Author + information
- Torben Bjerregaard Larsen, MD, PhD,
- Lars Hvilsted Rasmussen, MD, PhD,
- Flemming Skjøth, MSc, PhD,
- Karen Margrete Due, MSc,
- Torbjörn Callréus, MD, PhD,
- Mary Rosenzweig, MSc and
- Gregory Y.H. Lip, MD∗ ()
- ↵∗University of Birmingham, Centre for Cardiovascular Sciences, City Hospital, Dudley Road, Birmingham B18 7QH, United Kingdom
Dr. Sipahi and colleagues express concern about the discrepancy of our observational study with randomized controlled studies, and point to residual confounding as a possible explanation. We have already discussed these issues in the paper (1), but will expand on our discussion in the following paragraph.
In observational studies of intended drug effects or safety, substantial confounding (by indication) is to be expected because the perceived risk is often closely related to the physician's choice of treatment (2). Where there is confounding, there is also the possibility of residual confounding. Taken to the extreme, heterogeneity in risk factors (measured or unmeasured) between treatment groups in key risk factors is a possible explanation for the observed associations. However, “possible” need not mean “plausible.” Indeed, a careful choice of methods and principles can mitigate confounding concerns in observational studies (3).
In our study, we adopted a new-user design to ensure that meaningful comparisons were made (4). We explored both propensity score matching and regression-based confounder adjustment and found no appreciable differences between these approaches. Last, we found reassurance in the fact that estimates changed only modestly upon adjustment for key risk factors for the outcome. Any unmeasured confounders would have to be very strongly associated with treatment and outcome in order for estimates and conclusions to change qualitatively (5). While it is possible that such unmeasured confounders exist, we do not consider it very plausible.
There can be other explanations for the discrepancies between observational and randomized studies (6), which seem more plausible here. For example, differences in length of follow-up may result in different conclusions. Also, randomized controlled trials are externally valid only for the type of patients included in that trial. Our observational study represents a “real world” population, in which patients had a lower stroke risk cohort compared with the RE-LY (Randomized Evaluation of Long-Term Anticoagulation Therapy) trial (7) participants (mean CHADS2 scores of 1.2 and 2.1, respectively), and correspondingly, a lower myocardial infarction risk as well. Also, our patients had a lower prevalence of prior myocardial infarction or fewer risk factors for the same (e.g., diabetes mellitus, hypertension), compared with the RE-LY trial.
In their letter, Sipahi et al. proclaim randomized trials as the gold standard for drug efficacy and safety assessments. While randomized controlled trials are indeed the gold standard in the sense of providing “fair” comparisons, they may not always provide the most relevant comparisons. Well-designed observational studies can address the question whether drug treatment works in daily clinical practice, not just whether the drug by itself works in ideal settings. Careful consideration of all the available evidence (randomized and observational) should be considered the gold standard for post-marketing drug evaluations (8).
Please note: Dr. Lip has served as a consultant for Bayer, Astellas, Merck, AstraZeneca, Sanoﬁ, BMS/Pfizer, and Boehringer Ingelheim; and has been on the Speaker's Bureaus for Bayer, BMS/Pfizer, Boehringer Ingelheim, and Sanoﬁ. Drs. Larsen and Rasmussen have been on the Speaker's Bureaus for Bayer, BMS/Pfizer, and Boehringer Ingelheim. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- American College of Cardiology Foundation
- Larsen T.B.,
- Rasmussen L.H.,
- Skjoth F.,
- et al.
- Greenland S.,
- Neutra R.
- Ray W.A.
- Dreyer N.A.,
- Tunis S.R.,
- Berger M.,
- Ollendorf D.,
- Mattox P.,
- Gliklich R.