Intracranial Hemorrhage Among Patients With Atrial

1304

Most individuals with atrial fibrillation (AF) are at suf-ficient risk of thromboembolic ischemic stroke to war-rant prophylactic oral anticoagulation therapy.1 The most feared complication of anticoagulation is intracranial hemor-rhage (ICH) because it is responsible for most of the death

and disability attributable to anticoagulant-associated bleed-ing.2 A burning clinical question is how to predict reliably which patients with AF are at high (and low) risk of ICH if anticoagulated, and which factors reliably discriminate risk of ICH (which may be caused by anticoagulation) from risk

Background and PurposeIntracranial hemorrhage (ICH) is a life-threatening complication of anticoagulation.MethodsWe investigated the rate, outcomes, and predictors of ICH in 14 264 patients with atrial fibrillation from

Rivaroxaban Once Daily, Oral, Direct Factor Xa Inhibition Compared With Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation (ROCKET AF). Cox proportional hazards modeling was used.

ResultsDuring 1.94 years (median) of follow-up, 172 patients (1.2%) experienced 175 ICH events at a rate of 0.67% per year. The significant, independent predictors of ICH were race (Asian: hazard ratio, 2.02; 95% CI, 1.392.94; black: hazard ratio, 3.25; 95% CI, 1.437.41), age (1.35; 1.131.63 per 10-year increase), reduced serum albumin (1.39; 1.121.73 per 0.5 g/dL decrease), reduced platelet count below 210109/L (1.08; 1.021.13 per 10109/L decrease), previous stroke or transient ischemic attack (1.42; 1.021.96), and increased diastolic blood pressure (1.17; 1.011.36 per 10 mm Hg increase). Predictors of a reduced risk of ICH were randomization to rivaroxaban (0.60; 0.440.82) and history of congestive heart failure (0.65; 0.470.89). The ability of the model to discriminate individuals with and without ICH was good (C-index, 0.69; 95% CI, 0.640.73).

ConclusionsAmong patients with atrial fibrillation treated with anticoagulation, the risk of ICH was higher among Asians, blacks, the elderly, and in those with previous stroke or transient ischemic attack, increased diastolic blood pressure, and reduced platelet count or serum albumin at baseline. The risk of ICH was significantly lower in patients with heart failure and in those who were randomized to rivaroxaban instead of warfarin. The external validity of these findings requires testing in other atrial fibrillation populations. (Stroke. 2014;45:1304-1312.)

Key Words: anticoagulation atrial fibrillation intracranial hemorrhage risk prediction

Intracranial Hemorrhage Among Patients With Atrial Fibrillation Anticoagulated With Warfarin or Rivaroxaban

The Rivaroxaban Once Daily, Oral, Direct Factor Xa Inhibition Compared With Vitamin K Antagonism for Prevention of Stroke and Embolism Trial

in Atrial FibrillationGraeme J. Hankey, MD; Susanna R. Stevens, MS; Jonathan P. Piccini, MD;

Yuliya Lokhnygina, PhD; Kenneth W. Mahaffey, MD; Jonathan L. Halperin, MD; Manesh R. Patel, MD; Gnter Breithardt, MD; Daniel E. Singer, MD; Richard C. Becker, MD;

Scott D. Berkowitz, MD; John F. Paolini, MD, PhD; Christopher C. Nessel, MD; Werner Hacke, MD, PhD; Keith A.A. Fox, MB, ChB; Robert M. Califf, MD;

on behalf of the ROCKET AF Steering Committee and Investigators

Received December 12, 2013; final revision received March 10, 2014; accepted March 11, 2014.From the School of Medicine and Pharmacology, The University of Western Australia, Perth, Australia (G.J.H.); Department of Neurology, Sir Charles Gairdner

Hospital, Perth, Australia (G.J.H.); Duke Clinical Research Institute (S.R.S., J.P.P., Y.L., M.R.P.) and Duke Translational Medicine Institute (R.M.C.), Duke University Medical Center, Durham, NC; Department of Medicine, Stanford University, CA (K.W.M.); Cardiovascular Institute, Mount Sinai Medical Center, New York (J.L.H.); Department of Cardiovascular Medicine, Hospital of the University of Mnster, Mnster, Germany (G.B.); Massachusetts General Hospital and Harvard Medical School, Boston (D.E.S.); University of Cincinnati College of Medicine, OH (R.C.B.); Department of Global Clinical Development, Bayer HealthCare Pharmaceuticals, Whippany, NJ (S.D.B.); Cerenis Therapeutics, Labege, France (J.F.P.); Janssen Research and Development, Raritan, NJ (C.C.N.); Ruprecht- Karls-University, Heidelberg, Germany (W.H.); and University of Edinburgh and Royal Infirmary of Edinburgh, Edinburgh, United Kingdom (K.A.A.F.).

Guest Editor for this article was Kazunori Toyoda, MD.Presented in part at the International Stroke Conference of the American Heart Association, New Orleans, LA, January 31February 3, 2012.The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.

113.004506/-/DC1.Correspondence to Graeme J. Hankey, MD, School of Medicine and Pharmacology, The University of Western Australia, Room 222, Harry Perkins

Institute of Medical Research, QQ Block, QEII Medical Centre, 6 Verdun St, Nedlands, Perth 6009, Australia. E-mail [email protected] 2014 American Heart Association, Inc.Stroke is available at http://stroke.ahajournals.org DOI: 10.1161/STROKEAHA.113.004506

Hankey et al Predictors of ICH in ROCKET AF 1305

of thromboembolic ischemic stroke (which may be prevented by anticoagulation).

