Improving prediction of mortality, stroke and major bleeding in patients with atrial fibrillation: external validation of the GARFIELD-AF tool in UK primary care electronic records

Abstract

Abstract Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National Institute for Health Research (NIHR) Introduction The GARFIELD-AF tool is a novel risk tool that simultaneously assesses the risk of mortality, stroke or systemic embolism, and major bleeding in patients with atrial fibrillation (AF). The tool was derived from the global GARFIELD-AF registry in 2017 and a new version was published in 2021. Initial evaluations indicate that the tools are superior to CHA2DS2VASc in predicting ischemic stroke/systemic embolism and HAS-BLED in predicting the risk of major bleeding, however it has not been validated in the UK population. Purpose To validate the GARFIELD-AF risk prediction tools for all cause-mortality, stroke and major bleeding in patients with AF in UK primary care electronic records and compare with current tools (CHA2DS2VASc and HAS-BLED). Methods We used Clinical Practice Research Datalink (CPRD), an electronic primary care database comprising of anonymised patient medical records from general practitioners, linked with Hospital Episode Statistics data and Office for National Statistics mortality data. The covariates for the GARFIELD-AF models include demographic and lifestyle factors, clinical observations at diagnosis, and medical history. The performance of the models was assessed by examining discrimination and calibration at 1 month, 1 year and 2 years in the CPRD cohort. Discrimination was evaluated using the C-index and calibration was evaluated using calibration-in-the-large regression and calibration plots. Subgroup analysis was conducted according to risk stratification of stroke and bleeding and for individuals receiving anticoagulation or no anticoagulation at baseline. The discrimination of the models was compared with the current risk stratification tools CHA2DS2VASc and HAS-BLED. Results The validation cohort comprised of 486,818 adults aged ≥18 years, with incident diagnosis of non-valvular AF between 2 January 1998 and 31 July 2020. 83% had a CHA2DS2VASc≥2 and 50.6% received anticoagulation at diagnosis. The GARFIELD-AF models outperformed the CHA2DS2VASc and HAS-BLED tools in discrimination ability of death, stroke, and major bleeding at all the time points (Figure 1). The c-statistics for events at one year of the 2017 models were: death 0.747 (95% CI, 0.744 to 0.751) vs 0.635 (95% CI, 0.631 to 0.639) for CHA2DS2VASc; stroke 0.666 (95% CI, 0.663 to 0.669) vs 0.625 (95% CI, 0.622 to 0.628) for CHA2DS2VASc; and major bleeding 0.602 (95% CI, 0.598 to 0.606) vs 0.558; 95% CI, 0.554 to 0.562). The 2021 models had a similar performance to the 2017 models (Figure 1). Calibration between predicted and Kaplan-Meier observed events was inadequate (Figure 2). Conclusions The GARFIELD models were superior to the CHA2DS2VASc score for discriminating stroke and death and to the HAS-BLED score for discriminating major bleeding. The GARFIELD-AF models consistently under-predicted the level of risk, suggesting that a recalibration is needed to optimise its prediction in the UK population.