A Statistical Method Giving Early and Unbiased Conclusions from Clinical Trials

This article summarizes research in statistical methodology contained in a doctoral thesis by John G. Spritzler. A new statistical method is developed for comparing two response rates in a randomized clinical trial, taking into account two common problems. First, how can the trial be analyzed at multiple interim analysis times while still limiting the probability of falsely declaring a treatment difference when there is, in fact, none? Second, how can unbiased interim analyses be done which use all of the available information, if the time, referred to as “lag time,” between when a patient enrolls in the trial and when his or her response status is reported to the statistical center varies randomly from patient to patient, in an unknown manner? The first question is addressed by well-known methods of doing what are referred to as “group sequential” tests. The second question requires the new “lag adjusted” methodology presented in this article. The nature of these two problems in the randomized clinical trial setting is explained, and how a lag time in reporting clinical responses can introduce a bias into the standard group sequential statistical methods is discussed. The “lag adjusted” method is introduced, the principles upon which it is based explained, and then the results of computer simulated clinical trials that demonstrate how the “lag adjusted” method compares with the standard method of analysis, under various clinical trial conditions, are presented.