All-cause ICH in the general population has been linked to Asian and black ethnicity; the presence of an apolipoprotein E2 or E4 allele; decreased low-density lipoprotein cholesterol and triglycerides; and increasing age, blood pressure, and alcohol consumption.36 All-cause ICH in the anticoagulated population occurs at a rate of 0.2% to 1.0% per year79 and has likewise been associated not only with increasing age and blood pressure but also with previous ischemic stroke, chronic kidney disease, the early period of anticoagulation use, higher intensity (ie, poorly controlled) anticoagulation, antiplatelet use in addition to anticoagulation, and brain imaging evidence of cerebral leukoaraiosis and microbleeds.926 Difficulty arises in clinical practice, because the risk factors for ICH with anti-coagulation are also risk factors for ischemic stroke that could be prevented with anticoagulation. A recent study suggests that increasing age and previous stroke are more often associ-ated with ischemic stroke than with ICH, whereas a history of hypertension, diabetes mellitus, renal impairment, and alcohol intake are equally associated.26 In the Rivaroxaban Once Daily, Oral, Direct Factor Xa Inhibition Compared with Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation (ROCKET AF) cohort of 14 264 patients with nonvalvular AF and creatinine clearance 30 mL/min who were randomized to rivaroxaban or dose-adjusted warfarin, the factors at randomization that were independently associated with the occurrence of all stroke (ischemic and hemorrhagic) or noncentral nervous system embolism among 575 patients (4.0%) over a median follow-up of 1.94 years were reduced creatinine clearance, previous stroke or transient ischemic attack (TIA), elevated diastolic blood pressure and heart rate, as well as vascular disease of the heart and limbs.27 However, the predictors of hemorrhagic and ischemic stroke subtypes in the ROCKET AF cohort have not been reported.

We aimed to determine the rate, outcomes, and indepen-dent, significant predictors of ICH in the large, international cohort of 14 264 patients with AF who were enrolled in the ROCKET AF trial and followed-up prospectively for the occurrence of ICH.28,29 We also aimed to determine whether there are any predictors of ICH that are not also predictors of ischemic stroke and thereby may help identify patients for whom anticoagulants may be more hazardous than helpful.

MethodsThe design, methods, and primary results of the ROCKET AF trial have been described.28,29 Briefly, this was a multinational, random-ized, double-blind, double-dummy clinical trial comparing fixed-dose rivaroxaban (20 mg daily; 15 mg daily in patients with creatinine clearance, 3049 mL/min) with dose-adjusted warfarin (target inter-national normalized ratio, 2.5; range, 2.03.0) in participants with nonvalvular AF to prevent all stroke (ischemic or hemorrhagic) or systemic embolism.28,29 The trial protocol was approved by appropri-ate national regulatory authorities and ethics committees at the partici-pating centers, and all participants provided written informed consent.

ParticipantsEligible participants had electrocardiographically documented AF and increased risk of stroke as determined by a history of stroke, TIA, or systemic embolism or 2 of the following risk factors: heart

failure or left ventricular ejection fraction 35%, hypertension, age 75 years, or diabetes mellitus. Participants were excluded if they had any condition associated with increased bleeding risk.29

Follow-UpPatients were evaluated prospectively at 1, 2, and 4 weeks and month-ly thereafter for the duration of the study for study drug management and surveillance for primary end point events. A standardized ques-tionnaire and examination were used to screen for stroke symptoms and potential clinical events during follow-up.

Outcomes and Their DefinitionsThe primary outcome for this analysis was ICH, which was ascer-tained by local investigators and reported for central adjudication by an independent clinical events committee. The committee adjudicat-ed all suspected ICHs, masked to baseline characteristics and treat-ment allocation of all participants, and applied the protocol definition and classification of ICH.

ICH was defined as any primary bleed into the cranial cavity that was clinically overt (ie, caused symptoms or signs) and was con-firmed by brain imaging (computed tomography or MRI brain scan) or autopsy. If the hemorrhage was intraparenchymal and caused focal neurological symptoms or signs, it was classified as a hemorrhagic stroke. If the intraparenchymal hemorrhage was secondary hemor-rhagic transformation of a focal brain infarct (ie, primary ischemic stroke), it was not considered an ICH.

ICH was categorized as a major bleed and classified according to the site of hemorrhage as any bleed into the brain parenchyma or ventricular system (intracerebral hemorrhage), subarachnoid space (subarachnoid hemorrhage), subdural space (subdural hemorrhage), or extradural space (extradural hemorrhage) in the skull.

ICH was classified as spontaneous or not spontaneous. Intraparenchymal (intracerebral) hemorrhage was classified as trau-matic or nontraumatic, but extracerebral (subdural, subarachnoid, and extradural) hemorrhage was not classified as traumatic or nontrau-matic. Deaths after ICH were adjudicated as secondary to trauma if trauma clearly precipitated the ICH and death (eg, assault and motor vehicle accident).

Outcome after ICH because of hemorrhagic stroke (ie, intraparen-chymal or subarachnoid) was classified according to the modified Rankin Scale score (06) and residence (home, rehabilitation center, and long-term care) at 3 months after ICH (and 1 month after ICH if the ICH occurred at the end of the study). The modified Rankin Scale score was not measured if the ICH did not cause a clinical stroke syndrome of sudden onset of focal neurological dysfunction (eg, sub-dural hematoma and extradural hematoma).

Statistical AnalysisAll patients were included in the analysis regardless of study drug exposure (ie, intention-to-treat population). Data were analyzed using SAS version 9.2 (SAS Institute; Cary, NC).

Baseline characteristics were presented separately for participants with ICH, ischemic stroke, stroke of unknown pathological type, and no ICH or stroke. Baseline characteristics were summarized as num-bers (percentages) for categorical variables and medians (25th, 75th percentiles) for continuous variables.

Survival free of ICH was calculated, and survival curves were gen-erated by means of the KaplanMeier product limit technique.

