Customize your JAMA Network experience by selecting one or more topics from the list below.
Oral anticoagulation is highly effective at preventing stroke in patients with atrial fibrillation (AF). Warfarin and other vitamin K antagonist oral anticoagulants reduce stroke by roughly two-thirds compared with placebo.1 Four recently developed non–vitamin K antagonist oral anticoagulants (NOACs) offer important advantages over warfarin, including an approximate 50% reduction in intracranial hemorrhage and no requirement for routine monitoring.2-5 In trials comparing these 4 NOACs with warfarin in patients with AF, the relative effects on stroke, bleeding, and intracranial hemorrhage have been consistent across many subgroups defined by baseline patient factors. Nevertheless, given the cost of NOACs to patients and health care systems, it is valid to question whether there are cohorts of patients with AF for whom warfarin results in good enough outcomes that the absolute incremental benefit of NOACs would be small and the standard of care (ie, warfarin) remains the preferred option.
Warfarin is a complex drug with many food and drug interactions and requires regular therapeutic monitoring of the standardized international normalized ratio (INR). A target INR of between 2 and 3 is considered ideal to optimize safety and efficacy, and the time in therapeutic range has been developed as an indicator of the quality of warfarin management.6 In both clinical trials and practice, there is substantial heterogeneity in time in therapeutic range across countries, centers, and patients. The availability of INR data for warfarin invites the question, “Do patients with consistently excellent INR control have good outcomes on warfarin—perhaps good enough that the absolute benefit of NOACs would be small?”
In this issue of JAMA Cardiology, Björck and colleagues7 report an observational analysis of several large clinical and administrative data sets from Sweden investigating, among other things, the safety and effectiveness of well-managed warfarin in patients with AF and the association between individual patient time in therapeutic range and outcomes. They found relatively low overall rates of death and intracranial hemorrhage, particularly in the absence of renal failure or concomitant aspirin and among patients with good INR control (time in therapeutic range ≥70%). Based on these findings, the authors conclude that “well-managed warfarin therapy is…still a valid alternative for AF-associated stroke prevention.”7 This statement is almost certainly true for some patients; however, whether the current analysis helps to identify those patients or to inform decisions regarding choice of anticoagulation strategy is less clear.
Historically, among patients with AF, Sweden has had the best INR control in the world.8,9 Given the strong association between country and INR control, one might question how these results apply to patients outside of Sweden.10 The findings that renal dysfunction and concomitant antiplatelet therapy are associated with worse outcomes in anticoagulated patients with AF replicate prior analyses and are likely generalizable.10-12 These important subgroups do help inform treatment decisions, particularly given the randomized comparisons of event rates among patients receiving NOACs and warfarin.10,11 Perhaps the most attractive finding is that patients with consistently well-controlled warfarin have very low rates of subsequent complications or death.
The analysis relating INR control and outcomes, however, is not designed in a way that can be used to inform patient prognosis nor treatment decisions regarding anticoagulation for patients with AF. First, it is unlikely that this association is entirely causal. It is well known that patients with better INR control are different, and generally lower risk, than patients with worse INR control.13-15 Patients with better INR control might be expected to have better outcomes with NOACs as well. Without a NOAC comparator group, this cannot be assessed. Second, the observation likely involves a self-fulfilling prophecy. The cohort of patients with well-controlled warfarin was defined retrospectively across the same time frame in which outcomes were evaluated. Periods of good INR control during follow-up are more likely in the absence of procedures, interruptions, and clinical events. Finally, even if the association between good INR control and low outcomes is entirely causal, it cannot be prognostic. It requires information that can only be known after the outcomes themselves are known.
The analysis of outcomes in a retrospective cohort with well-controlled warfarin cannot inform treatment decisions that are made without knowledge of the future. The real danger is that physicians and patients will use these findings to guide treatment decisions among warfarin-experienced or, worse, warfarin-naive patients. It would be easy to inappropriately extrapolate these data because the results are consistent with the expectation that INR control matters and because other sources of information related to INR control (ie, historic measurements or center-level INR control) are commonly available.
Rather than extrapolate, future analyses on this topic should carefully align the question of interest with the analysis (ie, covariates, effect modifiers, comparators, outcomes, and duration of follow-up). When deciding whether to initiate anticoagulation, we might ask, “Can I identify patients who would do well while taking warfarin?” At this baseline, no individual INR data are available. Risk stratification can be performed based on traditional risk factors (CHA2DS2-VASc [cardiac failure or dysfunction, hypertension, age ≥75 years (doubled), diabetes mellitus, and stroke (doubled)–vascular disease, age 65-74 years, and sex category (female)] and biomarkers) as well as predicted INR, for which prediction is based on patient-level and/or center-level covariates measured at baseline.
Randomized comparisons of a NOAC with warfarin according to predicted INR levels have been conducted.13,15 For patients already taking warfarin, we might ask, “Can I identify patients who will continue to do well while taking warfarin?” In this population, the experience with warfarin (including historic INR data) may be one of the most important determinants. Given a well-defined duration of warfarin experience and adequate historical INR data, a new baseline or landmark can be established. A prediction model for subsequent outcomes might include classic risk factors (CHA2DS2-VASc and biomarkers) and/or INR data measured prior to the landmark (or series of pooled landmarks). Finally, while these analyses might inform which patients are likely to have good outcomes while taking warfarin, they will not answer the question of whether these patients would have better outcomes while taking NOACs. If outcomes while taking warfarin are sufficiently good, perhaps the likely relative benefits of NOACs are just not worth it. However, features related to good INR control, such as adherence, lack of interruptions, and better health care, might also contribute to good outcomes with NOACs. To definitively answer the comparative question of NOACs vs warfarin in patients with good INR control taking warfarin would require a randomized trial including patients with either predicted or observed good INR control while taking warfarin who are subsequently randomized to either warfarin (new or continued) or a NOAC and followed up for clinically relevant outcomes. Given the postulated good outcomes among patients with good INR control, such a trial would be large and, until we get better at doing very large, very simple randomized trials, will likely never be done. In the meantime, observational studies should be rigorously designed and evaluated, not only with respect to issues of confounding, but also in terms of what questions they are designed to answer.
Corresponding Author: John H. Alexander, MD, MHS, Duke Clinical Research Institute, Duke Medicine, PO Box 3850, Durham, NC 27710 (firstname.lastname@example.org).
Published Online: April 20, 2016. doi:10.1001/jamacardio.2016.0198.
Conflict of Interest Disclosures: Dr Alexander reports institutional research grants from Boehringer Ingelheim, Bristol-Myers Squibb, CSL Behring, Pfizer, Sanofi, Regado Biosciences, Tenax, and Vivus and consulting fees/honoraria from Bristol-Myers Squibb, CSL Behring, Daiichi Sankyo, GlaxoSmithKline, Janssen, Pfizer, Portola, Sohmalution, and Xoma. Dr Thomas reports institutional research grants from Bristol-Myers Squibb, Pfizer, and Janssen Scientific Affairs.
Alexander JH, Thomas LE. Using Data to Guide Anticoagulation in Patients With Atrial Fibrillation: Does the Analysis Fit the Question? JAMA Cardiol. 2016;1(2):121–122. doi:10.1001/jamacardio.2016.0198
Coronavirus Resource Center
Create a personal account or sign in to: