Introduction Probabilistic record linkage is an alternative to different study designs because it generates, at lower operating cost, better database completeness. The aim of this study was to assess the performance of the probabilistic record linkage between PBCR-SP for cases of stomach and intestine cancer and São Paulo death database.

Methods From 1997 through 2005, 18511 cases were recorded by PBCR-SP. From the 1997 through 2007, 767 752 deaths were recorded by Mortality Information System of São Paulo city. The fetal deaths were excluded. Exceeding spaces between characters were excluded. Type and size of field, and different spellings with same phoneme were standardised. Prepositions were removed. Scores for ranking true match pairs and probabilistic record linkage was performed using Reclink 3.2 software. The descriptive statistics and Kaplan-Meier test were conducted before and after linkage.

Results Of 18 511 cancer cases, of both sexes, aged between 0 and 100 years and older, 7328 were true matched pairs. There was an 85.8% gain in residential address and a 11.0% gain in death date. In 4086 cases, mother's name was aggregated and for all paired registries, cause of death. In the survival analysis, it was verified a 14.4% underestimation in the one-year survival probability and a 31.6% underestimation for the three-year.

Conclusion Probabilistic record linkage has shown to be effective in the identification of deaths. An expressive gain in information quality of PBCR-SP database was observed.