Estimating double burden of malnutrition among rural and urban children in Amazonia using Bayesian latent models

The double burden of malnutrition (DBM) in the same individual is a neglected public health concern, especially in low- and middle-income countries (LMICs). The DBM is associated with increased risks of non-communicable diseases, childbirth complications, and healthcare costs related to obesity in adulthood. However, evaluating low prevalence outcomes in relatively small populations is challenging using conventional frequentist statistics. Our study used Bayesian latent models to estimate DBM prevalence at the individual-level in small populations located in remote towns and rural communities in the Brazilian Amazon.

We employed a cross-sectional survey of urban and rural children aged 6-59 months, considering DBM as the coexistence of stunting and overweight in the same individual. We evaluated four river-dependent municipalities, sampling children in randomly selected households in each town and a total of 60 riverine forest-proximate communities. Through Bayesian modeling we estimated the latent double burden of malnutrition (LDBM) and credible intervals (CI).

The exceedance probability of LDBM was used to quantify this form of malnutrition at the population level. Rural prevalence of LDBM was significantly higher in Jutai (3.3%; CI: 1.5% to 6.7%) compared to Maues and Caapiranga. The likelihood that LDBM rural prevalence exceeded 1% was very high in Jutai (99.7%), and Ipixuna (63.2%), and very low (< 2%) in rural communities elsewhere. Exceedance probabilities (at 1%) also varied widely among urban sub-populations, from 6.7% in Maues to 41.2% in Caapiranga. The exceedance probability of LDBM prevalence being above 3.0% was high in rural Jutai (59.7%).

Our results have important implications for assessing DBM in vulnerable and marginalized populations, where health and nutritional status are often poorest, and public health efforts remain focused on undernutrition. Our analytical approach could enable more accurate estimation of low prevalence health outcomes, and strengthen DBM monitoring of hard-to-reach populations.