Establishing and Predicting Quality: Process Validation - Stage 1 Brad Evans / Kim Vukovinsky Pfizer May 20, 2015 Outline* • What statistical tools are used in PV Stage 1 and how do the results influence PV Stage 2? • How is the difference in scale addressed? • How are design space verification and PPQ related? • What is the role of variability in determining readiness for PV? (hmm, what about measurement uncertainty?) • Is it relevant to combine and analyze PV Stage 1 data with PV Stage 2 data? • What data is needed from PV Stage 1 in preparation for PV Stages 2 & 3? * Tools and topics are not equally distributed across all applications, e.g. mAbs, Vaccines, DP, API, Parenterals 2 Stages of Process Validation Stage 1: Process Design Stage 2: Process Qualification Stage 3: Continued Process Verification Pfizer Confidential │ 3 Process Validation Guidance • Guidance for Industry – Process Validation: General Principles and Practices • For purposes of this guidance, process validation is defined as the collection and evaluation of data, from the process design stage through commercial production, which establishes scientific evidence that a process is capable of consistently delivering quality product http://www.fda.gov/downloads/Drugs/Guidances/UCM070336.pdf 4 What Statistical tools are used in PV Stage 1? At a high level: • Visualization (“I love a good plot” Steve Novick) • Simple Descriptive Statistics • Statistical Intervals (Confidence, Prediction, Tolerance) • Sampling Plans • Monte Carlo Simulation • Messy Data Analysis Tools • Hypothesis Testing • Modeling • Design of Experiments 5 … and how do the results influence PV Stage 2? Design Space and Control Strategy The ICH Q8 Guidance* defines “Design Space” as: “The multidimensional combination and interaction of input variables (e.g. material attributes) and process parameters that have been demonstrated to provide assurance of quality”. However, knowledge of the parameters and their impacts does not assure quality. It is the Control Strategy that is critical in Assuring Quality. * http://www.ich.org/fileadmin/Public_Web_Site/ICH_Products/Guidelines/Quality/Q8_R1/Step4/Q8_R2_Guideline.pdf Quality Assurance … Process Understanding, Control Strategy, Specifications Controls Parameters, attributes, GMP, business Holistic control strategy Boundaries/ limits PARs, design space, release limits Product Efficacy, Patient safety, Reduced Cost to Society … assurance from the total quality system including the process definition + control strategy + testing … tight specifications are not the only way 7 Pfizer’s Right First Time / QbD Process Statistical Component Risk Assessment Multifactor Understanding: DOE + Data Models + Requirements Impurity1 = f(B, C) Impurity2 = f(B) Impurity1 < 0.1% Impurity2 < 0.1% Analysis + Visualization + Decisions Contour Plots: Two Responses, Two Process Parameters Want to be less than 0.10 for both impurities Impurity1 (%) 1.00 Impurity2 (%) 1.00 0.50 0.50 0.00 C C 0.03 0.07 3 0.02 0.00 0.10 0.05 0.09 0.07 0.11 3 -0.50 0.14 -0.50 0.17 -1.00 -1.00 -0.50 0.00 0.50 1.00 -1.00 -1.00 -0.50 0.00 B B 0.50 1.00 Overlay Plot of Two Responses vs. Two PP’s • Easy to implement (Design Expert) • Lends to “Edge of Failure” Terminology • “EOF” is misleading • Edge represents mean 50% failure (if model is perfect) • Blue dots have very different OOS rates Overlay: Two Responses, two Process Parameters • The probability of simultaneously passing the specifications varies within in the orange region – in fact it varies throughout the entire region • “Boundary” provides no greater than 50% probability of passing • Probability of meeting ALL specs decrease in areas of intersecting requirements • Reliability used to describe passing all Specs ~50% Prob < 50% Prob ~50% Prob 11 Prospective Process Reliability Estimate (PPRE) These levels curves now show the Reliability, the chance that the batch can be released This takes into account the predictive Distribution, not simply the Mean Prospective Process Reliability Estimate (PPRE) John J. Peterson, Guillermo Miró-Quesada and Enrique del Castillo, “A Bayesian Reliability Approach to Multiple Response Optimization with Seemingly Unrelated Regression Models”, Quality Technology & Quantitative Management, Vol. 6, No. 4, pp. 353-369, 2009. Data points New Betas Data Dist Counting. Specification Increase to Achieve Quality Requirements Decision Making - End Process Attribute Estimated Probability of Passing Original 0.1% Spec Estimated Probability of Passing New 0.3% Spec based on Safety Set Point Moved to Achieve Cost Target Decision Making - In Process Attribute Process adjusted so in process response acceptability is 80%. Response acceptability at process end >99.9% - next unit