Author + information
- William Whang, MD, MS∗ ( and )
- David J. Shim, MD, PhD
- ↵∗Reprint requests and correspondence:
Dr. William Whang, Department of Cardiology, Columbia University Medical Center, PH 9-321, 622 West 168th Street, New York, New York 10032.
William Osler characterized medicine as a science of uncertainty and an art of probability (1), and this is especially true in making treatment decisions to prevent stroke in patients with atrial fibrillation (AF), the most common cardiac arrhythmia. For most cardiologists, the decision to initiate anticoagulation in patients with AF starts with an attempt to quantify the patient’s stroke risk. It has been recognized for some time that thromboembolic risk in patients with AF depends less on the “quantity” or “severity” of their AF and more on other clinical characteristics. Epidemiological and observational research into these characteristics has yielded a number of risk stratification schemes to help guide anticoagulation decisions.
The first risk stratification scheme to gain widespread acceptance was the CHADS2 score (2), which was particularly advantageous due to its relative ease of use. The CHADS2 system stratified patients into low-, moderate-, and high-risk categories and became widely used for estimates of stroke risk in the clinical setting. Additional thromboembolic risk factors were recognized and incorporated into the subsequent CHA2DS2-VASc score (congestive heart failure, hypertension, age ≥75, diabetes mellitus, prior stroke or transient ischemic attack, vascular disease, age 65 to 74, female score) (3), with modification of the age component and inclusion of vascular disease (myocardial infarction, peripheral vascular disease, or aortic plaque) and sex. More recently, the anticoagulation and risk factors in atrial fibrillation (ATRIA) score was developed and validated on the basis of a sample of patients from Kaiser Permanente Northern California (4). The ATRIA score was reported to have better performance characteristics on the basis of the c-index and net reclassification index than prior scoring systems, according to analyses of the ATRIA-Cardiovascular Research Network cohort. Compared with CHA2DS2-VASc, the main differences in the ATRIA score were inclusion of renal dysfunction and proteinuria, rather than vascular disease, and a revamped, albeit more complicated, scoring algorithm for age that incorporated an interaction with history of ischemic stroke. ATRIA scores ranged from 0 to 15, and were collapsed into a relatively expansive low-risk category (0 to 5), as well as moderate-risk (score of 6) and high-risk (scores of 7 to 15) categories in order to correspond to annualized stroke rates of <1%, 1 to <2%, and ≥2% per year.
In this issue of the Journal, Chao et al. (5) compare the performance of ATRIA and CHA2DS2-VASc risk stratification schemes in a large independent cohort from the Taiwanese National Health insurance research database, representing over 23 million enrollees (5). The authors sampled more than 186,000 patients with AF, who were not taking anticoagulant or antiplatelet medications. Risk scores were estimated on the basis of the ATRIA and CHA2DS2-VASc models and were related to ischemic stroke frequency during a mean follow-up of 3.4 years. Their findings demonstrate better discriminatory performance with CHA2DS2-VASc scores than with the ATRIA score, again according to c-index and net reclassification index (5).
One of the study’s major strengths is its impressively large sample size, allowing for further analyses that help illustrate the relative merits of the 2 risk stratification schemes specifically for identification of low-risk patients. Within the substantial group of patients considered low-risk on the basis of their ATRIA scores (39.3% of the sample), a higher CHA2DS2-VASc score was still associated with greater stroke risk (5). It is noteworthy that within this same subgroup, higher ATRIA scores were also associated with risk of stroke (5).
Like many interesting studies, this one leads to a number of questions and potential avenues for further research. Why exactly did one scoring system perform better than the other? Among the more obvious differences between the two schemes is the possibility that perhaps vascular disease was more predictive of stroke in this sample or perhaps renal disease was less predictive. The type of population studied is likely to have significant impact on the utility of either factor in stroke prediction. For instance, in ROCKET AF (Rivaroxaban Once-daily, oral, direct factor Xa inhibition Compared with vitamin K antagonism for prevention of stroke and Embolism Trial in Atrial Fibrillation), reduced creatinine clearance was independently predictive of stroke and thromboembolic risk after adjusting for standard risk factors (6). However, patients in ROCKET AF had at least two risk factors for stroke, and severe renal failure was an exclusion criterion. In the Loire Valley Atrial Fibrillation Project of patients with AF identified from discharge records, although renal impairment was associated with stroke and thromboembolism, it did not significantly improve risk prediction when added to CHA2DS2-VASc (7). Similar examples on both sides can be identified for vascular disease, although there is greater support for its inclusion in risk stratification schemes (8). In the current study from Chao et al. (5), the study sample excluded patients treated not only with anticoagulation with warfarin but also with any antiplatelet agents such as aspirin. One could postulate that this might have amplified the relationship between vascular disease and stroke risk in this analysis.
Can the results from the relatively homogeneous ethnic group in this analysis be extrapolated to other populations? The cohort in the present study was at relatively high risk for ischemic stroke, 12.7% during a mean follow-up of 3.4 years. This incidence rate is consistent with another recent study that reported even higher annual risk of stroke, approximately 9%, in a hospital-based cohort from Hong Kong (9). It is also substantially higher than the 2.0% annualized rate of ischemic stroke in the ATRIA derivation cohort (4). Perhaps the ATRIA score could be improved for populations with higher baseline risk of stroke by recalibrating the designation of low-risk and intermediate-risk categories. In the present study, an ATRIA score of 0 or 1 was still associated with a low stroke rate at 5-year follow-up (1.36 to 1.60 per 100 person-years), comparable to a CHA2DS2-VASc score of 0 or 1 (1.00 to 1.89 per 100 person-years).
It is also worth mentioning that patients’ perspectives on what constitutes “low-risk” may differ. One person may accept a stroke risk of 1% per year, whereas another may accept a higher risk depending on his or her preferences to avoid bleeding. For some patients, it may be more useful to provide quantitative risks with ranges to incorporate uncertainty, as opposed to categories of risk that may give a false sense of precision. Risk stratification schemes can aid in our clinical decision making only if we use them, and even with the benefit of validated risk scores, evidence suggests a tendency to withhold anticoagulation in multiple clinical arenas (10,11). Further research is needed to help define optimal ways to convey information to patients with AF and help improve clinical decision-making (12).
Although prescribing anticoagulant agents for patients with AF is ultimately a clinical decision to be made between physician and patient, there is a need to continue development of more accurate stroke risk estimators. The study from Chao et al. (5) provides insight into the relative performance of the ATRIA and CHA2DS2-VASc scoring systems in a higher risk cohort. Further research in a variety of populations will hopefully help us understand the reasons for differential performance of particular risk stratification schemes and allow for more targeted recommendations for individual patients.
↵∗ Editorials published in the Journal of the American College of Cardiology reflect the views of the authors and do not necessarily represent the views of JACC or the American College of Cardiology.
Both authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Osler W.
- Singer D.E.,
- Chang Y.,
- Borowsky L.H.,
- et al.
- Chao T.-F.,
- Liu C.-J.,
- Wang K.-L.,
- et al.
- Piccini J.P.,
- Stevens S.R.,
- Chang Y.,
- et al.,
- for the ROCKET AF Steering Committee and Investigators
- Banerjee A.,
- Fauchier L.,
- Vourc'h P.,
- et al.
- Sandhu R.K.,
- Bakal J.A.,
- Ezekowitz J.A.,
- et al.