Appropriateness of administrative data for vaccine impact evaluation: the case of pneumonia hospitalizations and pneumococcal vaccine in Brazil.

Ten-valent pneumococcal conjugate vaccine (PCV10) was recently introduced into the Brazilian Immunization Programme. Secondary data are used as a measurement of community-acquired pneumonia (CAP) burden, but their completeness and reliability need to be ascertained. We performed probabilistic linkage between hospital primary data from active prospective population-based surveillance (APS) and hospital secondary data from the Hospital Information System administrative database of the National Unified Health System (SIH-SUS). Children aged 2-23 months hospitalized during January-December 2012 were identified. Incidence rates of hospitalized CAP were estimated. Agreement of case identification was measured by kappa index. A total of 1639 (26%) CAP cases were identified in APS and 1714 (35%) in SIH-SUS. Of these 3353 records, 1127 CAP cases were present in both databases. Kappa on CAP case identification was 0·72 (95% confidence interval 0·69-0·75). CAP hospitalization incidence using administrative (5285/100 000) and hospital (5054/100 000) primary data were similar (P = 0·184). Our findings suggest that administrative databases of hospitalizations are reliable sources to assess PCV10 impact in time-series analyses.