Introduction

More than one in ten patients with end stage renal disease (ESRD) also has comorbid atrial fibrillation (AF) [1]. The prevalence of AF is greater among older adults on dialysis, ranging from 13.2% among those aged 65 to 74 years to 22.5% among those 85 years or older [2]. Patients with comorbid ESRD and AF are at high risk of poor outcomes such as ischemic stroke, all-cause mortality, myocardial infarction and heart failure compared to patients with ESRD but without AF [3]. In the general population, older patients have a higher risk of thromboembolism, but they derive a greater net clinical benefit from warfarin in terms of protection of thromboembolic events [4].

While warfarin is recommended for prevention of stroke and thromboembolism in the general population with AF who have one or two risk factors for stroke [5–9], experts have raised concerns about the effectiveness of warfarin treatment in patients with ESRD and AF [10–13]. Recently published systematic reviews and observational studies suggest that warfarin is not beneficial in reducing stroke risk but is associated with increased bleeding risk in patients with AF undergoing dialysis [14–17], especially among older dialysis patients [18, 19]. However, there was some emerging evidence suggesting that warfarin use is associated with a reduction in mortality, even in the absence of a reduction in stroke incidence [20–24].

Given the unclear survival benefit, lack of association with stroke reduction, and increased risk of bleeding associated with warfarin use in patients with ESRD [14–17], clinicians are faced with uncertainty about this treatment decision in this population [25]. Little is known about the use of warfarin in older adults with ESRD, especially about which patients are most likely to initiate this therapy after AF diagnosis and how their utilization changes after initiation. In order to quantify warfarin use and understand the clinical determinants of warfarin treatment in this population, we evaluated the characteristics of warfarin initiation and utilization using a comprehensive national registry of older adults undergoing hemodialysis (HD) and peritoneal dialysis (PD) in the US.

Subjects and methods

Study population and data

We determined the study sample from all available patients in the US Renal Data System (USRDS). We began with a cohort of 9784 older adults aged 65 years or older with ESRD who had an incident AF diagnosis from July 1, 2007 to November 30, 2011. We identified AF by 1 inpatient or 2 outpatient diagnosis codes within 30 days of each other indicating AF (International Classification of Diseases, Ninth Revision [ICD-9] code 427.31) [26]. The study population was limited to patients who: (1) were without pre-existing AF diagnosis from January 1, 1999 to June 30, 2007; (2) had continuous Part A, B, and D coverage from 6 months before through 30 days after AF discharge; (3) had no valvular disease associated with AF in 6 months prior to AF diagnosis [22]; (4) were undergoing HD or PD at the time of AF diagnosis; and (5) had no history of warfarin prescription in 6 months prior to AF diagnosis. After excluding 84 patients with missing Medical Evidence Report (Centers for Medicare & Medicaid Services [CMS] form 2728), the final analytic sample included 5730 older adults with ESRD and incident AF (Appendix Fig. 1). This study was exempted from the full institutional review because the sample was comprised of de-identified secondary data.

Patient characteristics

We gleaned information regarding patients’ demographics and dialysis modality from the Medical Evidence Report (Centers for Medicare & Medicaid Services [CMS] form 2728). We ascertained medication history in the 6 months before AF diagnosis from the Medicare prescription claims data. We identified comorbidities from the Medical Evidence Report or previously published ICD-9 based algorithms [27–29] using all available Medicare inpatient claims data [30] (i.e. from January 1, 1999 to AF diagnosis).

Stroke and bleeding risk scores

We calculated the CHA2DS2-VASc [31] and HAS-BLED [32] scores for risk stratification. The CHA2DS2-VASc score (ranges 0 to 9) was calculated by assigning 1 point for the presence of congestive heart failure, hypertension, diabetes mellitus, vascular disease, age 65–74 years, sex category (i.e. female gender), and assigning 2 points to age ≥75 years, and history of stroke/transient ischemic attack/thromboembolism [31]. Since our study population only included older adults, all patients had a minimum CHA2DS2-VASc score of 1 for age.

The HAS-BLED score (ranges 0 to 9) was calculated by assigning 1 point for the presence of hypertension, abnormal renal function, abnormal liver function, stroke, bleeding history, elderly (>65 years), concomitant drug use, and alcohol abuse [32]. Since labile international normalized ratio (INR) information was not available from the USRDS database, this factor was not included in our calculation [33]. Additionally, our study population included older adults undergoing dialysis; therefore, all patients had a minimum score of 2 for age and abnormal kidney function, and a maximum score of 8 due to lack of INR information.

