Tighter specifications result in limited ability of the sponsor to extend the shelf-life when additional long-term data become available and an increased risk of drug shortages caused by the rejection of safe and effective batches. The scenario of having limited data at the time of registration is a frequent situation for new products and lends itself to the consideration of alternative specification modes such as adopting an interim specification that is again reviewed when additional data are available. On the other hand, some tests that are not driven by process capability, including assay, content uniformity, and pharmacopeial tests, would not require this strategy and could be unambiguously set at the time of filing. In those cases, the development task is to determine whether the product can meet these standard specification requirements.

By contrast, the conventional approach used to establish acceptance criteria that are not clearly dictated by pharmacopeias or guidance is as follows:

• Establish a safety threshold or set of product performance-based requirements that provide outer boundaries for possible acceptance limits.
• Gather development information concerning the robustness of the proposed process to understand the stability behavior of the drug substance and product.
• Based on the above information and the manufacture of a relatively small number of development batches, propose a limit that reflects the process capability of the proposed process.

Several current initiatives are prompting considerable discussion on other approaches to setting specifications. However, because the conventional approach is often followed and was discussed extensively at the September 2003 AAP workshop, this paradigm is assumed for much of this article.

Ruggedness of acceptance criteria

In general terms, establishing a limit based on process capability starts with estimating both the mean and the variance (or some measure of variability) of the data for a particular quality attribute. In setting the limit, these estimates are then used to establish a criterion that provides both an acceptable operational window and addresses all safety and efficacy considerations. Several approaches for calculating a limit have been proposed. The ICH Q6A and Q6B guidances (1, 2) suggest using the mean plus three standard deviations for impurities or a related variant for degradation products. A tolerance limit or prediction-interval approach, which considers individual values rather than focusing on the mean, is preferred by industry representatives. (Tolerance limits or prediction intervals are concerned with inference about single observations. The concept is to establish limits on expected individual values based on knowledge of the population parameters. These statistics estimate a range in which future individual results are expected to fall within a specified confidence.) The most conservative approach is to set the limit not to exceed the values associated with lots in the development program. Note that the situation is more complex for specifications where some change over time is expected and must be accommodated, such as for degradation products.