Multivariable Cox proportional hazards models were developed using stepwise selection of predefined candidate variables that have previously been shown to predict ICH in other studies.926 Entry and exit criteria of the stepwise selection was

1306 Stroke May 2014

Table 1. Baseline Characteristics of the 14 264 Patients Enrolled in Rivaroxaban Once Daily, Oral, Direct Factor Xa Inhibition Compared With Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation (ROCKET AF)29

Characteristic ICH (n=172)Ischemic Stroke, No ICH (n=382)

Unknown Stroke, No ICH (n=32)

No ICH or Stroke (n=13 678)

Age, y 75 (66.5, 80) 74 (68, 79) 71 (66, 76) 73 (65, 78)

Men 105 (61.0) 199 (52.1) 17 (53.1) 8283 (60.6)

Race

White 122 (70.9) 313 (81.9) 29 (90.6) 11 415 (83.5)

Asian 39 (22.7) 53 (13.9) 3 (9.4) 1691 (12.4)

Black 6 (3.5) 6 (1.6) 0 (0) 168 (1.2)

Other 5 (2.9) 10 (2.6) 0 (0) 404 (3.0)

Hispanic or Latino 31 (18.0) 61 (16.0) 2 (6.3) 2240 (16.4)

Region

Asia/Pacific Islands 46 (26.7) 60 (15.7) 3 (9.4) 2000 (14.6)

Eastern Europe 33 (19.2) 153 (40.1) 23 (71.9) 5291 (38.7)

Latin America 29 (16.9) 51 (13.4) 1 (3.1) 1797 (13.1)

North America 41 (23.8) 52 (13.6) 4 (12.5) 2584 (18.9)

Western Europe 23 (13.4) 66 (17.3) 1 (3.1) 2006 (14.7)

Randomized to rivaroxaban 64 (37.2) 194 (50.8) 16 (50.0) 6857 (50.1)

Baseline CHADS2 score 3 (3, 4) 4 (3, 4) 4 (3, 4) 3 (3, 4)

CHADS2 score >3 (ie, 46), median 73 (42.4) 220 (57.6) 20 (62.5) 5873 (42.9)

Baseline BMI, kg/m2 26.9 (24.5, 30.6) 27.5 (24.8, 31.0) 26.6 (24.3, 31.6) 28.2 (25.1, 32.0)

BMI >28.17 kg/m2, median 69 (40.1) 171 (44.9) 12 (37.5) 6876 (50.3)

Baseline heart rate, bpm 76.5 (67.5, 85) 76 (68, 86) 80.5 (72, 90) 76 (67, 86)

Baseline SBP, mm Hg 131 (121, 144) 132 (124, 141) 135 (126, 145) 130 (120, 140)

Baseline DBP, mm Hg 80 (70, 89) 80 (71, 86) 82.5 (80, 88.5) 80 (70, 85)

DBP >80 mm Hg, median 67 (39.0) 152 (39.8) 18 (56.3) 4630 (33.9)

History of AF

Persistent 145 (84.3) 314 (82.2) 29 (90.6) 11 060 (80.9)

Paroxysmal 23 (13.4) 61 (16.0) 3 (9.4) 2427 (17.7)

New 4 (2.3) 7 (1.8) 0 (0) 191 (1.4)

Left bundle branch block 6 (3.5) 26 (7.0) 3 (9.4) 942 (6.9)

History of stroke or TIA 100 (58.1) 251 (65.7) 19 (59.4) 7098 (51.9)

History of hypertension 153 (89.0) 349 (91.4) 31 (96.9) 12 377 (90.5)

History of CHF 81 (47.1) 230 (60.2) 26 (81.3) 8571 (62.7)

History of diabetes mellitus 68 (39.5) 137 (35.9) 12 (37.5) 5478 (40.0)

History of COPD 22 (12.8) 43 (11.3) 4 (12.5) 1428 (10.4)

History of GI bleed 11 (6.4) 13 (3.4) 1 (3.1) 474 (3.5)

History of liver disease 5 (2.9) 19 (5.0) 1 (3.1) 722 (5.3)

Vascular disease variable for CHA

2DS

2-VASc

37 (21.5) 100 (26.2) 9 (28.1) 3181 (23.3)

History of sleep apnea 11 (6.4) 17 (4.5) 0 (0) 617 (4.5)

History of cigarette smoking 64 (37.4) 115 (30.2) 7 (21.9) 4606 (33.7)

Alcohol consumption in last 12 mo

None 114 (66.7) 259 (67.8) 26 (81.3) 8812 (64.4)

Light 49 (28.7) 102 (26.7) 5 (15.6) 4180 (30.6)

Moderate 7 (4.1) 17 (4.5) 1 (3.1) 587 (4.3)

Heavy 1 (0.6) 4 (1.0) 0 (0) 98 (0.7)

Aspirin 61 (35.5) 110 (28.8) 11 (34.4) 3920 (28.7)

Thienopyridine 9 (5.2) 6 (1.6) 0 (0) 224 (1.6)

(Continued )


Associations are reported as hazards ratios (HRs) with 95% confi-dence intervals (CIs) and P values. The linearity assumption of Cox proportional hazards regression modeling was tested for continuous variables. When deviations were found, linear splines or variable truncations were used.

To evaluate the possibility that the factors associated with the oc-currence of ICH may have differed according to the randomized treat-ment assignment (dose-adjusted warfarin versus rivaroxaban), we conducted interaction tests across all candidate variables.

The models were validated by 2 methods. One was by means of bootstrapping to obtain the optimism-corrected C-index. The second was summarizing how often each variable was selected in stepwise models fit with 1000 bootstrap samples of the data set. Bootstrap samples of the same size as the original were taken from the original by means of random sampling with replacement. A table showing the number of times each variable was chosen in the 1000 models was produced.

