Clinical Trials

Clinical Trials: A Practical Guide ■❚❙❘Figure 2. Variation of the test statistic with respect to time in a simulated clinical trial with no treatment effect.The standardized normal test statistic is computed and plotted after each pair of patients in the trial.Horizontal dashed lines are drawn at zero and at ±1.96, the two-sided 5% boundary for the test statistic.4 –Standardized normal statistic2 –0––2 –|0|20|40|60|80|100Number of patientsFigure 2 shows a plot of the test statistic over the course of a simulated study in whichthere is no treatment effect. The test statistic could be many things, eg, thestandardized difference in means, the log odds ratio, or the log hazard ratio.Note that, in this simulation, the test statistic is above the two-sided 5% boundaryearly in the trial. This trial might have been stopped prematurely if hypothesis testingwere done after each pair of patients.The examples in Figures 1 and 2 highlight the need to exercise caution whenrepeating statistical tests during the course of a clinical trial. Interim monitoringtherefore involves the problem of multiplicity risking a Type 1 error occurring (seeChapter 29). There are many possible solutions to this problem. The most commonlyused solutions, and those that we focus on here, are group sequential methods.A thorough review can be found in Group Sequential Methods with Applications toClinical Trials by Jennison and Turnbull [2]. Other less common methods are basedon Bayesian statistics, decision theory, or conditional power analysis.357