Premium
Instrumental variable analysis as a complementary analysis in studies of adverse effects: venous thromboembolism and second‐generation versus third‐generation oral contraceptives
Author(s) -
Boef Anna G. C.,
Souverein Patrick C.,
Vandenbroucke Jan P.,
Hylckama Vlieg Astrid,
Boer Anthonius,
Cessie Saskia,
Dekkers Olaf M.
Publication year - 2016
Publication title -
pharmacoepidemiology and drug safety
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.023
H-Index - 96
eISSN - 1099-1557
pISSN - 1053-8569
DOI - 10.1002/pds.3956
Subject(s) - medicine , venous thromboembolism , third generation , first generation , adverse effect , instrumental variable , pharmacoepidemiology , intensive care medicine , population , pharmacology , thrombosis , statistics , medical prescription , environmental health , telecommunications , mathematics , computer science
Purpose A potentially useful role for instrumental variable (IV) analysis may be as a complementary analysis to assess the presence of confounding when studying adverse drug effects. There has been discussion on whether the observed increased risk of venous thromboembolism (VTE) for third‐generation oral contraceptives versus second‐generation oral contraceptives could be (partially) attributed to confounding. We investigated how prescribing preference IV estimates compare with conventional estimates. Methods Women in the Clinical Practice Research Database who started a second‐generation or third‐generation oral contraceptive from 1989 to 2013 were included. Ordinary least squares and two‐stage least squares regression were used to estimate risk differences in VTE. Cox regression and IV for Cox proportional hazards regression were used to calculate hazard ratios (HR). The instrument used was the proportion of prescriptions for third‐generation oral contraceptives by the general practitioner in the year preceding the current prescription. Results All analyses pointed in the direction of an increased VTE risk for third‐generation oral contraceptives. The adjusted HR from the conventional Cox regression was 1.62 (95% confidence interval 1.16–2.27) and the fully adjusted HR from the IV Cox regression was 3.45 (95% confidence interval; 0.97–11.7), showing a larger risk and wider confidence intervals in the IV analysis. Conclusions The similarity in direction of results from the IV analyses and conventional analyses suggests that major confounding is unlikely. IV analysis can be a useful complementary analysis to assess the presence of confounding in studies of adverse drug effects in very large databases. Copyright © 2016 John Wiley & Sons, Ltd.