CLINICAL RESEARCH STUDY
Stroke and Bleeding Risk Co-distribution in Real-world Patients with Atrial Fibrillation: The Euro Heart Survey Maura Marcucci, MD,a,b Gregory Y.H. Lip, MD,c Robby Nieuwlaat, PhD,a,d Ron Pisters, PhD,e Harry J.G.M. Crijns, MD,e Alfonso Iorio, MDa a Department of Clinical Epidemiology & Biostatistics, McMaster University, Hamilton, Ontario, Canada; bFoundation IRCCS Ca’ Granda e Ospedale Maggiore Policlinico, Geriatrics & Department of Clinical Sciences and Community Health, University of Milan, Milan, Italy; cUniversity of Birmingham Centre for Cardiovascular Sciences, City Hospital, Birmingham, United Kingdom; dPopulation Health Research Institute, Hamilton, Ontario, Canada; eDepartment of Cardiology, Maastricht University Medical Centre, Maastricht, The Netherlands.
ABSTRACT BACKGROUND: The choice to recommend antithrombotic therapy to patients with atrial fibrillation should rely on cardioembolic and bleeding risk stratification. Sharing some risk factors, schemes to predict thrombotic and bleeding risk are expected not to be independent, yet the degree of their association has never been clearly quantified. METHODS: We described the cardioembolic (Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack [CHADS2]/Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females [CHA2DS2-VASc]) and bleeding risk (Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly [HAS-BLED]) co-distribution among patients of the Euro Heart Survey on atrial fibrillation. We measured the within-patient correlation (Spearman) and concordance between the 2 types of score and score-based risk categorization (low, intermediate, high). The score-based predicted risk co-classification was then related to the observed 1-year stroke and bleeding occurrence. RESULTS: In 3920 patients, we found a between-scores correlation of 0.416 (P < .001) between HAS-BLED and CHADS2, and 0.512 (P < .001) between HAS-BLED and CHA2DS2-VASc. In 89% (CHADS2/ HAS-BLED) and 97% (CHA2DS2-VASc/HAS-BLED) of patients, the bleeding risk category was equal to or lower than their cardioembolic risk category (P < .001 for symmetry test). A complete concordance between risk categories was found in 39.6% (CHADS2/HAS-BLED) and 21.7% (CHA2DS2-VASc/HAS-BLED) of patients; 4.4% (CHADS2/HAS-BLED) and 7.7% (CHA2DS2-VASc/HAS-BLED) of patients had high cardioembolic risk/low bleeding risk or vice versa. A tendency for an increasing frequency of stroke was observed for increasing bleeding risk within cardioembolic risk categories and vice versa. CONCLUSIONS: In a real-world population with atrial fibrillation, we confirmed that the cardioembolic and bleeding risk classifications are correlated but not exchangeable. It is then worth verifying the advantages of a strategy adopting a combined risk assessment over a strategy relying only on the cardioembolic risk evaluation. Ó 2014 Elsevier Inc. All rights reserved. The American Journal of Medicine (2014) 127, 979-986 KEYWORDS: Atrial fibrillation; Cardioembolic and bleeding risk co-stratification
Funding: See last page of article. Conflict of Interest: See last page of article. Authorship: See last page of article. Requests for reprints should be addressed to Maura Marcucci, MD, Foundation IRCCS Ca’ Granda e Ospedale Maggiore Policlinico, 0002-9343/$ -see front matter Ó 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.amjmed.2014.05.003
Geriatrics & Department of Clinical Sciences and Community Health, University of Milan, via Pace 9, 20122 Milan, Italy. E-mail address: [email protected]
980
The American Journal of Medicine, Vol 127, No 10, October 2014
Patients with atrial fibrillation may have different benefits measured the within-patient correlation and concordance from long-term anticoagulation according to the balance of between the 2 types of score and score-based risk categotheir baseline risk for stroke and bleeding. With the availrization. Third, we related outcomes of stroke and bleeding ability of newer anticoagulants, the choice might be not only to the score-based risk co-classification. to treat or not to treat, but also which treatment choice better fits the risk profile. METHODS Strategies for treatment individualization, based on a trade-off Study Population CLINICAL SIGNIFICANCE between the treatment-related inWe analyzed the Euro Heart Surdividual benefit and harm, are When deciding on antithrombotic thervey (EHS) on atrial fibrillation facilitated by the availability of database, which included data apy in patients with atrial fibrillation, clinical tools to predict the patient on 5333 inpatients or outpatients, the usefulness of a joint assessment of risk for the target and adverse aged 18 years, who were repatient cardioembolic and bleeding risks event. The Congestive heart failferred to 182 university, nonunimight be limited by the interdependence ure, Hypertension, Age >75, versity, and specialized hospitals between the 2 risks. Diabetes mellitus, and prior Stroke among 35 member countries of the or transient ischemic attack The degree of correlation and concorEuropean Society of Cardiology, (CHADS2)1 and the Congestive with an electrocardiogram or dance between validated cardioembolic heart failure, Hypertension, Age Holter-proven diagnosis of atrial and bleeding risk schemes has never >75, Diabetes mellitus, and prior fibrillation during the qualifying been explicitly described. Stroke or transient ischemic attack, admission or in the preceding In a real-world population, we confirmed Vascular disease, Age 65-75, Sex year.17 Paper-based medical recategory i.e. females (CHA2DS2that the available cardioembolic and cords or data from medical inVASc)2 scores are 2 clinical risk bleeding risk classifications are signififormation systems were used to factor-based schemes to predict cantly correlated but not exchangeable. populate the database. We inthe risk of stroke in patients with cluded in the present analysis only atrial fibrillation, included in the the EHS patients with nonvalvular guidelines for the antithrombotic therapy of major scientific atrial fibrillation (ie, defined by the absence of a mitral valve societies.3-5 The CHA2DS2-VASc was developed to better stenosis or valvular surgery) and with data available for the identify patients “truly at low risk” of stroke,6 and this score calculation of the risk scores. is recommended in the European Society of Cardiology3 and Asia Pacific Heart Rhythm Society guidelines.7 Several scores for the assessment of bleeding risk in Risk Scores patients with atrial fibrillation are also