Completeness Assessment of Neonatal Deaths in a Region of Brazil: Linkage and Imputing Missing Data

Although coverage of neonatal deaths in Brazil is considered high, the completeness of death declaration items for several regions is a problem of concern and uncertainty, which can compromise the maternal and child health planning. Data set linkage offers considerable potential to address an extensive range of research questions, such as identifying risk factors, and quantifying mortality, morbidity and healthcare for infant health as in the neonatal period. This technique added to imputing missing data techniques is a feasible and cost-efficient way to recover data. The main aim of this paper is to evaluate the completeness of information on neonatal death declarations in the regions of Paraíba State from 2009 to 2017. The quality of data on neonatal deaths declaration was studied in two stages: in the first, the Mortality Information System and Birth Information System from the Brazilian Ministry of Health databases were matched using the deterministic linkage; in the second, the multiple imputation for missing data was carry out. In total, 5149 neonatal deaths were computed. The following variables were investigated: gender, race/color, mother’s age, weeks of gestation, birth weight, mother’s educational level, number of live children, number of dead children, type of pregnancy and type of delivery. There was an important decrease in neonatal death records over time, approximately 19%. Except for the variable mother’s educational level (20.0%) and gender (0,8%), all variables presented percentages of incompleteness ranging from 6.7% to 15.3%. The percentage of matched records ranged from 50.0% to 58.8% in the period. After five multiple imputations, the missing data were recovered. The Relative Efficiency of the variables with missing observations recovered was verified, whose efficiency for all variables ranged from 96.7% to 99.9%. The conclusion was an excellent and reliable imputation of missing data. The tools used here proved to be very efficient and useful for use in regions with deficient data, such as those deaths registered for Paraíba in Brazil.