Determination of Significance Levels for Stability Pooling Test:a Simulation Approach*

1. Introduction In the last 15 years, for the standard three-batch stability analyses, debates focused on what is the significance level to be used in the stability pooling tests (pooling intercepts and pooling slopes). Because of the lack of sample size and power consideration, FDA adopted Bancroft's [1] recommendation of using 0.25 as the significant level at any of the pooling tests in order to maintain some control of the false negative (pooling due to no power to reject null hypothesis) rate. The practice was often criticized by industry statisticians as non-scientific and over-conservative. In more recent times, the drug industries eagerly propose the matrixing and bracketing stability design to explore the drug stability. The goal of the matrixing and bracketing stability design tries to reduce sampling time points for each batch and use simpler model by removing the non-significant slope and intercept differences, formed by factor combinations, from the preliminary specified model to estimate the expiration dates for combinations of factor levels. Since the structure of the bracketing and matrixing stability design is much more complex than that of the three-batch design, the effect of the false negative result on the determination of the expiration dates for bracketing and matrixing stability design is more serious than that of three-batch stability design. Therefore, especially for the bracketing and matrixing stability design, a statistically sound methodology should be developed to resolve the issue on the false negative results of the pooling tests.