- Email: [email protected]
Comparison of the CHA2DS2-VASc, CHADS2, HAS-BLED, ORBIT, and ATRIA Risk Scores in Predicting Non–Vitamin K Antagonist Oral Anticoagulants-Associated Bleeding in Patients With Atrial Fibrillation
Comparison of the CHA2DS2-VASc, CHADS2, HASBLED, ORBIT, and ATRIA Risk Scores in Predicting Non–Vitamin K Antagonist Oral AnticoagulantsAssociated Bleeding in Patients With Atrial Fibrillation Xiaoxi Yao, PhDa,b, Bernard J. Gersh, MB, ChB, DPhil, FRCPc, Lindsey R. Sangaralingham, MPHa, David M. Kent, MD, CM, MSd,e, Nilay D. Shah, PhDa,b,f, Neena S. Abraham, MD, MSCEa,b,g, and Peter A. Noseworthy, MDa,c,* The increasing adoption of non–vitamin K antagonist oral anticoagulants (NOACs) for stroke prevention in atrial fibrillation (AF) necessitates a reassessment of bleeding risk scores. Because known risk factors for bleeding are largely the same as for stroke, we hypothesize that stroke risk scores could also be used to identify patients with high bleeding risks. We aimed to compare the performance of 2 stroke risk scores (Congestive Heart failure, hypertension, Age ≥75 [doubled], Diabetes, Stroke [doubled], Vascular disease, Age 65–74, and Sex [female] [CHA2DS2-VASc] and Cardiac failure, Hypertension, Age, Diabetes, Stroke [Doubled] [CHADS2]) and 3 bleeding risk scores (hypertension, abnormal renal/liver function [1 point each], stroke, bleeding history or predisposition, labile INR, elderly [.65 years], drugs/ alcohol concomitantly [1 point each] [HAS-BLED], Outcomes Registry for Better Informed Treatment of Atrial Fibrillation [ORBIT], and AnTicoagulation and Risk factors In Atrial fibrillation [ATRIA]) in predicting major and intracranial bleeding. Using a large US commercial insurance database, we identified 39,539 patients with nonvalvular AF who started NOACs between October 1, 2010 and June 30, 2015. The performance of risk scores was compared using C-statistic and net reclassification improvement (NRI). Over a total of 22,583 person-years, 665 patients (2.94% per year) had major bleeding, including 74 intracranial hemorrhages (0.33% per year). For the prediction of major bleeding, CHA2DS2-VASc had the highest C-statistic both as a continuous score (C-statistic 0.68) and as a categorical score (C-statistic 0.65). For the prediction of intracranial bleeding, CHADS2 had the highest C-statistic both as a continuous score (C-statistic 0.66) and as a categorical score (C-statistic 0.66). There were no statistically significant differences between scores based on NRI. In conclusion, CHA2DS2-VASc, CHADS2, HAS-BLED, ORBIT, and ATRIA had similar, albeit modest, performance in predicting NOAC-associated bleeding in patients with AF. Careful assessment and active management of bleeding risk factors may be warranted in all patients on NOACs who have high stroke risk scores. © 2017 Elsevier Inc. All rights reserved. (Am J Cardiol 2017;120:1549–1556)
a
Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery, Mayo Clinic, Rochester, Minnesota; bDivision of Health Care Policy and Research, Department of Health Sciences Research, Mayo Clinic, Rochester, Minnesota; cDepartment of Cardiovascular Medicine, Mayo Clinic, Rochester, Minnesota; dPredictive Analytics and Comparative Effectiveness (PACE) Center, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center/Tufts University School of Medicine, Boston, Massachusetts; eDepartment of Neurology, Tufts Medical Center/Tufts University School of Medicine, Boston, Massachusetts; fOptumLabs, Cambridge, Massachusetts; and gDivision of Gastroenterology and Hepatology, Department of Medicine, Mayo Clinic, Scottsdale, Arizona. Manuscript received June 2, 2017; revised manuscript received and accepted July 10, 2017. The work was performed at Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery, Mayo Clinic, Rochester, MN. This work was supported by Mayo Clinic Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery (Rochester, Minnesota). See page 1554 for disclosure information. *Corresponding author: Tel: 507 255 2446; fax: 507 255 2550. E-mail address: [email protected] (P.A. Noseworthy). 0002-9149/© 2017 Elsevier Inc. All rights reserved. https://doi.org/10.1016/j.amjcard.2017.07.051
The increasing adoption of NOACs for stroke prevention in AF necessitates a reassessment of bleeding risk scores. All the bleeding risk scores, except Outcomes Registry for Better Informed Treatment of Atrial Fibrillation (ORBIT), were developed and validated in patients receiving vitamin K antagonist.1–3 Little is known about how these scores perform in patients receiving NOACs. Regardless how the bleeding risk scores compare with each other, they all have at most modest ability to predict bleeding,4 and are not routinely used by clinicians.5 By contrast, stroke risk scores, such as the Congestive Heart failure, hypertension, Age ≥75 (doubled), Diabetes, Stroke (doubled), Vascular disease, Age 65–74, and Sex (female) (CHA2DS2-VASc) and Cardiac failure, Hypertension, Age, Diabetes, Stroke (Doubled) (CHADS2) scores, are recommended by guidelines and more commonly used in practice for stroke risk stratification.5 Because stroke and bleeding risks are closely related,6,7 we hypothesize that stroke risk scores could also be used to identify patients with high www.ajconline.org
1550
The American Journal of Cardiology (www.ajconline.org)
bleeding risks. Using a large heterogeneous cohort of patients with AF starting NOACs, we aimed to compare the performance of 2 stroke risk scores (CHA2DS2-VASc and CHADS2) and 3 bleeding risk scores (hypertension, abnormal renal/liver function [1 point each], stroke, bleeding history or predisposition, labile INR, elderly [.65 years], drugs/alcohol concomitantly [1 point each] [HAS-BLED], ORBIT, and AnTicoagulation and Risk factors In Atrial fibrillation [ATRIA]) in predicting major and intracranial bleeding. Methods The data source of this study is the OptumLabs Data Warehouse, a large administrative database that contains ≥100 million privately insured and Medicare Advantage enrollees of all ages and races and from all 50 states.8,9 We identified adult patients (≥18 years) with nonvalvular AF who started apixaban, dabigatran, edoxaban, or rivaroxaban between October 1, 2010 and June 30, 2015. Patients were followed until the earliest date of the following: end of study period (June 30, 2015), discontinuation of the medication, switch to another OAC, and the occurrence of major bleeding or stroke.
Our study was exempt by the Mayo Clinic Institutional Review Board for approval as we used only preexisting, deidentified data. Supplementary Table S1 presents an overview of the risk scores. The determination of cutoff points for risk categories (low, intermediate, and high) was mostly based on the original derivation studies of these scores. In a sensitivity analysis, we changed the cutoff points to make the event rates in each risk category similar across the risk scores (Figure 1). Cox proportional hazards regression was used to assess the relation between individual risk factors and major bleeding, and was also used to examine the prognostic value of the risk scores, with each score assessed separately. We evaluated the discrimination by comparing the C-statistics of the other 4 risk scores to CHA2DS2-VASc using the nonparametric method proposed by DeLong et al.10 We also calculated overall net reclassification improvement (NRI), a summary statistic for the incremental predictive value of each of the risk scores in comparison with CHA2DS2-VASc.11,12 In this NRI analysis, all the scores were analyzed as categorical variables. We evaluated calibration by plotting major bleeding events rates in the present study versus those previously published from
Figure 1. Proportions of patients and major bleeding event rates per 100 person-years in different risk strata and C-statistics for each risk score.
