Back to browse results
Spatial Quantile Regression with Application to High and Low Child Birth Weight in Malawi
Authors: Alfred Ngwira
Source: BMC Public Health, 19(1): 1593: DOI: 10.1186/s12889-019-7949-9
Topic(s): Birth weight
Spatial analysis
Country: Africa
Published: NOV 2019
Abstract: Background: Child low and high birth weight are important public health problems. Many studies have looked at factors of low and high birth weight using mean regression. This study aimed at using quantile regression to find out determinants of low and high birth weight. Methods: Spatial quantile regression models at 0.05 and 0.95 percentiles of birth weight were fitted to 13,087 children birth weight in kilograms using Malawi demographic health survey data of 2010 study. Full Bayesian method by integrated nested Laplace approximations (INLA) was used to estimate the model. Second order random walk priors were assigned for mother age and antenatal visits for pregnancy while Gaussian markov random field prior was used for district of the child. Results: Residual spatial patterns reveal areas in the southern region promoting high birth weight while areas in the central and northern region promote low birth weight. Most fixed effects findings are consistent with the literature. Richest family, normal mother body mass index (BMI), mother over weight (BMI > 25 kg/m2), birth order 2-3, mother secondary education and height (=150 cm) negate low birth weight while weight 45-70 kg promote low birth weight. Birth order category 6+, mother height (=150 cm) and poor wealth quintile, promote high birth weight, while richer and richest wealth quintiles and education categories: primary, secondary, and higher, and mother overweight (BMI > 25 kg/m2) reduce high birth weight. Antenatal visits for pregnancy reduce both low and high birth weight. Conclusion: Strategies to reduce low and high birth weight should simultaneously address mother education, weight gain during pregnancy and poverty while targeting areas increasing low and high birth weight. Keywords: Bayesian; INLA; Quantile; Spatial.