Low-Intensity Adjusted-Dose Warfarin for the Prevention of Hemodialysis Catheter Failure: A Randomized, Controlled Trial

Discussion

In a trial of 174 patients allocated to warfarin or to matching placebo, we found no evidence of efficacy of anticoagulation to a target INR 1.5 to 1.9. The study used concealed randomization and blinding of participants, researchers, and clinicians, prespecified outcomes, with no loss to follow-up, and analysis according to the intention-to-treat principle. The breadth of the inclusion criteria enhances the generalizability of results, borne out by the age and comorbidity in participants. Our data do not suggest that warfarin is effective prophylaxis against catheter malfunction.

This is the largest trial to date in people with end-stage renal disease to randomize patients to warfarin or placebo for any indication. Although it is possible that warfarin might be more effective if started the day of insertion, if patients received bridging heparin anticoagulation until the INR is therapeutic, or if a higher INR range had been examined, the absence of any tendency toward benefit in either intention-to-treat or on-treatment analyses argues for true lack of effect.

Four previous prospective studies examined oral anticoagulation with warfarin in the prevention of catheter malfunction. Warfarin (1 mg daily) was ineffective in a randomized trial of 85 patients, although there was a decrease in thrombosis in the subgroup that achieved INR greater than 1.0 (12). An observational study in which 35 patients with previous catheter malfunction were anticoagulated to INR 1.5 to 2.0 and compared with 35 controls showed a trend toward lower rates of malfunction in the anticoagulated patients, particularly in those in whom anticoagulation targets were met (13). A prospective observational study of 62 patients showed fewer episodes of thrombosis with warfarin (INR 2.0 to 3.0) compared with no therapy (14). Finally, in a randomized trial of 144 patients by Coli et al. (15), patients were treated with ticlopidine and randomized to receive warfarin (INR 1.8 to 2.5) combined with low molecular weight heparin until the INR was therapeutic or to no additional treatment, starting within 12 hours of central–venous-catheter insertion. At 1 year, 12% of anticoagulated and 52% of control patients had experienced a thrombosis (P < 0.01), an absolute risk reduction of 40%. Compared with these studies, ours used a randomized design, is the largest to date, and used concealed allocation, blinding of patient, clinical team, investigators, and outcome adjudicators. Our finding of no effect may reflect more rigorous design. Type II error is also possible. We had 80% power to detect a risk reduction of 40%: power for detecting more modest but clinically-important differences was more limited. However, since the design and initiation of our study, lower-intensity anticoagulation has also been shown not to be effective prophylaxis for catheter malfunction in patients with cancer (16).

No randomized studies have examined the efficacy of anti-platelet agents in preventing mechanical malfunction of dialysis catheters, and our study is the largest observational study of this issue. For the primary outcome we observed a clinically-important effect size in favor of anti-platelet agents that did not reach statistical significance (HR = 0.68 [0.43, 1.08], P = 0.11). This tendency is in keeping with other nonrandomized observations in dialysis (14) and nondialysis patients (17).

Bleeding was common in our study, but the risk of bleeding attributable to warfarin was not prohibitive had the intervention been effective. We found 12 major bleeds in 722 months in participants assigned to warfarin (0.20 per patient-year) and seven in 709 months (0.12 per patient-year) in those assigned to placebo (rate ratio, 1.67); the relative risk associated with any major bleed was 1.43 (95% CI, 0.57, 3.58; P = 0.61). Confidence intervals are wide, but the point estimate is in line with the accumulating literature: absolute risk of bleeding of 0.10 to 0.54 per patient-year, and relative risk 1.46 to infinity have been reported (18–20). In the general population, lower targets for INR have been found to be no safer than target INR 2 to 3 (21,22). Our sample size is inadequate to determine whether the difference in fatal bleeding (three in the warfarin group versus one in the placebo group) resulted from chance. A retrospective study of hemodialysis patients, using administrative data, found a tendency toward increased risk of fatal bleeding associated with warfarin (HR, 1.77; 95% CI, 0.82 to 3.82) that did not reach statistical significance (20).

We recognize limitations to our work. Following the Dialysis Outcomes Quality Initiative 2000 guideline that defined adequate blood flow as 300 ml/min (23), the availability of catheter-locking doses of tPA (24) and the accumulating evidence of its efficacy (25), the use of tPA became more liberal in dialysis units, and some study patients who had not met study failure definitions were given tPA. It is unlikely that these protocol violations biased our results: (1) the number of protocol violations was low, 15 out of 81 events; (2) of these, seven occurred in situations where the severity of the catheter problem was well established; (3) the decisions were made by blinded clinical staff; (4) similar numbers of protocol violations occurred in each group; and (5) “need for tPA” is a reasonable outcome in itself. We considered a post-hoc analysis ignoring tPA administration that did not meet criteria but rejected this because of tPA restores catheter patency in about 90% of patients (25) and alters the clinical history of catheter survival. Second, discontinuation of study drug was high: 44 of 83 on warfarin and 34 of 82 on placebo, 47% overall. We believe this reflects the risks of anticoagulation and patients' and clinicians' perceptions of those risks. For example in a recent randomized trial in patients with nonvalvular atrial fibrillation, who are at much lower risk than the patients we studied, discontinuations occurred in 640 of 1962 patients assigned to warfarin (33%), albeit over a longer time period (26). Third, audit revealed that seven patients received incorrect monitoring: two participants on placebo were monitored as if on warfarin and five participants on warfarin were monitored as if on placebo. Of these five, four participants had INRs within the therapeutic or subtherapeutic range, one of whom experienced nonfatal major bleeding (INR 1.6 at time of bleed). The remaining patient was inappropriately prescribed an increase in warfarin dose when the true INR was supratherapeutic (because the sham INR was subtherapeutic), followed by nonfatal major bleeding with INR 10.0. Although warfarin monitors received training and support, this problem arose twice, illustrating the difficulties of conducting double-blind randomized trials of warfarin. This difficulty is acknowledged by the design of recent major randomized trials that have blinded other interventions but used open-label warfarin (27,28). Finally, our sample-size calculation was based on a large 40% reduction in events, which seemed plausible on the basis of available literature from other catheters. We cannot exclude a benefit of smaller magnitude, which might still be clinically important, or that our results are incorrect through the play of chance. However, all our results are internally congruent, with no curve separation in the first 200 days in either intention-to-treat or on-treatment analyses and no benefit in terms of secondary clinical outcomes or the surrogate outcomes of flow and urea-reduction ratio.