The search for the ideal atrial fibrillation stroke risk prediction schema: is ATRIA a contender?

This editorial refers to ‘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’, by S. Aspberg et al., on page 3203. Analyses of the Framingham cohort established the independent contribution of non-valvular atrial fibrillation (NVAF) to the risk of stroke and estimated an annual stroke incidence of 4.1%.1 In an overview of the five randomized trials of warfarin for the primary prevention of stroke among patients with NVAF, in the control group there was an annual incidence of 4.5% for stroke and 5.0% for stroke or systemic embolism.2 Investigators from the Stroke Prevention in Atrial Fibrillation (SPAF) trials and the Atrial Fibrillation Investigators (AFI) who had conducted the overview established several independent risk factors for stroke, allowing subdivision of the risk of stroke over a range from 1% to ∼20%, depending upon which and how many risk factors were present.2,3 The CHADS2 schema was derived from the observations of these two groups and was validated in a cohort of 1733 American patients with NVAF who were not receiving warfarin on hospital discharge and were followed for a median of 1 year.4 A point score based upon only five clinical parameters was used to predict annual stroke risk, ranging from 1.9% to 18.2%. The CHADS-VASc schema, developed subsequently, added additional risk factors and refined the age criteria, and was validated in a cohort of 1084 patients in the Euro Heart Survey.5 A point score based upon seven clinical parameters allowed predictions of stroke risk ranging from 0 to 14.2% per year. The CHADS2 had come into wide use, and although C-statistics were only slightly better with the CHA2DS2-VASc, it appeared to allow discrimination of a range of stroke risks among patients with a CHADS2 = 0.6,7 The CHA2DS2-VASc was strongly promoted by the European Society of Cardiology,7 and was eventually adopted by the American Heart Association for use in their clinical guidelines recommendations for stroke prevention.8 The recommendations of the Canadian Cardiovascular Society for antithrombotic therapies have incorporated elements of both CHADS2 and CHA2DS2-VASc into an algorithm for selection of antithrombotic therapy which first identifies patients aged ≥65 years, then applies the CHADS2 criteria, and then considers the presence or absence of vascular disease.9 The American College of Chest Physicians guidelines have continued to utilize the CHADS2 schema.10 Recent controversy has centred on what annual risk of stroke warrants the use of oral anticoagulation and the role of aspirin. In this issue of the journal, Aspberg et al.11 report a comparison of the predictive values of the ATRIA, CHADS2, and CHA2DS2-VASc stroke risk scores in a large Swedish cohort of 152 153 AF patients (July 2005 to December 2010) not receiving warfarin. Their findings prompt further consideration of the relative values of available risk schemas, the many determinants of differing performance, and the rationales for use of these schemas. The ATRIA score for prediction of stroke risk in NVAF was initially derived and then validated among 3643 patients in a Northern California cohort prior to the development of both the CHADS2 and the CHA2DS2-VASc,12 but it has not had widespread use nor has it been incorporated in the guidelines of any national societies. This cohort had a median follow-up of 2.4 years off warfarin and an annualized thrombo-embolism rate of 2.1% (either ischaemic stroke or arterial embolus). The risk factors included diabetes, heart failure, hypertension, and prior stroke as in CHADS2 and CHA2DS2-VASc, with the addition of female sex, proteinuria, and chronic renal failure, and the further partition of age into categories of <65, 65–74, 75–84, and ≥85 years. Using a second Northern California cohort of 25 306 patients with a thrombo-embolism rate of 1.9% per year,12 the ATRIA schema was further validated, compared with the CHADS2 and CHA2DS2-VASc schemas, and was determined to perform better than either of these widely used schemas. In the Swedish cohort comparison of the three risk schemas, the sole outcome was acute ischaemic stroke, occurring after a blanking period of 14 days following the index AF diagnosis.11 The comparisons of schema performance were done using the C-index and the net reclassification improvement. These were determined for each schema using first the full range of point scores, then cut-off points for low, moderate, and high risk of ischaemic stroke based on published ranges (ATRIA, 1–5, 6, and ≥7; CHADS2, 0, 1, and ≥2; and CHA2DS2-VASc, 0, 1, and ≥2 points) and finally using cut-off points adjusted to optimize