Arrhythmias and Conduction Disturbances/Risk Score for Bleeding on NOACs
the original derivation cohorts for HAS-BLED, ORBIT-AF, and ATRIA.1–3 We did not plot CHA2DS2-VASc or CHADS2, because they were derived for stroke risk prediction and do not have bleeding rates reported from derivation cohorts. For these 2 stroke risk scores, we plotted the event rates of major bleeding observed in this study versus the event rates of stroke in patients not treated with OAC in the derivation cohorts to illustrate the correlation between stroke and bleeding risks.13,14 We performed a number of sensitivity analyses to investigate the robustness of the findings. These sensitivity tests included changing the reference group from CHA2DS2-VASc to CHADS2, using the laboratory data instead of diagnosis codes, using Harrell’s c (an extension of C-statistic that applies to time-to-event data), considering death and stroke as competing risks, including bleeding events after discontinuation of treatment, assessing the risk at a fixed time point (at the end of 1 year), stratified analysis by different NOACs, by NOAC dosage, by the presence of vascular disease (to assess the impact of incomplete aspirin data because of the strong correlation between vascular disease and antiplatelet use), and whether patients had at least 2 prescription fills, and by age. Further details of the methods, including study population, ascertainment and validation of major bleeding, ascertainment of OAC treatment, statistical analysis, and sensitivity tests, are provided in the Appendix. Results We identified 39,539 patients with a total of 22,583 personyears of follow-up (mean follow-up 0.6 ± 0.7 year). Among these, 20%, 38%, 0.04%, and 43% started apixaban, dabigatran, edoxaban, and rivaroxaban, respectively (Table 1). We observed 665 major bleeding events (2.94 per 100 person-years), including 74 intracranial bleeding events (0.33 per 100 person years). The event rates and the proportions of patients categorized by the different risk scores are shown in Figure 1 and Table 2. For the prediction of major bleeding, CHA2DS2-VASc had the highest C-statistic both as a continuous score (C-statistic 0.68; p <0.001 compared with CHADS2 and p <0.01 compared with HAS-BLED, but not statistically significantly different in comparison with ORBIT or ATRIA) and as a categorical score (C-statistic 0.65; p <0.001 compared with ORBIT or ATRIA, but not statistically significantly different in comparison with CHADS2 or HAS-BLED). For the prediction of intracranial bleeding, CHADS2 had the highest C-statistic both as a continuous score (C-statistic 0.66; not statistically significant in comparison with any other score) and as a categorical score (C-statistic 0.66; p <0.05 compared with CHA2DS2-VASc, p <0.01 compared with ORBIT or ATRIA, not statistically significant in comparison with HAS-BLED) (Table 3 and Figure 2). We found ORBIT and HAS-BLED had better calibration than ATRIA (Figure 3), and stroke and bleeding risks are highly correlated (Figure 4). There was no statistically significant difference comparing any other score with CHA2DS2-VASc (Table 3), or CHADS2 based on the overall NRI (Supplementary Table S3). Compared with CHA2DS2-VASc, all other scores had negative event NRI, indicating lower sensitivity, but positive nonevent NRI, indicating higher specificity (Supplementary Table S4).
1551
All other sensitivity analyses showed similar results to the main analysis (Supplementary Tables S5 to S12). Discussion Using a large cohort of patients treated with NOACs in routine clinical practice, we demonstrate that CHA2DS2-VASc, CHADS2, HAS-BLED, ORBIT, and ATRIA may have similar performance in predicting major and intracranial bleeding. This observation underscores the fact that patients at high risk of stroke are also at high risk of bleeding. As such, every effort should be made to address modifiable risk factors for bleeding in all patients on anticoagulation treatment who have high CHA2DS2-VASc or CHADS2 scores. To the best of our knowledge, our study is the first study comparing these scores in patients treated with NOACs. A comparison in NOAC-treated patients is very important as NOACs, and vitamin K antagonists have different bleeding risk profile,15 and NOACs have become the first choice for many patients.16 Our study is unique as few previous studies had compared stroke and bleeding risk scores.17,18 We found the stroke risk scores had similar performance to the bleeding risk scores, likely because the stroke and bleeding risks factors overlap.6,7 In our study, the strongest predictor of major bleeding was age. CHA2DS2-VASc assigns 1 point to age 65 to 74 years, and 2 points to ≥75 years, and thus has better resolution to capture the increasing risk across the age spectrum. Because patients at high risk of stroke are also at high risk of bleeding, careful assessment and active management of bleeding risk factors may be warranted in all patients on NOACs who have high stroke risk scores, not just those with high bleeding risk scores. These findings are particularly important in light of the change in the 2016 European guidelines that no longer recommend calculating a specific bleeding risk score.19 Our study suggests that it appeared to be no incremental benefit of any bleeding score in comparison with stroke risk scores, and thus, provides support to this guideline change. As being emphasized by practice guidelines, a high bleeding risk score should generally not result in withholding oral anticoagulation.19 Rather, the focus should be on identifying modifiable risk factors for bleeding. Future efforts are needed to increase awareness of the collinear risk of stroke and bleeding, and motivate clinicians to prescribe oral anticoagulants in high-risk patients and address reversible risk factors. The strength of our study is the use of a large contemporary cohort of patients treated in diverse clinical practice settings. The bleeding risk scores are meant to apply to patients in clinical care, but many of the previous derivation and validation studies used patients in clinical trials or registries that were very different from those treated in everyday clinical practice. Moreover, the large number of events allows us to assess the predictive accuracy not only for major bleeding, but also for intracranial bleeding, the most feared complication of OAC therapy. Another strength is that using pharmacy claims data, we can better determine whether patients were continuously treated with NOACs. However, we acknowledge that prescription fills are an imperfect surrogate for adherence and persistence, because patients may not always take the medications filled. Our study has some limitations that warrant discussion. First, we rely on billing codes for patient characteristics and
1552
The American Journal of Cardiology (www.ajconline.org)
Table 1 Baseline characteristics of patients initiating NOACs, by presence or absence of major bleeding events during the on-treatment study period Characteristics
Age, years Median (IQR) 18–64 65–74 ≥75 Female Non-white Race Heart Failure Hypertension Diabetes Previous stroke/TIA Vascular disease Stage III or IV chronic kidney disease Abnormal liver function Previous major bleeding Anemia Alcoholism Antiplatelet use NSAIDs use Previous warfarin exposure CHA2DS2-VASc Median (IQR) 0,1 2,3 ≥4 CHADS2 Median (IQR) 0,1 2,3 ≥4 HAS-BLED Median (IQR) 0,1 2 ≥3 ORBIT Median (IQR) 0–2 3 ≥4 ATRIA Median (IQR) 0–3 4 ≥5
