Journal of Pediatric and Neonatal Sciences
Download PDF
ABSTRACT
Objective
To demonstrate the impact of correcting measurement errors in child malnutrition prevalence estimates using a Bayesian hierarchical model, and to quantify the bias introduced by ignoring outcome misclassification. This study is based on simulated data designed to mirror the structure of Demographic and Health Surveys (DHS) from Cameroon (2004, 2011, 2018, and 2022); no real survey data were used, and all reported prevalence figures are simulation results, not actual national estimates.
Methods
Hierarchical Bayesian logistic regression was employed to evaluate the prevalence of malnutrition, accounting for misclassification in stunting status. Simulated data were generated to mimic pooled cross-sectional DHS surveys, including temporal random effects and known sensitivity/specificity parameters. Child’s age, gender, familial wealth, maternal education, prenatal care appointments, and water source were included as covariates. Markov Chain Monte Carlo (MCMC) was used to estimate posterior distributions. Corrected estimates were compared with an uncorrected (naive) logistic regression model.
Results
In all simulated survey years, the corrected model gave higher prevalence estimates than the uncorrected model. For example, in the simulated 2022 survey, the corrected prevalence was 30.7% compared with 26.5% from the uncorrected model. The corrected model improved AUC (from 0.928 to 0.930), accuracy, and precision. Larger simulated sample sizes reduced standard deviations and coefficients of variation. Maternal education, antenatal care, and water source emerged as important predictors.
Conclusion
Ignoring measurement error in binary outcomes such as stunting can lead to systematic underestimation of malnutrition prevalence. Our Bayesian correction approach provides more reliable estimates in simulation settings that reflect real-world survey conditions. These findings have methodological implications for health policy and intervention design, but they do not provide direct prevalence estimates for Cameroon. Future work should apply this framework to real DHS data once validation studies supply context-specific misclassification parameters.
Keywords: Child Malnutrition; Measurement Error Correction; Bayesian Hierarchical Modeling; Simulated DHS Data; Prevalence Estimation; Health Policy