This study used probabilistic record linkage to estimate underreporting of stillbirths and infant deaths in Maranhão State, Northeast Brazil, in 2008. Records were linked between the Hospital Information System (SIH) (N = 374,418) and the Mortality Information System (SIM) (N = 26,597), using data extraction, identification, and processing software. Under-recording rates were 9.7% for stillbirths, 12% for neonatal deaths, and 5.3% for post-neonatal deaths. In the larger municipalities, the correction of the infant mortality rate was greater than 19%. The superiority of information on infant deaths in the SIH (as compared to the SIM) was 6.5% in municipalities with < 25 thousand inhabitants, 2% in municipalities with 25-80 thousand inhabitants, and 8.3% in those with > 80 thousand inhabitants. Inconsistencies involved non-completion of items/variables, discordant data on the same event, and in some cases even total absence of death records. The method proved useful for retrieving mortality data for the SIM from the SIH.
Underregistration; Information Systems; Mortality
