Bayesian rank likelihood-based estimation: An application to low birth weight in Ethiopia

Background: Low birth weight is a significant risk factor associated with high rates of neonatal and infant mortality, particularly in developing countries. However, most studies conducted on this topic in Ethiopia have small sample sizes, often focusing on specific areas and using standard models employing maximum likelihood estimation, leading to potential bias and inaccurate coverage probability. Methods: This study used a novel approach, the Bayesian rank likelihood method, within a latent traits model, to estimate parameters and provide a nationwide estimate of low birth weight and its risk factors in Ethiopia. Data from the Ethiopian Demographic and Health Survey (EDHS) of 2016 were used as a data source for the study. Data stratified all regions into urban and rural areas. Among 15, 680 representative selected households, the analysis included complete cases from 10, 641 children (0-59 months). The evaluation of model performance considered metrics such as the root mean square error, the mean absolute error, and the probability coverage of the corresponding 95% confidence intervals of the estimates. Results: Based on the values of root mean square error, mean absolute error, and probability coverage, the estimates obtained from the proposed model outperform the classical estimates. According to the result, 40.92% of the children were born with low birth weight. The study also found that low birth weight is unevenly distributed across different regions of the country with the highest amounts of variation observed in the Afar, Somali and Southern Nations, Nationalities, and Peoples regions as represented by the latent trait parameter of the model. In contrast, the lowest low birth weight variation was recorded in the Addis Ababa, Dire Dawa, and Amhara regions. Furthermore, there were significant associations between birth weight and several factors, including the age of the mother, number of antenatal care visits, order of birth and the body mass index as indicated by the average posterior beta values of (β1= -0.269, CI=-0.320, -0.220), (β2= -0.235, CI=-0.268, -0.202), (β3= -0.120, CI=-0.162, -0.074) and (β5= -0.257, CI=-0.291, -0.225). Conclusions: The study showed that the low birth weight estimates obtained from the latent trait model outperform the classical estimates. The study also revealed that the prevalence of low birth weight varies between different regions of the country, indicating the need for targeted interventions in areas with a higher prevalence. To effectively reduce the prevalence of low birth weight and improve maternal and child health outcomes, it is important to concentrate efforts on regions with a higher burden of low birth weight. This will help implement interventions that are tailored to the unique challenges and needs of each area. Health institutions should take measures to reduce low birth weight, with a special focus on the factors identified in this study.