Anticoagulation for Chronic Atrial Fibrillation in Hemodialysis Patients: Which Fruit From the Decision Tree?

Atrial fibrillation (AF) is an increasingly common condition that highlights the dual risk of thrombosis and bleeding complications among patients with severe chronic kidney disease (CKD). As the long-term dialysis patient population grows older, the prevalence of AF will also most likely increase from its current estimated prevalence of 14% of hemodialysis patients,1 a value which is already several fold higher than the 2% to 4% prevalence in the general population aged 60 and over.2, 3, 4 Randomized controlled trials have shown that warfarin prophylaxis is more effective than aspirin in reducing subsequent stroke risk associated with AF, although also more likely to lead to bleeding complications. However, in a caveat familiar to nephrologists, these trials excluded patients with severe CKD. Nevertheless, conditions contributing to the risk of stroke in AF5 are highly prevalent in long-term dialysis patients. It might therefore be reasonable to conclude that hemodialysis patients are at very high risk for thromboembolic complications in a setting of AF and would benefit from therapy.

However, we also must acknowledge that dialysis patients are at greater risk of bleeding complications than the general population, especially gastrointestinal bleeding, based on observational data.6 Additionally, in contrast to patients in the general population, hemodialysis patients receive heparin anticoagulation with each treatment, presumably increasing their bleeding risk. Further, in studies conducted in hemodialysis patients, warfarin use has proved disappointing (and hazardous) in other circumstances where anticoagulation might have been expected to provide benefit, such as prophylaxis for thrombosis of polytetrafluoroethylene dialysis grafts.7 In fact, the findings of randomized controlled trials performed in the general population and those conducted in the dialysis population often conflict, including the recent Deutsche Diabetes Dialyse Studie (4D) trial.8 We also know that many interventions that are considered the standard of care are considerably underutilized in dialysis patients.9 In fact, even interventions shown to be directly beneficial from randomized controlled trials performed in dialysis patients are still underutilized.10, 11, 12 This suggests that, in the absence of additional guidance, providers might be unlikely to administer anticoagulation for hemodialysis patients with AF. All things considered, the decision on whether to use anticoagulation for AF in long-term dialysis patients is not straightforward. This issue’s publication of a report from Quinn et al13 on a decision analysis of anticoagulation of dialysis patients with AF is thus both timely and useful. The authors found that in hemodialysis patients with AF, compared with no anticoagulation, aspirin use was associated with improved quality-adjusted life-years (QALYs) at a cost of over $82,100 per QALY, while warfarin use provided additional benefits for an additional cost of $88,400 relative to aspirin. As the authors themselves ask, can we trust these results, and are they worthwhile?

These questions actually go far beyond the direct issue at hand; they take us to the issue of how to apply existing medical knowledge to our unique patient population, which, as all nephrologists know, is understudied. If we were to rely only on the results available from randomized clinical trials performed on dialysis patients, we would have scarce information indeed.

The authors’ study used decision tree/decision analysis techniques, which are being used increasingly in clinical research but have not (at least until recently) been emphasized in training programs. Therefore, many practicing nephrologists may not be familiar with their strengths and limitations, and although a discussion of them is beyond the scope of this editorial, in a nutshell these techniques can take available evidence for a “test drive” over a range of values and situations that would be impossible to assess individually in clinical trials. However, they do not eliminate the need for a high-quality randomized controlled trial of anticoagulation therapy for AF in dialysis patients, and the authors make clear the present study should not be considered a substitute for one. Of necessity, these studies make some sort of extrapolation from known to unknown circumstances, in essence, they assume the road continues around the unseen curve.

Any study dealing with a unique subpopulation faces the choice of whether to use data from the general population (since much of the information for dialysis patients is not available), and how to reconcile data for dialysis patients that conflict with the general population. Additionally, the hemodialysis population is unique in the magnitude of the costs for continuing care, now approaching $100,000 per patient per year, which are actually used by other disciplines as the limit for the maximum amount society would be “willing to pay” for 1 year of life. Quinn and colleagues dealt with these issues in a balanced and complete fashion which may serve as a model for future investigators tackling different issues.

