• atrial fibrillation
• stroke

A randomised controlled trial (RCT) is a research method that focuses on comparing interventions. Currently, RCTs are the gold standard study design to compare the efficacy and safety of different therapeutic modalities. Random allocation can control the measured and unmeasured confounding factors and manage the risk of ‘confounding by indication’, which cannot be solved in observational studies. These characteristics could result in controlling internal and external validity, intervention and outcomes of the study. Based on these, RCTs are placed above observational studies in the hierarchy of study design strength. However, RCTs also have several limitations: strict enrolment criteria, which would not cover and answer all of the questions that arise in everyday clinical practice; expensive and long study duration. Furthermore, the clinical equipoise or uncertainty principle is a basic assumption in the design of RCTs.1 The prior probability of RCTs is 50–50, which is higher than that of observational studies.2 In addition, the uncertainty principle can be violated in research sponsored by pharmaceutical companies, which are likely to have positive results.1

Observation is one of the traditional research methods used to discover new findings and to explain the cause of disease. Observational study results are then validated by other scientists in different subgroups and under various conditions using diverse analytic methods. Compared with RCTs, observational studies have weaknesses in terms of risk of confounding, multiplicity of analysis and subgroups. Replication is needed to solve the problems of multiplicity and subgroup analysis. Different study designs, different study populations, different data sets and analyses would be …