A sensitivity analysis was undertaken by deriving models based on: (1) the population of patients who adhered to their randomized treatment allocation (the safety and on-treatment population); (2) allowing geographical region to enter the model; and (3) including medication recorded at baseline, taken before randomization. We also developed models for secondary outcomes of intracerebral hemor-rhage, ischemic stroke, and hemorrhagic stroke.

ResultsParticipant CharacteristicsA total of 14 264 participants were randomized between December 18, 2006, and June 17, 2009. Overall, the median age was 73 years, the median congestive heart failure, hyper-tension, age >75 years, diabetes mellitus, prior stroke or TIA (CHADS2) score was 3.0, and 52% had a history of previous stroke or TIA.27 The median (25th, 75th) duration of follow-up was 1.94 years (1.42, 2.41) and was 99.8% complete (n=32 [0.2%] lost to follow-up).

Number of ICHsA total of 172 (1.2%) participants (intracerebral hemorrhage [n=128], subarachnoid hemorrhage [n=5], subdural hemor-rhage [n=38], and extradural hemorrhage [n=1]) experienced 1 (n=175) ICH events. Hence, 3 of the participants experienced >1 ICH event. The characteristics of all patients according to the occurrence of ICH, ischemic stroke, unknown pathological type of stroke, and no ICH or stroke are shown in Table 1.

Rate of ICHThe average rate of ICH for the duration of follow-up was 0.67% per 100 patient-years. Table I in the online-only Data Supplement shows the rates of ICH in different regions of the world, highest in Asia Pacific (1.21 per 100 patient-years) and lowest in Eastern Europe (0.33 per 100 patient-years). A KaplanMeier plot is presented showing the rates by random-ized treatment (Figure I in the online-only Data Supplement).

Causes and Treatment of ICHTrauma caused 9 (7%) intracerebral hemorrhages. Among 135 ICHs for which treatments were reported, medical or surgical intervention was undertaken in 92 (68%) cases, and transfu-sion of fresh, frozen plasma in 27 (20%) cases.

Outcome of ICHAmong the 172 participants with ICH, 75 (43%) died within 30 days (all attributed to the ICH), 90 were still alive at 30 days, and 7 had


from ICH among participants assigned warfarin (50%) and rivaroxaban (48%).

Among 90 participants who experienced an ICH because of hemorrhagic stroke, and in whom the modified Rankin Scale score was measured at 3 months after stroke or 1 month after stroke if the stroke occurred at the end of the study, the median time from hemorrhagic stroke to modified Rankin Scale score measurement in surviving patients was 92.5 days (89, 107). A total of 15 (17%) survived free of disability, 13 (14%) were disabled, and 62 (69%) were dead at 3 months after hemor-rhagic stroke.

Among the remaining 82 participants with ICHs that were not hemorrhagic stroke, 59 (72%) survived (followed up for a median of 328 [101, 528] days after hemorrhage) and 23 (28%) died (followed up for a median of 5 [2, 18] days after hemorrhage).

Predictors of ICHTable 3 shows that the significant, independent baseline pre-dictors of an increased risk of ICH were race (Asian HR, 2.02; 95% CI, 1.392.94; black HR, 3.25; 95% CI, 1.437.41), age (1.35; 1.131.63 per 10-year increase), reduced

serum albumin (1.39; 1.121.73 per 0.5 g/dL decrease), reduced platelet count below 210109/L (1.08; 1.021.13 per 10109/L decrease), previous stroke or TIA (1.42; 1.021.96), and increased diastolic blood pressure (1.17; 1.011.36 per 10 mm Hg increase). The predictors of a reduced risk of ICH were randomization to rivaroxaban (0.60; 0.440.82) and his-tory of congestive heart failure (CHF; 0.65; 0.470.89). The ability of the model to discriminate individuals with and with-out ICH was good (C-index, 0.69; 95% CI, 0.640.73).30,31

The results of interaction tests for all candidate variables are shown in Table II in the online-only Data Supplement. There was no significant interaction between treatment allo-cation (rivaroxaban or warfarin) and any of the variables listed in Table 1 and, therefore, no evidence that different factors would be selected for the model depending on whether the subject was taking rivaroxaban or warfarin.

Table III in the online-only Data Supplement shows the results of internal validation by means of taking 1000 boot-strap samples from the data and summarizing the percentage of bootstrap samples in which each variable was selected in the 1000 stepwise models. In the case of highly correlated variables, the validation was run, including only 1 of these variables and was repeated with the other. These variables are shown as the contribution in a model that does not include the correlated variable. Interval validation by means of bootstrap-ping realized the optimism-corrected C-index as 0.669.

Table 4 shows the platelets, albumin, no CHF, warfarin, age, race, diastolic blood pressure, stroke (PANWARDS) nomo-gram for predicting absolute risk of ICH in this cohort, which is based on the independent, significant prognostic factors for ICH and the strength of their association with risk of ICH. Table 5 and the Figure show the predicted probabilities of ICH at 2.5 years according to the score derived from the variables within the PANWARDS nomogram. There was evidence of good calibration between predicted and observed ICH rates (Figure II in the online-only Data Supplement).

Sensitivity analyses showed that the primary model in the intention-to-treat population (Table 3) was internally consis-tent in the safety, on-treatment population (Table IV in the online-only Data Supplement), except diastolic blood pres-sure and history of CHF did not enter.

The multivariate analysis of the effect of geographical region on the risk of ICH revealed that the adjusted HRs for ICH among residents of Asia Pacific (HR, 3.27; 95% CI, 2.065.21), Latin America (2.74; 1.644.56), North America (2.76; 1.704.47), and Western Europe (1.99; 1.163.41) were similar. Consequently, these regions were combined (versus Eastern Europe) in a second model (Table V in the online-only Data Supplement). In the second model, with geographical region entered into the model, residence in Eastern Europe (0.37; 0.250.55) and reduced creatinine clearance (1.11; 1.041.19 per 10 mL/min decrease) emerged as additional significant, independent predictors of ICH, whereas race, age, and history of CHF did not.

