Effect of Warfarin Treatment on Survival of Patients With Pulmonary Arterial Hypertension (PAH) in the Registry to Evaluate Early and Long-Term PAH Disease Management (REVEAL)

Supplemental Digital Content is available in the text.

outcome in patients with IPAH; 6 of these studies reported positive outcomes. [9][10][11][12][13][14][15][16] Most of these observational studies were published more than a decade ago, before the introduction of many current PAH therapies. However, despite the limitations of the available evidence, current treatment algorithms recommend warfarin in patients with IPAH. 17 Data regarding anticoagulation in patients with associated forms of PAH such as SSc-PAH are even scarcer; however, such a strategy has been extrapolated from IPAH studies and current treatment algorithms recommend warfarin in these patients, based mostly on consensus opinion. 17 Expert opinions on the benefit of warfarin in SSc-PAH and IPAH patients vary widely in regard to the effectiveness of treatment. 18 Recently, the retrospective analysis from the European database Comparative, Prospective Registry of Newly Initiated Therapies for Pulmonary Hypertension (COMPERA) examined survival in patients with IPAH and SSc-PAH who received anticoagulation in comparison with patients who did not receive anticoagulation. 15 In this analysis, survival improved in IPAH patients who received anticoagulation; however, anticoagulation was not associated with a survival benefit in patients with SSc-PAH. 15 These observations are of particular concern because chronic anticoagulation can be associated with major fatal bleeding. 19 Specifically, SSc-PAH patients with arteriovenous malformations of the gastrointestinal tract are at increased risk of serious bleeding. 20 Last, all but 1 of the aforementioned studies were conducted before the availability of PAH-specific therapies, 7 some of which have an antiplatelet effect and may alter the thrombotic profile of the pulmonary blood and vasculature. Therefore, in an era when multiple effective PAH therapies are available, the role of anticoagulation in the treatment of PAH is even more uncertain.
In an effort to reconcile the variable results and the limitations of previous studies, we used the large, multicenter, observational, US-based, longitudinal registry of patients with group I PAH, the Registry to Evaluate Early and Long-term PAH Disease Management (REVEAL Registry), to assess the association between chronic anticoagulation use with warfarin and outcomes in contemporary patients with IPAH and SSc-PAH.

REVEAL Registry
The design and baseline characteristics of patients enrolled in the REVEAL Registry have been described previously. 21 In brief, the REVEAL Registry is a multicenter (55 sites, university-affiliated and community hospitals), observational, US-based study designed to provide information about demographics, disease course, and management of 3515 consecutively enrolled patients with newly or previously diagnosed group I PAH. The study was conducted in accordance with the amended Declaration of Helsinki and the protocol was reviewed by the institutional review board of each participating center with written informed consent obtained from all patients. 21 Patients were followed for at least 5 years from the time of enrollment. Diagnosis was confirmed by right heart catheterization within 3 months before enrollment for newly diagnosed and >3 months before enrollment for previously diagnosed patients. PAH was defined as a mean pulmonary artery pressure >25 mm Hg at rest or >30 mm Hg with exercise, pulmonary capillary wedge pressure or left ventricular end-diastolic pressure ≤18 mm Hg, and pulmonary vascular resistance ≥240 dyn·sec·cm -5 . The IPAH group included only patients with IPAH and did not include patients with heritable or anorexigen-induced PAH. SSc-PAH included patients with both limited and diffuse SSc-PAH.
Data were collected at the time of enrollment, followed by quarterly updates including information from any clinic visits or other patient contact in the previous 90 days. Exact dates of PAH-specific medications were collected throughout the study, whereas concomitant medications such as warfarin were identified at quarterly updates.

Patients Who Initiated Warfarin On-Study and Patients Never on Warfarin
The patient cohort for this analysis was enrolled from 2006 to 2009. We identified 4 groups based on etiologic cohort and warfarin initiation. A nested design was used to match patients who initiated warfarin after REVEAL Registry enrollment to patients never on warfarin within each etiologic cohort (IPAH and SSc-PAH). These new warfarin users and patients never on warfarin were matched by a 1:1 ratio with the same enrollment center and exact etiology. Patients were also matched by diagnosis status (newly or previously diagnosed) at study enrollment and were followed on a quarterly basis. To adjust for warfarin discontinuation, a separate analogous time-varying covariate Cox proportional hazards model including all warfarin data within each etiologic cohort was performed as a sensitivity analysis. This model accounts for warfarin starts and stops at each quarter of data collection and includes all warfarin data until the last available quarter. To avoid immortal time bias, 22 the time at risk was defined as the initiation of warfarin or at the matching quarterly update so that all survival follow-up was prospective (online-only Data Supplement Figure I). Thus, patients were excluded if they were already on warfarin at enrollment in both the matching-designed analysis and the sensitivity analysis.

