Wang Suzhen, Xia Jielai, Yu Lili, Li Chanjuan, Xu Li
Department of Health Statistics, Faculty of Preventative Medicine, Fourth Military Medical University, Xi'an, Shaanxi 710032, China.
Comput Methods Programs Biomed. 2008 Apr;90(1):66-88. doi: 10.1016/j.cmpb.2007.11.016. Epub 2008 Jan 14.
An unnecessarily high or inadequately low sample size often occurs in clinical trials if the planned variance of the trials is overestimated or underestimated. Internal pilot study which utilizes the information from the patients recruited up to interim stage can solve this problem well by re-estimating the variance and re-calculating the sample size. The trial may get a satisfactory power but the type I error rate may be inflated while the t-test is adopted to make hypothesis test because condition of t-distribution is not sufficed any more with variation of the sample size resulted from internal pilot design. If blind variance estimators of the internal pilot are used for sample size recalculation and randomization test is used to accomplish the final hypothesis test, not only the blindness of the internal pilot is preserved but also the ability to control the type I error rate is guaranteed. A SAS macro is programmed to simulate the process of sample size adjustment and randomization test.
如果临床试验计划的方差被高估或低估，往往会出现样本量过高或过低的情况。利用中期之前招募患者的信息进行的内部预试验研究，可以通过重新估计方差和重新计算样本量来很好地解决这个问题。虽然该试验可能具有令人满意的检验效能，但在采用t检验进行假设检验时，由于内部预试验设计导致样本量变化，t分布条件不再满足，可能会使I型错误率膨胀。如果使用内部预试验的盲法方差估计量重新计算样本量，并使用随机化检验完成最终的假设检验，不仅可以保持内部预试验的盲法，还能保证控制I型错误率的能力。编写了一个SAS宏来模拟样本量调整和随机化检验的过程。
