Background: the primary prevention of ischaemic stroke in chronic non-valvular atrial fibrillation (AF) typically involves consideration of aspirin or warfarin. CHA₂DS₂-VASc estimates annual stroke rates for untreated AF patients, which are reduced by 60% with warfarin and by 20% with aspirin. HAS-BLED estimates annual rates of major bleeding on warfarin. The latter risk with aspirin is 0.5–1.2% per year.

Hypothesis: given a ‘warfarin, aspirin or no therapy’ choice, AF patients will prefer the option that maximises the annual probability of not having a stroke and not having a major bleed.

Methods: decision tree applied to the 60 possible combinations of CHA₂DS₂-VASc and HAS-BLED scores.

Results: according to the pre-specified hypothesis, when CHA₂DS₂-VASc is <2, the balance of risk and benefit would advise no treatment; when CHA₂DS₂-VASc is 2 or 3, warfarin would be best when HAS-BLED <2, otherwise no treatment would be advised; for CHA₂DS₂-VASc =4, warfarin would be best when HAS-BLED <3, otherwise no treatment would be advised and for CHA₂DS₂-VASc ≥5, warfarin would be the preferred option if HAS-BLED <4, otherwise aspirin would be advised.

Conclusion: this theoretical exercise illustrates the potential benefit of decision analysis in an area where high complexity and uncertainty still remain.