Risk based approach for batch release

Abstract Release of manufactured lots is a major decision point in reaching the goal of a customer – the product quality is reasonably good – and of manufacture – no rework/scrap/lost revenue. In this study we will focus on pharmaceutical industry (solid dosage form). Adaptive Bayesian sample size determination. In this study we used adaptive Bayesian approach to determine the number of lots needed to estimate inter/intra lot variability. This was dictated by limited number of lots available in the study. Specifically, a linear random effects model was built and fitted to tablet data from multiple lots. The estimate was considered accurate enough when change in signal (or variability estimates) disappeared within noise. In our case we considered 1% of label claim as noise level. However, if signal to noise ratio was high, additional lot was sampled. Sampling included traditional content uniformity and dissolution testing. In order to be consistent with pharmacopeia 10 tablets from a lot were used for content uniformity testing while 12 tablets from the same lot were used for dissolution testing. Manufacturer, however, may not need this step since historical production information may be used to estimate the inter/intra lot variability by employing Bayesian methodology. It is important to note that analytical variability component was factored out which is not easily done with traditional frequentist methodology. Estimation of probability that product is within specifications at given confidence level. The next step was to estimate the probability of product being within specifications at a given confidence level which is metrics used for lot release by ASTM E2709–12 standard. However, the approach delineated in the standard was extended by Bayesian treatment. Specifically, inter/intra lot variability estimates were used to determine the probability while confidence level was replaced by the Bayesian credible interval concept. As delineated in ASTM E2709–12 standard this non-trivial task using frequentist methodology was very straightforward when using Bayesian methodology. We propose that lot can be released (consistently with ASTM E2709–12 standard) when the probability of product being within specifications at a given confidence level is high enough not to cause any harm to the patient. In order to determine the probability threshold, we propose to perform virtual clinical study with Simcyp or PK-Sim to determine if there is no harm to the patient and drug product is efficacious, i.e. drug plasma levels are within therapeutic window for prescribed dosing regimen. We verified the proposed methodology on seven immediate release products from different manufactures. We found that there are large differences in inter/intra lot variability between different manufacturers. Based on our experience more than six lots are typically required to establish accurate variability estimate. The current pharmacopeia doesn’t incorporate inter lot variability and therefore our proposed approach is more restrictive in the sense that it allows releasing lots that contain safer and more efficacious product than using existing lot release criteria. It is based on FDA’s risk based approach. From manufacturers perspective it supports the FDA’s quality metrics initiative to assure that only safe and quality medicine is being produced. While this study considered solid dosage form the proposed methodology could be easily generalized for different dosage forms and any batch processing industries.