We modified both the CHA2DS2-VASc and HAS-BLED risk categories, whereby a score of 1 corresponded to low, 2–3 corresponded to intermediate, and 4–8 corresponded to high risk. We adjusted the score cut points, because no older adult dialysis patients had a CHA2DS2-VASc score of 0 due to age, or a HAS-BLED score of 0–1 due to age and abnormal kidney function.

Warfarin initiation

We identified claims for warfarin using Medicare Part D claims of filled prescriptions. We defined warfarin initiation as any warfarin prescription claims within 30 days of AF discharge (i.e. date of discharge for patients who had a one inpatient diagnosis of AF or date of first diagnosis for patients who had two or more diagnoses of AF), and non-initiation as the absence of a warfarin prescription claims within 30 days of AF discharge. We then calculated the proportion of dialysis patients with warfarin initiation among the cohort of older adults undergoing dialysis with newly diagnosed AF, and by CHA2DS2-VASc and HAS-BLED risk categories.

Warfarin utilization

We described medication utilization characteristics including number of warfarin prescriptions per user, total quantity dispensed per user, and total days of supply per user for those who initiated warfarin between AF discharge and the end of follow-up (the earliest day of death date, first kidney transplantation date, loss of Medicare coverage, or December 31, 2011).

We also assessed medication utilization measures such as discontinuation and switching. For these analyses, we excluded patients who were diagnosed with AF on or after October 1, 2011 to ensure patients analyzed had enough follow up to assess discontinuation or switching. We defined discontinuation as no additional refill for warfarin within 60 days after the end of days of supply of warfarin prescription. We defined switching as any dabigatran, rivaroxaban, or apixaban prescription between the initial warfarin prescription date and the end of follow-up for those who initiated warfarin, and as any warfarin prescription between 30 days after AF discharge and the end of follow-up for those who did not initiate warfarin.

Statistical analysis for association of patient characteristics with warfarin initiation

We described patient characteristics using median and interquartile range (IQR) for continuous variables, and counts and proportions for categorical variables. We compared the patient characteristics of warfarin initiators and non-warfarin initiators, and tested the difference using the two-sample t-test for continuous variables and the Pearson Chi square test for categorical variables. We calculated the proportion of warfarin initiators by CHA2DS2-VASc and HAS-BLED risk categories. We tested the independent associations of demographic, dialysis, and clinical factors with warfarin initiation by fitting multiple logistic regression models and reported the odds ratio (OR) and 95% confidence interval (CI). We selected the most parsimonious model by backward stepwise selection based on lowest the Akaike information criterion, and assessed the model fit by the Hosmer–Lemeshow goodness-of-fit test. We performed similar analyses among the older PD patients with incident AF. All statistical analyses were performed using Stata 14.0 (StataCorp, College Station, TX).

Sensitivity analysis

We conducted two sensitivity analyses using alternate definition of initiation and discontinuation. First, we defined warfarin initiation as any prescription within 60 days of AF discharge instead of 30 days. Second, we defined warfarin discontinuation using a gap of 30 days instead of 60 days for no refill after the end of days’ supply. We also conducted an additional sensitivity analysis to assess the determinants of warfarin initiation excluding patients identified by a single inpatient diagnosis of AF.

Results

Study population

We identified 5730 older adults undergoing HD or PD who were newly diagnosed with AF from July 1, 2007 to November 30, 2011 (Table 1). These patients had a median age of 74 years (IQR 69–80) at AF diagnosis and had a median time on dialysis of 3.9 years (IQR 2.3–5.9); 56.8% were female and 62.7% were white. A history of comorbid conditions was common: 21.6% had cerebrovascular disease, 71.5% had diabetes mellitus, 68.8% had congestive heart failure, and 98.6% had hypertension.

