Candidate variables were chosen as variables associated with current anticoagulation treatment at admission in the bivariate analysis, with a value of p < 0.25, and consisting of clinical variables assessable before stroke onset: age (years), male sex, previously known atrial fibrillation, mitral stenosis, previous systemic emboli, depression, peripheral artery disease, arterial hypertension, hyperlipidaemia, other high risk cardiopathies, cardioversion, digoxin, β blockers, calcium channel blockers, and centre = Paris (all quoted 0 = no and 1 = yes). Colinearity between variables (defined as r > 0.6) was excluded. Overall prediction of the model 84.98%.

β, regression coefficient; SE, standard error of the mean; constant = −1.7602.