Benefits and risks of therapies are best addressed by clinical trials; unfortunately, clinical trials in the CKD population have been the exception rather than the rule. Results from observational studies, particularly large registries such as the US Renal Data System, often contradict data from the general population. This phenomenon has been called by some “reverse epidemiology,” which does not mean that the normal rules do not apply to the CKD population, but rather that the high prevalence of comorbid conditions, malnutrition, inflammation, and the resulting high death rates muddy the waters of many retrospective (and even prospective) trials. Perhaps the most difficult contradiction was estimating the rates of thromboembolic events among hemodialysis patients with AF. Here, 2 important studies conflict, 1 showing that stroke rates were lower among patients with AF than among those with sinus rhythm,14 contrary to findings in the general population, and another which found thromboembolic event rates much higher for AF patients (24/100 patient-years) than those in sinus rhythm (5/100 patient-years).15 Quinn and colleagues do a commendable job of synthesizing the evidence available from the general population with that available from the dialysis population, showing the results of sensitivity analyses among the range of plausible values.

Further reinforcing the need to carefully screen hemodialysis patients for potential bleeding risk, a systematic review by Elliott et al16 in this same issue of the journal found that in comparison to dialysis patients not on anticoagulation other than intradialytic heparin, hemodialysis patients who received warfarin anticoagulation had approximately double the risk of bleeding episodes. Further, in comparison to studies of patients in the general population who received warfarin, the risk of bleeding episodes for dialysis patients treated with warfarin is increased from 3- to 10-fold. The authors implicated lack of appropriate monitoring as a substantial contributor to this risk. Many dialysis satellite units do not have quick turnaround time for laboratory results. Further, a high percentage (25%) of dialysis patients receive warfarin despite a lack of clear indications.17 While the detailed reasons for suboptimal monitoring are not available, this suggests the possibility that communication/handoff between hospitalization to the long-term dialysis unit setting, a consistent emphasis of Joint Commission on Accreditation of Healthcare Organizations inspections, should be improved in this population.

The other dilemma is the high cost of ongoing care for hemodialysis patients. In economic analyses, this actually sets up the perverse situation in which it is less expensive to society for patients to die than to remain alive, unless they receive a less expensive type of therapy such as kidney transplantation. How to handle costs associated with patient death on hemodialysis remains unresolved, especially after invasive procedures or surgeries, since analysis of Medicare claims indicates this varies with the timing from the procedure.18 This situation makes interpretation of therapies very complex. However, the authors address this point by showing analyses both taking into account and disregarding costs of ongoing therapy (which may be the more realistic approach since providers would not likely discontinue hemodialysis in the United States for purely economic reasons). As shown vividly in their figures, warfarin use was more cost effective over the range of “willingness to pay” ranges most likely in clinical practice. In fact, warfarin became even more attractive if the future costs of hemodialysis were not included, which therefore does not penalize patients for their costs of remaining alive (Fig 3 in13), and actually was of similar attractiveness to interventions widely used in the general population. The authors point out that the $50,000/QALY threshold, based on old estimates of annual hemodialysis costs, are likely outdated. They state that even though convention has yet to be established on this point, “a rational decision maker should be willing to pay at least as much for a QALY as is already being paid.”13 Since the cost of hemodialysis is now above $90,000 annually, we would be somewhat inconsistent to say therapies costing less are too expensive. Perhaps the most noteworthy aspect of the study is setting the standard for future investigators to perform an analysis that does not penalize patients or providers for the future costs of ongoing hemodialysis care. The high cost of maintenance hemodialysis makes cost-effectiveness comparisons between hemodialysis and the general population very difficult. This is more than a technical point, because the “background noise” of the high cost of hemodialysis may obscure important cost differences between alternative therapies, as the authors point out, and thus analysis excluding the costs of long-term hemodialysis would also make comparisons with the general population more relevant and understandable.

In their Fig 6, the authors present a very compelling case on the need for more information about AF in hemodialysis patients. The “expected value of perfect information,” in a manner somewhat analogous to calculus determining a result as a particular predictive variable approaches infinity, gives us an educated guess as to how much “perfect information” about this condition and its complications would be a worthwhile investment, considering how much society already pays for ongoing treatment. While we should not be willing to pay more for imperfect information, as the authors point out, given current assumptions such information would be worth more than $76 million,13 well worth a clinical trial. In the meantime, given the current state of information, the authors have provided useful guidance to providers and future investigators, and a very solid justification for future studies.

Until more data are available, providers should consider standard warfarin anticoagulation in hemodialysis patients with chronic atrial fibrillation, adhering to screening and monitoring as in the general population. Providers should be alert to review medications for patients who have recently been discharged from hospitalization for use of warfarin, and to arrange appropriate monitoring.