On the detection of outlier clinics in medical and surgical trials: I. Practical considerations

In the conduct of multicenter clinical trials, comparisons among the clinical centers are often made with respect to such measures as incidence of diagnosis of various events, operative mortality (in the case of a surgical trial), and patient adherence to the treatment regimen. If one or two clinics seem to stand out from all the other clinics with respect to any such measures, this may indicate-among other things-differences in skill levels or possible deficiencies in understanding and following the protocol among these clinics. On the other hand, when clinics are ranked on any measure, one clinic must of necessity be first and one last; thus, it is necessary to determine whether the extreme values represent true differences from the values of the other clinics or whether they can be attributed to random variation alone. A statistical test is described for detecting clinics that are outliers-i.e., truly different from all the rest of the clinics in the trial. Critical values for this test are given for various numbers of clinics, numbers of patients enrolled per clinic, and probabilities of a patient incurring a given event under the null hypothesis of no differences among the clinics. Use of this test in detecting outlier clinics with respect to incidence of various endpoints is illustrated for the case of a large, multicenter drug trial.