Minimization of masking in signal detection from Chinese spontaneous reporting databases based on data removal strategy

Abstract

This study aimed to develop an experimental method for minimizing masking in signal detection using a data removal strategy. Reports in the Chinese Spontaneous Reporting Database (CSRD) between 2010 and 2011 were selected as the initial database. A reference database including known signals was used for performance evaluation. The data removal strategy was as follows: 1) the data were sorted according to the frequency of drug–event combinations (DECs), and the top n% of DECs was removed from the initial database; 2) signals of disproportionate reporting were detected using the MHRA for each new database; and 3) the performance was evaluated based on the reference database before and after data removal. The five adverse events (AEs) of interest: renal failure acute, skin exfoliation, syncope, leucopenia, and tetany were selected to test the result. Our experimental results showed that the value of F index increased first and then decreased with data removal, and the value of benefit rate (BR) rose in the new database constantly. In the sixth experiment, the F index reached a peak value (50.63%), and the performance of unmasking achieved the best, where the value of BR was changed from 10.72% to 52.12% and the number of known signals exposed was changed from 6314 to 6787. The performance of unmasking achieved the best when the top 6% of DECs were removed from the CSRD.