available in the The CHADS2, CHA2DS2-VASc, and HAS-BLED scores literature.8-11 The Hypertension, Abnormal renal/liver were calculated retrospectively for each patient using function, Stroke, Bleeding history or predisposition, Labile the data collected at discharge from the hospital or at the end international normalized ratio, Elderly (>65 years), Drugs/ of the index outpatient visit. A reduced version of the alcohol concomitantly (HAS-BLED) score was shown to HAS-BLED score was used because of the unavailability have a good predictive ability both in patients taking and in of data for liver dysfunction and labile international patients not taking oral anticoagulants,12 and its use is recnormalized ratio. ommended in the most recent European and Canadian guidelines.3,4 Statistical Analysis Schemes for thrombotic and bleeding risk assessment have some risk factors in common, and there is evidence Both the CHADS2/HAS-BLED and the CHA2DS2-VASc/ that these scores are associated with both the risk of stroke HAS-BLED combinations of scores were considered. The and the risk of bleeding.13-16 Whether and to what extent the co-classification of the study population according to each of those combinations was described, using the raw scores joint use of a score for the cardioembolic risk and a score for or the score-based risk categories as defined in the literthe bleeding risk offers an advantage in making clinical ature2,18 (for CHADS2 and CHA2DS2-VASc: low risk ¼ decisions in clinical practice, in terms of risk stratification and treatment individualization, over a strategy relying only score 0; intermediate risk ¼ score 1; high risk ¼ score 2; on the cardioembolic risk assessment are not definitely clear. for HAS-BLED: low risk ¼ score 0; intermediate risk ¼ Surely it is affected by the co-distribution of the 2 scores, score 1-2; high risk ¼ score 3). Cross tabulation and which needs to be assessed in a “real-word” population. box-plots were used to represent the co-distribution. With these objectives, we first described how a populaA Spearman rank correlation coefficient was calculated tion of patients with atrial fibrillation who were referred to to express the extent of association/dependence between the the hospital were classified by a combined use of a score for patient cardioembolic and bleeding risk scores. We then cardioembolic risk and a score for bleeding risk. Second, we used 2 measures of concordance/exchangeability between
Marcucci et al
Joint Risk Distribution in Atrial Fibrillation
the 2 risk categorizations (high, intermediate, and low): a) the percentage agreement, that is, percentage of patients classified into the same cardioembolic and bleeding risk category; and b) a weighted percentage agreement, also taking into account the “partial concordance,” assigning a weight of 2 for perfect agreement, 1 for a 1-category disagreement, and 0 for 2-category disagreement. Symmetry and marginal homogeneity (Stuart-Maxwell) tests also were performed to determine, in case of discordance, whether there was a statistically significant tendency for the cardioembolic risk category to be higher than the bleeding risk category or vice versa.
Secondary Analyses Score co-distribution by antithrombotic therapy. We described the score co-distribution in subgroups of EHS patients defined according to the antithrombotic therapy they were prescribed at the time of the discharge from the hospital or at the end of the index outpatient visit, qualitatively comparing the degree of correlation in each group. The following subgroups of patients were defined: patients not receiving any antithrombotic treatment, patients receiving an antiplatelet agent, and patients receiving a vitamin K antagonist (including patients receiving both a vitamin K antagonist and an antiplatelet agent). Patients receiving heparin were excluded from the current analyses. Our hypothesis for this secondary analysis was that the distribution of each risk score would be different in each bytreatment defined group (according to the KruskaleWallis rank-test), but that the level of within-patient correlations between the 2 types of score would be similar in each treatment group as in the whole population. Score co-distribution and outcome. We explored whether there was any association between the score co-distribution and the outcome. We did that by investigating whether, for each cardioembolic risk category, there was a trend in the occurrence of adverse events according to the bleeding risk category and vice versa. Also, we investigated whether, regardless of the level of risk, the concordance/discordance between cardioembolic and bleeding risk categories was associated with a different outcome (considering stroke, bleeding, or any); logistic regression analysis was used for this purpose. All analyses were stratified by or adjusted for antithrombotic treatment. Definitions used for the outcomes in the EHS on atrial fibrillation have been published.2
RESULTS Of the 5272 patients with atrial fibrillation in the EHS of atrial fibrillation who were discharged alive,17 3920 without mitral valve stenosis or valvular surgery and with necessary information to calculate the risk scores were included in this analysis. Table 1 shows the relevant characteristics of the study population.
981 Table 1 Characteristics of the 3920* Patients with Atrial Fibrillation from the Euro Heart Survey Included in the Current Analyses Characteristic
Description
Age, median (Q1, Q3) Women, n (%) CHADS2, median (Q1, Q3) CHA2DS2-VASc, median (Q1, Q3) HAS-BLED, median (Q1, Q3) Drugs prescribed at discharge, n (%) Warfarin Aspirin No antithrombotic therapy Stroke at 1 y of follow-up, n (%) Any patient In patients on warfarin In patients on aspirin In patients off antithrombotic therapy Major bleeding at 1 y of follow-up, n (%) Any patient In patients on warfarin In patients on aspirin In patients off antithrombotic therapy Stroke and major bleeding at 1 y of follow-up, n (%) Any patient In patients on warfarin In patients on aspirin In patients off antithrombotic therapy
68 1587 1 3 1
(59, 75) (40.5) (1, 2) (2, 4) (1, 2)
2476 (63.2) 1079 (27.5) 365 (9.3) 47 27 19 1
(1.6) (1.4) (2.5) (0.4)
44 35 8 1
(1.5) (1.8) (1.0) (0.4)
4 4 0 0
(0.1) (0.2) (0) (0)
CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; CHA2DS2-VASc ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/ liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly; Q1 ¼ first quartile; Q3 ¼ third quartile. *Data on outcome at 1 year of follow-up available for 2934 of 3920 patients.
