Author + information
- Robert S. Epstein, MD, MS⁎ (, )
- Thomas P. Moyer, PhD,
- Ronald E. Aubert, PhD,
- Dennis J. O'Kane, PhD,
- Fang Xia, PhD and
- J. Russell Teagarden, DMH, RPh
- ↵⁎Medco Health Solutions, Inc., 100 Parsons Pond Drive, Franklin Lakes, New Jersey 07417
Dr. Lefevre and colleagues express concern about selection bias in our study due to low patient acceptance rates (1), a concern also expressed by Ginsburg and Voora in their accompanying editorial (2). In fact, 46% of subjects contacted agreed to participate. We compared subjects declining to participate with subjects agreeing to participate on 29 baseline variables. Participating subjects had more prior use of nonsteroidal anti-inflammatory drugs (19.9% vs. 15.4%) and more frequent prior history of gastrointestinal bleed (4.0% vs. 2.2%), suggesting that participants would have biased against our findings, because participants were at a slightly higher risk of adverse outcomes compared with nonparticipants.
The concerns Dr. Lefevre and colleagues have expressed about potential confounders such as socioeconomic status, ethnicity, and educational level were addressed in the study design. Both the intervention group and the control group were derived from the same participating payer organizations. Therefore, both groups had the same insurance coverage and access to care. Additionally, because both groups were constructed from the same payer organizations in the same regions and from the same industries, it is reasonable to assume they were similar in socioeconomic status and in other unmeasured factors as well.
The larger difference we showed in “all-cause” hospital stay compared with bleeds or thromboembolic events was anticipated and expected. Prior studies describe imprecision inherent in International Classification of Diseases of the World Health Organization-9th Edition claims coding. We used validated codes from the peer-reviewed published data to identify the specific events, because the specificity of these codes has been shown to be nearly 100% (3). However, the sensitivity to identify these events has been shown to be 30% to 50%, indicating that we miss events by using these validated codes alone. For example, the all-cause category included hospital stays for problems such as anemia secondary to blood loss, unspecified anemia, and iatrogenic pulmonary embolism with infarction. We could not use these coded events, because the codes themselves were not included in prior validated studies. Like others before us in health services research, we anticipated and thus reported both the specific and all-cause event codes.
Dr. Lefevre and colleagues suggest that nonrandomized study designs be used only when randomized trials are neither feasible nor ethical. In our case, this was not feasible, because benefit payer organizations would not participate in a study where they would have to withhold a covered health care benefit to those potentially randomized to usual care. Additionally, we purposefully did not want to conduct a study in a controlled environment in which the patients and providers and care itself do not necessarily reflect typical practice conditions.
Warfarin treatment, although efficacious, is treacherous for patients and clinicians alike; the status quo features an alarming rate of hospital stay due to complications—up to 30% during the first 6 months of use—and a top ranking among the leading causes of emergency department admissions for adverse drug events. Allelic variances have a strong association with warfarin sensitivity and testing is widely available, inexpensive, and low risk to patients. We believe clinicians evaluating our study should focus on the frequency and consequences of warfarin adverse events and the considerable evidence supporting testing rather than on methodological grounds in isolation.
- American College of Cardiology Foundation
- Epstein R.S.,
- Moyer T.P.,
- Aubert R.E.,
- et al.
- Ginsburg G.S.,
- Voora D.