Patients further were excluded if they switched antiplatelet therapy between aspirin and clopidogrel during the follow-up period to make the analyses straightforward. The Taiwan National Health Insurance Bureau provides reimbursement for the use of clopidogrel in patients with ischaemic stroke who are allergic to aspirin or have peptic ulcer (the latter confirmed by prior or current pan-endoscopy results). Although ‘aspirin treatment failure’ is not one of the prespecified criteria for clopidogrel use, the Bureau typically provides reimbursement in these circumstances. As such, physicians generally have broad latitude to prescribe clopidogrel

or aspirin based on their personal preferences. Patients were excluded if their medication possession ratio (number of days drug supplied divided by the number of days in the follow-up period) was <80% or clopidogrel or aspirin was not prescribed within 30 days before an end point to reduce bias from poor drug adherence or antiplatelet-discontinuation

effects.11 12 Main outcome measures The primary end point was the first event of a new-onset major adverse cardiovascular event (MACE: composite of any stroke (ischaemic or haemorrhagic) or myocardial infarction). The leading secondary end point was the first event of any recurrent stroke (ischaemic or haemorrhagic) alone. Additional secondary end points were ischaemic stroke, intracranial haemorrhage (codes 430–432), fatal stroke, myocardial infarction (code 410) and all-cause

mortality. Follow-up was from time of the index stroke to admission for the first event of recurrent stroke (codes 430–434, 436) or myocardial infarction, death, or the end of 2010. National Health Insurance is a compulsory programme in Taiwan, and moving out of the country, which is supposed to be scarce among patients with stroke, is almost the only reason, besides death, for being withdrawn from this programme. A previous study from the Taiwan NHIRD also used ‘withdrawn’ from this programme to define death.13 Therefore, we defined death as in-hospital death or withdrawal of the patient from the National Health Insurance programme. Statistical analysis The baseline characteristics of two treatment groups were compared using student t test for continuous variables and χ2 test for categorical variables. Kaplan-Meier plots were generated, Brefeldin_A and the log-rank test was used to evaluate the difference between curves. We employed Cox’s proportional hazard model to estimate the unadjusted and adjusted HRs and 95% CIs, which considered the aspirin group as the reference group. The model was adjusted for baseline age, gender, hypertension, diabetes, prior stroke, prior ischaemic heart disease, hyperlipidaemia, gastrointestinal bleeding or peptic ulcer, Charlson index, statin use, other antiplatelet drugs use, ACE inhibitors or angiotensin receptor blockers use, calcium channel blockers use and diuretics use during the follow-up period.