ABSTRACT
OBJECTIVE: To evaluate the change in tuberculosis (TB) notification rates due to the removal of unwanted duplicate records from the Brazilian notification system (2001-2007, data extracted in October 2008), and therefore extending the period of investigation of the previous study using the same methodology (2000-2004, data extracted in February 2006). METHODS: Repeat records were identified using a probabilistic record linkage, classified into six mutually exclusive categories, and then kept, combined or removed from the database. RESULTS: In the TB database, 22.7% of all records belonged to patients with multiple records. When we excluded the first record of every patient in this group, 43.7% were classified as transfers, 29% as returns after default, 16.3% as relapses and 6.6% as true duplicates, while 2.9% were inconclusive and 1.5% had missing data. Removal of unwanted duplicate records reduced the notification rates of new cases by 4% to 6.3%, and increased the proportion cured by 3.4% to 4.9%. DISCUSSION: Linkage of records within the TB notification database and the implementation of procedures to distinguish between new and retreatment or transfer-in records yielded better data. Recommendations are provided on how to prevent duplicates and misclassifications in national TB databases.