Analysis of severity of childhood anemia in Malawi: a Bayesian ordered categories model

Abstract

Purpose: This study proposes an ordered categories model, using multinomial cumulative logistic regression, to investigate the risk factors affecting the severity of childhood anemia in Malawi. Patients and methods: We generated a four-category outcome based on the categorization of child hemoglobin (Hb) level: nonanemia (Hb ≥11 g/dL), mild anemia (10.0 g/dL ≤ Hb ≤ 10.9 g/dL), moderate anemia (7.0 g/dL ≤ Hb ≤ 9.9 g/dL), and severe anemia (Hb <7.0 g/dL), using the 2010 Malawi Demographic and Health Survey data. We fitted a cumulative logistic threshold model, permitting nonlinear effects for continuous variables and spatial effects for district of residence. Inference was based on the empirical Bayes framework, with continuous covariates modeled by the penalized (P) splines and spatial effects smoothed by the two-dimensional P-spline. Results: Findings reveal substantial spatial variation, with increased risk of anemia observed in the districts of Nsanje, Chikwawa, Salima, Nkhotakota, Mangochi, Machinga, and Balaka. On the other hand, reduced risk was estimated in the districts of Karonga, Chitipa, Rumphi, Mzimba, Zomba, Chiradzulu, and Thyolo. All known determinants, such as maternal anemia, child stunting, wasting, fever, and being underweight, increased the likelihood of childhood anemia. Furthermore, infant anemia decreased with child's age and wealth index. In addition, there was a U relationship between childhood anemia and mother's age. Conclusion: Strategies for minimizing infant anemia must include optimized iron intake but should also simultaneously address maternal anemia, food insecurity, poverty, and child fever, particularly targeting districts identified to have a high risk of anemia.