Table 2 and Figure 1A describe the co-distribution of CHADS2 and HAS-BLED; Table 3 and Figure 1C describe the co-distribution of CHA2DS2-VASc and HASBLED. The Spearman correlation coefficient was 0.416 (P < .001) between CHADS2 and HAS-BLED, and 0.512 (P < .001) between CHA2DS2-VASc and HAS-BLED. Those co-distributions translated into the joint scorebased risk categorizations shown in Tables 4 and 5, and Figure 1B and D. The most prevalent risk group was the one at high cardioembolic risk and intermediate bleeding risk (37.5% if CHADS2, 60.8% if CHA2DS2-VASc was used), with 89% (CHADS2/HAS-BLED) and 97% (CHA2DS2-VASc/HAS-BLED) of patients classified into a bleeding risk category equal to or lower than their cardioembolic risk category (P < .001 for symmetry and marginal homogeneity tests). With the use of the CHADS2 and HAS-BLED, 39.6% of patients were classified into
982 Table 2
The American Journal of Medicine, Vol 127, No 10, October 2014 CHADS2 and HAS-BLED Scores Co-distribution HAS-BLED
n (% of Total) CHADS2
0
0 1 2 3 4 5 6 Any CHADS2
331 314 129 28 4 0 0 806
1 (8.44) (8.01) (3.29) (0.71) (0.10) (0.00) (0.00) (20.56)
264 641 472 213 71 17 5 1683
2 (6.73) (16.35) (12.04) (5.43) (1.81) (0.43) (0.13) (42.93)
83 318 369 196 77 42 8 1093
3 (2.12) (8.11) (9.41) (5.00) (1.96) (1.07) (0.20) (27.88)
6 64 83 60 52 25 6 296
4 (0.15) (1.63) (2.12) (1.53) (1.33) (0.64) (0.15) (7.55)
0 5 7 5 10 9 3 39
5 (0.00) (0.13) (0.18) (0.13) (0.26) (0.23) (0.08) (0.99)
0 0 0 0 1 2 0 3
Any HAS-BLED (0.00) (0.00) (0.00) (0.00) (0.03) (0.05) (0.00) (0.08)
684 (17.45) 1342 (34.23) 1060 (27.04) 502 (12.81) 215 (5.48) 95 (2.42) 22 0.56 3920
CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
the same cardioembolic and bleeding risk category, whereas 21.7% was the agreement between CHA2DS2-VASc and HAS-BLED; 16.3% were classified into the same cardioembolic and bleeding risk category by both the
CHADS2/HAS-BLED and the CHA2DS2-VASc/HASBLED combination. By also considering the partial agreement, the weighted percentage agreement was 67.7% and 57.0%, respectively.
Figure 1 Cardioembolic and bleeding risk scores co-distribution. A, CHADS2 and HAS-BLED scores. B, CHADS2 and HAS-BLED risk categories. C, CHA2DS2-VASc and HAS-BLED scores. D, CHA2DS2-VASc and HAS-BLED risk categories. Red line: median of HAS-BLED scores. CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; CHA2DS2-VASc ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
Marcucci et al Table 3
Joint Risk Distribution in Atrial Fibrillation
983
CHA2DS2-VASc and HAS-BLED Scores Co-distribution HAS-BLED
n (% of Total) CHA2DS2-VASc
0
0 1 2 3 4 5 6 7 8 9 Any CHA2DS2-VASc
218 284 168 104 26 5 1 0 0 0 806
1 (5.56) (7.24) (4.29) (2.65) (0.66) (0.13) (0.03) (0.00) (0.00) (0.00) (20.56)
83 239 356 390 318 187 78 20 9 3 1683
2 (2.12) (6.10) (9.08) (9.95) (8.11) (4.77) (1.99) (0.51) (0.23) (0.08) (42.93)
8 64 139 269 273 170 101 49 18 2 1093
3 (0.20) (1.63) (3.55) (6.86) (6.96) (4.34) (2.58) (1.25) (0.46) (0.05) (27.88)
0 7 25 43 73 61 40 33 12 2 296
4 (0.00) (0.18) (0.64) (1.10) (1.86) (1.56) (1.02) (0.84) (0.31) (0.05) (7.55)
0 0 2 3 8 3 8 10 4 1 39
5 (0.00) (0.00) (0.05) (0.08) (0.20) (0.08) (0.20) (0.26) (0.10) (0.03) (0.99)
0 0 0 0 0 0 1 0 2 0 3
Any HAS-BLED (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.03) (0.00) (0.05) (0.00) (0.08)
309 594 690 809 698 426 229 112 45 8 3920
(7.88) (15.15) (17.60) (20.64) (17.81) (10.87) (5.84) (2.86) (1.15) (0.20)
CHA2DS2-VASc ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
Score Co-distribution and Antithrombotic Therapy Table 1 describes the frequencies and modalities of antithrombotic therapy prescribed at discharge. As expected, the distribution of each individual risk score significantly differed among the 3 by-treatment groups (P < .001). When looking at the score co-distribution, positive Spearman correlation coefficients were obtained in each group similar to those found in the whole population (Supplementary Table 1). The score co-distribution across by-treatment groups is described in Supplementary Figure 1.
category. Likewise, a tendency for an increasing frequency of events in accordance with an increasing cardioembolic risk score was found within the intermediate and high bleeding risk categories. The same results for the entire population are provided in Supplementary Table 2. Finally, a trend of an increasing frequency of stroke for a stronger concordance between cardioembolic and bleeding risk category was observed (Table 7), before and after accounting for the antithrombotic therapy, even if without reaching a statistical significance. The trend was not confirmed for the occurrence of major bleedings.
