To drive change, medicine requires hard data to supply evidence of clinical benefit. However, the studies we rely on to make decisions about a drug's efficacy are often statistically underpowered - that is, therapeutic trials may fail to show the benefit of agents or devices when a benefit does, in fact, exist. This is due to limited data from smallpatient populations or too much variability in the data.
We performed analyses of studies of anticoagulation in electrical cardioversion to examine this problem more clearly. We also show how proactive data pooling could help to mitigate limitations in statistical power.
Electrical cardioversion of atrial fibrillation (AF) to normal heart rhythm abruptly increases stroke risk because an atrial blood clot (thrombus) could be dislodged. The standard approach is to start the anticoagulant, warfarin, for at least 3 weeks prior to cardioversion. Anticoagulation with warfarin is still provided after cardioversion. A recent trend is to avoid anticoagulation pretreatment with warfarin if transesophageal echocardiographic (TEE) imaging demonstrates that no clots are present.
The pivotal trial leading to guideline acceptance of this strategy was underpowered. Our analysis shows less than 50% of the study population required for a properly powered study was recruited and randomized to the two study arms (warfarin pretreatment or TEE imaging).
Physicians are beginning to substitute a novel oral anticoagulant (NOAC), either dabigatran or a factor Xa inhibitor (apixaban, rivaroxaban or edoxaban), for warfarin. While NOACs have been shown to reduce embolic stroke in chronic AF studies, there is scarce data to show that the same is true in the setting of electrical cardioversion.
Some retrospective analyses have been performed and additional prospective trials are ongoing, all based on documenting non-inferiority (NOAC versus warfarin). Unfortunately, none of the studies have been designed with enough statistical power to come even close to determining this with a strong degree of confidence.
We recently analyzed statistical power in several studies that tested this hypothesis in AF cardioversion (McKenzie D et al. J Am Coll Cardiol. 2015;65(10_S). While a robust study would normally maintain statistical power of 80-90%, the individual 7 completed or ongoing non-inferiority trials had statistical powers of only 4-16% for ruling out a doubling of thromboembolic risk and 8-34% for excluding a two-fold difference in major bleeding.
We then estimated how many subjects would be required to demonstrate rigorous statistical power for ruling out a risk doubling (hazard ratio of 2.0). The numbers we determined cause concern - these studies would need 16,300 and 21,800 subjects for 80 and 90% power, respectively, to demonstrate a doubling of risk.
They would need even more subjects - 47,600 and 63,700 subjects for 80% and 90% power, respectively -to exclude a smaller difference, 1.5 times the risk, between treatment arms.
The current ENSURE-AF study, the largest study of this type ever attempted, is evaluating approximately 2,200 subjects, far fewer than the numbers needed. Even ENSURE will be dramatically underpowered to answer the clinical non-inferiority question with a high degree of certainty.
Large-scale studies are often financially impractical, yet small-scale studies are statistically underpowered and inadequate to guide major changes in medical practice. We see a strong need for collaboration within the pharmaceutical industry to enable data pooling for statistical analyses.
For example, pooling data in the 7 preceding or ongoing trials above would yield 6,518 subjects and a 43% power to demonstrate non-inferiority at a margin of 2.0 times the risk. While not perfect, a meta-analysis like this would provide substantial additional confidence to practitioners about switching to factor Xa inhibitors from warfarin.
Pooling data could lead to greatly reduced medical costs and improvement in quality of life for patients by improving the statistical power of clinical studies. We, therefore, advocate for the prospective standardization of data by sponsor companies with a goal of eventual data pooling. This approach will improve the quality and dependability of the results and provide an improved confidence in this therapeutic class that will benefit both patients and industry alike.
About Jonathan Plehn, MD
Jonathan Plehn, MD earned his Doctorate in Medicine from New York University School of Medicine and completed an internship and residency at Montefiore Hospital, University of Pittsburgh before completing his cardiology fellowship at Rush-Presbyterian-St. Luke's Medical Center in Chicago with further training at Boston University School of Public Health. Dr. Plehn has a wide range of therapeutic trial experience in the areas of lipid therapy, congestive heart failure and the design and management of cardiovascular outcomes trials.
About David McKenzie, MS
David McKenzie earned his Master of Science degree from the University of Minnesota and has over 27 years of clinical trial experience. He has been with Covance for over 6 years where he supports DMCs as the independent statistician, analyzing and presenting interim analysis.