Vietnam has experienced a demographic transition characterized by decreasing fertility rates in the past two decades and marked improvements in child survival [1, 2]. Historical demographic trends have shown that such changes lead to increasing longevity and population aging, accompanied by a shift in disease burden from infectious to non-communicable diseases [3]. Reliable data on patterns of mortality and causes of death is critical to monitor such transitions, and to support the development of evidence-based health policy to minimise avoidable adult mortality. Unfortunately, accurate nationally representative statistics on total and cause-specific mortality in Vietnam are lacking due to inadequate routine information systems [4]. The only available information on mortality levels and patterns is from the Fila Bavi, a demographic surveillance site covering a population of roughly 50,000 located in a rural district about 100 km north of Hanoi [5, 6]. These data are limited to a small sample size, narrow geographic coverage, and pertain to a population that does not represent the socio-economic characteristics of Vietnam in general.

To monitor population and demographic indicators in Vietnam, the General Statistics Office (GSO) has been operating an annual population change survey (PCS) in a nationally representative population sample since 2000 [2]. Analysis of 2004, 2005, and 2006 mortality data from this survey indicated severe deficiency in death reporting, ranging between 39% to 59% for the whole country [7]. During 2007 - 2008, a research project was undertaken to improve the completeness of death reporting from the PCS conducted in 2007, and to use verbal autopsy (VA) methods to ascertain the cause of reported deaths. This project is a step towards the expanded use of VA to develop a long-term, national mortality data collection system in Vietnam. This paper presents total and cause-specific mortality results from the project, highlights methodological and logistical challenges in conducting a nation-wide assessment of mortality based on surveys, and discusses future strategies for routine mortality data collection in Vietnam.

The annual PCS is implemented in a nationally representative population sample derived through multi-stage stratified cluster sampling. The primary sampling unit is the Enumeration Area (EA) defined in the 1999 population census, which on average comprises approximately 100 households. Each year, a fresh sample of EAs is drawn from these pre-defined EAs. A total of 3840 EAs in the 64 provinces of Vietnam were included in the 2007 survey, covering about 2% of the national population.

The PCS in April 2007 collected information on deaths reported by households to have occurred between 1^st April 2006 and 31^st March, 2007. For each death, demographic information of the deceased (e.g., age, sex, date of death) and household address were used to locate the household for subsequent VA interviews with the principal care taker of the deceased. For national application of VA, the WHO standard VA questionnaire [8] was adapted to the Vietnamese context, based on experience with VA in the Fila Bavi demographic surveillance site [5, 6]. (see Additional file 1).

Support for training, fieldwork supervision and data management, was provided by 5 medical universities, including Thai Nguyen, Hanoi in the North, Hue in the centre, and Ho Chi Minh and Can Tho in the South. For logistical convenience, 64 provinces were divided into 5 clusters (Figure 1), corresponding with the location of each medical university. VA interviews were completed in 6 months from September 2007 to April 2008, excluding 2 months of New Year (January and February 2008). Filled questionnaires were reviewed by a team of experienced medical doctors at each medical university, who then assigned the causes of death following the standard death certification form [8]. Underlying causes of death were selected and coded using ICD Version10 (ICD-10).

This paper presents the first ever findings on nation-wide total and cause specific mortality patterns in Vietnam. Although there are several constraints in the quality of the data, the broad findings from the study have important implications for Vietnam. Firstly, the adjusted estimates of life expectancy at birth and levels of adult mortality are consistent with previously observed mortality time trends in Vietnam [12] and demonstrate steady improvements in population health status in Vietnam over the past three decades. Secondly, important differences were noted when our results on cause-specific mortality are compared with the findings from the VA study of 189 deaths in Fila-Bavi in 1999 [5]. While diseases of the circulatory system, cancers, and accidents were the three prominent causes of deaths among all age groups in both studies, infectious diseases, prenatal and neonatal causes were not among 10 leading causes in our study. These differences are important, but unsurprising, given the relatively small sample and rural setting of the Fila-Bavi demographic surveillance site.

Overall, the use of VA has resulted in about 10% of deaths being classified to ill-defined conditions among males, and 15% in females, which is acceptable given the challenges in diagnosing causes of death from this method. In old age (60 years or older), the proportion of deaths classifiable is lowest compared to that in other age groups, a similar finding to studies in India that reported VA is a less reliable to ascertain cause of death for older age [13, 14].

Despite the incompleteness of death recording, the proportionate mortality by cause at different ages provides an empirical basis for understanding health priorities. The findings suggest a clear need to improve health services to control perinatal mortality, and the need for evidence based interventions to reduce deaths from traffic accidents and drowning. Among adults, the observation that cerebrovascular disease causes about 5 times the number of deaths as ischemic heart diseases (IHD)calls for more detailed research into the epidemiology of these conditions, and the implementation of primary and secondary prevention strategies. This ratio could be skewed on account of IHD deaths being misclassified as deaths from hypertensive diseases, which is observed to cause about double the proportion of deaths from IHD at ages 60 and over. This requires careful evaluation through assessment of the reliability and/or validity of VA application in Vietnam. Additional findings for males include the emergence of HIV/AIDS as a leading cause of death at ages 15-59 years, and tuberculosis as the seventh leading cause of death at all ages. All these findings signify the need to improve surveillance and treatment programs for these conditions.

From an operational perspective, integration of activities to measure cause-specific mortality with the existing annual national PCS offered several advantages. Firstly, it saved resources that otherwise would be needed for identification of deaths in a nationally representative sample, given incomplete vital registration in Vietnam. Secondly, mortality data provided by GSO is legally recognized by the government of Vietnam, and therefore, the results from the study can inform policy development. In this regard, activities in this project to strengthen death recording by GSO staff did result in improvement in the completeness of data. More importantly, however, the activity enabled 5 medical universities throughout the country to develop experience in the systematic investigation of cause of death on a nation-wide scale, using the VA method. This has created an institutional network in different regions, which is a sound platform for routine and sustained implementation of cause-specific mortality data collection systems in all parts of Vietnam.

However, despite the operational advantages described above, the GSO annual PCS presented significant challenges in translating the data collected into reliable information for routine mortality measurement. Firstly, a principal issue with the data is the low completeness of deaths recorded by surveyors (see Additional File 2). Secondly, there were several challenges in the implementation of VA interviews in conjunction with the survey (see Additional File 1). Hence, while this exercise has proved to be a valuable experience as the first ever national level mortality and cause of death data collection in Vietnam, more effective and sustainable options are required for routine implementation. To meet this challenge, sentinel mortality surveillance sites are being tested in 2009. The proposed sentinel sites comprise 192 communes in 16 provinces, covering an optimal sample population given the mortality profile for Vietnam [15]. At the commune level, data collection efforts involve the collaboration of the official civil registration system, the commune health station, and local 'population collaborators' from the Population and Family Planning Department of the Ministry of Health. The tools, methods, and experiences from the current study would be used together with technical support from the five medical universities. Continuous data collection in these sites over the next 5-10 years, supported by appropriately designed studies to validate reported causes of death, would yield valuable evidence on current mortality patterns and trends, to inform health policy and epidemiological research in Vietnam.

This study provides evidence that verbal autopsy can be a viable tool to identify cause of death in Vietnam. However, its use in conjunction with the national sample survey is limited by incomplete death reporting as well as logistical challenges in implementing household VA interviews. The proposed sentinel mortality surveillance sites being tested in 2009, using tools, methods and experiences from the current study, will help better understand the applicability of the VA method and develop effective and sustainable systems for routine mortality assessment in Vietnam.