No impact of adjusting for lifestyle factors or general practice on risk estimates for the association between antidepressants and hip/femur fracture

Background: Routinely collected data from electronic health record databases often lack information on relevant risk factors, like lifestyle-factors (LSF, smoking, alcohol use, body mass index) or socioeconomic factors that may be needed for confounder adjustment in epidemiogical studies. Objectives: In the context of the Pharmacoepidemiological Research on Outcomes of Therapeutics by a European Consortium (PROTECT) project, the impact of confounder adjustment on the risk of antidepressant (AD) use on hip/femur fracture (HF) and compared results across three primary care databases was assessed. Methods: We conducted a case-control study nested within 3 new AD user cohorts of adult patients (2001-2009) in three databases (Spanish BIFAP, Dutch Mondriaan and UK THIN. Cases were defined as a first HF during the study period. Up to 4 controls were matched by sex, age (+/-2 years) and time since cohort entry (+/- 6 months). Exposure to AD was classified into current, recent and past use. We adjusted for comedication and comorbidities, using same models for all data sources. The impact of matching on practice (marker for socioeconomic factors) and additional adjustment for LSF was done in THIN. Odds ratios (OR) were estimated using conditional logistic regression analysis. Results: Current use of AD was associated with an significantly increased risk of HF in all data sources. Adjusted ORs were 1.52 in BIFAP (1535 cases), 1.59 in THIN (3756 cases) and 3.32 in Mondriaan (79 cases). In BIFAP/THIN, adjustment resulted in