Total (n = 39,539)
Major Bleeding
Univariable Hazard Ratio (95% CI)
Multivariable Hazard Ratio (95% CI)
Ref 1.93*** (1.44, 2.58) 3.52*** (2.67, 4.64) 1.25** (1.06, 1.46) 1.15 (0.96, 1.38) 1.56*** (1.32, 1.83) 1.63** (1.15, 2.31) 1.03 (0.88, 1.21) 1.02 (0.84, 1.24) 1.43*** (1.21, 1.70) 1.30* (1.03, 1.64) 1.08 (0.73, 1.59) 1.61*** (1.30, 1.98) 1.41*** (1.19, 1.66) 1.74* (1.13, 2.69) 1.47** (1.15, 1.87) 1.24 (0.88, 1.74) 1.02 (0.87, 1.20)
No (n = 38,874)
Yes (n = 665)
71 (63–79) 11,863 (30%) 12,335 (31%) 15,341 (39%) 16,612 (42%) 7,775 (20%) 11,100 (28%) 33,825 (86%) 13,434 (34%) 5,467 (14%) 18,838 (48%) 2,784 (7%) 1,484 (4%) 3,406 (9%) 8,581 (22%) 1,048 (3%) 2,635 (7%) 1,910 (5%) 10,990 (28%)
71 (63–79) 11,798 (30%) 12,166 (31%) 14,910 (38%) 16,259 (42%) 7,619 (20%) 10,802 (28%) 33,195 (85%) 13,167 (34%) 5,339 (14%) 18,407 (47%) 2,698 (7%) 1,457 (4%) 3,294 (9%) 8,335 (21%) 1,026 (3%) 2,556 (7%) 1,875 (5%) 10,747 (28%)
78 (72–84) 65 (10%) 169 (25%) 431 (65%) 353 (53%) 156 (24%) 298 (45%) 630 (95%) 267 (40%) 128 (19%) 431 (65%) 86 (13%) 27 (4%) 112 (17%) 246 (37%) 22 (3%) 79 (12%) 35 (5%) 243 (37%)
Ref 2.33*** (1.75, 3.11) 4.94*** (3.80, 6.41) 1.55*** (1.33, 1.80) 1.31** (1.10, 1.57) 2.20*** (1.89, 2.56) 2.77*** (1.97, 3.90) 1.29** (1.10, 1.50) 1.46*** (1.20, 1.77) 2.10*** (1.79, 2.46) 2.19*** (1.75, 2.75) 1.16 (0.79, 1.71) 2.24*** (1.82, 2.74) 2.26*** (1.93, 2.65) 1.45 (0.95, 2.21) 1.89*** (1.49, 2.39) 1.10 (0.78, 1.54) 1.21* (1.03, 1.42)
4 (2–5) 5,010 (13%) 12,853 (33%) 21,676 (55%)
4 (2–5) 4,999 (13%) 12,744 (33%) 21,131 (54%)
5 (4–6) 11 (2%) 109 (16%) 545 (82%)
Ref 3.22*** (1.73, 5.98) 9.97*** (5.49, 18.10)
2 (1–3) 13,340 (34%) 20,428 (52%) 5,771 (15%)
2 (1–3) 13,250 (34%) 20,045 (52%) 5,579 (14%)
3 (2–4) 90 (14%) 383 (58%) 192 (29%)
Ref 2.63*** (2.09, 3.30) 4.96*** (3.86, 6.37)
2 (1–3) 10,503 (27%) 14,722 (37%) 14,314 (36%)
2 (1–3) 10,447 (27%) 14,501 (37%) 13,926 (36%)
3 (2–3) 56 (8%) 221 (33%) 388 (58%)
Ref 2.58*** (1.93, 3.46) 4.96*** (3.75, 6.56)
1 (0–2) 32,735 (83%) 4,194 (11%) 2,610 (7%)
1 (0–2) 32,312 (83%) 4,075 (11%) 2,487 (6%)
1 (1–3) 423 (64%) 119 (18%) 123 (19%)
Ref 2.22*** (1.81, 2.72) 4.07*** (3.33, 4.97)
2 (1–3) 30,393 (77%) 3,729 (9%) 5,417 (14%)
2 (1–3) 30,000 (77%) 3,665 (9%) 5,209 (13%)
3 (2–6) 393 (59%) 64 (10%) 208 (31%)
Ref 1.39* (1.07, 1.81) 3.14*** (2.66, 3.72)
* p <0.05, **p <0.01, ***p <0.001. CI = confidence interval; IQR = interquartile range; other abbreviations as in Table 1. Anemia was defined using diagnosis codes. In the sensitivity test using laboratory data, anemia was defined as abnormal hemoglobin (<13 mg/dl for males and <12 mg/dl for females) or hematocrit (<40% for males and <36% for females).
outcome adjudication, which may be prone to inaccuracies. However, the algorithms used to capture patient characteristics and outcomes have been validated and used in numerous previous studies,2,20 and demonstrated good performance in our internal validation of bleeding outcomes. Certain risk factors, such as alcoholism, may be undercoded in claims data— just as it tends to be underdiagnosed in the clinical setting—so the performance measured in this study may well reflect the
usefulness in clinical care. Second, we lack certain clinical characteristics, such as blood pressure, so we used the diagnosis of hypertension instead of uncontrolled hypertension for HAS-BLED. We included patients with all types of AF, but it is difficult to use claims data to ascertain AF type, burden, or time since the first AF diagnoses. However, these characteristics are not included in any of the risk stratification schemes; thus, lack of these data did not affect our findings. Third, we
Arrhythmias and Conduction Disturbances/Risk Score for Bleeding on NOACs
1553
Table 2 Observed major and intracranial bleeding rates by risk score Patients
PY
n (%) Total CHA2DS2-VASc 0 1 2 3 4 5 6 7 8–9 Risk Low (0–1) Intermediate (2–3) High (≥4) CHADS2 0 1 2 3 4 5 6 Risk Low (0–1) Intermediate (2–3) High (≥4) HAS-BLED 0 1 2 3 4 5 6–8 Risk Low (0–1) Intermediate (2) High (≥3) ORBIT 0 1 2 3 4 5 6–7 Risk Low (0–2) Intermediate (3) High (≥4) ATRIA 0 1 2 3 4 5 6 7 8–10 Risk Low (0–3) Intermediate (4) High (≥5) PY = person-years.
Major bleeding
Intracranial bleeding
Events
Events/100 PY
Events
Events/100 PY
39,539 (100%)
22,583
665
2.94
74
0.33
1,497 (4%) 3,513 (9%) 5,576 (14%) 7,277 (18%) 8,256 (21%) 6,504 (16%) 3,953 (10%) 1,953 (5%) 1,010 (3%)
600 1,813 3,382 4,424 4,932 3,723 2,168 1,068 472
1 10 36 73 168 140 122 68 47
0.17 0.55 1.06 1.65 3.41 3.76 5.63 6.37 9.96
0 2 4 12 17 14 12 8 5
0.00 0.11 0.12 0.27 0.34 0.38 0.55 0.75 1.06
5,010 (13%) 12,853 (33%) 21,676 (55%)
2,413 7,806 12,363
11 109 545
0.46 1.40 4.41
2 16 56
0.08 0.20 0.45
3,251 (8%) 10,089 (26%) 12,422 (31%) 8,006 (20%) 3,660 (9%) 1,623 (4%) 488 (1%)
1,484 5,896 7,495 4,558 2,042 885 223
6 84 208 175 120 55 17
0.40 1.42 2.78 3.84 5.88 6.21 7.63
0 10 24 13 18 8 1
0.00 0.17 0.32 0.29 0.88 0.90 0.45
13,340 (34%) 20,428 (52%) 5,771 (15%)
7,379 12,053 3,150
90 383 192
1.22 3.18 6.09
10 37 27
0.14 0.31 0.86
2,346 (6%) 8,157 (21%) 14,722 (37%) 9,595 (24%) 3,750 (10%) 864 (2%) 105 (0.3%)
1,069 4,653 8,896 5,540 1,970 410 45
4 52 221 222 120 39 7
0.37 1.12 2.48 4.01 6.09 9.52 15.58
1 8 21 28 11 4 1
0.09 0.17 0.24 0.51 0.56 0.98 2.23
10,503 (27%) 14,722 (37%) 14,314 (36%)
5,723 8,896 7,964
56 221 388
0.98 2.48 4.87
9 21 44
0.16 0.24 0.55
17,469 (44%) 10,182 (26%) 5,084 (13%) 4,194 (11%) 1,519 (4%) 840 (2%) 251 (1%)
10,166 6,003 2,727 2,391 775 413 107
120 222 81 119 59 50 14
1.18 3.70 2.97 4.98 7.61 12.09 13.03
20 22 11 14 4 2 1
0.20 0.37 0.40 0.59 0.52 0.48 0.93
32,735 (83%) 4,194 (11%) 2,610 (7%)
18,895 2,391 1,296
423 119 123
2.24 4.98 9.49
53 14 7
0.28 0.59 0.54
3,899 (10%) 15,350 (39%) 1,964 (5%) 9,180 (23%) 3,729 (9%) 850 (2%) 3,520 (9%) 872 (2%) 175 (0.4%)
1,948 9,255 1,147 5,295 2,029 463 1,967 417 62
11 138 27 217 64 30 118 52 8
0.56 1.49 2.35 4.10 3.15 6.47 6.00 12.48 13.00
1 20 1 27 8 2 12 2 1
0.05 0.22 0.09 0.51 0.39 0.43 0.61 0.48 1.62
30,393 (77%) 3,729 (9%) 5,417 (14%)
17,645 2,029 2,908
393 64 208
2.23 3.15 7.15
49 8 17
0.28 0.39 0.58
1554
The American Journal of Cardiology (www.ajconline.org)
Table 3 C-statistic and net reclassification improvement in comparison with CHA2DS2-VASc (n = 39,539) Major bleeding
CHA2DS2-VASc CHADS2 HAS-BLED ORBIT ATRIA
Intracranial bleeding
C-statistic (95% CI) for continuous scores
C-statistic (95% CI) for categorical scores
NRI for categorical scores
C-statistic (95% CI) for continuous scores
C-statistic (95% CI) for categorical scores
NRI for categorical scores
0.68 (0.66, 0.70) 0.65*** (0.63, 0.67) 0.66** (0.64, 0.67) 0.66 (0.64, 0.68) 0.67 (0.65, 0.69)
0.65 (0.63, 0.66) 0.64 (0.62, 0.65) 0.64 (0.62, 0.66) 0.60*** (0.58, 0.62) 0.60*** (0.58, 0.62)
Ref −0.04 0.02 0.01 0.05
0.65 (0.59, 0.71) 0.66 (0.60, 0.72) 0.64 (0.58, 0.70) 0.60 (0.54, 0.66) 0.63 (0.57, 0.68)
0.61 (0.57, 0.66) 0.66* (0.60, 0.72) 0.63 (0.58, 0.69) 0.55 (0.50, 0.61) 0.56 (0.50, 0.61)
Ref 0.09 0.07 −0.06 −0.04
CI = confidence interval; NRI = net reclassification improvement. * p <0.05, **p <0.01, ***p <0.001 for each score compared with CHA2DS2-VASc.
