|
Abstract:
|
In single-arm studies of time to event data, hypothesis testing can be done by estimating the failure rate using the Kaplan-Meier method, calculating the variance using Greenwood's formula, and using a Wald test to compare the estimated failure rate against a prespecified performance goal. If there is no right censoring, this test is equivalent to a binomial test on the observed proportion of failures using a normal approximation. Because of the discreteness of the binomial distribution, the type-1 error and power of the Wald test can fluctuate with the sample size, even decreasing locally as the sample size increases.
Additional properties of the Wald test were explored in the setting of a group sequential design with an interim analysis after half of the subjects have completed the full follow-up Specifically, the behavior of the power and type-1 error of the test are examined as a function of the right censoring rate and the enrollment rate. The power and type-1 error are not dependent on the enrollment rate. In contrast, the power tends to decrease and the type-1 error tends to increase as the right censoring rate increases.
|
