Risk of acute myocardial infarction with NSAIDs in real world use: bayesian meta-analysis of individual patient data