Pooled prevalence and associated factors of diarrhea among under-five years children in East Africa: A multilevel logistic regression analysis

Background Worldwide, diarrhea is the second most common cause of death and morbidity among under-five years’ children. In sub-saran Africa, access to water, sanitation, and hygiene are very scanty and the burden of diarrhea diseases is countless relative to the rest of the world. Prior studies conducted in East Africa vary in design, sample size, and other data collection tools. Through those studies, it is hard to make regional comparisons. Combining datasets that are studied on similar people and having common variable identified enhances statistical power due to the large sample size, advance the ability to compare outcomes, and create the opportunity to develop new indicators. Hence, this study aimed to assess the prevalence and associated factors of diarrhea among under five years’ children using the most recent national representative Demographic and Health Surveys from 12 East African countries. The information generated from this pooled datasets will give good insight into the sub-regional prevalence of diarrhea. Methods This study utilized secondary data from 12 East African countries’ most recent demographic health survey. Variables were extracted and appended together to assess the pooled prevalence of diarrhea and associated factors. A total of 90,263 under-five years of age children were encompassed in this study. STATA version was used to cross-tabulate and fit the models. To account for the hierarchical nature of the demographic health survey, multilevel logistic regression was calibrated. BIC, AIC, deviance, and LLR were used as Model comparison parameters. Variables with a p-value of <0.2 were considered for multivariable analysis. Adjusted odds ratio with 95% CI and p-value <0.05 were used to declare statistical significances of factors. Results The pooled prevalence of diarrhea in under five years children was 14.28% [95%CI; 14.06%, 14.51%]. Being child whose mother age is 15–24 years [AOR = 1.41, 95% CI; 1.33, 1.49], 25–34 years[AOR = 1.17, 95%CI; 1.10, 1.23], being 7–12 months child [AOR = 3.10, 95%CI; 2.86, 3.35], being 12–24 months child [AOR = 2.56, 95%CI; 2.38, 3.75], being 25–59 months child [AOR = 0.88, 95%CI; 0.82, 0.95], being child from poor household [AOR = 1.16, 95%CI; 1.09, 1.23], delayed breast feeding initiation (initiated after an hour of birth) [AOR = 1.15, 95%CI; 1.10, 1.20], and being a child from community with low educational status [AOR = 1.10, 95%CI; 1.03, 1.18] were factors associated with diarrheal diseases. Conclusion The pooled prevalence of diarrhea among under five years of children in East African countries is high. Maternal age, child’s age, wealth status of the household, the timing of breast feeding initiation, sex of the child, community level of educational status, working status of the mother, and the number of under five children were factors that were associated with diarrheal diseases. Scaling up of maternal and child health services by government and other concerned bodies should consider those economically marginalized communities. Additionally, awareness should be created for those uneducated mothers concerning the nature of childhood diarrhea. As the majority of the population of East African countries are rural residents, more than fifty percent of the resident of East African countries lacks improved WASH indicators [20]. East African countries are countries with the highest prevalence of diarrheal diseases among under-five children when compared to the rest of the world [17]. This study used data from 12 Eastern African countries of most up to dated demographic health surveys. Eastern African countries embodied in this study were Burundi, Ethiopia, Comoros, Uganda, Rwanda, Tanzania, Mozambique, Madagascar, Zimbabwe, Kenya, Zambia, and Malawi. Mayotte, Reunion, South Sudan, Djibouti, Seychelles, and Mauritius were omitted because of no history of DHS conduction. Additionally, Eritrea and Sudan were also not included due to the long period since their last conduction of DHS, i.e. Eritrea in 2002 and Sudan in 1989/90 (Table 1). It was conducted using the principle of a two-stage stratified sampling procedure. In the first stage, Enumeration Areas (EAs) were randomly selected proportionally to their respected clusters. In the second stage, households were selected. The primary objective of conducting DHS is to provide up-to-date information about health and health-related indicators for planning, policy formulation, monitoring, and evaluation of population and health programs in the respective countries. Variables were extracted after a deep literature review and appended together to assess the pooled prevalence of diarrhea and associated factors in East Africa among under five children. In this study, the children’s dataset (KR file) was used. Ultimately, a total of 129,651(weighted) children under the age of five were encompassed in this study. The outcome variable was binary, children who had diarrhea at any time during the 2 weeks preceding the interview. The response variable diarrhea is recoded as follows: Those mothers/caregivers who responded yes to the question “had diarrhea in the last two weeks?” were coded as 1 and those who answered no were coded as 0 [21]. We sub-portioned the independent variables into two groups; level-1 (individual-level factors) and level -2(community-level factors). Child’s age, child’s sex, number of under five years children, immunization status, duration of breast feeding in months, age of the mother/caregiver, education of the mother, mother’s working status, mass media exposure of the mother, household wealth status, type of latrine, type of drinking water source and timing of breast feeding initiation after birth were considered for this study. The place of residence, community level of poverty, and community-level of educational status were variables assigned as community-level factors. The variable community level of poverty and community-level of educational status were generated by aggregating individual