Table 1 Characteristics of older adults (≥ 65 years) undergoing dialysis with newly diagnosed atrial fibrillation by warfarin initiation
Full size table

Warfarin initiation patterns

Among these 5730 patients, 15.5% initiated warfarin within 30 days of AF discharge. Half of those who initiated warfarin did so in the first week after AF discharge (median 6 days, IQR 3–11) (Fig. 1). Warfarin initiators tended to be healthier than non-initiators, with lower median age at AF diagnosis (73 vs 74 years, P < 0.001), lower rates of liver disease (4.7 vs 8.3%, P < 0.001), myocardial infarction (10.8 vs 16.8%, P < 0.001), ischemic stroke/TIA/TE history (10.0 vs 13.8%, P = 0.002) and bleeding history (21.0% vs 31.1%, P < 0.001) (Table 1).

Fig. 1
figure 1

Time to warfarin initiation among older adults (≥65 years) undergoing dialysis within 30 days of atrial fibrillation discharge

Full size image

Warfarin initiation by CHA2DS2-VASc and HAS-BLED risk categories

The CHA2DS2-VASc and HAS-BLED scores were statistically different between warfarin initiators and non-initiators (Table 1). Among 4824 patients with high risk of stroke predicted by CHA2DS2-VASc score, 14.9% were warfarin initiators (Table 2). Among 2906 patients with high risk of bleeding predicted by HAS-BLED score, 12.7% were warfarin initiators.

Table 2 Warfarin initiation by CHA2DS2-VASc and HAS-BLED risk score categories among older adults (≥65 years) undergoing dialysis with newly diagnosed atrial fibrillation
Full size table

Independent association of patient characteristics with warfarin initiation

After adjustment for other patient characteristics, age (OR 0.745 per 10-year increase, 95% CI 0.66–0.83), but not sex (OR 1.00, 95% CI 0.85–1.17), race (OR 0.88, 95% CI 0.75–1.03) or dialysis modality (OR 1.05, 95% CI 0.72–1.54), was independently associated with warfarin initiation (Table 3). Patients with diabetes (OR 0.75, 95% CI 0.63–0.90), liver disease (OR 0.58, 95% CI 0.41–0.83), myocardial infarction (OR 0.64, 95% CI 0.50–0.80), stroke history (OR 0.76, 95% CI 0.58-1.00) and bleeding history (OR 0.63, 95% CI 0.50–0.80) were significantly less likely to initiate warfarin. Among PD patients, 18.3% initiated warfarin within 30 days of AF diagnosis. None of the patient characteristics was independently associated with warfarin initiation in the subset of 178 patients. (Appendix Table 1).

Table 3 Adjusted association of patient characteristics with warfarin initiation among older adults (≥65 years) undergoing dialysis, who initiated warfarin within 30 days of atrial fibrillation discharge
Full size table

Warfarin utilization patterns

Warfarin initiators had a median of 6 (IQR 2–14) warfarin prescriptions, covering a median of 231 (IQR 90–532) days of supply (Table 4). Among warfarin initiators, 46.8% discontinued warfarin after a median duration of 257 days (IQR 165–444) after the initial warfarin prescription (Table 4). Only 0.5% warfarin initiators switched to dabigatran, and 15.4% patients who did not used warfarin initially started warfarin use after a median duration of 149 days (IQR 66–358).

Table 4 Warfarin utilization among older adults (≥65 years) undergoing dialysis with newly diagnosed atrial fibrillation, who initiated warfarin within 30 days of atrial fibrillation discharge
Full size table

Sensitivity analysis

In the sensitivity analysis of warfarin initiation using 60 day for initiation period, we excluded 44 patients diagnosed with AF between November 1 and November 30, 2011. Among 5,686 older adults on dialysis, 17.3% initiated warfarin prescription within 60 days of AF discharge. Similar to the primary analysis, patient age and year of AF diagnosis were significantly associated with warfarin initiation (Appendix Table 2). In the sensitivity analysis of discontinuation using 30 days instead of 60 days as discontinuation window, 60.1% of warfarin initiators discontinued warfarin after a median duration of 194 days (IQR 125–335). In the sensitivity analysis excluding 408 patients with a single diagnosis of AF, 16.5% initiated warfarin within 30 days after AF discharge and 45.8% of them discontinued warfarin after a median of 166 days (IQR 259–446). Similar to the main analysis, patients age (OR 0.73, 95% CI 0.64–0.82), diabetes (OR 0.76, 95% CI 0.64–0.92), liver disease (OR 0.61, 95% CI 0.43–0.87), myocardial infarction (OR 0.60, 95% CI 0.48–0.76) and bleeding history (OR 0.61, 95% CI 0.48–0.77) were negatively associated with warfarin initiation, whereas CHA2DS2-VASc and HAS-BLED scores were not associated (Appendix Table 3).

