Exploration of Drug Safety and Administrative Claims Databases for Signal Detection in Pharmacovigilance

Signal detection processes in pharmacovigilance aim to identify potential associations between drugs and adverse events (AE). There are various data sources and methods used for signal detection. Within pharmaceutical manufacturers, an increasing number of AE reports are received from their patient support programs (PSP). PSPs promote interaction between the manufacturer and patients or healthcare professionals, which may lead to increased AE reports that would not have been received in the absence of PSPs. The impact of PSPs on signal detection is unclear. Administrative claims databases are an unconventional, albeit possible resource with important advantages for signal detection. The tree-based scan statistic (TBSS) method is a signal detection method that can be used on administrative claims database to detect multiple signals simultaneously. The TBSS method is relatively new and its ability to detect signals that led to regulatory actions such as labeling changes has not been assessed. This dissertation aimed to explore the use of a manufacturer drug safety database and administrative claims database in search of more efficient and effective ways to detect signals. The first study aimed to compare the signal detection performance using data from PSPs and data from non-PSP sources within a manufacturer drug safety database and also among PSPs providing different types of services. While data from PSPs were worse at detecting signals compared to data from non-PSP sources, PSPs providing certain types of services performed better than others at detecting signals. The second study evaluated the impact of completeness of reports on signal detection by incorporating measures of completeness into the signal detection process through application of frequency weights and restricting the dataset. The study found that incorporating the completeness of AE reports did not significantly impact signal detection performance. Finally, the third study evaluated whether the TBSS method was able to detect signals that led to subsequent labeling updates for four medications using administrative claims database. The TBSS method was able to detect the signal of interest for one of the four medications. Findings from this research can help pharmacovigilance experts to better understand these data sources and interpret the signal detection results accordingly.