A third model, in which baseline use of antithrombotic agents (aspirin, vitamin K antagonism, and thienopyridines) was allowed to enter the model (Table VI in the online-only Data Supplement), showed that the baseline use of a thieno-pyridine significantly increased the hazard of ICH (HR, 2.57;

Table 2. Case Fatality of ICH According to Location and Treatment Allocation

Site of ICH n

Fatal, n (%) P ValueTotal Rivaroxaban Warfarin

Intracerebral 128 73 (57) 25 (53) 48 (59) 0.504

Hemorrhagic stroke 90 62 (69) 22 (67) 40 (70) 0.729

Subdural hemorrhage 38 11 (29) 5 (38) 6 (24) 0.457

Subarachnoid hemorrhage 5 1 (20) 1 (25) 0 (0) 1.000

Extradural hemorrhage 1 0 (0) 0 (0) 0 (0) ...

Total ICH 172 85 (49) 31 (48) 54 (50) 0.843

Hemorrhagic stroke is a subcomponent of intracerebral hemorrhage. The percentages are those of patients with ICH whose ICH was fatal. ICH indicates intracranial hemorrhage.

Table 3. Primary Model Showing the Factors Independently and Significantly Associated With ICH

Factors 2 HR 95% CI P Value

Race (vs white or other) 19.18


95% CI, 1.305.07; P=0.006), and previous experience with a vitamin K antagonist was associated with a significantly lower risk of ICH (0.68; 0.500.94; P=0.018).

Predictors of Intracerebral Hemorrhage and Hemorrhagic StrokeThe independent, significant predictors of intracerebral hem-orrhage (ie, intraparenchymal hemorrhage only [n=128]; excluding subarachnoid, subdural, and extradural hemor-rhage) were the same as those for all ICH (Table 3), except for the inclusion of creatinine clearance rather than age (HR, 1.09; 95% CI, 1.011.17 per 10 mL/min decrease) and the absence of stroke history in the model (Table VII in the online-only Data Supplement).

The independent, significant predictors of hemorrhagic stroke (ie, intraparenchymal hemorrhage causing focal

neurological symptoms or signs [n=90]) were the same as those for all ICH (Table 3), except for the exclusion of previ-ous stroke as a significant, independent predictor of hemor-rhagic stroke (Table VIII in the online-only Data Supplement).

Predictors of Ischemic StrokeTable IX in the online-only Data Supplement shows the inde-pendent, significant predictors of ischemic stroke. Three of these 6 factors were also predictors of ICH (ie, previous stroke or TIA, reduced albumin, and increased diastolic blood pres-sure), whereas 3 were not predictors of ICH (reduced creati-nine clearance [HR, 1.14; 95% CI, 1.091.19 per 10 mL/min decrease], increasing blood glucose [1.03; 1.011.05 per 10 mg/dL increase], and increasing platelet count [1.02; 1.011.04

Table 5. Predicted Probabilities of ICH at 2.5 Years According to Total Score on PANWARDS Nomogram

Total Score on Nomogram n Probability of ICH at 2.5 y

10 20 0.002

15 161 0.003

20 679 0.004

25 1465 0.006

30 2185 0.008

35 3001 0.012

40 2788 0.017

45 1889 0.025

50 994 0.035

55 406 0.049

60 166 0.070

65 62 0.098

70 12 0.137

75 4 0.190

ICH indicates intracranial hemorrhage; n, number of patients in the Rivaroxaban Once Daily, Oral, Direct Factor Xa Inhibition Compared With Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation (ROCKET AF) cohort with PANWARDS score; and PANWARDS, platelets, albumin, no congestive heart failure, warfarin, age, race, diastolic blood pressure, stroke.

Figure. Predicted probabilities of intracranial hemorrhage (ICH) at 2.5 years according to total score on platelets, albumin, no con-gestive heart failure, warfarin, age, race, diastolic blood pressure, stroke (PANWARDS) nomogram.

Table 4. PANWARDS Nomogram for Predicting Risk of Intracranial Hemorrhage

Points

Platelets, 109/L


per 10109/L increase]). Indeed, a decreasing platelet count below 210109/L was associated with an increased hazard of ICH (Table 3; Figure III in the online-only Data Supplement), and an increasing platelet count above 210109/L was associ-ated with an increased hazard of ischemic stroke (Table IX and Figure IV in the online-only Data Supplement).

The predictors of a reduced risk of ICH that were not pre-dictors of ischemic stroke were history of CHF (HR, 0.65; 95% CI, 0.470.89) and randomization to rivaroxaban (0.60; 0.440.82). Predictors of an increased risk of ICH that were not predictors of ischemic stroke were age (1.35; 1.131.63 per 10-year increase) and Asian (2.02; 1.392.94) or black race (3.25; 1.437.41; Table 3).

DiscussionPrincipal FindingsIn an international cohort of 14 264 patients with AF at moder-ate to high risk of stroke who were treated with anticoagula-tion, the overall rate of ICH was 0.67% per 100 patient-years, and the overall case fatality rate was 49%. There was no dif-ference in case fatality from ICH among participants assigned warfarin or rivaroxaban. The risk of ICH was significantly higher in Asians, blacks, the elderly, those with a history of stroke or TIA, those with increased diastolic blood pressure, and those with reduced platelet count and serum albumin at baseline. The risk of ICH was significantly lower among those who had a history of heart failure and who were randomized to rivaroxaban instead of warfarin. Internal validation by boot-strapping revealed that race and region were highly correlated, as were diastolic and systolic blood pressure. The discrimina-tive capacity of the model was good (C-index, 0.69; 95% CI, 0.640.73).30,31 The PANWARDS nomogram has been derived from the model for predicting the probability of ICH at 2.5 years for any individual patient.