Score Co-distribution and Outcome
DISCUSSION
Data on 1-year outcome were available for 2934 of 3920 patients (74.8%). The score co-distribution and correlation in these patients resembled those of the initial population. Table 6 shows, among patients on warfarin, the tendency for an increasing frequency of events (stroke or bleedings, or both) as a function of an increasing bleeding risk score, mainly evident within the high cardioembolic risk
We described the joint distribution of scores for the cardioembolic risk (CHADS2 or CHA2DS2-VASc) and scores for the bleeding risk (HAS-BLED) in patients with atrial fibrillation. We found a positive within-patient correlation of moderate strength between the 2 types of scores (w0.5) in the whole population and in subgroups defined by the prescribed antithrombotic therapy. A perfect concordance
Table 4
CHADS2 and HAS-BLED Co-distribution According to the Score-based Risk Categories HAS-BLED Risk Category n (% of the Total in Each CHADS2 Risk Category)
CHADS2 Risk Category
Low Risk
Intermediate Risk
High Risk
n (% of the Total)
Low risk Intermediate risk High risk n (% of the total)
331 314 161 806
347 959 1470 2776
6 69 263 338
684 (17.5) 1342 (34.2) 1894 (48.3) 3920
(48.4) (23.4) (8.5) (20.6)
(50.7) (71.5) (77.6) (70.8)
(0.9) (5.1) (13.9) (8.6)
CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
984 Table 5
The American Journal of Medicine, Vol 127, No 10, October 2014 CHA2DS2-VASc and HAS-BLED Co-distribution According to the Score-based Risk Categories HAS-BLED Risk Category n (% of the Total in Each CHA2DS2-VASc Risk Category)
CHA2DS2-VASc Risk Category
Low Risk
Intermediate Risk
High Risk
n (% of Total)
Low risk Intermediate risk High risk n (% of total)
218 284 304 806
91 303 2382 2776
0 7 331 338
309 (7.9) 594 (15.1) 3017 (77.0) 3920
(70.5) (47.8) (10.1) (20.6)
(29.5) (51.0) (78.9) (70.8)
(0.0) (1.2) (11.0) (8.6)
CHA2DS2-VASc ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
between cardioembolic and bleeding risk category was found in less than 50% of patients, but only 4.4% (CHADS2/HAS-BLED) to 7.7% (CHA2DS2-VASc/HASBLED) of patients were in the cells corresponding to a full discordance between the cardioembolic and bleeding risk categories (ie, high cardioembolic risk/low bleeding risk or vice versa). An increasing bleeding risk score was associated with an increasing frequency of stroke events within the intermediate and high cardioembolic risk score categories; the trend was less evident for bleeding events. A higher degree of within-patient concordance between cardioembolic and bleeding risk categories appeared to be associated with a higher frequency of stroke. The current literature has been focusing on verifying the predictive ability of proposed cardioembolic and bleeding
Table 6
risk scores in real-word atrial fibrillation cohorts, but scarce attention has been paid to addressing the question of how those scores co-distribute and correlate, and so how the assumption of their independence is justifiable and their separate use necessary. For example, Friberg et al14 recently used the score-based co-stratification to assess the net benefit with warfarin associated with different CHADS2 (or CHA2DS2-VASc) and HAS-BLED scores in a large Swedish atrial fibrillation population; their cross-tabulations resemble ours (Tables 2 and 3), with empty cells at the top right and bottom left corners, and the highest concentration in the cells along the diagonal. Yet, they did not focus on how those strata were differently populated. Lopes et al19 reported the cardioembolic and bleeding score-based categories co-distribution among the participants of the
Cardioembolic and Bleeding Risk Categories Co-distribution and Outcomes: Patients Prescribed Warfarin
Cardioembolic Risk Category
HAS-BLED Risk Category
% of Patients Experiencing a Stroke
% of Patients Experiencing a Major Bleeding
% of Patients Experiencing a Stroke or Major Bleeding
CHADS2 Low risk
Low Intermediate High Low Intermediate High Low Intermediate High Low Intermediate High Low Intermediate High Low Intermediate High
1.0 0.0 0.0 1.5 1.1 0.0 0.9 1.4 6.5 0.7 0.0 e 1.6 0.9 0.0 1.0 1.2 5.8
0.5 0.8 0.0 1.0 1.8 9.1 1.8 1.9 6.5 0.0 0.0 e 0.5 0.9 0.0 2.0 1.9 6.7
1.0 0.8 0.0 2.0 2.7 9.1 2.6 3.3 12.0 0.7 0.0 e 1.6 1.8 0.0 2.5 3.0 11.5
CHADS2 Intermediate risk CHADS2 High risk CHA2DS2VASc Low risk CHA2DS2VASc Intermediate risk CHA2DS2VASc High risk
CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; CHA2DS2-VASc ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
Marcucci et al Table 7
Joint Risk Distribution in Atrial Fibrillation
985
Within-Patient Concordance Between Cardioembolic and Bleeding Risk Categories and Outcomes
Cardioembolic and Bleeding Risk Scores
Concordance Between Cardioembolic and Bleeding Risk Category
% of Patients Experiencing a Stroke OR (95% CI)*
% of Patients Experiencing a Major Bleeding OR (95% CI)*
% of Patients Experiencing a Stroke or Major Bleeding OR (95% CI)*
CHADS2/HAS-BLED
Discordance
0.8 1.0 1.5 1.6 1.8 2.0 0.9 1.0 1.4 1.3 2.3 2.1
1.7 1.0 1.4 1.0 1.7 1.2 1.8 1.0 1.3 0.8 1.8 1.2
2.5 1.0 2.8 1.1 3.2 1.3 2.3 1.0 2.7 1.1 4.0 1.7
Partial concordance† Concordance CHA2DS2VASc/HAS-BLED
Discordance Partial concordance† Concordance
(0.2-11.9) (0.3-14.9)
(0.3-5.8) (0.4-9.4)
(0.2-4.2) (0.3-5.3)
(0.3-2.4) (0.4-4.0)
(0.3-3.7) (0.4-4.3)
(0.4-2.9) (0.6-4.7)
CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; CI ¼ confidence interval; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly; OR ¼ odds ratio. *From logistic regressions adjusted for antithrombotic therapy. †1-category disagreement.
Apixaban for Reduction in Stroke and Other Thromboembolism Events in Atrial Fibrillation study. They calculated a weighted kappa (ie, a measure of agreement adjusted for chance, which is usually used for evaluating the agreement between observers) but did not report any measure of correlation between scores. The relationship between cardioembolic and bleeding risk scores has 2 important implications. First, if the 2 scores are used together (in clinical practice or a risk-benefit analysis), their nonindependence has to be accounted for.20 Second, one type of score may be predicted from the other using appropriate statistics. Moreover, if there is not only an association but also a large degree of concordance, one score may be used as the surrogate for the other. According to our findings, a patient at low, intermediate, or high cardioembolic risk would have an approximately 30% chance of being at low, intermediate, or high bleeding risk. On the other hand, a generic patient would have a very low chance (75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly; No ¼ no therapy; VKA ¼ vitamin K antagonist.