their fits for ischaemic stroke incidences of <1%/year (low risk), 1–2%/year (moderate risk), and >2%/year (high risk). Using the entire point score, the ATRIA score had a C-index which was minimally but significantly higher than that of either the CHADS2 or the CHA2DS2-VASc schema, and using the published cut-off points the C-index for the ATRIA score was still higher than that of the CHA2DS2-VASc schema. When the categorical scores were altered to optimize the fits for ischaemic stroke incidences, all C-indices were similar. Using the net reclassification index approach, the ATRIA score was favoured over both the CHADS2 and the CHA2DS2-VASc schemas, but when the categorical scores were optimized there was no longer an improvement with the ATRIA score. Do these findings mean that national guidelines groups should consider replacing the CHADS2 and the CHA2DS2-VASc schemas with the ATRIA schema? The clinician is faced by the necessity to evaluate the risks of ischaemic stroke compared with the risk of major bleeding with antithrombotic therapy and to discuss these competing risks with the patient to reach a decision regarding appropriate antithrombotic therapy. A risk schema based upon parameters readily obtained at the bedside and that is easily remembered by the clinician is desirable. The CHADS2 has the advantage of simplicity, whereas the CHA2DS2-VASc is more complex and the ATRIA is even more so, probably requiring access to the electronic health record and the use of a computing device. An annual stroke risk of 2% would prompt most clinicians to prescribe an oral anticoagulant, a risk of <1% would be unlikely to do so, and a risk of 1–2% might result in a range of actions, with patient input being particularly pertinent. A shortcoming of the CHADS2 is its relative insensitivity to differences in risk within the category of a CHADS2 = 0, where the annual risk of stroke may vary from ∼0.5% to >2%.7,9 Both the CHA2DS2-VASc and the ATRIA schemas allow the delineation of individual patient risks within this range, primarily because of the inclusion of additional age categories. The national guidelines groups have taken these factors into consideration in their most recent recommendations, incorporating use of either the CHA2DS2-VASc,7,8 factors from both CHA2DS2-VASc and CHADS2,9 or the CHADS2 alone.10 Although in the Aspberg study, for most of the comparisons the performance of the ATRIA was slightly better than that of the CHADS2 and the CHA2DS2-VASc schemas,11 other reports have found the CHA2DS2-VASc performs slightly better than the ATRIA.13 It appears unlikely that the ATRIA score will be adopted by national guidelines groups. What else can we learn from perusal of the Aspberg paper? (i) There are marked differences in stroke risk between the various age categories, providing a rationale assigning increasing risk as age increases (<65, 65–74, 75–84, and ≥85) and also among various risk factors which may each confer only a single point in a risk score, but signify quite different risks.11,12,14 (ii) Comparisons of performance among various risk schemas are likely to be more useful (and the differences perhaps less) if the cut-off points are adjusted to optimize their fits with ischaemic stroke incidences of <1%/year (low risk), 1–2%/year (moderate risk), and >2%/year (high risk). (iii) The outcome chosen for the assessment of stroke risk varies importantly among the various validations, resulting in substantial differences in estimated annual risk for the various risk schemas. The ATRIA validation in the study of Aspberg et al. has selected an outcome of ischaemic stroke only, whereas the CHA2DS2-VASc was derived using a composite outcome of ischaemic stroke, systemic embolism, or pulmonary embolism.5 (iv) It appears that the rates of the various thrombo-embolic outcomes vary considerably among populations, with immediate implications for the clinician wishing to estimate stroke risk in a given patient.11–14 As the population ages, the prevalence of AF continues to rise, as does the population incidence of disabling and fatal stroke. National clinical practice guidelines organizations strive to provide risk schemas that provide clinicians with practical tools for accurately estimating annual stroke risk from AF, balancing them against major bleeding risks from antithrombotic therapies [which remain substantial even with the advent of the novel oral anticoagulants (NOACs)] and providing the context for meaningful discussion with their patients. Studies such as that of Aspberg et al. are needed to provide the evidential basis for developing such guidelines. Conflict of interest: none declared.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.