A new risk score, based on data from GARFIELD-AF (Global anticoagulation registry in the field), may be superior to the CHA2DS2-VASc score at predicting which patients truly have a low-risk of stroke. Therefore, the score—a simplified form of which can be used online—may be more able to identify which patients with a traditionally low-risk of stroke would benefit from anticoagulation.
Speaking at the European Society of Cardiology Congress (ESC; 27–31 August, Rome, Italy), Keith Fox (Centre for Cardiovascular Science, University of Edinburgh, Edinburgh, UK) reported that identifying patients with a truly low-risk of stroke is “an ongoing clinical challenge”. He noted that while, according to data from GARFIELD-AF, which has now enrolled 57,262 patients, the use of anticoagulation has increased over time, there are some critical questions—such as whether the use of anticoagulation had increased “in the right patients” or whether patients with a truly low-risk of stroke (and thus, do not need anticoagulation) are “well characterised”. “Data from GARFIELD-AF show that, contrary to guideline recommendations, 28% of high-risk patients, based on the CHA2DS2-VASc score, are not anticoagulated. Even more frightening, 51% of low-risk patients are anticoagulated. This suggests to us that there are some factors beyond those in current risk scores that appear to influence prescribing decisions on anticoagulation, including bleeding,” Fox added.
Therefore, Fox and colleagues hypothesised that a more accurate and “more user friendly” score would enable more guideline-based prescribing habits—ie. fewer patients at low-risk of score would be on anticoagulation. With the GARFIELD-AF score, using a computer-generated machine learning model, patients were stratified according to their risk of death, stroke and bleeding; this is the first time a stroke risk score for atrial fibrillation has incorporated bleeding. Furthermore, they used a coalescent regression model rather than a multivariate regression model. Fox explained: “Coalescent regression avoids the need to specify levels of relatedness in the statistical model; it allows joint modelling of all outcomes. So it means each of the factors can be independently analysed in relation to the multivariate analysis. Also, it can be used in electronic systems to automatically extract the risk score.”
Including data from 38,984 patients from the GARFIELD-AF registry, Fox et al compared the predictive ability of the GARFIELD-AF score to that of the CHA2DS2-VASc score. They found it better predicted all-cause mortality in all patients (0.78 C statistic vs. 0.66 for the CHA2DS2-VASc score) and haemorrhagic stroke/major bleeding (0.67 C statistic vs. 0.61 C statistic, respectively) whereas its predictive ability of ischaemic stroke/embolism was similar (both 0.63). However, for low-risk patients, the GARFIELD-AF score had superior prediction ability for all-cause mortality (0.72 C statistic vs. 0.56 C statistic), ischaemic stroke/systemic embolism (0.62 vs. 0.56), and haemorrhagic stroke/major bleeding (0.72 vs. 0.57).
A simplified GARFIELD-AF score, which can be used online or via mobile app, was evaluated using data from GARFIELD-AF and from ORBIT-I (Outcomes registry for better informed treatment of atrial fibrillation). This analysis showed that for all patients, the score had: a C statistic of 0.7 for ischaemic stroke/systemic embolism in GARFIELD-A; a C statistic of 0.69 for any stroke/systemic embolism in ORBIT-I; a C statistic of 0.68 for major bleeding in GARFIELD-A; and a C statistic of 0.61 for major bleeding in ORBIT-I. Fox commented that the HAS BLEED score has a C statistic of 0.64 and, therefore, “it is inferior” to the C statistic seen with the GARFIELD-AF score (for the GARFIELD-AF data).
Concluding his presentation, Fox commented that the GARFIELD-AF score “may help physicians to assess the appropriateness of anticoagulation in low-risk patients”. He added: “If we have one score that gives us mortality, stroke/systemic embolism and gives us bleeding, our hypothesis is that this will be user friendly and perhaps better able to guide clinical practice.”
Gerhard Hindricks (University of Leipzig Heart Center, Leipzig, Germany) Chairman of the Registries atrial fibrillation session, in which Fox presented the GARFIELD-AF data, told Cardiac Rhythm News that the GARFIELD-AF score was “a very interesting approach” and that the ability of identifying patients who require anticoagulation, and those who did not, was “a continuous process”. He added: “Further data are needed but I take this score very seriously. I think it is a significant contribution to the field and may be a significant improvement.”