Our Results in Context With Other StudiesThe rate and outcomes of ICH observed in ROCKET AF were similar to those observed among anticoagulated participants in large cohorts32,33 and in the Randomized Evaluation of Long-Term Anticoagulant Therapy (RE-LY) trial.9 The simi-larly high case fatality rates from ICH among patients taking warfarin or rivaroxaban in ROCKET AF and warfarin or dabi-gatran in RE-LY are consistent with other studies that report a poor prognosis in anticoagulant-associated ICH.34

Our findings of a higher risk of anticoagulant-associated ICH in Asians, blacks, the elderly, and patients with a history of stroke and elevated blood pressure are consistent with other observational studies and schema for predicting an increased risk of major (intracranial and extracranial) hemorrhage among anticoagulated individuals in other populations, sup-porting the external validity of our results.926,3537 The more novel findings from this study are the association of declining platelet count and albumin with an increased risk of ICH and the association of a history of heart failure and taking rivar-oxaban (instead of warfarin) with a reduced risk of ICH.

Declining platelet count has also been reported as a com-ponent of the hepatic or renal disease, ethanol abuse, malig-nancy, older age, rebleeding, reduced platelet count or function,

hypertension, anemia, genetic factors (CYP2C9), excessive fall risk, and stroke (HEMORRHAGES) scheme for predicting all major bleeding, but it was not reported whether reduced platelet count actually added independent, significant predictive power for major bleeding.38 Furthermore, a low platelet count, particu-larly a low-normal platelet count, has not been reported previ-ously as an independent, significant predictor of ICH.

The higher risk of ICH with declining serum albumin may reflect that both warfarin and rivaroxaban are highly protein bound. The lower risk of ICH in patients with a history of CHF may reflect a hypercoagulable state in heart failure.39

The significantly lower risk of ICH in patients with rivarox-aban when compared with those with warfarin, irrespective of age, is consistent with the lower risk of ICH with dabigatran when compared with warfarin.9 This may reflect the possibil-ity that warfarin compromises a normal hemostatic mecha-nism in the brain. In the event of injury to a vessel wall in the brain, tissue factor, which is found in high concentrations in the brain, interacts with activated factor VII (VIIa) to initiate coagulation and provide hemostatic protection.40,41 Warfarin blocks vitamin Kdependent -carboxylation of coagulation factors II, VII, IX, and X; suppresses the production of fac-tor VIIa; and compromises the formation of tissue factorVIIa complexes. In contrast, rivaroxaban selectively inhibits factor Xa, and dabigatran inhibits thrombin. Both agents do not com-promise the formation of tissue factorVIIa complexes, which are primary cellular initiators of coagulation. Our study did not have the statistical power to identify or exclude reliably a difference in rates of subdural hemorrhage between partici-pants assigned rivaroxaban and warfarin, which, if real, would invalidate the theory proposed above. Other mechanisms, such as the fact that rivaroxaban does not substantially penetrate the bloodbrain barrier, may also be important.42

The finding in our second model of a consistently lower rate of ICH among residents of Eastern Europe, when compared with other parts of the world, most likely reflects ascertain-ment or diagnostic bias for several reasons. First, the valid-ity of the diagnosis of ICH in Eastern Europe is less robust because the pathological diagnosis of stroke (ischemic versus hemorrhagic) was based on definitive brain imaging less often in Eastern Europe (computed tomography brain scan, 65% and MRI brain scan, 11%) when compared with other regions of the world (computed tomography brain scan range, 77%84% and MRI brain scan range, 12%26%). Second, the preva-lence of hypertension and mean level of diastolic blood pres-sure (major causal risk factors for hemorrhagic stroke) were highest among residents of Eastern Europe when compared with other regions of the world. Third, the magnitude of the effect of residence in Eastern Europe on ICH, as the strongest predictor (2, 24.3; HR, 0.37; 95% CI, 0.250.55), is extraor-dinary, not previously reported, and unlikely to be plausible. Fourth, there was no significant association between residence in Eastern Europe and risk of ischemic stroke.

One of the challenges in selecting patients with AF for anti-coagulant therapy is that the predictors of harm (ie, high risk of ICH) are frequently the same predictors of benefit (ie, high risk of ischemic stroke).26 Our prediction models for ICH and isch-emic stroke reinforce this impression to some extent. Previous stroke or TIA, increased diastolic blood pressure, and reduced


serum albumin were independent, significant predictors of both ICH and ischemic stroke. However, the independent predictors of an increased risk of ICH that were not predic-tors of an increased risk of ischemic stroke were race (Asian, black), increased age, and reduced platelet count. The indepen-dent predictors of a reduced risk of ICH that were not predic-tors of an increased risk of ischemic stroke were a history of CHF and randomization to rivaroxaban instead of warfarin. A reduced platelet count below 210109/L was associated with an increased hazard of ICH, and an elevated platelet count was associated with an increased hazard of ischemic stroke.

The PANWARDS nomogram is derived from the indepen-dent, significant prognostic factors for ICH and the strength of their association with risk of ICH. It is designed to enable cli-nicians to predict the probability of ICH for the next 2.5 years for any individual patient with AF who is anticoagulated. In the ROCKET AF cohort, most patients scored between 20 and 50 on the nomogram, which corresponded to a probability of ICH at 2.5 years ranging from 0.4% (score of 20) to 3.5% (score of 50). The predicted risk of ICH correlated closely with the observed rates of ICH, but the predictive ability of the PANWARDS nomogram awaits external validation in other cohorts of anticoagulated patients with AF.

