Stability models for sequential storage

Some drugs are intended for sequential storage under two different storage conditions. If the data for each condition are analyzed separately, predicting assay and other responses after T1 months at one condition followed by T2 months at the other condition is non-trivial for several reasons. First, the two analyses will give different intercept terms. What should one do about that? Second, how would one calculate the confidence limits for combined storage? Third, what if prior storage at one condition affects the slope at the other condition? This paper proposes a simple ANCOVA model containing two slope terms, one for each storage condition. When multiple batches and/or packages are involved, it is easily generalized to two sets of slope terms. Confidence limits are straightforward and can be calculated using existing commercial software. With properly designed data, one can test whether prior storage at one condition affects the slope at the other condition. If no such effect is significant, very useful extrapolations can be made. Temperature excursions, model reduction and curvilinear dependencies are discussed.