Clinical background
The ATRIA stroke risk score was developed to estimate the risk of ischaemic stroke and thromboembolism in patients with atrial fibrillation as a basis for decisions on anticoagulation. The score was developed at a time when CHADS₂ and CHA₂DS₂-VASc dominated, but both showed only moderate discrimination. CHA₂DS₂-VASc in particular has a very low threshold for its high-risk category, which means that a large proportion of patients are classified as high risk without the individual variation in risk within that group being well captured. The ATRIA stroke risk score addresses this through a finer age stratification, an interaction between age and previous stroke, and the inclusion of renal function and proteinuria as risk markers [1].
Although the score has shown better discrimination than both CHADS₂ and CHA₂DS₂-VASc in several cohorts, it has not gained traction in international guidelines. The ESC guidelines on atrial fibrillation recommend CHA₂DS₂-VASc as the primary tool, and the ATRIA stroke risk score is not mentioned as an alternative in either the ESC or the ACC/AHA guidelines [3]. This is essential to bear in mind when the score is used clinically.
Calculating the ATRIA stroke risk score
The ATRIA stroke risk score consists of a combined age-and-stroke component together with six binary risk factors. The score is calculated as:
where is an interaction term depending on both the age category and the presence of previous ischaemic stroke or TIA, and the other variables are:
| Variable | Value |
|---|---|
| Female sex: 1 if yes, otherwise 0 | |
| Diabetes mellitus: 1 if yes, otherwise 0 | |
| Heart failure: 1 if yes, otherwise 0 | |
| Hypertension: 1 if yes, otherwise 0 | |
| Proteinuria: 1 if yes, otherwise 0 | |
| eGFR <45 mL/min/1.73 m² or end-stage renal disease: 1 if yes, otherwise 0 |
The combined age-and-stroke component is scored as follows:
| Age | No previous stroke | Previous stroke |
|---|---|---|
| <65 years | 0 | 8 |
| 65–74 years | 3 | 7 |
| 75–84 years | 5 | 7 |
| ≥85 years | 6 | 9 |
The interaction reflects the fact that the effect of age on stroke risk is strong in patients without previous stroke but attenuated in those who have already had one, since these patients have an increased baseline risk irrespective of age [1].
The derivation cohort consisted of 10,927 patients with non-valvular atrial fibrillation from Kaiser Permanente Northern California, diagnosed between 1996 and 1997. The patients contributed 32,609 person-years off warfarin and 685 thromboembolic events (643 ischaemic strokes and 42 other thromboembolic events), corresponding to an annual incidence of 2.1% [1]. A split-sample method was used: two-thirds of the cohort formed the derivation set (7,284 patients, 456 events) and one-third was reserved for internal validation (3,643 patients, 229 events). The variables were chosen by bootstrap selection over 1,000 samples, in which eight variables were selected in more than 60% of samples. Vascular disease (coronary artery disease and peripheral arterial disease) was tested but did not remain significant in the multivariable model [1].
Interpretation in practice
The ATRIA stroke risk score is divided into three risk categories based on the observed annual thromboembolic rates in the derivation cohort [1]:
| Score | Risk category | Annual thromboembolic rate | Clinical action |
|---|---|---|---|
| 0–5 | Low risk | <1.0% | Anticoagulation generally not indicated on the score. Individual assessment is required where the bleeding risk is increased or other factors apply. |
| 6 | Intermediate risk | 1.0–<2.0% | The decision on anticoagulation should be individualised. Bleeding risk and patient preference are weighed in. |
| 7–15 | High risk | ≥2.0% | Anticoagulation indicated unless contraindicated. |
The thresholds were based on a formal decision analysis in which an annual stroke risk of 1% was taken as the point at which the benefit of anticoagulation begins to exceed the risk, and 2% as the point at which the benefit is clear [1]. It is important to note that these thresholds were derived in an American cohort from the late 1990s and that the absolute level of risk at a given score may vary between populations.
Validation and performance
In the external validation cohort ATRIA-CVRN, comprising 25,306 patients with incident atrial fibrillation diagnosed between 2006 and 2009 within Kaiser Permanente in northern and southern California, the c-index was 0.70 (95% CI 0.67–0.72) for all thromboembolic events and 0.75 (95% CI 0.72–0.78) for severe events [1]. In the derivation cohort the corresponding figures were 0.73 (95% CI 0.71–0.75) and 0.76 (95% CI 0.74–0.79). Calibration was satisfactory in both the derivation and the validation part of the derivation cohort (goodness-of-fit P = 0.14 and P = 0.78 respectively) [1].
