Conceived and designed the experiments: CLFW REB MJA CB-P. Performed the experiments: CLFW MJA. Analyzed the data: CLFW MJA. Contributed reagents/materials/analysis tools: MJA. Wrote the paper: CLFW MJA REB CB-P.
The authors have declared that no competing interests exist.
Diarrhea remains one of the leading causes of morbidity and mortality among children under 5 years of age, but in many low and middle-income countries where vital registration data are lacking, updated estimates with regard to the proportion of deaths attributable to diarrhea are needed.
We conducted a systematic literature review to identify studies reporting diarrhea proportionate mortality for children 1–59 mo of age published between 1980 and 2009. Using the published proportionate mortality estimates and country level covariates we constructed a logistic regression model to estimate country and regional level proportionate mortality and estimated uncertainty bounds using Monte-Carlo simulations.
We identified more than 90 verbal autopsy studies from around the world to contribute data to a single-cause model. We estimated diarrhea proportionate mortality for 84 countries in 6 regions and found diarrhea to account for between 10.0% of deaths in the Americas to 31.3% of deaths in the South-east Asian region.
Diarrhea remains a leading cause of death for children 1–59 mo of age. Published literature can be used to create a single-cause mortality disease model to estimate mortality for countries lacking vital registration data.
Although child survival rates have made improvements in the past decade, 8.8 million children still die every year before reaching their 5th birthday
The Child Health Epidemiology Reference Group (CHERG) has developed standards for systematic literature reviews and a cross validation process including both single and multi-cause modeling approaches for estimating causes of child deaths where vital registration data are not available
We conducted a systematic review of all papers reporting diarrhea proportionate mortality among children less than 5 years of age with the objective of estimating diarrhea proportionate mortality. We thoroughly reviewed the child diarrhea mortality literature and utilize 2008 country-specific indicators to produce diarrhea mortality estimates for countries where vital registration data are not available based on a single cause modeling approach. This methodological alternative has the benefit of using all possible diarrhea-specific data and thus provides an important comparison to the previously published multi-cause approach.
We conducted a systematic review of the published literature to identify studies with diarrhea as a cause of death among children less than 5 years of age. We searched PubMed, CAB abstracts, the WHO Regional Databases, and SIGLE for all studies containing data on diarrhea proportionate mortality published between 1980 and 2008 using all combinations of “diarrh(o)ea”, “mortality”, and “cause of death” both as keywords and MeSH terms. Studies conducted in low and middle-income countries and published in all languages were included. We included community based mortality studies with at least 12 months of mortality monitoring. All included studies used standardized verbal autopsy methodologies. Studies were judged to be of poor quality and unsuitable for inclusion if methods were poorly described, if the study was conducted in a non-representative population (such as low birth weigh infants), or if the cause of death was attributable to more than one cause unless the death could be proportionately reallocated to single causes. Because we sought to estimate diarrhea proportionate mortality for all children 1–59 mo of age we excluded verbal autopsy studies conducted among non-standardized age groups (i.e. those with upper age limits other than 11/12 mo, 35/36 mo, or 59/60 mo). Data from included studies were abstracted into a standard data abstraction sheet by two trained abstractors and all discrepancies were resolved by one of the study investigators. To minimize bias we also sought unpublished data from investigators who were able to provide a full report of study methods and results.
Our outcome of interest was diarrhea proportionate mortality. When studies did not report proportionate mortality, this was calculated using diarrhea-specific and all-cause mortality rates. For studies that attributed a proportion of deaths to multiple causes, these deaths were distributed to single causes according to the relative single cause proportions. Because we did not find representative mortality studies for all countries of the world where vital registration data are lacking, we used study and country level covariates to create a regression model to predict diarrhea proportionate mortality for all countries lacking vital registration data. We chose a logistic model because it allows modeling of the outcome variable with estimates constrained between 0 and 1. Observations were weighted by the square root of sample size to prevent studies with extremely large samples sizes from overweighting the model. The model included under 5 mortality risk (5q0), malaria risk, Gross National Income (GNI), year of data collection, and WHO region (
Parameter | Estimate |
WHORegion AFR | −9.29E−01 |
WHORegion AMR | −1.95E−01 |
WHORegion EMRSEAR | −5.96E−01 |
WHORegion WPR | −1.60E+00 |
GNI | −1.70E−04 |
Post 1990 | −1.16E−01 |
lowAge 0 | −2.69E−01 |
lowAge 2 | −2.66E−01 |
highAge 1 | −2.22E−01 |
highAge 2 | 3.00E−01 |
Our analysis sought to estimate the proportion of mortality attributable to diarrhea among children 1–59 mo of age; however, some studies included neonatal deaths that could not be excluded during the data abstraction period. Because diarrhea deaths are rare in the neonatal period, we included a neonatal dummy variable in the model to help account for the residual effect of these neonatal deaths being included in the final model. Similarly, we sought to estimate diarrhea mortality for children up to 59 mo of age. However, many studies included only infants up to 12 or 36 mo of age. Thus, we created dummy variables to differentiate studies that included children up to the full 59 mo of age from those with truncated age cohorts up to 12 or 36 mo of age.
