Back to Results

< Previous   Next >

Misspecification of a nuisance parameter can lead to study power far from the desired level. Internal pilots for Gaussian data protect study power by allowing sample size re-estimation based on an interim power analysis using a revised estimate of the variance parameter, but without any data analysis. In order to reduce study time and cost, researchers and sponsors of studies often desire early decision possibilities that the internal pilot design lacks but that group sequential methods allow. Combining early stopping rules with internal pilot methods would increase study flexibility, scope, and efficiency for general linear models. An internal pilot with an interim analysis (IPIA) design for Gaussian linear models is introduced and defined. The design allows for early stopping for efficacy and futility while also re-estimating sample size needs based on an interim variance estimate. In order for accurate study planning in small samples, exact theory is derived for both the one or two group [tau] test setting, as well as more complex multiple degree of freedom hypothesis tests within the general linear univariate model framework. Exact and computable forms of distributions allow accurate calculations of power, type I error rate, and expected sample size. In general, the IPIA design maintains and controls power to the desired level and also provides sample size savings. However, it can also inflate type I error rate, especially in smaller studies. By utilizing the exact theory, planning procedures associated with the design are examined and refined to create a working method for planning sound studies. A bounding method successfully controls the type I error rate while maintaining the benefits of the design. Explicit recommendations are detailed that achieve the combined goals of an internal pilot and a two-stage group sequential design. The results can be used during planning to create an efficient two-stage study with early stopping rules and predictable power properties, even in small samples.