Information-based monitoring of clinical trials

Abstract

When designing a clinical trial to compare the effect of different treatments on response, a key issue facing the statistician is to determine how large a study is necessary to detect a clinically important difference with sufficient power. This is the case whether the study will be analysed only once (single-analysis) or whether it will be monitored periodically with the possibility of early stopping (group-sequential). Standard sample size calculations are based on both the magnitude of difference that is considered clinically important as well as values for the nuisance parameters in the statistical model. For planning purposes, best guesses are made for the value of the nuisance parameters and these are used to determine the sample size. However, if these guesses are incorrect this will affect the subsequent power to detect the clinically important difference. It is argued in this paper that statistical precision is directly related to Statistical Information and that the study should continue until the requisite statistical information is obtained. This is referred to as information-based design and analysis of clinical trials. We also argue that this type of methodology is best suited with group-sequential trials which monitor the data periodically and allow for estimation of the statistical information as the study progresses. Copyright © 2006 John Wiley & Sons, Ltd.