Demographic and health surveys for Bayesian network meta-analysis of childhood diarrhea cases (2000–2023)

The resulting dataset was constructed from Demographic and Health Surveys (DHS) conducted across multiple countries in Eastern and Southern Africa between 2000 and 2023. DHS data are collected under strict ethical oversight, including approval by relevant institutional review boards and informed consent from participants; this analysis uses only de-identified, aggregated secondary data in accordance with DHS data use policies, with no additional ethical approval required. The original data comprised aggregated counts of childhood diarrhea cases (children aged five years or younger) stratified by environmental and infrastructural conditions, including access to improved and unimproved water sources, improved and unimproved sanitation, and urban or rural household classification, with 50 survey-level observations defined by country–region–year combinations.Because these exposure categories are not mutually exclusive at the individual level and may co-occur within households, the original structure does not satisfy the independence assumptions required for standard network meta-analysis. To address this, the data were transformed into composite exposure states designed to approximate mutually exclusive intervention arms: a baseline category of unimproved water and unimproved sanitation and three comparator categories representing improved water only, improved sanitation only, and improved water plus improved sanitation. Event and non-event counts were aggregated accordingly within each survey, and the dataset was reshaped into an arm-based long format suitable for Bayesian network meta-analysis using the BUGSnet package, yielding 150 arm-level observations. Although the transformation improves interpretability and enables network connectivity, residual dependence between exposure arms may remain due to the aggregated nature of the data; the dataset is therefore intended primarily for methodological demonstration rather than strict causal inference.