We selected twelve countries based on the following criteria: 1) being in sub-Saharan Africa given that it is the region with the greatest proportion of diarrhea deaths; 2) being high-burden: over 10 % of country’s deaths among children 1–59 months caused by diarrhea; and 3) having recent population-based survey data available (2010 or later) with sufficient sample size for analysis. We reviewed available nationally representative population data for countries meeting the above criteria from Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Survey (MICS), but retained only DHS surveys because until 2013, MICS did not collect data on care seeking for childhood diarrhea which was a key variable for this analysis .
The 12 countries included in the analysis are: Burkina Faso (year of survey: 2010; % deaths among 1–59 month-old caused by diarrhea in 2013: 14 %), Burundi (2010; 18 %), Cameroon (2011; 16 %), Cote d’Ivoire (2011–12; 15 %), Democratic Republic of the Congo (DRC) (2013–14; 15 %), Ethiopia (2011; 16 %), Mali (2012–13; 16 %), Niger (2012; 16 %), Nigeria (2013; 14 %), Sierra Leone (2013; 18 %), United Republic of Tanzania (2010; 12 %), and Uganda (2011; 12 %) [7, 21]. Together these countries accounted for about one-third of all under-five deaths worldwide due to diarrhea in 2013 and about two-thirds (62 %) in sub-Saharan Africa . Table 4 in the Appendix presents information about the evolution of the adoption and implementation of the low-osmolarity and zinc policy in sub-Saharan African countries as well as the number and percentage of childhood deaths caused by diarrhea.
DHS collects data on childhood diarrhea treatment from nationally representative samples. In the surveys, mothers are asked if their children under five had diarrhea in the past 2 weeks. If the answer is positive, they are asked follow-up questions about care seeking and treatment, including the four recommended management interventions (ORS, zinc, increased fluids, continued feeing) as well as other potentially harmful practices including antimotility drugs.
We classified the quality of diarrhea management practice as good, fair, or poor based on the  guidelines (Table 1). We consulted with the Ministry of Health in each country to categorize the reported sources of care as facility, community based, traditional, or no care outside the home. Facility care refers to care sought from health facilities, whether public or private. Community-based care relates to care sought from community health workers, mobile clinics, village health posts, government dispensaries or health centers and health posts located at the community level, as well as pharmacies. Traditional sources of care include traditional healers, traditional practitioners as well as shops, stores, informal drug sellers and markets (Appendix Table 5). No care outside the home refers to children who were treated at home for diarrhea but for whom care was not sought outside the home.
‘Good’ diarrhea management is the main outcome of interest in this analysis, and source of care is the main independent variable. We first described prevalence of diarrhea management practices across countries, then examined the country specific unadjusted association between ‘good’ diarrhea management and type of care. We finally fitted logistic regression models of ‘good’ diarrhea management practice on source of care for all countries, adjusting for the following known confounders as described below. Table 6 in the Appendix presents the estimated regression coefficients (log odds ratios for adjusted factors) from logistic regression models for the factors from Anderson’s conceptual framework, for the probability of children under five receiving good diarrhea management.
We used Andersen’s conceptual framework of access to medical care to help identify variables to control for in the regression model . We identified these factors based on three main categories outlined in the Andersen’ framework. The predisposing characteristics included child age, mother’s age, child gender, mother’s marital status and education, partner's education, parity, number of children under five living in the household; the enabling resources included wealth quintile, rural or urban location, distance from health care reported by the mother as a problem in receiving health care, mother’s participation in decision making, household improved water access, open defecation and the need characteristics included whether there was blood in the child’s stool, as a measure of severity. However, all reported cases of diarrhea were included in the analysis and having had ‘blood in stool’ was not used for classification of diarrhoea but instead as a variable in the final regression model, thus avoiding potential misclassification issues. We assessed multicollinearity in these independent variables and retained only one of two or more variables that were highly correlated. The final model reported here includes adjustment for the following variables or potential confounders: child age, mother’s age, child gender, mother’s marital status and education, number of children under five living in household, wealth quintile, rural or urban location, if distance is a problem in receiving health care, mother’s participation in decision making, whether household has improved water, and whether there was blood in the child’s stool. These variables were all of the ones included in the initial Andersen’s model, except for mother’s participation in decision making and open defecation which were highly correlated with other variables and therefore not included in the final model. All regression analysis took into account the survey complex multi-stage sample design and sample weights. We used Stata 13 for analysis.