986.e1
986.e2 Supplementary Table 1
The American Journal of Medicine, Vol 127, No 10, October 2014 Correlation Between Cardioembolic and Bleeding Scores in the Whole Population and in Treatment Groups
Cardioembolic and Bleeding Risk Scores CHADS2/HAS-BLED
CHA2DS2VASc/HAS-BLED
Treatment Group
Trend for Raw Scores Spearman Correlation Coefficient (P Value)*
Any (whole population) No antithrombotic therapy On antiplatelet agent On VKA Any (whole population) No antithrombotic therapy On antiplatelet agent On VKA
0.416 0.603 0.420 0.438 0.512 0.673 0.563 0.582
(
Stroke and Bleeding Risk Co-distribution in Real-world Patients with Atrial Fibrillation: The Euro Heart Survey Maura Marcucci, MD,a,b Gregory Y.H. Lip, MD,c Robby Nieuwlaat, PhD,a,d Ron Pisters, PhD,e Harry J.G.M. Crijns, MD,e Alfonso Iorio, MDa a Department of Clinical Epidemiology & Biostatistics, McMaster University, Hamilton, Ontario, Canada; bFoundation IRCCS Ca’ Granda e Ospedale Maggiore Policlinico, Geriatrics & Department of Clinical Sciences and Community Health, University of Milan, Milan, Italy; cUniversity of Birmingham Centre for Cardiovascular Sciences, City Hospital, Birmingham, United Kingdom; dPopulation Health Research Institute, Hamilton, Ontario, Canada; eDepartment of Cardiology, Maastricht University Medical Centre, Maastricht, The Netherlands.
ABSTRACT BACKGROUND: The choice to recommend antithrombotic therapy to patients with atrial fibrillation should rely on cardioembolic and bleeding risk stratification. Sharing some risk factors, schemes to predict thrombotic and bleeding risk are expected not to be independent, yet the degree of their association has never been clearly quantified. METHODS: We described the cardioembolic (Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack [CHADS2]/Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females [CHA2DS2-VASc]) and bleeding risk (Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly [HAS-BLED]) co-distribution among patients of the Euro Heart Survey on atrial fibrillation. We measured the within-patient correlation (Spearman) and concordance between the 2 types of score and score-based risk categorization (low, intermediate, high). The score-based predicted risk co-classification was then related to the observed 1-year stroke and bleeding occurrence. RESULTS: In 3920 patients, we found a between-scores correlation of 0.416 (P < .001) between HAS-BLED and CHADS2, and 0.512 (P < .001) between HAS-BLED and CHA2DS2-VASc. In 89% (CHADS2/ HAS-BLED) and 97% (CHA2DS2-VASc/HAS-BLED) of patients, the bleeding risk category was equal to or lower than their cardioembolic risk category (P < .001 for symmetry test). A complete concordance between risk categories was found in 39.6% (CHADS2/HAS-BLED) and 21.7% (CHA2DS2-VASc/HAS-BLED) of patients; 4.4% (CHADS2/HAS-BLED) and 7.7% (CHA2DS2-VASc/HAS-BLED) of patients had high cardioembolic risk/low bleeding risk or vice versa. A tendency for an increasing frequency of stroke was observed for increasing bleeding risk within cardioembolic risk categories and vice versa. CONCLUSIONS: In a real-world population with atrial fibrillation, we confirmed that the cardioembolic and bleeding risk classifications are correlated but not exchangeable. It is then worth verifying the advantages of a strategy adopting a combined risk assessment over a strategy relying only on the cardioembolic risk evaluation. Ó 2014 Elsevier Inc. All rights reserved. The American Journal of Medicine (2014) 127, 979-986 KEYWORDS: Atrial fibrillation; Cardioembolic and bleeding risk co-stratification
Funding: See last page of article. Conflict of Interest: See last page of article. Authorship: See last page of article. Requests for reprints should be addressed to Maura Marcucci, MD, Foundation IRCCS Ca’ Granda e Ospedale Maggiore Policlinico, 0002-9343/$ -see front matter Ó 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.amjmed.2014.05.003
Geriatrics & Department of Clinical Sciences and Community Health, University of Milan, via Pace 9, 20122 Milan, Italy. E-mail address: [email protected]
980
The American Journal of Medicine, Vol 127, No 10, October 2014
Patients with atrial fibrillation may have different benefits measured the within-patient correlation and concordance from long-term anticoagulation according to the balance of between the 2 types of score and score-based risk categotheir baseline risk for stroke and bleeding. With the availrization. Third, we related outcomes of stroke and bleeding ability of newer anticoagulants, the choice might be not only to the score-based risk co-classification. to treat or not to treat, but also which treatment choice better fits the risk profile. METHODS Strategies for treatment individualization, based on a trade-off Study Population CLINICAL SIGNIFICANCE between the treatment-related inWe analyzed the Euro Heart Surdividual benefit and harm, are When deciding on antithrombotic thervey (EHS) on atrial fibrillation facilitated by the availability of database, which included data apy in patients with atrial fibrillation, clinical tools to predict the patient on 5333 inpatients or outpatients, the usefulness of a joint assessment of risk for the target and adverse aged 18 years, who were repatient cardioembolic and bleeding risks event. The Congestive heart failferred to 182 university, nonunimight be limited by the interdependence ure, Hypertension, Age >75, versity, and specialized hospitals between the 2 risks. Diabetes mellitus, and prior Stroke among 35 member countries of the or transient ischemic attack The degree of correlation and concorEuropean Society of Cardiology, (CHADS2)1 and the Congestive with an electrocardiogram or dance between validated cardioembolic heart failure, Hypertension, Age Holter-proven diagnosis of atrial and bleeding risk schemes has never >75, Diabetes mellitus, and prior fibrillation during the qualifying been explicitly described. Stroke or transient ischemic attack, admission or in the preceding In a real-world population, we confirmed Vascular disease, Age 65-75, Sex year.17 Paper-based medical recategory i.e. females (CHA2DS2that the available cardioembolic and cords or data from medical inVASc)2 scores are 2 clinical risk bleeding risk classifications are signififormation systems were used to factor-based schemes to predict cantly correlated but not exchangeable. populate the database. We inthe risk of stroke in patients with cluded in the present analysis only atrial fibrillation, included in the the EHS patients with nonvalvular guidelines for the antithrombotic therapy of major scientific atrial fibrillation (ie, defined by the absence of a mitral valve societies.3-5 The CHA2DS2-VASc was developed to better stenosis or valvular surgery) and with data available for the identify patients “truly at low risk” of