Roaming through methodology. III. Randomization at the level of the physicians

In some studies randomization at the patient level is not possible; randomization at the doctor's level can then be applied. However, statistical analysis at the level of the individual patient data can lead to wrong conclusions in such a study design. By basing the analysis on the averages per doctor a statistical test can be performed which guarantees the correct significance level. The correct sample size can be found by means of a correction factor that depends on the variation between the doctors. The data of a study with randomization at the doctor's level can be analysed in a more sophisticated way by means of so-called 'mixed models' that take account of possible variation between doctors.