(175) The Use and Implementation of Instrumental Variables in Pharmacoepidemiology: A Scoping Review

Background: Unmeasured confounding is a concern when examining the safety and effectiveness of medicines in real-world studies. Instrumental variable (IV) analyses offer a potential solution to this problem if several assumptions are fulfilled. However, IV analyses are still relatively uncommon and there are substantial gaps in our understanding of their use in pharmacoepidemiological research.

Objectives: To describe how IV analyses have been applied in prescription drug research, including how IVs have been defined and operationalized, whether evidence of their validity was demonstrated in published studies, and characteristics of the studies for which these methods were used.

Methods: We searched MEDLINE, Embase, PsycINFO, CINAHL Complete, and National Bureau of Economic Research databases for relevant articles published between January 1, 2000 to August 1, 2022. Studies were included if they applied an IV to assess the safety or effectiveness of a drug or combination of drugs, defined as pharmacologically actives molecules such as a medicine or vaccine. Studies were included where the drug was either the intervention or comparator; or where the same drug was evaluated at different doses, schedules, routes, or formulations. Information extracted on the IVs included the target instrument, operationalization of the instrument, role of the IV analysis in the study (primary/sensitivity), and whether evidence was provided for the IV assumptions including: relevance, exclusion restriction, and exchangeability. All IVs used in a single study were extracted separately. This study was registered with Open Science Framework (DOI: 10.17605/OSF.IO/V96MG).

Results: From 4,368 unique studies, 241 eligible studies were identified. There has been a steady increase in studies employing IVs over time, with most (n=217) published after 2009 and 41% (n=100) in the last 5 years (2018-2022). Two countries (USA, Japan) generated nearly half (47%) of all studies. From data extracted to date on IVs (n=33), use of facility-prescribing pattern has been the most (n=21) common instrument and assumptions were infrequently verified (33%).

Conclusions: IV analyses are increasingly being implemented in prescription drug research, but the preliminary findings indicate authors often neglect to test assumptions necessary for valid IV estimation.