StrengthsA strength of our study is that it complies with the Strengthening the Reporting of Observational Studies in Epidemiology guidelines for reporting an observational study.43 The study design was prospective and follow-up was prospective, regular, and nearly complete (99.8%), thus optimizing ascertainment of ICH events. ICH was diagnosed by means of standard-ized diagnostic criteria, including brain imaging and autopsy, and audited by a committee blinded to the study hypothesis and treatment allocation. The number of outcome events was reasonably large (n=175 ICH events in 172 patients), thus enabling 17 prognostic variables to be examined reliably in the prognostic model, and the effect of potential confounding by associated variables that were recorded was adjusted for by means of multiple regression analysis.

LimitationsA limitation of our study is that we were unable to adjust for variables that may influence ICH risk but were not measured at baseline or at the time of ICH, such as the presence or absence of the apolipoprotein E2 or E4 allele,4 CYP2C9 poly-morphism,18 cerebral leukoaraiosis,12 cerebral microbleeds,21,22 cerebrovascular pathology such as amyloid angiopathy,21,22 creatinine clearance


2. Fang MC, Go AS, Chang Y, Hylek EM, Henault LE, Jensvold NG, et al. Death and disability from warfarin-associated intracranial and extracra-nial hemorrhages. Am J Med. 2007;120:700705.

3. Woo D, Sauerbeck LR, Kissela BM, Khoury JC, Szaflarski JP, Gebel J, et al. Genetic and environmental risk factors for intracerebral hem-orrhage: preliminary results of a population-based study. Stroke. 2002;33:11901195.

4. Ariesen MJ, Claus SP, Rinkel GJ, Algra A. Risk factors for intracere-bral hemorrhage in the general population: a systematic review. Stroke. 2003;34:20602065.

5. Sturgeon JD, Folsom AR, Longstreth WT Jr, Shahar E, Rosamond WD, Cushman M. Risk factors for intracerebral hemorrhage in a pooled pro-spective study. Stroke. 2007;38:27182725.

6. ODonnell MJ, Xavier D, Liu L, Zhang H, Chin SL, Rao-Melacini P, et al; INTERSTROKE Investigators. Risk factors for ischaemic and intracerebral haemorrhagic stroke in 22 countries (the INTERSTROKE study): a case-control study. Lancet. 2010;376:112123.

7. Flaherty ML, Kissela B, Woo D, Kleindorfer D, Alwell K, Sekar P, et al. The increasing incidence of anticoagulant-associated intracerebral hem-orrhage. Neurology. 2007;68:116121.

8. Huhtakangas J, Tetri S, Juvela S, Saloheimo P, Bode MK, Hillbom M. Effect of increased warfarin use on warfarin-related cerebral hemorrhage: a longitudinal population-based study. Stroke. 2011;42:24312435.

9. Hart RG, Diener HC, Yang S, Connolly SJ, Wallentin L, Reilly PA, et al. Intracranial hemorrhage in atrial fibrillation patients during anticoagulation with warfarin or dabigatran: the RE-LY trial. Stroke. 2012;43:15111517.

10. Hylek EM, Singer DE. Risk factors for intracranial hemorrhage in outpa-tients taking warfarin. Ann Intern Med. 1994;120:897902.

11. The Stroke Prevention in Atrial Fibrillation Investigators. Bleeding dur-ing antithrombotic therapy in patients with atrial fibrillation. Arch Intern Med. 1996;156:409416.

12. Gorter JW. Major bleeding during anticoagulation after cerebral isch-emia: patterns and risk factors. Stroke Prevention In Reversible Ischemia Trial (SPIRIT). European Atrial Fibrillation Trial (EAFT) study groups. Neurology. 1999;53:13191327.

13. Sjblom L, Hrdemark HG, Lindgren A, Norrving B, Fahln M, Samuelsson M, et al. Management and prognostic features of intrace-rebral hemorrhage during anticoagulant therapy: a Swedish multicenter study. Stroke. 2001;32:25672574.

14. Gage BF, Birman-Deych E, Kerzner R, Radford MJ, Nilasena DS, Rich MW. Incidence of intracranial hemorrhage in patients with atrial fibrilla-tion who are prone to fall. Am J Med. 2005;118:612617.

15. Shen AY, Yao JF, Brar SS, Jorgensen MB, Chen W. Racial/ethnic differ-ences in the risk of intracranial hemorrhage among patients with atrial fibrillation. J Am Coll Cardiol. 2007;50:309315.

16. Hylek EM, Evans-Molina C, Shea C, Henault LE, Regan S. Major hem-orrhage and tolerability of warfarin in the first year of therapy among elderly patients with atrial fibrillation. Circulation. 2007;115:26892696.

17. Shen AY, Chen W, Yao JF, Brar SS, Wang X, Go AS. Effect of race/eth-nicity on the efficacy of warfarin: potential implications for prevention of stroke in patients with atrial fibrillation. CNS Drugs. 2008;22:815825.

18. Limdi NA, McGwin G, Goldstein JA, Beasley TM, Arnett DK, Adler BK, et al. Influence of CYP2C9 and VKORC1 1173C/T genotype on the risk of hemorrhagic complications in African-American and European- American patients on warfarin. Clin Pharmacol Ther. 2008;83:312321.

19. Poli D, Antonucci E, Grifoni E, Abbate R, Gensini GF, Prisco D. Bleeding risk during oral anticoagulation in atrial fibrillation patients older than 80 years. J Am Coll Cardiol. 2009;54:9991002.

20. Reinecke H, Brand E, Mesters R, Schbitz WR, Fisher M, Pavenstdt H, et al. Dilemmas in the management of atrial fibrillation in chronic kidney disease. J Am Soc Nephrol. 2009;20:705711.