stroke,6 and this score calculation of the risk scores. is recommended in the European Society of Cardiology3 and Asia Pacific Heart Rhythm Society guidelines.7 Several scores for the assessment of bleeding risk in Risk Scores patients with atrial fibrillation are also available in the The CHADS2, CHA2DS2-VASc, and HAS-BLED scores literature.8-11 The Hypertension, Abnormal renal/liver were calculated retrospectively for each patient using function, Stroke, Bleeding history or predisposition, Labile the data collected at discharge from the hospital or at the end international normalized ratio, Elderly (>65 years), Drugs/ of the index outpatient visit. A reduced version of the alcohol concomitantly (HAS-BLED) score was shown to HAS-BLED score was used because of the unavailability have a good predictive ability both in patients taking and in of data for liver dysfunction and labile international patients not taking oral anticoagulants,12 and its use is recnormalized ratio. ommended in the most recent European and Canadian guidelines.3,4 Statistical Analysis Schemes for thrombotic and bleeding risk assessment have some risk factors in common, and there is evidence Both the CHADS2/HAS-BLED and the CHA2DS2-VASc/ that these scores are associated with both the risk of stroke HAS-BLED combinations of scores were considered. The and the risk of bleeding.13-16 Whether and to what extent the co-classification of the study population according to each of those combinations was described, using the raw scores joint use of a score for the cardioembolic risk and a score for or the score-based risk categories as defined in the literthe bleeding risk offers an advantage in making clinical ature2,18 (for CHADS2 and CHA2DS2-VASc: low risk ¼ decisions in clinical practice, in terms of risk stratification and treatment individualization, over a strategy relying only score 0; intermediate risk ¼ score 1; high risk ¼ score 2; on the cardioembolic risk assessment are not definitely clear. for HAS-BLED: low risk ¼ score 0; intermediate risk ¼ Surely it is affected by the co-distribution of the 2 scores, score 1-2; high risk ¼ score 3). Cross tabulation and which needs to be assessed in a “real-word” population. box-plots were used to represent the co-distribution. With these objectives, we first described how a populaA Spearman rank correlation coefficient was calculated tion of patients with atrial fibrillation who were referred to to express the extent of association/dependence between the the hospital were classified by a combined use of a score for patient cardioembolic and bleeding risk scores. We then cardioembolic risk and a score for bleeding risk. Second, we used 2 measures of concordance/exchangeability between
Marcucci et al
Joint Risk Distribution in Atrial Fibrillation
the 2 risk categorizations (high, intermediate, and low): a) the percentage agreement, that is, percentage of patients classified into the same cardioembolic and bleeding risk category; and b) a weighted percentage agreement, also taking into account the “partial concordance,” assigning a weight of 2 for perfect agreement, 1 for a 1-category disagreement, and 0 for 2-category disagreement. Symmetry and marginal homogeneity (Stuart-Maxwell) tests also were performed to determine, in case of discordance, whether there was a statistically significant tendency for the cardioembolic risk category to be higher than the bleeding risk category or vice versa.
Secondary Analyses Score co-distribution by antithrombotic therapy. We described the score co-distribution in subgroups of EHS patients defined according to the antithrombotic therapy they were prescribed at the time of the discharge from the hospital or at the end of the index outpatient visit, qualitatively comparing the degree of correlation in each group. The following subgroups of patients were defined: patients not receiving any antithrombotic treatment, patients receiving an antiplatelet agent, and patients receiving a vitamin K antagonist (including patients receiving both a vitamin K antagonist and an antiplatelet agent). Patients receiving heparin were excluded from the current analyses. Our hypothesis for this secondary analysis was that the distribution of each risk score would be different in each bytreatment defined group (according to the KruskaleWallis rank-test), but that the level of within-patient correlations between the 2 types of score would be similar in each treatment group as in the whole population. Score co-distribution and outcome. We explored whether there was any association between the score co-distribution and the outcome. We did that by investigating whether, for each cardioembolic risk category, there was a trend in the occurrence of adverse events according to the bleeding risk category and vice versa. Also, we investigated whether, regardless of the level of risk, the concordance/discordance between cardioembolic and bleeding risk categories was associated with a different outcome (considering stroke, bleeding, or any); logistic regression analysis was used for this purpose. All analyses were stratified by or adjusted for antithrombotic treatment. Definitions used for the outcomes in the EHS on atrial fibrillation have been published.2
RESULTS Of the 5272 patients with atrial fibrillation in the EHS of atrial fibrillation who were discharged alive,17 3920 without mitral valve stenosis or valvular surgery and with necessary information to calculate the risk scores were included in this analysis. Table 1 shows the relevant characteristics of the study population.
981 Table 1 Characteristics of the 3920* Patients with Atrial Fibrillation from the Euro Heart Survey Included in the Current Analyses Characteristic
Description
Age, median (Q1, Q3) Women, n (%) CHADS2, median (Q1, Q3) CHA2DS2-VASc, median (Q1, Q3) HAS-BLED, median (Q1, Q3) Drugs prescribed at discharge, n (%) Warfarin Aspirin No antithrombotic therapy Stroke at 1 y of follow-up, n (%) Any patient In patients on warfarin In patients on aspirin In patients off antithrombotic therapy Major bleeding at 1 y of follow-up, n (%) Any patient In patients on warfarin In patients on aspirin In patients off antithrombotic therapy Stroke and major bleeding at 1 y of follow-up, n (%) Any patient In patients on warfarin In patients on aspirin In patients off antithrombotic therapy
68 1587 1 3 1
(59, 75) (40.5) (1, 2) (2, 4) (1, 2)
2476 (63.2) 1079 (27.5) 365 (9.3) 47 27 19 1
(1.6) (1.4) (2.5) (0.4)
44 35 8 1
(1.5) (1.8) (1.0) (0.4)
4 4 0 0
(0.1) (0.2) (0) (0)
CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; CHA2DS2-VASc ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/ liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly; Q1 ¼ first quartile; Q3 ¼ third quartile. *Data on outcome at 1 year of follow-up available for 2934 of 3920 patients.