Figure 2. Receiver operating characteristic curves of the risk scores.
were not able to accurately assess aspirin and nonsteroidal anti-inflammatory drugs use, because they generally do not need a prescription. Patients may use aspirin or nonsteroidal anti-inflammatory drugs for a short period or as needed, which makes it difficult to assess in any studies. We conducted numerous sensitivity tests to evaluate whether these limitations could affect our findings. Such sensitivity tests include using laboratory data instead of diagnosis codes to assess renal function and anemia, and stratified analysis by vascular disease to assess to impact of incomplete aspirin data. The results of these sensitivity analyses support the robustness of our findings. The 2 stroke risk scores (CHA2DS2-VASc and CHADS2) and 3 bleeding risk scores (HAS-BLED, ORBIT, and ATRIA)
had similar, albeit modest, performance in predicting major bleeding and intracranial bleeding in patients with AF treated with NOACs. A single stroke risk score appears to be a convenient tool in identifying low-risk patients who may not require anticoagulation therapy, whereas at the same time highlights patients at elevated bleeding risk for closer followup. Careful assessment and active management of bleeding risk factors may be warranted in all patients on NOACs who have high CHA2DS2-VASc or CHADS2 scores. Disclosures The authors have no conflicts of interest to disclose.
Arrhythmias and Conduction Disturbances/Risk Score for Bleeding on NOACs
1555
Figure 3. Calibration plots for bleeding risk scores.
Figure 4. Major bleeding event rates per 100 person-years in the present study versus stroke event rates per 100 person-years in derivation studies.
Supplementary Data Supplementary data associated with this article can be found, in the online version, https://doi.org/10.1016/ j.amjcard.2017.07.051. 1. 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;138:1093–1100. 2. Fang MC, Go AS, Chang Y, Borowsky LH, Pomernacki NK, Udaltsova N, Singer DE. A new risk scheme to predict warfarin-associated hemorrhage: the ATRIA (Anticoagulation and Risk Factors in Atrial Fibrillation) Study. J Am Coll Cardiol 2011;58:395–401. 3. O’Brien EC, Simon DN, Thomas LE, Hylek EM, Gersh BJ, Ansell JE, Kowey PR, Mahaffey KW, Chang P, Fonarow GC. The ORBIT bleeding score: a simple bedside score to assess bleeding risk in atrial fibrillation. Eur Heart J 2015;36:3258–3264. 4. Donzé J, Rodondi N, Waeber G, Monney P, Cornuz J, Aujesky D. Scores to predict major bleeding risk during oral anticoagulation therapy: a prospective validation study. Am J Med 2012;125:1095– 1102. 5. Angaran P, Dorian P, Tan MK, Kerr CR, Green MS, Gladstone DJ, Mitchell LB, Fournier C, Cox JL, Talajic M, Lin PJ, Langer A, Goldin L, Goodman SG. The risk stratification and stroke prevention therapy care gap in Canadian atrial fibrillation patients. Can J Cardiol 2016;32:336–343. 6. Poli D, Antonucci E, Marcucci R, Fatini C, Alterini B, Mannini L, Falciani M, Abbate R, Gensini GF, Prisco D. Risk of bleeding in very old atrial fibrillation patients on warfarin: relationship with ageing and CHADS 2 score. Thromb Res 2007;121:347–352. 7. Hylek EM, Evans-Molina C, Shea C, Henault LE, Regan S. Major hemorrhage and tolerability of warfarin in the first year of therapy among elderly patients with atrial fibrillation. Circulation 2007;115:2689– 2696.
8. Wallace PJ, Shah ND, Dennen T, Bleicher PA, Crown WH. Optum Labs: building a novel node in the learning health care system. Health Aff (Millwood) 2014;33:1187–1194. 9. Optum. Optum research data assets. 2015; Available at: https:// www.optum.com/content/dam/optum/resources/productSheets/ 5302_Data_Assets_Chart_Sheet_ISPOR.pdf. Accessed June 22, 2015. 10. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988;837–845. 11. Pencina MJ, D’Agostino RB, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med 2008;27:157–172. 12. Leening MJ, Vedder MM, Witteman JC, Pencina MJ, Steyerberg EW. Net reclassification improvement: computation, interpretation, and controversies: a literature review and clinician’s guide. Ann Intern Med 2014;160:122–131. 13. Gage BF, Waterman AD, Shannon W, Boechler M, Rich MW, Radford MJ. Validation of clinical classification schemes for predicting stroke: results from the National Registry of Atrial Fibrillation. JAMA 2001;285:2864–2870. 14. Lip GYH, Nieuwlaat R, Pisters R, Lane DA, Crijns HJGM. Refining clinical risk stratification for predicting stroke and thromboembolism in atrial fibrillation using a novel risk factor-based approach: the Euro Heart Survey on atrial fibrillation. Chest 2010;137:263–272. 15. Yao X, Abraham NS, Sangaralingham LR, Bellolio MF, McBane RD, Shah ND, Noseworthy PA. Effectiveness and safety of dabigatran, rivaroxaban, and apixaban versus warfarin in nonvalvular atrial fibrillation. J Am Heart Assoc 2016;5:e003725. 16. Gadsboll K, Staerk L, Fosbol EL, Sindet-Pedersen C, Gundlund A, Lip GY, Gislason GH, Olesen JB. Increased use of oral anticoagulants in patients with atrial fibrillation: temporal trends from 2005 to 2015 in Denmark. Eur Heart J 2017;38:899–906. 17. Apostolakis S, Lane DA, Buller H, Lip GY. Comparison of the CHADS2, CHA2DS2-VASc and HAS-BLED scores for the prediction of clinically relevant bleeding in anticoagulated patients with atrial fibrillation: the AMADEUS trial. Thromb Haemost 2013;110:1074–1079.