Discussion

In this national study of older adults with ESRD and newly diagnosed AF, nearly one in six initiated warfarin prescription within 30 days of their discharge. Patient characteristics such as age and geographic region, as well as comorbid conditions including diabetes, liver disease, myocardial infarction, and bleeding history, were independently associated with warfarin initiation. Half of warfarin initiators discontinued use of warfarin after a median treatment length of 8.6 months.

Although several studies have examined the characteristics of patients using warfarin for AF in the US, these studies tended to focus on the general population [34] or older adults with AF not on dialysis [35, 36]. We found the overall proportion of warfarin initiation was 15.5%, which was the same as that (15%) among US adults with AF on HD reported by Shen et al. [22]. We found that bleeding history was inversely associated with warfarin initiation, but bleeding risk score and stroke risk score were not associated after adjustment for other patient characteristics. Genovesi et al. found that previous bleedings and hemorrhagic risk score was inversely related to warfarin use, but stroke risk score was not associated with warfarin use [37]. Such discrepancy might be due to differences in the study population and the definition of warfarin use: Genovesi et al. studied a mixed population of new and chronic warfarin among chronic HD patients with AF in Italy [37], whereas we studied new users among older adults undergoing HD or PD with incident AF in the US.

We found that several predictors of stroke risk including advanced age, history of diabetes and myocardial infarction were inversely associated with warfarin initiation, which differed from Genovesi et al.’s observation that these factors were not significantly associated with warfarin prescription [37]. The magnitude of these factors reported by Genovesi et al. were in the same direction as ours (i.e. OR < 1), but they may lacked statistical power to detect significant associations due to small sample size (N = 290). Given the high prevalence of frailty among HD patients [38, 39], and the high risk of adverse outcomes such as falls in this population [40, 41], physicians are reluctant to prescribe warfarin [25].

Among PD patients, we observed that warfarin was used among 18.3% older patients on PD, higher than that (15.4%) among older patients on HD; however, dialysis modality was not a significant predictor of warfarin initiation. In addition to warfarin initiation, we extended the previous findings by assessing utilization measures such as discontinuation and switching. We found that 46.8% warfarin initiators discontinued use, which was higher than that in the general population with AF (41.5%) reported by Nelson et al. [34].

Our study has several limitations. First, we determined warfarin treatment based on prescription claims, which indicates how medications were dispensed rather than how patients consumed the medications. Due to inherent limitations of claims analysis, we could not rule out the presence of unmeasured confounding such as factors affecting physicians’ prescribing behaviors, other indications for warfarin use, or over-the-counter medication use. Secondly, information such as the type of AF (persistent or paroxysmal) and INR values was lacking due to inherent limitations of administrative claims, so we were not able to study how these factors influence warfarin prescription. Although the presence of AF was not directly documented, we tried to minimize the misclassification bias by using a previously validated algorithm to identify AF [22, 26]. Finally, we used a dummy variable approach instead of a multilevel model to examine factors associated with warfarin initiation. We did not assess whether such relationships vary across group-level factors because we were mainly interested in individual-level factors.

Our study has strengths in that we studied the patterns and determinants of warfarin initiation among a national cohort of older adults with AF undergoing dialysis. To our knowledge, our study is the first to describe the detailed utilization characteristics of warfarin initiation after AF diagnosis in this vulnerable population. Identifying the initiation and utilization patterns of warfarin among older adults with AF undergoing dialysis is a first step in understanding the real world practice of stroke prophylaxis in this under-studied population, which could help inform future studies in evaluating the benefits and risks of warfarin treatment.

Our findings suggest that warfarin was initiated among 15.5% older adults undergoing dialysis within 30 days of AF discharge, even among 12.7% of those with high bleeding risk. Patients’ age, history of diabetes, liver disease, myocardial infarction, or bleeding were significantly associated with warfarin initiation. Warfarin was discontinued among 46.8% of those who initiated warfarin after 8.6 months, reflecting the challenges in warfarin management requiring careful monitoring and dosage adjustment. In light of the increasing number of older dialysis patients with AF and the unclear clinical benefit of warfarin treatment, more research is needed to determine the complex interplay of factors that affect AF epidemiology, warfarin use, and outcomes in this vulnerable population.