Impact of varying control moment selection in a case-crossover (CCO) study on antidepressant drug (AD) use and hip/femur fracture (HFF) in protect

Abstract

Background: The CCO design is appealing because it eliminates time-invariant person related confounding. A prerequisite is that exposure in real life drug use is sufficiently transient to allow for independence of exposure states. The impact of variation in time of control moment selection is relatively unknown. Objectives: To assess the influence of selection of control moments at different times in a CCO study of AD and HFF on variation in effect estimates. Methods: Adult patients with HFF who received an AD prescription during 2001-2009 were identified from the Dutch Mondriaan GP database. For each patient, a case moment (the date of HFF) and four control moments at 3, 6, 9, and 12 months before the HFF (M3, M6, M9, M12) were defined. Each AD prescription had a pre-defined duration of 90 days.AD treatment episodes were constructed and divided into current, recent (0-2 months following current use) and past use (>2months follow current use).We used conditional logistic regression to compute odds ratios (ORs) and 95% confidence intervals CI between AD use and HFF. Results: Pairwise (1:1) comparisons of 82 case moments to varied control moments for current versus no use resulted in ORs for HFF-M3 of 16.3 (95%CI: 2.2-123), M6: 7.8 (2.3-26), M9: 5.9 (2.1-16.1), and M12: 4.1 (1.8-9.4). Including all (1:4),M3-M12, resulted in OR 7.0 (3.2-15.2). For recent use even higher ORs were found; M3: 49.7 (3.9-637), M6: 17.6 (2.5-136), M9: 2.6 (0.7-9.5), M12: 3.7 (0.7-20), All 8.6 (2.7-27). Discordancy of exposure and thus number of strata contributing to the analyses increased from 32% in M3 to 50% in M12. Conclusions: Selection of control moments at different times in CCO has considerable impact on effect estimates in this particular setting. CCO studies should be designed with sufficient time between case and control moments to allow for sufficient discordancy in exposure to get reliable estimates.