1556
The American Journal of Cardiology (www.ajconline.org)
18. Roldán V, Marín F, Manzano-Fernández S, Gallego P, Vílchez JA, Valdés M, Vicente V, Lip GY. The HAS-BLED score has better prediction accuracy for major bleeding than CHADS2 or CHA2DS2-VASc scores in anticoagulated patients with atrial fibrillation. J Am Coll Cardiol 2013;62:2199–2204. 19. Kirchhof P, Benussi S, Kotecha D, Ahlsson A, Atar D, Casadei B, Castella M, Diener HC, Heidbuchel H, Hendriks J, Hindricks G, Manolis AS, Oldgren J, Popescu BA, Schotten U, Van Putte B, Vardas P, Agewall S, Camm J, Baron Esquivias G, Budts W, Carerj S, Casselman F, Coca A, De Caterina R, Deftereos S, Dobrev D, Ferro JM, Filippatos G, Fitzsimons D, Gorenek B, Guenoun M, Hohnloser
SH, Kolh P, Lip GY, Manolis A, McMurray J, Ponikowski P, Rosenhek R, Ruschitzka F, Savelieva I, Sharma S, Suwalski P, Tamargo JL, Taylor CJ, Van Gelder IC, Voors AA, Windecker S, Zamorano JL, Zeppenfeld K. 2016 ESC guidelines for the management of atrial fibrillation developed in collaboration with EACTS. Europace 2016;18:1609– 1678. 20. Thigpen JL, Dillon C, Forster KB, Henault L, Quinn EK, Tripodis Y, Berger PB, Hylek EM, Limdi NA. Validity of international classification of disease codes to identify ischemic stroke and intracranial hemorrhage among individuals with associated diagnosis of atrial fibrillation. Circ Cardiovasc Qual Outcomes 2015;8:8–14.
a
Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery, Mayo Clinic, Rochester, Minnesota; bDivision of Health Care Policy and Research, Department of Health Sciences Research, Mayo Clinic, Rochester, Minnesota; cDepartment of Cardiovascular Medicine, Mayo Clinic, Rochester, Minnesota; dPredictive Analytics and Comparative Effectiveness (PACE) Center, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center/Tufts University School of Medicine, Boston, Massachusetts; eDepartment of Neurology, Tufts Medical Center/Tufts University School of Medicine, Boston, Massachusetts; fOptumLabs, Cambridge, Massachusetts; and gDivision of Gastroenterology and Hepatology, Department of Medicine, Mayo Clinic, Scottsdale, Arizona. Manuscript received June 2, 2017; revised manuscript received and accepted July 10, 2017. The work was performed at Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery, Mayo Clinic, Rochester, MN. This work was supported by Mayo Clinic Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery (Rochester, Minnesota). See page 1554 for disclosure information. *Corresponding author: Tel: 507 255 2446; fax: 507 255 2550. E-mail address: [email protected] (P.A. Noseworthy). 0002-9149/© 2017 Elsevier Inc. All rights reserved. https://doi.org/10.1016/j.amjcard.2017.07.051
The increasing adoption of NOACs for stroke prevention in AF necessitates a reassessment of bleeding risk scores. All the bleeding risk scores, except Outcomes Registry for Better Informed Treatment of Atrial Fibrillation (ORBIT), were developed and validated in patients receiving vitamin K antagonist.1–3 Little is known about how these scores perform in patients receiving NOACs. Regardless how the bleeding risk scores compare with each other, they all have at most modest ability to predict bleeding,4 and are not routinely used by clinicians.5 By contrast, stroke risk scores, such as the Congestive Heart failure, hypertension, Age ≥75 (doubled), Diabetes, Stroke (doubled), Vascular disease, Age 65–74, and Sex (female) (CHA2DS2-VASc) and Cardiac failure, Hypertension, Age, Diabetes, Stroke (Doubled) (CHADS2) scores, are recommended by guidelines and more commonly used in practice for stroke risk stratification.5 Because stroke and bleeding risks are closely related,6,7 we hypothesize that stroke risk scores could also be used to identify patients with high www.ajconline.org
1550
The American Journal of Cardiology (www.ajconline.org)
bleeding risks. Using a large heterogeneous cohort of patients with AF starting NOACs, we aimed to compare the performance of 2 stroke risk scores (CHA2DS2-VASc and CHADS2) and 3 bleeding risk scores (hypertension, abnormal renal/liver function [1 point each], stroke, bleeding history or predisposition, labile INR, elderly [.65 years], drugs/alcohol concomitantly [1 point each] [HAS-BLED], ORBIT, and AnTicoagulation and Risk factors In Atrial fibrillation [ATRIA]) in predicting major and intracranial bleeding. Methods The data source of this study is the OptumLabs Data Warehouse, a large administrative database that contains ≥100 million privately insured and Medicare Advantage enrollees of all ages and races and from all 50 states.8,9 We identified adult patients (≥18 years) with nonvalvular AF who started apixaban, dabigatran, edoxaban, or rivaroxaban between October 1, 2010 and June 30, 2015. Patients were followed until the earliest date of the following: end of study period (June 30, 2015), discontinuation of the medication, switch to another OAC, and the occurrence of major bleeding or stroke.
Our study was exempt by the Mayo Clinic Institutional Review Board for approval as we used only preexisting, deidentified data. Supplementary Table S1 presents an overview of the risk scores. The determination of cutoff points for risk categories (low, intermediate, and high) was mostly based on the original derivation studies of these scores. In a sensitivity analysis, we changed the cutoff points to make the event rates in each risk category similar across the risk scores (Figure 1). Cox proportional hazards regression was used to assess the relation between individual risk factors and major bleeding, and was also used to examine the prognostic value of the risk scores, with each score assessed separately. We evaluated the discrimination by comparing the C-statistics of the other 4 risk scores to CHA2DS2-VASc using the nonparametric method proposed by DeLong et al.10 We also calculated overall net reclassification improvement (NRI), a summary statistic for the incremental predictive value of each of the risk scores in comparison with CHA2DS2-VASc.11,12 In this NRI analysis, all the scores were analyzed as categorical variables. We evaluated calibration by plotting major bleeding events rates in the present study versus those previously published from
Figure 1. Proportions of patients and major bleeding event rates per 100 person-years in different risk strata and C-statistics for each risk score.
Arrhythmias and Conduction Disturbances/Risk Score for Bleeding on NOACs
the original derivation cohorts for HAS-BLED, ORBIT-AF, and ATRIA.1–3 We did not plot CHA2DS2-VASc or CHADS2, because they were derived for stroke risk prediction and do not have bleeding rates reported from derivation cohorts. For these 2 stroke risk scores, we plotted the event rates of major bleeding observed in this study versus the event rates of stroke in patients not treated with OAC in the derivation cohorts to illustrate the correlation between stroke and bleeding risks.13,14 We performed a number of sensitivity analyses to investigate the robustness of the findings. These sensitivity tests included changing the reference group from CHA2DS2-VASc to CHADS2, using the laboratory data instead of diagnosis codes, using Harrell’s c (an extension of C-statistic that applies to time-to-event data), considering death and stroke as competing risks, including bleeding events after discontinuation of treatment, assessing the risk at a fixed time point (at the end of 1 year), stratified analysis by different NOACs, by NOAC dosage, by the presence of vascular disease (to assess the impact of incomplete aspirin data because of the strong correlation between vascular disease and antiplatelet use), and whether patients had at least 2 prescription fills, and by age. Further details of the methods, including study population, ascertainment and validation of major bleeding, ascertainment of OAC treatment, statistical analysis, and sensitivity tests, are provided in the Appendix. Results We identified 39,539 patients with a total of 22,583 personyears of follow-up (mean follow-up 0.6 ± 0.7 year). Among these, 20%, 38%, 0.04%, and 43% started apixaban, dabigatran, edoxaban, and rivaroxaban, respectively (Table 1). We observed 665 major bleeding events (2.94 per 100 person-years), including 74 intracranial bleeding events (0.33 per 100 person years). The event rates and the proportions of patients categorized by the different risk scores are shown in Figure 1 and Table 2. For the prediction of major bleeding, CHA2DS2-VASc had the highest C-statistic both as a continuous score (C-statistic 0.68; p <0.001 compared with CHADS2 and p <0.01 compared with HAS-BLED, but not statistically significantly different in comparison with ORBIT or ATRIA) and as a categorical score (C-statistic 0.65; p <0.001 compared with ORBIT or ATRIA, but not statistically significantly different in comparison with CHADS2 or HAS-BLED). For the prediction of intracranial bleeding, CHADS2 had the highest C-statistic both as a continuous score (C-statistic 0.66; not statistically significant in comparison with any other score) and as a categorical score (C-statistic 0.66; p <0.05 compared with CHA2DS2-VASc, p <0.01 compared with ORBIT or ATRIA, not statistically significant in comparison with HAS-BLED) (Table 3 and Figure 2). We found ORBIT and HAS-BLED had better calibration than ATRIA (Figure 3), and stroke and bleeding risks are highly correlated (Figure 4). There was no statistically significant difference comparing any other score with CHA2DS2-VASc (Table 3), or CHADS2 based on the overall NRI (Supplementary Table S3). Compared with CHA2DS2-VASc, all other scores had negative event NRI, indicating lower sensitivity, but positive nonevent NRI, indicating higher specificity (Supplementary Table S4).
