Background: Large phase III clinical trials convey a lot of important information besides the main analysis of the treatment effect. For example, the use of multicenter clinical trial data to identify prognostic indices is now common. In addition, the study of heterogeneity in patient outcome between centers has received considerable attention in recent years. In this paper, we explain and illustrate a method used to investigate such heterogeneity with data from an early breast cancer clinical trial.
Methods: The inclusion of a random effect for center in a Cox proportional hazards model allows us to study the heterogeneity in time-to-event outcomes between centers. Such a model has the major advantage that it provides a measure of the spread of outcomes over centers. This technique is illustrated using data from EORTC trial 10854, a randomized phase III trial comparing perioperative chemotherapy with no perioperative chemotherapy for early breast cancer; 2793 patients were entered by 14 centers.
Results: Substantial heterogeneity between centers was detected for disease-free survival. This can be explained by the geographical area in which the center is located, with better outcomes achieved in France as compared with southern Europe and South Africa. None of the prognostic factors considered could explain this heterogeneity.
Conclusion: Although clinical trials are run with the objective of removing as much heterogeneity as possible, some heterogeneity in the outcome of patients between centers may remain, as was the case in our study. The use of a random effect for center within a Cox PH model is an excellent method to investigate this heterogeneity. Such types of analyses, although exploratory, provide further insight into possible factors which may have an impact on the patient's outcome.