Statistical Analysis
All statistical comparisons were made between patients receiving and not receiving warfarin within etiology. Categorical data were presented as percentages and were analyzed using the χ 2 or Fisher exact test where appropriate. Continuous data are summarized as mean±standard deviation or percentiles and were analyzed using the Student t test or Wilcoxon 2-sample rank sum test. Kaplan-Meier curves and estimates of survival were presented from the initiation of warfarin or the matching quarter to 36 months. Matching does not address potential confounding associated with differences in PAH severity. Therefore, univariable comparisons based on Kaplan-Meier curves and the log-rank test were supplemented with multivariable Cox proportional hazards models adjusting for differences in the risk profile. Risk adjustment to account for selection bias 23 was conducted based on the REVEAL Registry prognostic equation 24 at the quarterly update corresponding to warfarin initiation and match.
In addition to the main analysis, a multivariable Cox proportional hazards model including all warfarin data within each etiologic cohort was performed as a sensitivity analysis within both the IPAH and SSc-PAH subgroups. Warfarin use was modeled as timeinterval time-varying covariates. In 2 different models, (1) past use (any on-study warfarin use) or (2) current use (any on-study warfarin use within the previous year) was compared with all patients who had no on-study warfarin use or patients who had no on-study warfarin use within the past year, respectively, as the main predictors in each model. PAH severity at baseline (study enrollment) was also accounted for after adjusting the models for diagnosis status, PAH medications, and risk profile at baseline. Survival was based on allcause mortality, with patients censored only for loss to follow-up or at the close of study in December 2012. A P value of <0.05 was considered statistically significant.

Patient Characteristics
A total of 2197 IPAH and SSc-PAH patients were enrolled in the REVEAL Registry: 1645 IPAH patients and 552 SSc-PAH patients. In this registry, 922 IPAH patients and 194 SSc-PAH patients started warfarin before or at enrollment. Of these, 163 IPAH patients and 55 SSc-PAH patients started warfarin after enrollment (postbaseline) and were included in the sensitivity analysis ( Figure 1). Based on the matching criteria, we identified 144 IPAH and 43 SSc-PAH patients who initiated warfarin after enrollment ( Figure 1) and were matched to 144 IPAH and 43 SSc-PAH patients, respectively, who were never on warfarin. These patients were included in the main analysis. Most patients were previously diagnosed with PAH at the time of enrollment into the REVEAL Registry. Table 1 describes the demographics and clinical characteristics of the 4 groups of PAH patients. With the exception of New York Heart Association (NYHA)/World Health Organization (WHO) functional class III/IV status and brain natriuretic peptide levels in the warfarin groups, characteristics were similar between the 2 pairs of groups. Notably, the REVEAL Risk Score was higher in the SSc-PAH warfarin cohort. Overall, the REVEAL Registry cohort was 72.4% white and 77.6% female. In the current analysis, no sex-or ethnicity-specific effects were observed. However, because of the small representation of some subgroups in this study, it is difficult to interpret whether these effects are truly absent.
Comorbid conditions were similar between the 2 sets of groups, including a low prevalence of thromboembolism ( Table 2). PAH-specific medications are listed in Table 3. More patients in the warfarin groups received parenteral prostacyclins; (46.4% versus 15.1% for IPAH patients who started warfarin and never on warfarin, respectively, and 34.1% versus 14.6% SSc-PAH patients who started warfarin and never on warfarin, respectively). Moreover, patients in the warfarin groups received combination PAH-specific therapies more frequently (50.0% versus 40.2% for IPAH patients who started warfarin and never on warfarin, respectively, and 55.9% versus 25.6% SSc-PAH patients who started warfarin and never on warfarin, respectively). The mean international normalized ratio (INR) was 1.9 for IPAH and 2.0 for SSc-PAH patients ( Table 4). In both IPAH and SSc-PAH warfarin-treated groups, the mean time on warfarin was 1 year. Time on warfarin ranged from 3 months to 42 months. Nearly two-thirds of patients discontinued warfarin before the last assessment (65.3% and 62.8% for IPAH and SSc-PAH, respectively) ( Table 4).
Matching analysis design results in the loss of information for unmatched patients and does not account for the time of warfarin discontinuation. Because there was a high warfarin discontinuation rate in the REVEAL Registry, a time-varying covariate model of the longitudinal effect of warfarin use was used to overcome this limitation. This model accounts for warfarin starts and stops and was used as a sensitivity analysis (online-only Data Supplement Table I). After adjusting for diagnosis status, PAH medication, and REVEAL Risk Score in the sensitivity analysis, the model affirmed the results in the main analysis in the IPAH cohort: there was no survival advantage in IPAH patients who were current (HR, 0.96; 95% CI 0.66-1.39, P=0.82) or past warfarin users (HR, 0.84; 95% CI, 0.59-1.20, P=0.34) in comparison with IPAH patients who had no on-study warfarin use within the past year or who had no on-study warfarin use, respectively (online-only Data Supplement Table I  than patients who had no on-study warfarin use within the past year or who had no on-study warfarin use, respectively, even after adjusting for diagnosis status, PAH medication, and REVEAL Risk Score (online-only Data Supplement Table I).
Causes of death are presented in Table 5. The majority of deaths were attributed to worsening of PAH.