1551
All other sensitivity analyses showed similar results to the main analysis (Supplementary Tables S5 to S12). Discussion Using a large cohort of patients treated with NOACs in routine clinical practice, we demonstrate that CHA2DS2-VASc, CHADS2, HAS-BLED, ORBIT, and ATRIA may have similar performance in predicting major and intracranial bleeding. This observation underscores the fact that patients at high risk of stroke are also at high risk of bleeding. As such, every effort should be made to address modifiable risk factors for bleeding in all patients on anticoagulation treatment who have high CHA2DS2-VASc or CHADS2 scores. To the best of our knowledge, our study is the first study comparing these scores in patients treated with NOACs. A comparison in NOAC-treated patients is very important as NOACs, and vitamin K antagonists have different bleeding risk profile,15 and NOACs have become the first choice for many patients.16 Our study is unique as few previous studies had compared stroke and bleeding risk scores.17,18 We found the stroke risk scores had similar performance to the bleeding risk scores, likely because the stroke and bleeding risks factors overlap.6,7 In our study, the strongest predictor of major bleeding was age. CHA2DS2-VASc assigns 1 point to age 65 to 74 years, and 2 points to ≥75 years, and thus has better resolution to capture the increasing risk across the age spectrum. Because patients at high risk of stroke are also at high risk of bleeding, careful assessment and active management of bleeding risk factors may be warranted in all patients on NOACs who have high stroke risk scores, not just those with high bleeding risk scores. These findings are particularly important in light of the change in the 2016 European guidelines that no longer recommend calculating a specific bleeding risk score.19 Our study suggests that it appeared to be no incremental benefit of any bleeding score in comparison with stroke risk scores, and thus, provides support to this guideline change. As being emphasized by practice guidelines, a high bleeding risk score should generally not result in withholding oral anticoagulation.19 Rather, the focus should be on identifying modifiable risk factors for bleeding. Future efforts are needed to increase awareness of the collinear risk of stroke and bleeding, and motivate clinicians to prescribe oral anticoagulants in high-risk patients and address reversible risk factors. The strength of our study is the use of a large contemporary cohort of patients treated in diverse clinical practice settings. The bleeding risk scores are meant to apply to patients in clinical care, but many of the previous derivation and validation studies used patients in clinical trials or registries that were very different from those treated in everyday clinical practice. Moreover, the large number of events allows us to assess the predictive accuracy not only for major bleeding, but also for intracranial bleeding, the most feared complication of OAC therapy. Another strength is that using pharmacy claims data, we can better determine whether patients were continuously treated with NOACs. However, we acknowledge that prescription fills are an imperfect surrogate for adherence and persistence, because patients may not always take the medications filled. Our study has some limitations that warrant discussion. First, we rely on billing codes for patient characteristics and
1552
The American Journal of Cardiology (www.ajconline.org)
Table 1 Baseline characteristics of patients initiating NOACs, by presence or absence of major bleeding events during the on-treatment study period Characteristics
Age, years Median (IQR) 18–64 65–74 ≥75 Female Non-white Race Heart Failure Hypertension Diabetes Previous stroke/TIA Vascular disease Stage III or IV chronic kidney disease Abnormal liver function Previous major bleeding Anemia Alcoholism Antiplatelet use NSAIDs use Previous warfarin exposure CHA2DS2-VASc Median (IQR) 0,1 2,3 ≥4 CHADS2 Median (IQR) 0,1 2,3 ≥4 HAS-BLED Median (IQR) 0,1 2 ≥3 ORBIT Median (IQR) 0–2 3 ≥4 ATRIA Median (IQR) 0–3 4 ≥5
Total (n = 39,539)
Major Bleeding
Univariable Hazard Ratio (95% CI)
Multivariable Hazard Ratio (95% CI)
Ref 1.93*** (1.44, 2.58) 3.52*** (2.67, 4.64) 1.25** (1.06, 1.46) 1.15 (0.96, 1.38) 1.56*** (1.32, 1.83) 1.63** (1.15, 2.31) 1.03 (0.88, 1.21) 1.02 (0.84, 1.24) 1.43*** (1.21, 1.70) 1.30* (1.03, 1.64) 1.08 (0.73, 1.59) 1.61*** (1.30, 1.98) 1.41*** (1.19, 1.66) 1.74* (1.13, 2.69) 1.47** (1.15, 1.87) 1.24 (0.88, 1.74) 1.02 (0.87, 1.20)
No (n = 38,874)
Yes (n = 665)
71 (63–79) 11,863 (30%) 12,335 (31%) 15,341 (39%) 16,612 (42%) 7,775 (20%) 11,100 (28%) 33,825 (86%) 13,434 (34%) 5,467 (14%) 18,838 (48%) 2,784 (7%) 1,484 (4%) 3,406 (9%) 8,581 (22%) 1,048 (3%) 2,635 (7%) 1,910 (5%) 10,990 (28%)
71 (63–79) 11,798 (30%) 12,166 (31%) 14,910 (38%) 16,259 (42%) 7,619 (20%) 10,802 (28%) 33,195 (85%) 13,167 (34%) 5,339 (14%) 18,407 (47%) 2,698 (7%) 1,457 (4%) 3,294 (9%) 8,335 (21%) 1,026 (3%) 2,556 (7%) 1,875 (5%) 10,747 (28%)
78 (72–84) 65 (10%) 169 (25%) 431 (65%) 353 (53%) 156 (24%) 298 (45%) 630 (95%) 267 (40%) 128 (19%) 431 (65%) 86 (13%) 27 (4%) 112 (17%) 246 (37%) 22 (3%) 79 (12%) 35 (5%) 243 (37%)
Ref 2.33*** (1.75, 3.11) 4.94*** (3.80, 6.41) 1.55*** (1.33, 1.80) 1.31** (1.10, 1.57) 2.20*** (1.89, 2.56) 2.77*** (1.97, 3.90) 1.29** (1.10, 1.50) 1.46*** (1.20, 1.77) 2.10*** (1.79, 2.46) 2.19*** (1.75, 2.75) 1.16 (0.79, 1.71) 2.24*** (1.82, 2.74) 2.26*** (1.93, 2.65) 1.45 (0.95, 2.21) 1.89*** (1.49, 2.39) 1.10 (0.78, 1.54) 1.21* (1.03, 1.42)
4 (2–5) 5,010 (13%) 12,853 (33%) 21,676 (55%)
4 (2–5) 4,999 (13%) 12,744 (33%) 21,131 (54%)
5 (4–6) 11 (2%) 109 (16%) 545 (82%)
Ref 3.22*** (1.73, 5.98) 9.97*** (5.49, 18.10)
2 (1–3) 13,340 (34%) 20,428 (52%) 5,771 (15%)
2 (1–3) 13,250 (34%) 20,045 (52%) 5,579 (14%)
3 (2–4) 90 (14%) 383 (58%) 192 (29%)
Ref 2.63*** (2.09, 3.30) 4.96*** (3.86, 6.37)
2 (1–3) 10,503 (27%) 14,722 (37%) 14,314 (36%)
2 (1–3) 10,447 (27%) 14,501 (37%) 13,926 (36%)
3 (2–3) 56 (8%) 221 (33%) 388 (58%)
Ref 2.58*** (1.93, 3.46) 4.96*** (3.75, 6.56)
1 (0–2) 32,735 (83%) 4,194 (11%) 2,610 (7%)
1 (0–2) 32,312 (83%) 4,075 (11%) 2,487 (6%)
1 (1–3) 423 (64%) 119 (18%) 123 (19%)
Ref 2.22*** (1.81, 2.72) 4.07*** (3.33, 4.97)
2 (1–3) 30,393 (77%) 3,729 (9%) 5,417 (14%)
2 (1–3) 30,000 (77%) 3,665 (9%) 5,209 (13%)
3 (2–6) 393 (59%) 64 (10%) 208 (31%)
Ref 1.39* (1.07, 1.81) 3.14*** (2.66, 3.72)
* p <0.05, **p <0.01, ***p <0.001. CI = confidence interval; IQR = interquartile range; other abbreviations as in Table 1. Anemia was defined using diagnosis codes. In the sensitivity test using laboratory data, anemia was defined as abnormal hemoglobin (<13 mg/dl for males and <12 mg/dl for females) or hematocrit (<40% for males and <36% for females).