Time was included as a dichotomous interaction variable with the AMR region differentiating studies with a mid-point prior to 1995 and those including and post January 1995. Exploratory analysis revealed that other regions did not have a significant time interaction. The possibility of using a continuous time variable was considered but decided against, due to unrealistically low 2008 estimates of diarrhea mortality as a result of linear extrapolation of the reduction observed during the 1990s.
We sought to include studies from all regions of the world, but failed to identify any studies from the WHO European region, a region that now includes several middle-income countries without complete vital registration data. For the purposes of the model, we substituted the Americas regional variable for these selected countries because this region had the next highest GNI and GNI proved to be an important predictor in the model.
We generated country level proportionate mortality estimates using a predictive logistic regression model for 84 countries without vital registration data (
Regional level uncertainty bounds were estimated in a manner similar to that described previously
We compared our country and regional specific diarrhea proportionate mortality estimates with those recently published for 2008
All statistical analyses were carried out using R version 2.13
The funding source had no involvement in the study design, analysis, or interpretation of results.
We reviewed more than 10,000 titles and included 66 studies representing 90 data points in our final review (
The data were used to estimate proportion of deaths among children 1–59 mo of age in 84 low- and middle-income countries. Overall the countries included in this model represent 92–99% of the deaths among 1–59 mo old children in Africa, South-east Asia, Eastern Mediterranean, and the Western Pacific regions (
Single-Cause | Multi-cause Mixed Methods Approach |
|||
WHO region | Weighted Mean Diarrhea-Proportionate Mortality (%) [number of data points contributing to the estimate] | Uncertainty range | % of all deaths represented by the countries included in the model | Diarrhea-Proportionate Mortality (%) |
Africaa | 25.2 [33] | 20.2, 34.7 | 98.1 | 25.3 |
Americasb | 10.0 [19] | 6.6, 15.7 | 28.6 | 13.5 |
Eastern Mediterraneanc | 30.4 [36] | 20.9, 47.1 | 91.6 | 30.9 |
Europed | 10.6 | 6.5, 17.2 | 39.2 | 10.6 |
South-East Asiae | 31.3 [36] | 18.0, 60.3 | 99.0 | 26.1 |
Western Pacificf | 11.0 [3] | 5.9, 15.3 | 94.1 | 8.3 |
*Countries Included in Single Cause Model: a) Algeria, Angola, Benin, Burkina Faso, Burundi, Cameroon, Cape Verde, Central African Republic, Chad, Comoros, Congo, Cote d'Ivoire, Democratic Republic of the Congo, Equatorial Guinea, Eritrea, Ethiopia, Gabon, Gambia, Ghana, Guinea, Guinea-Bissau, Kenya, Lesotho, Liberia, Madagascar, Malawi, Mali, Mauritania, Mozambique, Namibia, Niger, Nigeria, Rwanda, Sao Tome and Principe, Senegal, Sierra Leone, Swaziland, Tanzania, Togo, Uganda, Zambia, Zimbabwe; b) Bolivia, Dominican Republic, Guatemala, Haiti, Honduras, Jamaica, Nicaragua, Paraguay; c) Afghanistan, Djibouti, Iraq, Morocco, Pakistan, Somalia, Sudan, Yemen; d) Azerbaijan, Georgia, Kyrgyz Republic, Tajikistan, Turkmenistan, Uzbekistan; e) Bangladesh, Bhutan, DPR Korea, India, Indonesia, Maldives, Myanmar, Nepal, Timor-Leste; f) Cambodia, China, Lao, People's Democratic Republic of Micronesia, Mongolia, Nauru, Papua New Guinea, Philippines, Samoa, Solomon Islands, Vanuatu.
