Database size and power to detect safety signals in pharmacovigilance

Abstract

Most regulatory agencies and pharmaceutical companies focus the majority of their pharmacovigilance on safety signal identification in large databases. GlaxoSmithKline (GSK) has > 100 drugs marketed worldwide. In order to determine which database has the highest statistical power to detect safety signals in three large global databases, ten GSK marketed drugs were randomly selected for review in the three databases. At the time of data lock, the FDA database (Adverse Event Reporting System [AERS]) contained approximately 6.2 million total records of adverse drug reactions (ADRs). The WHO database (VIGIBASE) contained 7.2 million total records of ADRs. GSK's global safety database (OCEANS) contained approximately 2 million total ADRs for all of its marketed drugs. For the ten drugs selected, there was an average of 7566 reports found in AERS, 8661 reports found in VIGIBASE and 15,496 reports in OCEANS. The information from all three databases was used in pairs (AERS/OCEANS; AERS/VIGIBASE; and OCEANS/VIGIBASE) to calculate power using the maximum likelihood estimation. The OCEANS database contained more ADRs for all 10 drugs than AERS. OCEANS also contained more ADRs for 8/10 drugs than VIGIBASE. The highest statistical power to detect safety signals was determined by the pair of databases which had the greatest number of reports for the given drug. Based on this data, it was concluded that the highest power may be achieved by combining those databases with the most drug-specific data. It is also believe that early safety signal detection should involve the use of multiple large global databases because this permits the use of the largest number of reports for a given drug, and that reliance on a single database may reduce statistical power and diversity of ADRs.