Spatial clustering of stunting cases in Indonesia: A Bayesian approach

Abstract

Stunting is one of the major public health problems, especially in developing countries such as Indonesia. In the Southeast Asian region, Indonesia has the third highest prevalence with an average prevalence of 36.4% (2005-2017). Indonesian Nutritional Status Study (SSGI) in 2021 stated that the percentage of stunting in Indonesia is 24.4%. Research on spatial modelling of stunting has been done, but the use of the Bayesian spatial Conditional Autoregressive (CAR) model is still rare. This article aims to provide the most appropriate Bayesian spatial CAR localised (clustering) model and identify the relative risk (RR) of stunting in each province in Indonesia. Data on the number of toddlers 0-59 months whose height is measured and the number of stunted toddlers in each province in Indonesia in 2021 were used. The best model is based on the Deviance Information Criterion, Watanabe Akaike Information Criterion and Modified Moran's I value for residuals. The results indicated that the Bayesian spatial CAR Localised with hyperprior Inverse-Gamma (0.5, 0.05) and Inverse-Gamma (1, 0.1) are preferred for two and three clusters, respectively. Our results identified the high-risk areas for stunting. Approximately 56% of provinces in Indonesia are at a high risk of stunting. Sulawesi Barat has the highest RR for stunting followed by Nusa Tenggara Timur dan Papua Barat. In contrast, Jakarta has the lowest RR of stunting followed by Sulawesi Utara and Sumatera Selatan. Government should pay more attention to areas that are most at high risk of stunting.