level factors at the cluster/community level. Media exposure. This variable is composite which consisted of watching television, listening to the radio, and reading magazines. Watching television (those who watch television less than once a week, at least once a week and every day are coded as = yes, otherwise = no), frequency of listening to the radio (listening less than once a week, at least once a week and every day are coded as = yes, otherwise = no) and frequency of reading Newspaper or magazine (reading less than once a week, at least once a week and every day are coded as = yes, otherwise = no) [22]. Visits to health facility or visited by health worker. Women either visited by health worker or had visited health facility in the last 12 months are categorized under “yes” and those who neither visited health facility nor visited by health worker were categorized under “no”. Type of toilet. Population using toilet characterized by flushing to somewhere else, pit latrine—without slab, bucket toilet, hanging toilet or other toilet were coded as “unimproved toilet” and population using toilet which flush—to piped sewer system, flush—to septic tank, flush—to pit latrine, flush—don’t know where, pit latrine—ventilated improved pit, pit latrine—with slab or composting toilet were coded as “improved toilet” [22]. Drinking water type. Household using drinking water which is, piped into dwelling piped to yard/plot public tap/standpipe, piped to a neighbor, tube well or borehole, protected well, protected spring, rainwater, tanker truck, cart with small tank or bottled water were coded as “improved drinking water” and household categorized under unprotected well, unprotected spring, surface water or other sources of drinking water was coded as “unimproved drinking water” [22]. Timing of BF initiation. Children who initiated BF within one hour of birth are labeled as “early” and coded 1, apart from that labeled as “delayed” and coded as 0 [23]. Community level of poverty. Proportion of households assigned to poorest and poorer wealth index. Those fall at the median value and above are categorized under the high poverty level, and those who fall below the median value of the variables are categorized under the low poverty level. Median is used as a cut point because of skewed distribution. The same way of categorization was used for community-level educational status. Community-level of educational status. Proportion of mother’s/caregiver’s of the child who is educated primary and above are categorized as having “high level of educational status” and otherwise “low level of educational status”. Perceived distance from health facility. The DHS program asks caregivers or mothers their perception whether the distance from health facility is a “big problem “or “no a big problem” when they were seeking medical advice or treatment for themselves when they are sick. Immunization status. Fully vaccination definition is adopted from the number of children aged 12–23 months who received one dose of BCG vaccine, three doses of polio vaccine, three doses of pentavallent vaccine (DTP-hepB-Hib), three-dose of pneumococcal conjugate vaccine (PCV), two-doses of virus vaccine, and one dose of measles vaccine was considered as “fully vaccinated” otherwise “not fully vaccinated [24]. Cross tabulations and summary statistics were done using STATA version 16 software. The forest plot technique was utilized to display the prevalence of diarrhea across countries. To plot 95% CI of the coefficient of each variable of the best-fitted model, STATA command “coefpot” was applied. AS the DHS datasets have hierarchical nature (sample is not taken randomly), non-independencies of observations and violation of equal variance assumption of the single level statistical model like logistic regression are inevitable. In the multistage stratified clustered sampling of DHS, children within a cluster are more likely to relate to certain characteristics as compared to children between the clusters. To overcome those problems, to draw reliable inferences, we calibrated so what sophisticated model called the multilevel logistic model to identify factors associated with diarrhea. We first calibrated the null model (model with only constant/intercept) in order to declare nesting of observation within clusters and to determine the use of multilevel analysis. To warrant the use of multilevel analysis, ICC (intra-class coefficient) was checked. Intra-class coefficient takes the value between 0 and 1. If the intraclass coefficient value approaches value one, then it indicates observations within the cluster are more similar than observations between clusters. Therefore, it implies that a multilevel model is necessary for that specific dataset. It also shows how much of the response’s total variation is explained by clustering. Deviance Information Criterion (DIC), Log-Likelihood Ratio (LLR), Akaike information criteria (AIC), and Bayesian information criteria (BIC) were used as a model comparison and selection parameters. The model with the lowest values of those parameters was selected as the best-fitted model. The model comparison was done among the null model (a model with no independent variables), model I (a model with only individual-level factors), model II (a model with only community-level factors) and model III (a model with both individual and community level independent variables). Variables with a p-value <0.2 in the bi-variable analysis were considered in the multivariable mixed-effect logistic regression model. Adjusted Odds Ratios (AOR) with a 95% Confidence Interval (CI) and p-value ≤ 0.05 in the multivariable model were used to declare significant factors associated with diarrhea. This study used datasets of national representative demographic health surveys. Therefore, ethical is approval not required. But, datasets for this study were requested by providing a clear explanation about the objectives and necessity of this study. We registered and requested the DHS dataset to the online database (www.dhsprogram.com) and received an authorization letter to download the requested datasets.