outcome adjudication, which may be prone to inaccuracies. However, the algorithms used to capture patient characteristics and outcomes have been validated and used in numerous previous studies,2,20 and demonstrated good performance in our internal validation of bleeding outcomes. Certain risk factors, such as alcoholism, may be undercoded in claims data— just as it tends to be underdiagnosed in the clinical setting—so the performance measured in this study may well reflect the
usefulness in clinical care. Second, we lack certain clinical characteristics, such as blood pressure, so we used the diagnosis of hypertension instead of uncontrolled hypertension for HAS-BLED. We included patients with all types of AF, but it is difficult to use claims data to ascertain AF type, burden, or time since the first AF diagnoses. However, these characteristics are not included in any of the risk stratification schemes; thus, lack of these data did not affect our findings. Third, we
Arrhythmias and Conduction Disturbances/Risk Score for Bleeding on NOACs
1553
Table 2 Observed major and intracranial bleeding rates by risk score Patients
PY
n (%) Total CHA2DS2-VASc 0 1 2 3 4 5 6 7 8–9 Risk Low (0–1) Intermediate (2–3) High (≥4) CHADS2 0 1 2 3 4 5 6 Risk Low (0–1) Intermediate (2–3) High (≥4) HAS-BLED 0 1 2 3 4 5 6–8 Risk Low (0–1) Intermediate (2) High (≥3) ORBIT 0 1 2 3 4 5 6–7 Risk Low (0–2) Intermediate (3) High (≥4) ATRIA 0 1 2 3 4 5 6 7 8–10 Risk Low (0–3) Intermediate (4) High (≥5) PY = person-years.
Major bleeding
Intracranial bleeding
Events
Events/100 PY
Events
Events/100 PY
39,539 (100%)
22,583
665
2.94
74
0.33
1,497 (4%) 3,513 (9%) 5,576 (14%) 7,277 (18%) 8,256 (21%) 6,504 (16%) 3,953 (10%) 1,953 (5%) 1,010 (3%)
600 1,813 3,382 4,424 4,932 3,723 2,168 1,068 472
1 10 36 73 168 140 122 68 47
0.17 0.55 1.06 1.65 3.41 3.76 5.63 6.37 9.96
0 2 4 12 17 14 12 8 5
0.00 0.11 0.12 0.27 0.34 0.38 0.55 0.75 1.06
5,010 (13%) 12,853 (33%) 21,676 (55%)
2,413 7,806 12,363
11 109 545
0.46 1.40 4.41
2 16 56
0.08 0.20 0.45
3,251 (8%) 10,089 (26%) 12,422 (31%) 8,006 (20%) 3,660 (9%) 1,623 (4%) 488 (1%)
1,484 5,896 7,495 4,558 2,042 885 223
6 84 208 175 120 55 17
0.40 1.42 2.78 3.84 5.88 6.21 7.63
0 10 24 13 18 8 1
0.00 0.17 0.32 0.29 0.88 0.90 0.45
13,340 (34%) 20,428 (52%) 5,771 (15%)
7,379 12,053 3,150
90 383 192
1.22 3.18 6.09
10 37 27
0.14 0.31 0.86
2,346 (6%) 8,157 (21%) 14,722 (37%) 9,595 (24%) 3,750 (10%) 864 (2%) 105 (0.3%)
1,069 4,653 8,896 5,540 1,970 410 45
4 52 221 222 120 39 7
0.37 1.12 2.48 4.01 6.09 9.52 15.58
1 8 21 28 11 4 1
0.09 0.17 0.24 0.51 0.56 0.98 2.23
10,503 (27%) 14,722 (37%) 14,314 (36%)
5,723 8,896 7,964
56 221 388
0.98 2.48 4.87
9 21 44
0.16 0.24 0.55
17,469 (44%) 10,182 (26%) 5,084 (13%) 4,194 (11%) 1,519 (4%) 840 (2%) 251 (1%)
10,166 6,003 2,727 2,391 775 413 107
120 222 81 119 59 50 14
1.18 3.70 2.97 4.98 7.61 12.09 13.03
20 22 11 14 4 2 1
0.20 0.37 0.40 0.59 0.52 0.48 0.93
32,735 (83%) 4,194 (11%) 2,610 (7%)
18,895 2,391 1,296
423 119 123
2.24 4.98 9.49
53 14 7
0.28 0.59 0.54
3,899 (10%) 15,350 (39%) 1,964 (5%) 9,180 (23%) 3,729 (9%) 850 (2%) 3,520 (9%) 872 (2%) 175 (0.4%)
1,948 9,255 1,147 5,295 2,029 463 1,967 417 62
11 138 27 217 64 30 118 52 8
0.56 1.49 2.35 4.10 3.15 6.47 6.00 12.48 13.00
1 20 1 27 8 2 12 2 1
0.05 0.22 0.09 0.51 0.39 0.43 0.61 0.48 1.62
30,393 (77%) 3,729 (9%) 5,417 (14%)
17,645 2,029 2,908
393 64 208
2.23 3.15 7.15
49 8 17
0.28 0.39 0.58
1554
The American Journal of Cardiology (www.ajconline.org)
Table 3 C-statistic and net reclassification improvement in comparison with CHA2DS2-VASc (n = 39,539) Major bleeding
CHA2DS2-VASc CHADS2 HAS-BLED ORBIT ATRIA
Intracranial bleeding
C-statistic (95% CI) for continuous scores
C-statistic (95% CI) for categorical scores
NRI for categorical scores
C-statistic (95% CI) for continuous scores
C-statistic (95% CI) for categorical scores
NRI for categorical scores
0.68 (0.66, 0.70) 0.65*** (0.63, 0.67) 0.66** (0.64, 0.67) 0.66 (0.64, 0.68) 0.67 (0.65, 0.69)
0.65 (0.63, 0.66) 0.64 (0.62, 0.65) 0.64 (0.62, 0.66) 0.60*** (0.58, 0.62) 0.60*** (0.58, 0.62)
Ref −0.04 0.02 0.01 0.05
0.65 (0.59, 0.71) 0.66 (0.60, 0.72) 0.64 (0.58, 0.70) 0.60 (0.54, 0.66) 0.63 (0.57, 0.68)
0.61 (0.57, 0.66) 0.66* (0.60, 0.72) 0.63 (0.58, 0.69) 0.55 (0.50, 0.61) 0.56 (0.50, 0.61)
Ref 0.09 0.07 −0.06 −0.04
CI = confidence interval; NRI = net reclassification improvement. * p <0.05, **p <0.01, ***p <0.001 for each score compared with CHA2DS2-VASc.
