A Predictive Distribution Approach to QbD: Going Beyond the Flaw of Averages to Assess Risk

• Activity Number: 411
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): John Peterson*+
• Companies: GlaxoSmithKline
• Keywords: Response Surface Methodology ; Bayesian ; Quality by Design ; Nonclinical Statistics

Abstract

Quality improvement has been described as "reduction in variation about a target". Such reduction is driven by the desire to have a high probability of meeting process specifications. Proper quantification of such a probability requires construction of a predictive distribution. Unfortunately, many statistical quantifications and decisions are focused only on means, without careful thought to risk assessment (e.g. the design space example in the ICH Q8(R2) Figure 2c). This is curious, as the ICH Quality Guidances are filled with phrases related to 'risk' and 'risk-based'. This talk will discuss some recent examples of QbD inference related to mean process performance and contrast them with those that utilize predictive distributions to assess process reliability and risk. In addition, this presentation will discuss the use of Bayesian methodology to construct predictive distributions for process optimization and risk assessment. This talk will also address modeling building to incorporate various sources of variability, such as batch-to-batch variation for more realistic risk assessment.