Table 2 and Figure 1A describe the co-distribution of CHADS2 and HAS-BLED; Table 3 and Figure 1C describe the co-distribution of CHA2DS2-VASc and HASBLED. The Spearman correlation coefficient was 0.416 (P < .001) between CHADS2 and HAS-BLED, and 0.512 (P < .001) between CHA2DS2-VASc and HAS-BLED. Those co-distributions translated into the joint scorebased risk categorizations shown in Tables 4 and 5, and Figure 1B and D. The most prevalent risk group was the one at high cardioembolic risk and intermediate bleeding risk (37.5% if CHADS2, 60.8% if CHA2DS2-VASc was used), with 89% (CHADS2/HAS-BLED) and 97% (CHA2DS2-VASc/HAS-BLED) of patients classified into a bleeding risk category equal to or lower than their cardioembolic risk category (P < .001 for symmetry and marginal homogeneity tests). With the use of the CHADS2 and HAS-BLED, 39.6% of patients were classified into
982 Table 2
The American Journal of Medicine, Vol 127, No 10, October 2014 CHADS2 and HAS-BLED Scores Co-distribution HAS-BLED
n (% of Total) CHADS2
0
0 1 2 3 4 5 6 Any CHADS2
331 314 129 28 4 0 0 806
1 (8.44) (8.01) (3.29) (0.71) (0.10) (0.00) (0.00) (20.56)
264 641 472 213 71 17 5 1683
2 (6.73) (16.35) (12.04) (5.43) (1.81) (0.43) (0.13) (42.93)
83 318 369 196 77 42 8 1093
3 (2.12) (8.11) (9.41) (5.00) (1.96) (1.07) (0.20) (27.88)
6 64 83 60 52 25 6 296
4 (0.15) (1.63) (2.12) (1.53) (1.33) (0.64) (0.15) (7.55)
0 5 7 5 10 9 3 39
5 (0.00) (0.13) (0.18) (0.13) (0.26) (0.23) (0.08) (0.99)
0 0 0 0 1 2 0 3
Any HAS-BLED (0.00) (0.00) (0.00) (0.00) (0.03) (0.05) (0.00) (0.08)
684 (17.45) 1342 (34.23) 1060 (27.04) 502 (12.81) 215 (5.48) 95 (2.42) 22 0.56 3920
CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
the same cardioembolic and bleeding risk category, whereas 21.7% was the agreement between CHA2DS2-VASc and HAS-BLED; 16.3% were classified into the same cardioembolic and bleeding risk category by both the
CHADS2/HAS-BLED and the CHA2DS2-VASc/HASBLED combination. By also considering the partial agreement, the weighted percentage agreement was 67.7% and 57.0%, respectively.
Figure 1 Cardioembolic and bleeding risk scores co-distribution. A, CHADS2 and HAS-BLED scores. B, CHADS2 and HAS-BLED risk categories. C, CHA2DS2-VASc and HAS-BLED scores. D, CHA2DS2-VASc and HAS-BLED risk categories. Red line: median of HAS-BLED scores. CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; CHA2DS2-VASc ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
Marcucci et al Table 3
Joint Risk Distribution in Atrial Fibrillation
983
CHA2DS2-VASc and HAS-BLED Scores Co-distribution HAS-BLED
n (% of Total) CHA2DS2-VASc
0
0 1 2 3 4 5 6 7 8 9 Any CHA2DS2-VASc
218 284 168 104 26 5 1 0 0 0 806
1 (5.56) (7.24) (4.29) (2.65) (0.66) (0.13) (0.03) (0.00) (0.00) (0.00) (20.56)
83 239 356 390 318 187 78 20 9 3 1683
2 (2.12) (6.10) (9.08) (9.95) (8.11) (4.77) (1.99) (0.51) (0.23) (0.08) (42.93)
8 64 139 269 273 170 101 49 18 2 1093
3 (0.20) (1.63) (3.55) (6.86) (6.96) (4.34) (2.58) (1.25) (0.46) (0.05) (27.88)
0 7 25 43 73 61 40 33 12 2 296
4 (0.00) (0.18) (0.64) (1.10) (1.86) (1.56) (1.02) (0.84) (0.31) (0.05) (7.55)
0 0 2 3 8 3 8 10 4 1 39
5 (0.00) (0.00) (0.05) (0.08) (0.20) (0.08) (0.20) (0.26) (0.10) (0.03) (0.99)
0 0 0 0 0 0 1 0 2 0 3
Any HAS-BLED (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.03) (0.00) (0.05) (0.00) (0.08)
309 594 690 809 698 426 229 112 45 8 3920
(7.88) (15.15) (17.60) (20.64) (17.81) (10.87) (5.84) (2.86) (1.15) (0.20)
CHA2DS2-VASc ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
Score Co-distribution and Antithrombotic Therapy Table 1 describes the frequencies and modalities of antithrombotic therapy prescribed at discharge. As expected, the distribution of each individual risk score significantly differed among the 3 by-treatment groups (P < .001). When looking at the score co-distribution, positive Spearman correlation coefficients were obtained in each group similar to those found in the whole population (Supplementary Table 1). The score co-distribution across by-treatment groups is described in Supplementary Figure 1.
category. Likewise, a tendency for an increasing frequency of events in accordance with an increasing cardioembolic risk score was found within the intermediate and high bleeding risk categories. The same results for the entire population are provided in Supplementary Table 2. Finally, a trend of an increasing frequency of stroke for a stronger concordance between cardioembolic and bleeding risk category was observed (Table 7), before and after accounting for the antithrombotic therapy, even if without reaching a statistical significance. The trend was not confirmed for the occurrence of major bleedings.
