ABSTRACT
As indicated earlier, to establish an expiration dating period, the FDA stability guidelines require that at least three batches, and preferably more, be tested in stability analysis to account for batch-to-batch variation so that a single shelf-life is applicable to all future batches manufactured under similar circumstances. Under the assumption that the drug characteristic decreases linearly over time, the FDA stability guidelines indicate that if there is no documented evidence for batch-to-batch variation (i.e., all the batches have the same shelf-life), the single shelf-life can be determined, based on the ordinary least squares method, as the time point at which the 95% lower confidence bound for the mean degradation curve of the drug characteristic intersects the approved lower specification limit. Several methods for determination of drug shelf-life have been proposed, as discussed in detail in the previous chapter. As indicated in the 1987 FDA stability guideline, the batches used in long-term stability studies for establishment of drug shelf-life should constitute a random sample from the population of future production batches. In addition, the guidelines require that all estimated expiration dating periods be applicable to all future batches. In this case the statistical methods discussed in the previous chapter, which are derived under a fixed effects model, may not be appropriate. This is because statistical inference about the expiration dating period obtained from a fixed effects model can only be made to the batches under study and cannot be applied to future batches. Since the ultimate goal of a stability study is to apply the established expiration dating period to the population of all future production batches, statistical methods based on a random effects model seem more appropriate. In the past two decades several statistical methods for determining drug shelf-life with random batches have been proposed. See, for example, Chow and Shao (1989, 1991), Murphy and Weisman (1990), Chow (1992), Ho, Liu, and Chow (1993), Shao and Chow (1994), and Shao and Chen (1997).
