Data-driven methods for imputing national-level incidence in global burden of disease studies

摘要: Objective To develop transparent and reproducible methods for imputing missing data on disease incidence at national-level the year 2005. Methods We compared several models country-level rates two foodborne diseases – congenital toxoplasmosis aflatoxin-related hepatocellular carcinoma. Missing values were assumed to be random. Predictor variables selected using least absolute shrinkage selection operator regression. predictive performance of naive extrapolation approaches Bayesian random mixed-effects regression models. Leave-one-out cross-validation was used evaluate model accuracy. Findings The accuracy significantly better than that method one However, produced wider prediction intervals both sets. Conclusion Several are available national level. Strengths a hierarchical approach this type task ability derive estimates from other similar countries, transparency, computational efficiency ease interpretation. inclusion informative covariates may improve performance, but results should appraised carefully