The ATRIA stroke risk score outperformed both CHADS₂ and CHA₂DS₂-VASc in discrimination in the derivation cohort. Across the whole score range the c-index was 0.73 for ATRIA compared with 0.69 for CHADS₂ and 0.70 for CHA₂DS₂-VASc. When the scores were collapsed into three categories the difference was more marked still, since CHA₂DS₂-VASc then had a low c-index of 0.58 because of its very low threshold for the high-risk category [1].
A large Swedish national cohort study compared ATRIA, CHADS₂ and CHA₂DS₂-VASc in 152,153 patients with atrial fibrillation not taking warfarin. In this cohort the annual stroke risk was higher (3.2%) than in the Californian cohorts. ATRIA performed better than both comparators in discrimination, but the authors noted that the published thresholds for low/intermediate/high risk may need to be adapted to local stroke rates [2].
In a Korean multicentre study of 3,112 patients with a stroke already sustained and atrial fibrillation (secondary prevention), all the scores, including ATRIA, performed poorly in predicting recurrent ischaemic stroke, with c-indices between 0.544 and 0.558. The scores performed somewhat better for death and composite cardiovascular outcomes, but ATRIA did not even outperform the Essen stroke risk score in this population [4].
Limitations
The ATRIA stroke risk score was derived in a cohort of patients with non-valvular atrial fibrillation. With a mechanical valve prosthesis or significant mitral stenosis, anticoagulation is indicated whatever the score, and the ATRIA stroke risk score must not be used to justify withholding anticoagulation. The external validation cohort ATRIA-CVRN did, however, include 1.5% of patients with mitral stenosis or a previous valve prosthesis, so the performance of the score in this group is incompletely characterised [1].
The score includes two laboratory-based variables, proteinuria and eGFR, which complicates purely clinical bedside calculation if current laboratory values are not available. In the derivation cohort, proteinuria was imputed as normal for 22.2% of person-years, meaning that a substantial proportion of the data rests on assumptions rather than measurements [1].
The score is not validated for secondary prevention. In the Korean cohort of patients who had already had a stroke, predictive ability for recurrent stroke was close to chance [4]. In patients with a previous stroke the score should not be used to guide the choice of anticoagulation, since these patients by definition belong to a high-risk group in whom anticoagulation is indicated irrespective of further risk stratification.
Perhaps the most important limitation is the absence of guideline support. The ESC guidelines on atrial fibrillation recommend CHA₂DS₂-VASc as the primary tool for stroke risk stratification, and the ATRIA stroke risk score is not mentioned as an alternative [3]. This means that in practice the score functions as a complementary tool for those who want a more finely graded risk stratification than CHA₂DS₂-VASc offers, rather than as a substitute in routine management.
The risk is not static. A patient who today falls into the low-risk category may cross the threshold into intermediate or high risk by growing older, developing diabetes or losing renal function. The score should therefore be reassessed regularly, particularly when the patient's comorbidity changes.
External validation in a national cohort
The Swedish national cohort study by Aspberg et al. is to date the largest external validation of the ATRIA stroke risk score and showed that the score performed better than both CHADS₂ and CHA₂DS₂-VASc in a Swedish population as well [2]. The study included all patients treated at a Swedish hospital or specialist clinic with atrial fibrillation between 2005 and 2010 and was based on complete national registries. An important observation was that the annual stroke risk in the Swedish cohort was higher than in the American cohorts, which suggests that the thresholds for low and high risk may need to be adjusted upwards in a Swedish setting. Swedish guidelines nonetheless follow the ESC and recommend CHA₂DS₂-VASc as the primary tool, so in Swedish clinical practice the ATRIA stroke risk score should be seen as a complement in uncertain decisions rather than as a first choice.
References
- Singer DE, Chang Y, Borowsky LH, et al. A new risk scheme to predict ischemic stroke and other thromboembolism in atrial fibrillation: the ATRIA Study Stroke Risk Score. J Am Heart Assoc. 2013;2(3):e000250. PMID: 23782923
- Aspberg S, Chang Y, Atterman A, et al. Comparison of the ATRIA, CHADS2, and CHA2DS2-VASc stroke risk scores in predicting ischaemic stroke in a large Swedish cohort of patients with atrial fibrillation. Eur Heart J. 2016;37(42):3203–3210. PMID: 26941204
- Dzeshka MS, Lane DA, Lip GY. Stroke and bleeding risk in atrial fibrillation: navigating the alphabet soup of risk-score acronyms (CHADS2, CHA2DS2-VASc, R2CHADS2, HAS-BLED, ATRIA, and more). Clin Cardiol. 2014;37(10):634–644. PMID: 25168181
- Yu I, Song TJ, Kim BJ, et al. CHADS2, CHA2DS2-VASc, ATRIA, and Essen stroke risk scores in stroke with atrial fibrillation: A nationwide multicenter registry study. Medicine. 2021;100(3):e24000. PMID: 33545993