Discussion
Our study shows no significant survival advantage in IPAH patients treated with warfarin in comparison with matched controls never treated with warfarin. This is further supported by the sensitivity analysis, which shows no survival advantage after adjustment for diagnosis status, PAH medication, and REVEAL Risk Score. Although SSc-PAH patients treated with warfarin had decreased survival in comparison with matched controls in the unadjusted analysis, SSc-PAH patients receiving warfarin were inherently sicker; no survival difference associated with warfarin therapy was present with adjustment for disease severity. After accounting for the high warfarin discontinuation rate in our study, the sensitivity analysis confirmed the observation of increased SSc-PAH patient mortality even after adjusting for disease severity at enrollment. Thus, analyses of the REVEAL Registry data provided no conclusive evidence to support the use of warfarin to reduce mortality in PAH patients. These data are in concordance with a Bayesian analysis of observational data from a longitudinal cohort of patients with SSc-PAH or IPAH, which demonstrated a low probability of a survival benefit with warfarin. 25 In this Bayesian analysis, patients with SSc-PAH and IPAH who received warfarin were also determined to be sicker than patients not receiving warfarin, similar to COMPERA 15 and to our REVEAL Registry population. 25 In other disease states such as atrial fibrillation, warfarin has been shown to be effective in reducing the risk of morbid events (including stroke and systemic thromboembolism), 26 with a recommended INR generally between 2.0 and 3.0. 26,27 Importantly, the presumed goal of anticoagulant therapy in PAH is prevention of mortality, with a target INR of 1.5 to 2.5. The INR values for the patient population in the REVEAL Registry were within the American Heart Association recommended range for PAH 28 and were below the levels of anticoagulation recommended for other conditions. 26,27 Therefore, the risk of bleeding was likely minimized, because bleeding risk for patients treated with warfarin increases with increasing INR, particularly above a level of 4. 29 Indeed, we did not detect an increased risk of death from bleeding, although morbid events were not specifically collected to detect adverse events related to warfarin use. Surprisingly, adherence to warfarin therapy in this cohort was poor in both cohorts. During the 3-year study, mean warfarin exposure was 1 year with a high discontinuation rate for both IPAH and SSc-PAH patients, suggesting a particularly poor tolerance to anticoagulation in PAH patients enrolled in the REVEAL Registry, a registry that represents current practice for treatment of PAH in the United States.
The current analysis is specifically designed to include new warfarin starts and excludes previously anticoagulated patients. This design is deliberate; to address the issue of immortal time bias, 22 we focus on new warfarin starts and, hence, evaluate the possibility of anticoagulation-attributable survival. In contrast, the COMPERA analysis included  P value calculation uses χ 2 test for categorical data or Fisher exact test for categorical data with expected cell counts ≤5. All events were measured at quarterly intervals corresponding to warfarin start/match. IPAH indicates idiopathic pulmonary arterial hypertension; and SSc-PAH, pulmonary arterial hypertension associated with systemic sclerosis. Baseline is defined as the start date for Started-Warfarin-on-Study patients and the assessment date used to match the warfarin patients for Never-on-Warfarin patients. ERA indicates endothelin receptor antagonists; IPAH, idiopathic pulmonary arterial hypertension; IV, intravenous; PAH, pulmonary arterial hypertension; PDE5i, phosphodiesterase type 5 inhibitor; PGI 2, prostacyclin; SC, subcutaneous; and SSc-PAH, pulmonary arterial hypertension associated with systemic sclerosis. December 22/29, 2015 patients who were anticoagulated (mostly with warfarin) at the time of enrollment, and thus did not account for immortal time bias. Importantly, COMPERA showed a positive correlation between anticoagulation and survival in IPAH patients only. 15 Of note, both the REVEAL Registry and COMPERA revealed no positive association between anticoagulation and survival in SSc-PAH patients. This major difference in study design and immortal time bias might explain the discrepant results between the REVEAL Registry and the COMPERAderived analyses in IPAH patients. The extent of this immortal time bias depends on the length of time occurring between diagnosis and initiation of warfarin (ie, immortal time) and whether this immortal time was misclassified or excluded. Immortal time refers to the period of time patients had to survive to initiate warfarin and be included in the warfarin cohort rather than the nonwarfarin cohort. For example, survival estimates may be biased in favor of the warfarin group if a lengthy period of prewarfarin survival is instead attributed to longer survival for warfarin patients (online-only Data Supplement Figure I). On the other hand, if patients typically initiate warfarin quickly after diagnosis, immortal time bias would be minimal and unlikely to fully explain the differences between the REVEAL Registry and COMPERA results. 15 As such, immortal time bias could not definitively be excluded in COMPERA.
