| A number of issues have converged in recent years to focus increasing attention on the 'interval hypothesis test' following the improvement in medicine. 'Interval hypothesis test' is a new test method which is based on the traditional significance test and includes noninferiority test, equivalence test and superiority test. It is named with 'interval hypothesis test' or 'interval test' by Shuirmann in 1987 because its test hypothesis is not a point but a interval. It has played an important role in the development and evaluation of new drug since the last century and has being used widely in varied fields, such as Social Science, pedagogy and morphology, and so on. In the area of medicine and health statistics, suppose that a test formulation and referential formulation are to be tested whether one is noninferior / equivalent / superior against the other , the trditional hypothesis test is not sufficient. On the contrary, 'interval test' is more important to ensure the noninferiority /equivalence /superiority between test formulation and referential formulation and make sure their therapetic efficacy and safety.Based on the definition of 'interval test' and its difference and relationship with traditional test, we analyzed power of test and estimated sample size for interval test by Monte Carlo simulations: Simulated and affirmed whether 3 should be one-sided or two-sided for interval test ; Computed the power of test and sample size for equivelence test by varying methods and with different parameters; Calculated power and sample size for those three types of interval test, i.e. noninferiority test, equivelence test and superiority test, with identical designs and different parameters; Estimated primarily the influence of center and baseline on sample size and power of test in multicenter clinical trials. A series of SAS procedures had been provided for sampling and simulating. The main works and results of the study are as follows:1. We summarized the factors which affected power and sample size in traditional and interval test such as and , and introduced the meaning of a and 3 in detail, what's more, we pointed out the divergence in literatures about this topic. Then the Monte Carlo simulations were used to estimate power in a given sample size calculated by traditional formulation with different designs, different 9 and CV. The results indicated that: As far as noninferiorirty and superiority are concerned, 3 is one-sided; but for equivalence test:2. Phillips method and Monte Carlo simulations were chosen to estimate sample size for equivalence test with four different designs such as one sample equivalence design, paired equivalence design, two paralle sample design and 2X2 cross-over design. The sample size and power estimated by Phillips method was compared with those of simulations and traditional method and indicated that the results ofPhillips were closer to those of simulations and it can eliminate the traditional method's defect; In addition, Monte Carlo simulations were performed to estimate sample size for the three different interval tests, in general, the interval bound is in ascending in test: noninferiority test, equivalence test, superiority test, however, we can't conclude that sample sizes of them are in the same ascending order which is affected by the significance level and many other factors.3. Evaluated the influence of central effect and number of center on sample size and power of test in multicenter trial by simulations with varying number of center and varying parameters. The outcomes displayed that the central effect shouldn't be overlooked, otherwise, we would underestimate the correct sample size. On the contrary, the influnce of number of center seemed to be insignificant.4. Evaluated the effect of baseline in multicenter trial by simulating and computing power of test with different number of center, different parameters relying on a covariance analysis model. The outcomes suggested we shouldn't neglected t... |
PDF Full Text Request