OBJECTIVES: To augment the AF risk prediction tools CHA₂DS₂-VASc and HAS-BLED by adding more detail regarding the stroke event while maintaining the scoring structure of the original scale

METHODS: We linked individual-level hospital and prescribing datasets for all adult patients in Scotland with incident AF (2009 -2017.) Based on presumed differential risk, for CHA₂DS₂-VASc, we assigned ischaemic stroke two points, TIA one point, and haemorrhage zero points. For HAS-BLED we assigned haemorrhage two points, ischaemic stroke one point, and TIA zero points. We compared prognostic utility of original and augmented tools against outcomes of stroke, bleeding, mortality, and care home admission using discrimination and calibration plots.

RESULTS: We included 117,749 people with AF: mean age 73 years (SD:12.7), 48% female. Original and augmented tools performed similarly, for CHA₂DS₂-VASc prediction of stroke, AUC original:0.56 (95%CI:0.56-0.57), AUC augmented:0.57 (95%CI:0.56-0.57). For HAS-BLED prediction of bleeding, AUC original:0.53 (95%CI:0.51-0.54), AUC augmented:0.53 (95%CI:0.52-0.54). Patterns were similar for mortality and care-home outcomes.

CONCLUSIONS: It is possible to use routinely-recorded clinical data to augment AF risk prediction tools, However, improvements in prognostic utility were negligible. When applying CHA₂DS₂-VASc and HAS-BLED, any previous history of stroke is important regardless of pathology.