Other potential explanations for the discrepancy in results may include differences in patient characteristics between the 2 registries. The COMPERA cohort included more males than females and the patients were older. Moreover, the older age of patients in COMPERA is an important difference; patients 65 years or older had poorer survival despite having less severe (presumably advantageous) hemodynamic impairment. 30 In addition, the use of prostacyclins in PAH care differs greatly between Germany, France, and the United States. In France and the United States, parenteral prostacyclins are initiated routinely-and early-as long-term therapy for patients presenting with NYHA/WHO functional class III/IV PAH. In contrast, in Germany, parental prostacyclin therapy is reserved mainly as a rescue therapy for hospitalized patients who are then discharged on oral therapies. Similarly, differences exist in PAH anticoagulation practices between the United States and Europe. In Europe, the target INR for PAH anticoagulation is 2.0 to 3.0 6 and temporary interruptions are usually bridged with heparin or analogues. In contrast, the US PAH centers target an INR of 1.5 to 2.5 for PAH patients 28 with no bridging for temporary interruptions. Of note, INR levels were not reported in COMPERA, but are reported in the REVEAL Registry.
Limitations of our study include the inherent limitations of observational registry-derived data and the retrospective nature of the report. 31 Additionally, a high discontinuation rate of warfarin was observed, consistent with practice patterns in US PAH centers and suggestive of poor tolerance of anticoagulation in PAH patients. We attempted to account for this high rate of discontinuation by using a time-varying covariate model as a sensitivity analysis. Although both the matched and sensitivity analyses were adjusted for disease severity as captured by the REVEAL Risk Score, the anticoagulation arm was characterized by parameters consistent with disease severity. For example, a greater proportion of patients on anticoagulation received PAH parenteral and combination therapy, and had higher brain natriuretic peptide and NYHA/ WHO functional class, which may well reflect that sicker PAH patients were more likely to be anticoagulated at baseline. The REVEAL Risk Score encompasses PAH risk factors including demographic data, comorbidities, NYHA/WHO group I subgroup, NYHA/WHO functional class, vital signs, 6-minute walk distance, and brain natriuretic peptide along with echocardiographic, pulmonary function testing, and hemodynamic risk factors, in a risk score prognostic model. 32 Although it is evident that the REVEAL Risk Score captures most, if not all, known prognostic risk factors in PAH, the possibility of an unknown possible factor cannot be excluded. Furthermore, the REVEAL Risk Score has been validated in both incident and prevalent patients and has been cross-validated using the French registry database. 33 Although a predictive score combining right ventricular function and right atrial function indices for predicting outcome has been validated in PAH patients, this score was not shown to be statistically different from the REVEAL Risk Score. 34 Because no other general risk scores have been evaluated or validated in PAH, the use of the REVEAL Risk Score is currently the best tool that accounts for differences in disease severity.
Our study is novel in that it consists of a rigorous analysis of a large database with inclusion of new starts on warfarin, exclusion of immortal time bias, as well as adjustment for disease severity as measured by the REVEAL PAH-specific Risk Scores. This study also reports on the length of anticoagulant therapy and on the level of anticoagulation measured by INR during warfarin use, all of which contribute to the strength and accuracy of the results. In addition, statistical modeling was done to address the limitations as well as to validate the matching analysis, by performing a time-varying covariate model as an analogous sensitivity analysis. In conclusion, in IPAH patients in the REVEAL Registry, there is insufficient evidence to conclude that the initiation of warfarin is associated with improved survival. Furthermore, similar results were observed in SSc-PAH patients. Our findings differ from COMPERA in regard to IPAH patients but are concordant with COMPERA and other studies with regard to the use of anticoagulation in SSc-PAH patients. Further studies are required to resolve this ongoing controversy.