21. Greenberg SM, Vernooij MW, Cordonnier C, Viswanathan A, Al-Shahi Salman R, Warach S, et al; Microbleed Study Group. Cerebral microbleeds: a guide to detection and interpretation. Lancet Neurol. 2009;8:165174.

22. Orken DN, Kenangil G, Uysal E, Forta H. Cerebral microbleeds in isch-emic stroke patients on warfarin treatment. Stroke. 2009;40:36383640.

23. Flaherty ML. Anticoagulant-associated intracerebral hemorrhage. Semin Neurol. 2010;30:565572.

24. Poli D, Antonucci E, Testa S, Tosetto A, Ageno W, Palareti G; Italian Federation of Anticoagulation Clinics. Bleeding risk in very old patients

on vitamin K antagonist treatment: results of a prospective collaborative study on elderly patients followed by Italian Centres for Anticoagulation. Circulation. 2011;124:824829.

25. Olesen JB, Lip GY, Kamper AL, Hommel K, Kber L, Lane DA, et al. Stroke and bleeding in atrial fibrillation with chronic kidney disease. N Engl J Med. 2012;367:625635.

26. McGrath ER, Kapral MK, Fang J, Eikelboom JW, Conghaile A, Canavan M, et al; Investigators of the Registry of the Canadian Stroke Network. Which risk factors are more associated with ischemic stroke than intracerebral hemorrhage in patients with atrial fibrillation? Stroke. 2012;43:20482054.

27. Piccini JP, Stevens SR, Chang Y, Singer DE, Lokhnygina Y, Go AS, et al. Renal Dysfunction as a predictor of stroke and systemic embo-lism in patients with nonvalvular atrial fibrillation: validation of the R(2)CHADS(2) index in the ROCKET AF and ATRIA study cohorts. Circulation. 2013;127:224232.

28. ROCKET AF Study Investigators. Rivaroxaban-once daily, oral, direct factor Xa inhibition compared with vitamin K antagonism for prevention of stroke and Embolism Trial in Atrial Fibrillation: rationale and design of the ROCKET AF study. Am Heart J. 159:340347.

29. Patel MR, Mahaffey KW, Garg J, Pan G, Singer DE, Hacke W, et al; ROCKET AF Investigators. Rivaroxaban versus warfarin in nonvalvular atrial fibrillation. N Engl J Med. 2011;365:883891.

30. Tripep G, Japger KJ, Dekker FW, Zoccali C. Statistical methods for the assessment of prognostic biomarkers (part 1): discrimination. Nephrol Dial Transplant. 2010;25:13991401.

31. Cook NR. Statistical evaluation of prognostic versus diagnostic models: beyond the ROC curve. Clin Chem. 2008;54:1723.

32. Fang MC, Go AS, Chang Y, Borowsky LH, Pomernacki NK, Udaltsova N, et al. Thirty-day mortality after ischemic stroke and intracranial hemorrhage in patients with atrial fibrillation on and off anticoagulants. Stroke. 2012;43:17951799.

33. Singer DE, Chang Y, Fang MC, Borowsky LH, Pomernacki NK, Udaltsova N, et al. The net clinical benefit of warfarin anticoagulation in atrial fibrillation. Ann Intern Med. 2009;151:297305.

34. Dowlatshahi D, Butcher KS, Asdaghi N, Nahirniak S, Bernbaum ML, Giulivi A, et al; Canadian PCC Registry (CanPro) Investigators. Poor prognosis in warfarin-associated intracranial hemorrhage despite antico-agulation reversal. Stroke. 2012;43:18121817.

35. Pisters R, Lane DA, Nieuwlaat R, de Vos CB, Crijns HJ, Lip GY. A novel user-friendly score (HAS-BLED) to assess 1-year risk of major bleeding in patients with atrial fibrillation: the Euro Heart Survey. Chest. 2010;138:10931100.

36. Lip GY, Frison L, Halperin JL, Lane DA. Comparative validation of a novel risk score for predicting bleeding risk in anticoagulated patients with atrial fibrillation: the HAS-BLED (Hypertension, Abnormal Renal/Liver Function, Stroke, Bleeding History or Predisposition, Labile INR, Elderly, Drugs/Alcohol Concomitantly) score. J Am Coll Cardiol. 2011;57:173180.

37. Fang MC, Go AS, Chang Y, Borowsky LH, Pomernacki NK, Udaltsova N, et al. A new risk scheme to predict warfarin-associated hemorrhage: The ATRIA (Anticoagulation and Risk Factors in Atrial Fibrillation) Study. J Am Coll Cardiol. 2011;58:395401.

38. Gage BF, Yan Y, Milligan PE, Waterman AD, Culverhouse R, Rich MW, et al. Clinical classification schemes for predicting hemorrhage: results from the National Registry of Atrial Fibrillation (NRAF). Am Heart J. 2006;151:713719.

39. De Lorenzo F, Saba N, Kakkar VV. Blood coagulation in patients with chronic heart failure: evidence for hypercoagulable state and potential for pharmacological intervention. Drugs. 2003;63:565576.

40. Fleck RA, Rao LV, Rapaport SI, Varki N. Localization of human tissue factor antigen by immunostaining with monospecific, polyclonal anti-human tissue factor antibody. Thromb Res. 1990;59:421437.

41. Mackman N. The role of tissue factor and factor VIIa in hemostasis. Anesth Analg. 2009;108:14471452.

42. Haas S. Rivaroxaban an oral, direct Factor Xa inhibitor: lessons from a broad clinical study programme. Eur J Haematol. 2009;82:339349.

43. von Elm E, Altman DG, Egger M, Pocock SJ, Gtzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guide-lines for reporting observational studies. Lancet. 2007;370:14531457.

Documents

Intracranial Hemorrhage Among Patients With Atrial