Score Co-distribution and Outcome
DISCUSSION
Data on 1-year outcome were available for 2934 of 3920 patients (74.8%). The score co-distribution and correlation in these patients resembled those of the initial population. Table 6 shows, among patients on warfarin, the tendency for an increasing frequency of events (stroke or bleedings, or both) as a function of an increasing bleeding risk score, mainly evident within the high cardioembolic risk
We described the joint distribution of scores for the cardioembolic risk (CHADS2 or CHA2DS2-VASc) and scores for the bleeding risk (HAS-BLED) in patients with atrial fibrillation. We found a positive within-patient correlation of moderate strength between the 2 types of scores (w0.5) in the whole population and in subgroups defined by the prescribed antithrombotic therapy. A perfect concordance
Table 4
CHADS2 and HAS-BLED Co-distribution According to the Score-based Risk Categories HAS-BLED Risk Category n (% of the Total in Each CHADS2 Risk Category)
CHADS2 Risk Category
Low Risk
Intermediate Risk
High Risk
n (% of the Total)
Low risk Intermediate risk High risk n (% of the total)
331 314 161 806
347 959 1470 2776
6 69 263 338
684 (17.5) 1342 (34.2) 1894 (48.3) 3920
(48.4) (23.4) (8.5) (20.6)
(50.7) (71.5) (77.6) (70.8)
(0.9) (5.1) (13.9) (8.6)
CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
984 Table 5
The American Journal of Medicine, Vol 127, No 10, October 2014 CHA2DS2-VASc and HAS-BLED Co-distribution According to the Score-based Risk Categories HAS-BLED Risk Category n (% of the Total in Each CHA2DS2-VASc Risk Category)
CHA2DS2-VASc Risk Category
Low Risk
Intermediate Risk
High Risk
n (% of Total)
Low risk Intermediate risk High risk n (% of total)
218 284 304 806
91 303 2382 2776
0 7 331 338
309 (7.9) 594 (15.1) 3017 (77.0) 3920
(70.5) (47.8) (10.1) (20.6)
(29.5) (51.0) (78.9) (70.8)
(0.0) (1.2) (11.0) (8.6)
CHA2DS2-VASc ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
between cardioembolic and bleeding risk category was found in less than 50% of patients, but only 4.4% (CHADS2/HAS-BLED) to 7.7% (CHA2DS2-VASc/HASBLED) of patients were in the cells corresponding to a full discordance between the cardioembolic and bleeding risk categories (ie, high cardioembolic risk/low bleeding risk or vice versa). An increasing bleeding risk score was associated with an increasing frequency of stroke events within the intermediate and high cardioembolic risk score categories; the trend was less evident for bleeding events. A higher degree of within-patient concordance between cardioembolic and bleeding risk categories appeared to be associated with a higher frequency of stroke. The current literature has been focusing on verifying the predictive ability of proposed cardioembolic and bleeding
Table 6
risk scores in real-word atrial fibrillation cohorts, but scarce attention has been paid to addressing the question of how those scores co-distribute and correlate, and so how the assumption of their independence is justifiable and their separate use necessary. For example, Friberg et al14 recently used the score-based co-stratification to assess the net benefit with warfarin associated with different CHADS2 (or CHA2DS2-VASc) and HAS-BLED scores in a large Swedish atrial fibrillation population; their cross-tabulations resemble ours (Tables 2 and 3), with empty cells at the top right and bottom left corners, and the highest concentration in the cells along the diagonal. Yet, they did not focus on how those strata were differently populated. Lopes et al19 reported the cardioembolic and bleeding score-based categories co-distribution among the participants of the
Cardioembolic and Bleeding Risk Categories Co-distribution and Outcomes: Patients Prescribed Warfarin
Cardioembolic Risk Category
HAS-BLED Risk Category
% of Patients Experiencing a Stroke
% of Patients Experiencing a Major Bleeding
% of Patients Experiencing a Stroke or Major Bleeding
CHADS2 Low risk
Low Intermediate High Low Intermediate High Low Intermediate High Low Intermediate High Low Intermediate High Low Intermediate High
1.0 0.0 0.0 1.5 1.1 0.0 0.9 1.4 6.5 0.7 0.0 e 1.6 0.9 0.0 1.0 1.2 5.8
0.5 0.8 0.0 1.0 1.8 9.1 1.8 1.9 6.5 0.0 0.0 e 0.5 0.9 0.0 2.0 1.9 6.7
1.0 0.8 0.0 2.0 2.7 9.1 2.6 3.3 12.0 0.7 0.0 e 1.6 1.8 0.0 2.5 3.0 11.5
CHADS2 Intermediate risk CHADS2 High risk CHA2DS2VASc Low risk CHA2DS2VASc Intermediate risk CHA2DS2VASc High risk
CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; CHA2DS2-VASc ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly.
Marcucci et al Table 7
Joint Risk Distribution in Atrial Fibrillation
985
Within-Patient Concordance Between Cardioembolic and Bleeding Risk Categories and Outcomes
Cardioembolic and Bleeding Risk Scores
Concordance Between Cardioembolic and Bleeding Risk Category
% of Patients Experiencing a Stroke OR (95% CI)*
% of Patients Experiencing a Major Bleeding OR (95% CI)*
% of Patients Experiencing a Stroke or Major Bleeding OR (95% CI)*
CHADS2/HAS-BLED
Discordance
0.8 1.0 1.5 1.6 1.8 2.0 0.9 1.0 1.4 1.3 2.3 2.1
1.7 1.0 1.4 1.0 1.7 1.2 1.8 1.0 1.3 0.8 1.8 1.2
2.5 1.0 2.8 1.1 3.2 1.3 2.3 1.0 2.7 1.1 4.0 1.7
Partial concordance† Concordance CHA2DS2VASc/HAS-BLED
Discordance Partial concordance† Concordance
(0.2-11.9) (0.3-14.9)
(0.3-5.8) (0.4-9.4)
(0.2-4.2) (0.3-5.3)
(0.3-2.4) (0.4-4.0)
(0.3-3.7) (0.4-4.3)
(0.4-2.9) (0.6-4.7)
CHADS2 ¼ Congestive heart failure, Hypertension, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack; CI ¼ confidence interval; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly; OR ¼ odds ratio. *From logistic regressions adjusted for antithrombotic therapy. †1-category disagreement.
Apixaban for Reduction in Stroke and Other Thromboembolism Events in Atrial Fibrillation study. They calculated a weighted kappa (ie, a measure of agreement adjusted for chance, which is usually used for evaluating the agreement between observers) but did not report any measure of correlation between scores. The relationship between cardioembolic and bleeding risk scores has 2 important implications. First, if the 2 scores are used together (in clinical practice or a risk-benefit analysis), their nonindependence has to be accounted for.20 Second, one type of score may be predicted from the other using appropriate statistics. Moreover, if there is not only an association but also a large degree of concordance, one score may be used as the surrogate for the other. According to our findings, a patient at low, intermediate, or high cardioembolic risk would have an approximately 30% chance of being at low, intermediate, or high bleeding risk. On the other hand, a generic patient would have a very low chance (75, Diabetes mellitus, and prior Stroke or transient ischemic attack, Vascular disease, Age 65-75, Sex category i.e. females; HAS-BLED ¼ Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly; No ¼ no therapy; VKA ¼ vitamin K antagonist.
986.e1
986.e2 Supplementary Table 1
The American Journal of Medicine, Vol 127, No 10, October 2014 Correlation Between Cardioembolic and Bleeding Scores in the Whole Population and in Treatment Groups
Cardioembolic and Bleeding Risk Scores CHADS2/HAS-BLED
CHA2DS2VASc/HAS-BLED
Treatment Group
Trend for Raw Scores Spearman Correlation Coefficient (P Value)*
Any (whole population) No antithrombotic therapy On antiplatelet agent On VKA Any (whole population) No antithrombotic therapy On antiplatelet agent On VKA
0.416 0.603 0.420 0.438 0.512 0.673 0.563 0.582
(
