|
Abstract:
|
The specification of acceptable ranges for various critical quality attributes (CQAs) of a given compound should be built into the manufacturing process, based on accumulated process knowledge and specific scientific domain knowledge, using coherent statistical methods. The quality of the final drug product is measured via CQAs, which in are affected by known and unknown sources of variation. Acceptance limits should be primarily based on patient safety. However, statistical methods are needed to properly quantify the variation of a given CQA. JoS typically use tolerance intervals. We will discuss why this approach may be inappropriate, and suggest alternatively the use of the posterior predictive distribution of a CQA. Assuming data are available for batches of the product through the development process, the predictive distribution can be used to determine the quantiles that will contain the next reportable value of the CQA from a future batch of the product with a given probability. This approach incorporates previous measurable data in a mathematically coherent way via the specification of prior distributions on model parameters.
|
