this 2009-2010 assessment was based on a national census from 2000, which was clearly out of date.
Other aspects of Zambia's application of SAVVY likely contributed to high quality and completeness of data. For instance, a dedicated census allowed for shorter recall periods for our interviewees. Additionally, community health workers and traditional birth attendants were employed and trained to identify deaths in their own communities for autopsy interview. As community members, they also facilitated entry of SAVVY interviewers into their neighbors' households.
In part, because of the strengths and despite the weaknesses, these data can be used to determine needs and gaps in the health care system. Results could be used to develop community-based interventions to improve survival in the groups identified as most at risk for death. Based on our results, interventions could include improvements in HAART access for people with HIV; access to treated mosquito nets for malaria prevention and access to prompt treatment for those with malaria; clinically-attended birthing and nutritional support for females; access to information about preventing and treating circulatory diseases; and increased education to parents about knowledge of signs and symptoms that should prompt urgent medical attention of their young children. Further in-depth study is needed to develop interventions to avert deaths, especially those that are preventable and treatable.
While SAVVY data have not yet been linked with Zambia's national electronic health records (EHR) system, "SmartCare," identifiers used in the system are compatible with those used in SAVVY. With dedicated effort, appropriate approvals, and confidentiality protections, the health care, morbidity, and mortality data collected in health facilities that are captured in SmartCare could be linked with the verbal autopsy and census data captured in SAVVY. This linkage could identify clinical antecedents of mortality, providing a more comprehensive description of gaps in care and prevention.
Finally, discussion is also needed worldwide to determine whether the gains enjoyed from standardizing SAVVY methods across countries are worth the potential loss of cost efficiency and perhaps the additional sustainability gained by integrating it with other activities.