Hypothesis-free screening of large administrative databases for unsuspected drug-outcome associations

Jesper Hallas, Shirley V. Wang, Joshua J. Gagne, Sebastian Schneeweiss, Nicole Pratt, Anton Pottegård

Active surveillance for unknown or unsuspected adverse drug effects may be carried out by applying epidemiological techniques to large administrative databases. Self-controlled designs, like the symmetry design, have the advantage over conventional design of adjusting for confounders that are stable over time. The aim of this paper was to describe the output of a comprehensive open-ended symmetry analysis of a large dataset. All drug dispensings and all secondary care contacts in Denmark during the period 1995-2012 for persons born before 1950 were analyzed by a symmetry design. We analyzed all drug-drug sequences and all drug-disease sequences occurring during the study period. The identified associations were ranked according to the number of outcomes that potentially could be attributed to the exposure. In the main analysis, 29,891,212 incident drug therapies, and 21,300,000 incident diagnoses were included. Out of 186,758 associations tested in the main analysis, 43,575 (23.3%) showed meaningful effect size. For the top 200 drug-drug associations, 47% represented unknown associations, 24% represented known adverse drug reactions, 30% were explained by mutual indication or reverse causation. For the top 200 drug-disease associations the proportions were 31, 15, and 55%, respectively. Screening by symmetry analysis can be a useful starting point for systematic pharmacovigilance activities if coupled with a systematic post-hoc review of signals.