**100% of deaths included in multi-disease mixed approach Multi-cause mixed methods approach. Uncertainty ranges for total number of diarrhea deaths from MC model have been previously presented
Regional differences are large with proportionate mortality ranging from 10.03% (uncertainty range: 6.6 to 15.7%) in the Americas to 31.3% (uncertainty range: 18.0 to 60.3%) in the Southeast Asian regions. Diarrhea proportionate mortality rates were greater than 25% in the African, South-east Asian, and Eastern Mediterranean regions, where nearly 9 out of 10 child deaths occur.
We compared the country level results of the single-cause model with previously published results for 2008 and present both sets of estimates for proportionate mortality and deaths in
In this paper we estimate that diarrhea was responsible for a varying proportion of deaths among children 1–59 mo of age in 2008 in major world regions. We did not calculate diarrhea deaths among neonates thus are unable to calculate the proportion of diarrhea mortality among all children under 5 years of age. Diarrhea deaths during the neonatal period have been estimated to be 1% of all deaths among children 0–59 mo of age
This method ensures continued comparability across diseases for the single cause modeling approach and provides a comparative approach to the updated 2008 cause of death estimation process that included a verbal autopsy multi-cause model for a selected group of countries without vital registration data
We chose the wide publication dates because a significant number of child mortality studies, especially those with quality diarrhea mortality data were published pre-1990 and while overall child mortality has declined, diarrhea incidence rates have not, thus we felt maintaining a wide net for potential studies would provide the best data possible. The model we present here represents more than 650,000 child deaths, yet despite this improvement in overall data availability when compared to previous estimates
We included India and China in our regional estimates for 2008 diarrhea proportionate mortality. Given our literature review search strategies we identified a number of studies from both countries. However, in the recently published 2008 cause of death analysis these two countries were modeled separately and not included in regional models
Because overall child mortality and the proportion of deaths occurring during the 1–59 mo age group relative to neonatal deaths continue to decline it is expected that the proportion of deaths attributable to diarrhea will also continue to decline. In the 2008 model presented here we included 33 new data points (90 in current review compared to 57 in 2004) and now include 4 studies with data collection periods in the 2000's
Relying on published data to estimate proportionate mortality has a number of inherent weaknesses. By nature of using retrospective verbal autopsy to assign cause of death, measurement error is possible. Though not perfect, verbal autopsy techniques have been published and standard algorithms exist to provide the best possible data in areas where physician confirmed cause of death are not available. A number of studies included in our analysis are from the 1980s and 1990s and are lacking site-specific data such as population specific 5q0 and are too old to successfully track down additional information from authors. When these data were not available we used the best national data corresponding to the time and place of data collection, but often the identified 5q0 was from another year and only available at the country level whereas more site-specific data was desired. In addition, we would have liked to have been able to abstract a number of diarrhea specific indicators from the studies themselves for use in the single cause model such as site specific diarrhea prevention and treatment variables, but these data are rarely reported in verbal autopsy studies and using national level proxy indicators to describe extremely variable behavior and treatment practices is far from ideal. In early models we tested a number of these variables and found that because most were not available from the data sources, they created noise in the final model and thus were deliberately excluded. If these data were available from the study sites and reported in conjunction with the diarrhea proportionate mortality rates this may enhance the ability of models to predict diarrhea mortality in countries where data gaps remain.
The estimates we report here represent an overall improvement in the available data with an increase in the total number and geographic distribution of data points and the inclusion of more recent mortality reports as compared to previous single cause estimates. As newer data are released we will be able to capture the most recent mortality trends with regard to overall rates as well as cause of death for children less than 5 years of age. With new prevention and treatment interventions such as insecticide-treated bednets, pneumococcal and rotavirus vaccines and zinc for the treatment of diarrhea being scaled-up in low- and middle-income countries, it will become increasingly important to capture the effect of these interventions on all cause and cause-specific mortality. There will inevitably continue to be a time lag from when interventions are successfully rolling out and saving lives to when data are available, but with continued emphasis on the critical importance of these data we are hopeful this distance from intervention to availability of mortality estimates will decline.
Despite recent reductions in all cause mortality, diarrhea remains the 2nd leading cause of death among children under 5 years of age
(DOCX)
The authors thank Melinda Munos and Hilda Wangari for their assistance with conducting the literature searches and initial data abstraction.