operations will achieve goal. Affects cost but not quality. Sets up continuous improvement opportunity; for Development or Manufacturing. How is the difference in scale addressed*? Two types of parameters: • Scale dependent: need strategy to assess DOE at scale (and life cycle change management understanding) • Scale independent or scalable: parameter that is scale independent (by model, science, equipment design) - run DoE’s at lab scale and results apply to scale. Examples: • Pressure, temperature are scale independent • Mixing rpm is scale dependent, w/kg is scale independent • High Sheer Granulator is scale dependent, Gerties roller compactors are scale independent * Garcia, Thomas, et. al. “Verification of Design Space Developed at Subscale”, Journal of Pharmaceutical Innovation, Vol 7, pg. 13-18 (2012). 16 Design Space Verification* Option: verification at set-point Option: Verify as required Option: Verify a region around setpoint * Garcia, Thomas, et. al. “Verification of Design Space Developed at Subscale”, Journal of Pharmaceutical Innovation, Vol.7, pg. 13-18 (2012). 17 What is the role of variability in determining readiness for PV? • As a next step within the QbD process, data from relevant batches are analyzed. • Create a Process Reliability Assessment (PRA) plot: QA’s • Process Understanding + data used to assess risk and support decision to commercialize process • What coverage, with 90% Confidence, fills Spec window? pH example (not shown) 6.2-6.8 data recorded to tenth: insufficient granularity Is it relevant to combine and analyze PV Stage 1 &2 data? Maybe 18 Control Strategy Implementation Activities Holistic strategy mitigates any risk from a single unit operation: e.g. in the step, a downstream purge, or an analytical test. Could include: – Facility/equipment qualification/ verification – Validation • Analytical methods, manufacturing, packaging, cleaning – Training • Operators, analysts, engineering/maintenance, technical support… • Understanding of product, process and control strategy – What are the potential risks during processing? – Which control strategy elements are the most critical? 19 Example Control Strategy for Dissolution This equation opens up different control strategy options 20 Statistics – Design Space – Control Strategy - PPQ How are design space verification and PPQ related? Verification Statistics tools: Risk mitigation, confidence, process/ product performance Statistics Tools: Visualization, Intervals, Sampling, Simulation, Modeling, DoE QbD Process Understanding Holistic Control Strategy Engineering Mechanistic Models Science “Design Space” as a Mathematical Model PPQ Statistics Tools: Sampling Acceptance Criteria Batch Evaluation The area tools of theare design space Statistical useful Design space can be a to where we plan toconfidence operate could understand risk, mathematical expression of levels, be verified during PPQ, butwith process process performance, understanding,along which then otherwise PPQ remains essentially other science & risk feeds supporting into the development of an the samerationale as it should bedeciding driven by based when appropriate control strategy. the processcontrol understanding overall strategy. and the holistic control strategy. 21 Data Needed from PV Stage 1 in Preparation for PV Stages 2 & 3 • Product and process knowledge – Risk assessment, Cause & Effect matrix, experimental outcomes – Process performance data from development • High level knowledge management document with links to studies, reports etc – Should be maintained as a lifecycle document • Control Strategy – what to control 22 Final Thoughts… • Through PV Stage 1, R&D Science designs the quality level for the product • Statistics has an important contribution to Design Space • PPRE (and many other Statistical tools) are useful to understand risk, confidence levels, process performance in developing the control strategy • Assurance of quality is provided by the control strategy • Confidence in quality cannot be estimated based on data alone • Statistics is part of the solution but not the solution 23 Acknowledgements • Kim Vukovinsky • • • • • • • • Penny Butterell Eric Cordi Tom Garcia Fasheng Li Roger Nosal Greg Steeno Ke Wang Tim Watson 24 References http://www.fda.gov/downloads/Drugs/Guidances/UCM070336.pdf http://www.ispeboston.org/files/handouts_-_morrison.pdf 25 References http://www.mbswonline.com/presentationyear.php?year=2012 http://www.mbswonline.com/presentationyear.php?year=2013 http://www.mbswonline.com/presentationyear.php?year=2014 26 http://www.iabs.org/index.php/docs/doc_download/386-iabs-settingspecifications-2013t-schofield 27 http://www.ispe.org/2015-statistician-forum Pfizer Confidential │ 28 Pfizer Confidential │ 29