Copy the page URI to the clipboard
Hallas, Jesper; Cadarette, Suzanne; Pratt, Nicole; Delaney, Joseph A.; Whitaker, Heather and Maclure, Malcolm
(2019).
URL: https://onlinelibrary.wiley.com/doi/full/10.1002/p...
Abstract
Background: When self‐controlled designs are used to study the triggering of medication‐related adverse effects, time‐varying confounding by indication can occur if the indication or its severity varies over time.
Objectives: We aimed to describe how self‐controlled designs might mitigate or eliminate such confounding by indication by incorporating active comparators with similar indications, illustrated by an empirical example
Methods: Practical approaches to incorporating active comparators will be described for case‐crossover, case‐time‐control, self‐controlled case‐series and symmetry analyses.In the empirical example, we used nation‐wide data from Denmark to study the association between narrow‐spectrum penicillin and venous thromboembolism (VTE), using a case‐crossover design. Macrolide antibiotics were selected as active comparator. This example was chosen because upper respiratory infection ‐ the main indication for narrow‐spectrum penicillin and macrolides ‐ is a transient risk factor for venous thromboembolism, i.e., representing time‐dependent confounding by indication.We identified Danish VTE patients, born 1950 or earlier, during the period 1995–2012. If patients had more than one VTE, we included only the first. The focal window was the 14‐d period before VTE diagnosis. We compared the odds of exposure in that window with one reference window 29–42 days before the VTE. We counted a window as exposed if one of the two antibiotics (penicillin or macrolide) was dispensed within it. We used a Wald‐based method and an interaction term in a conditional logistic regression model to estimate the exposure odds ratio (OR) with 95% confidence limits (CI) for the narrow‐spectrum penicillin users, having the macrolide users as active comparators, i.e. adjusted for transient confounding by indication.
Results: We identified 57486 patients, of whom 4898 (8.5%) were dispensed penicillin during the focal window, and 2226 (3.9%) during the reference window. Corresponding figures were 1192 (2.1% and 572 (1.0%) for macrolide antibiotics. The case‐crossover estimate for penicillin was 2.45 (CI: 2.32–2.59) and 2.22 (CI: 2.00–2.47) for macrolide antibiotics. The Wald‐based estimate for penicillin with macrolide antibiotics as active comparator was 1.10 (CI: 0.98–1.24), and the interaction‐term based estimate was 1.22 (CI: 1.07–1.39).
Conclusions: The strong association of penicillin and macrolides with VTE suggests both are due mostly to time‐varying confounding by indication. Such confounding can be mitigated by applying an active comparator drug that has an similar indication.