[Probabilistic record linkage: an application to severe maternal morbidity (near miss) and maternal mortality]

This article presents an experience with record linkage from the Brazilian Hospital Information System (SIH) with the Mortality Information System (SIM), and the SIH with itself, applied to severe maternal morbidity (near miss) and maternal mortality. This was an empirical study using Brazilian data for the state capitals and Federal District in 2002. For the two linkages separately applied in each capital, a three simple step blocking strategy was established, plus related multiple steps and also two clerical review strategies. From the total number of true pairs found after the two linkages, simple steps failed to find fewer than 8%, while the multiple step strategy failed to find only 0.7%. This approach allowed exploring the issue of severe maternal morbidity and mortality in these databases. The number of pairs found and reviewed under the multiple steps strategy was lower than the sum of pairs obtained with the three simple steps, and fewer pairs were lost. However, for the record linkage of the SIH with itself, both strategies are suggested.