Author + information
- Frank Lefevre, MD⁎ (, )
- Steven N. Goodman, MD, MHS, PhD and
- Margaret A. Piper, PhD, MPH
- ↵⁎Northwestern University, General Internal Medicine, 750 North Lake Shore Drive, Chicago, Illinois 60618
The recently published MM-WES study (Medco-Mayo Warfarin Effectiveness Study) (1) addresses an urgent need for evidence in the area of pharmacogenomics, in this case the potential impact on clinical outcomes of pharmacogenetic testing for guiding warfarin dosing. However, serious design flaws preclude drawing reliable conclusions from this study. Modest changes in the original study design would have produced more convincing results.
A major concern is selection bias arising from the use of a historical control group. Outcomes such as hospital stays might be more strongly influenced by clinical and demographic characteristics and their variable distribution between groups than by genetic testing. Unfortunately, because key information such as the number of patients who were offered enrollment but declined was omitted, it is impossible to exclude the possibility that the results are heavily influenced by selection bias. A high patient refusal rate could produce a highly selected population, not comparable to unselected historical controls. Often patients who agree to enroll in trials are healthier than those who do not agree, the so-called “healthy volunteer” effect (2,3).
Even for the limited set of baseline clinical and demographic variables examined, there were differences between intervention and control groups in the history of hypertension and history of diabetes. The same limited set of factors was used for propensity matching, making it unlikely that baseline differences could be controlled. Important potential confounders such as ethnicity, socioeconomic status, and educational level—which are often powerful predictors of whether a patient agrees to enter the study and whether the patient is hospitalized—were not available.
Strong clues in these data suggest that outcome differences might be heavily influenced by factors other than genetic testing. Hospital stays related to bleeding/thromboembolism were reduced by 2.16% in absolute terms (from 8.13% to 5.97%), whereas all-cause hospital stays had a much larger reduction of 7.07% (from 25.52% to 18.45%). This is more than triple the absolute effect that would be expected from an improvement in hospital stays for bleeding/thromboembolism resulting from genetic testing. The authors do not provide an explanation for this large difference (4.91%) in hospital stays that are not for hematologic reasons. In addition, an increase in the difference between groups over time (Fig. 1 in Epstein et al. )—indicating an increase in the absolute effect—is not expected, because most of the benefit of testing should be seen early.
Nonrandomized study designs must be used in situations where randomized controlled trials (RCTs) are either not feasible or unethical. That is not the case for this clinical question. At least 2 RCTs are already underway, as noted in the accompanying editorial by Ginsburg and Voora (4). Moreover, this study could have been conducted as an RCT simply by sending the test results randomly to one-half of the physicians caring for patients tested. This would have eliminated selection bias while preserving the “naturalistic setting” (1) and the advantage of improved generalizability. Because of these limitations, the results of this study need to be validated in an RCT before we can conclude that genetic testing for warfarin dosing confers clinical benefits.
- American College of Cardiology Foundation
- Epstein R.S.,
- Moyer T.P.,
- Aubert R.E.,
- et al.
- Ginsburg G.S.,
- Voora D.