Figure 2. Receiver operating characteristic curves of the risk scores.
were not able to accurately assess aspirin and nonsteroidal anti-inflammatory drugs use, because they generally do not need a prescription. Patients may use aspirin or nonsteroidal anti-inflammatory drugs for a short period or as needed, which makes it difficult to assess in any studies. We conducted numerous sensitivity tests to evaluate whether these limitations could affect our findings. Such sensitivity tests include using laboratory data instead of diagnosis codes to assess renal function and anemia, and stratified analysis by vascular disease to assess to impact of incomplete aspirin data. The results of these sensitivity analyses support the robustness of our findings. The 2 stroke risk scores (CHA2DS2-VASc and CHADS2) and 3 bleeding risk scores (HAS-BLED, ORBIT, and ATRIA)
had similar, albeit modest, performance in predicting major bleeding and intracranial bleeding in patients with AF treated with NOACs. A single stroke risk score appears to be a convenient tool in identifying low-risk patients who may not require anticoagulation therapy, whereas at the same time highlights patients at elevated bleeding risk for closer followup. Careful assessment and active management of bleeding risk factors may be warranted in all patients on NOACs who have high CHA2DS2-VASc or CHADS2 scores. Disclosures The authors have no conflicts of interest to disclose.
Arrhythmias and Conduction Disturbances/Risk Score for Bleeding on NOACs
1555
Figure 3. Calibration plots for bleeding risk scores.
Figure 4. Major bleeding event rates per 100 person-years in the present study versus stroke event rates per 100 person-years in derivation studies.
Supplementary Data Supplementary data associated with this article can be found, in the online version, https://doi.org/10.1016/ j.amjcard.2017.07.051. 1. 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;138:1093–1100. 2. Fang MC, Go AS, Chang Y, Borowsky LH, Pomernacki NK, Udaltsova N, Singer DE. A new risk scheme to predict warfarin-associated hemorrhage: the ATRIA (Anticoagulation and Risk Factors in Atrial Fibrillation) Study. J Am Coll Cardiol 2011;58:395–401. 3. O’Brien EC, Simon DN, Thomas LE, Hylek EM, Gersh BJ, Ansell JE, Kowey PR, Mahaffey KW, Chang P, Fonarow GC. The ORBIT bleeding score: a simple bedside score to assess bleeding risk in atrial fibrillation. Eur Heart J 2015;36:3258–3264. 4. Donzé J, Rodondi N, Waeber G, Monney P, Cornuz J, Aujesky D. Scores to predict major bleeding risk during oral anticoagulation therapy: a prospective validation study. Am J Med 2012;125:1095– 1102. 5. Angaran P, Dorian P, Tan MK, Kerr CR, Green MS, Gladstone DJ, Mitchell LB, Fournier C, Cox JL, Talajic M, Lin PJ, Langer A, Goldin L, Goodman SG. The risk stratification and stroke prevention therapy care gap in Canadian atrial fibrillation patients. Can J Cardiol 2016;32:336–343. 6. Poli D, Antonucci E, Marcucci R, Fatini C, Alterini B, Mannini L, Falciani M, Abbate R, Gensini GF, Prisco D. Risk of bleeding in very old atrial fibrillation patients on warfarin: relationship with ageing and CHADS 2 score. Thromb Res 2007;121:347–352. 7. Hylek EM, Evans-Molina C, Shea C, Henault LE, Regan S. Major hemorrhage and tolerability of warfarin in the first year of therapy among elderly patients with atrial fibrillation. Circulation 2007;115:2689– 2696.
8. Wallace PJ, Shah ND, Dennen T, Bleicher PA, Crown WH. Optum Labs: building a novel node in the learning health care system. Health Aff (Millwood) 2014;33:1187–1194. 9. Optum. Optum research data assets. 2015; Available at: https:// www.optum.com/content/dam/optum/resources/productSheets/ 5302_Data_Assets_Chart_Sheet_ISPOR.pdf. Accessed June 22, 2015. 10. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988;837–845. 11. Pencina MJ, D’Agostino RB, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med 2008;27:157–172. 12. Leening MJ, Vedder MM, Witteman JC, Pencina MJ, Steyerberg EW. Net reclassification improvement: computation, interpretation, and controversies: a literature review and clinician’s guide. Ann Intern Med 2014;160:122–131. 13. Gage BF, Waterman AD, Shannon W, Boechler M, Rich MW, Radford MJ. Validation of clinical classification schemes for predicting stroke: results from the National Registry of Atrial Fibrillation. JAMA 2001;285:2864–2870. 14. Lip GYH, Nieuwlaat R, Pisters R, Lane DA, Crijns HJGM. Refining clinical risk stratification for predicting stroke and thromboembolism in atrial fibrillation using a novel risk factor-based approach: the Euro Heart Survey on atrial fibrillation. Chest 2010;137:263–272. 15. Yao X, Abraham NS, Sangaralingham LR, Bellolio MF, McBane RD, Shah ND, Noseworthy PA. Effectiveness and safety of dabigatran, rivaroxaban, and apixaban versus warfarin in nonvalvular atrial fibrillation. J Am Heart Assoc 2016;5:e003725. 16. Gadsboll K, Staerk L, Fosbol EL, Sindet-Pedersen C, Gundlund A, Lip GY, Gislason GH, Olesen JB. Increased use of oral anticoagulants in patients with atrial fibrillation: temporal trends from 2005 to 2015 in Denmark. Eur Heart J 2017;38:899–906. 17. Apostolakis S, Lane DA, Buller H, Lip GY. Comparison of the CHADS2, CHA2DS2-VASc and HAS-BLED scores for the prediction of clinically relevant bleeding in anticoagulated patients with atrial fibrillation: the AMADEUS trial. Thromb Haemost 2013;110:1074–1079.
1556
The American Journal of Cardiology (www.ajconline.org)
18. Roldán V, Marín F, Manzano-Fernández S, Gallego P, Vílchez JA, Valdés M, Vicente V, Lip GY. The HAS-BLED score has better prediction accuracy for major bleeding than CHADS2 or CHA2DS2-VASc scores in anticoagulated patients with atrial fibrillation. J Am Coll Cardiol 2013;62:2199–2204. 19. Kirchhof P, Benussi S, Kotecha D, Ahlsson A, Atar D, Casadei B, Castella M, Diener HC, Heidbuchel H, Hendriks J, Hindricks G, Manolis AS, Oldgren J, Popescu BA, Schotten U, Van Putte B, Vardas P, Agewall S, Camm J, Baron Esquivias G, Budts W, Carerj S, Casselman F, Coca A, De Caterina R, Deftereos S, Dobrev D, Ferro JM, Filippatos G, Fitzsimons D, Gorenek B, Guenoun M, Hohnloser
SH, Kolh P, Lip GY, Manolis A, McMurray J, Ponikowski P, Rosenhek R, Ruschitzka F, Savelieva I, Sharma S, Suwalski P, Tamargo JL, Taylor CJ, Van Gelder IC, Voors AA, Windecker S, Zamorano JL, Zeppenfeld K. 2016 ESC guidelines for the management of atrial fibrillation developed in collaboration with EACTS. Europace 2016;18:1609– 1678. 20. Thigpen JL, Dillon C, Forster KB, Henault L, Quinn EK, Tripodis Y, Berger PB, Hylek EM, Limdi NA. Validity of international classification of disease codes to identify ischemic stroke and intracranial hemorrhage among individuals with associated diagnosis of atrial fibrillation. Circ Cardiovasc Qual Outcomes 2015;8:8–14.