Drug Product Continuous Process Verification A Case Study

Transcription

1 Drug Product Continuous Process Verification A Case Study CASSS 2016 Summer CMC 19 July 2016 Tom Damratoski Bristol-Myers Squibb Director, Biologics Drug Product MS&T 1

2 Drug Product CPV In Practice - 3 Levels Product A Product B ACME ACME CMO % Ppk>1.3 80% % Ppk > Quarterly Portfolio- Sr. Management Scorecard looking at all sites and products Rejection Rate 2/50 0/20 Yield 2. End to End Product Reviews --Cross functional team across Drug Substance, Drug Product, Analytical, Stability and Quality --Deeper Statistical Analysis applied to Hot Spots Attribute X Drug Product Manufacturing Site -Approximately Monthly interbatch data reviews -Voice of the Process at the Shop Floor -Site Focused but also feeds data upward 10 BMS Proprietary 0 and Confidential 2

3 1. Manufacturing Site CPV- Getting Started 1. Procedures and training Statistical software and Training Data mining practices- Manual, Automated, or Hybrid Statistical Alert Rules And Procedures Data review frequency and participants 2. Initial Monitoring Beginning of Stage III Process Validation Monitoring Plan Established Based on Risk Assessment Monitor 1 st ~ 30 lots, generate Time Series plots (too soon for CpK) Lock Control Limits 3. Continuous Process Verification Visualize Data using Time series, histograms, and PpK/CpK analysis Generate statistical alert event reports where needed Utilize Heat Map, and apply multivariate analyses when needed (Generally multivariate utilized less for DP than in DS) Data Feed upwards into E2E Product Reviews and Portfolio scorecards 3

4 Drug Product Monitoring Plan- Typical Profile Critical Quality Attributes ph Concentration HMW/LMW Bioanalytical impurities Particulate Potency Moisture (lyo) Etc. In Process Control ph Fill weight Performance 100 % Visual inspection Process Valuable, often overlooked % Particle defects % Critical defects found % Overall Less is more at the start. Automated data capture and analysis enables larger data sets with faster alert response 4

5 ph 2 Learnings From a Basic Assay N=30 Batches PpK = 1.07 (Borderline Capable) 6.3 USL = 6.3 Individual Value ph Most Recent Batches PpK Increases to 1.63 UCL= _ X= LCL= LSL = 5.7 Learning 1: Data visualization is King PpK is a lagging indicator. Learning 2: Assays with low sig figs, while scientifically correct, often drive nonnormal data sets 5

6 Fill Weight Control Boxplot of Net Fill Weight Checks by Lot Yervoy 50 mg/vial 10.9 USL=10.9 Fill Weight, g/vial Target= H56677 AAB2789 Box Plots are ideal but data-intensive Per lot sigma is another measure LSL=10.5 Automated data acquisition from OEM check weigh system AAD4371 can be a challenge AAG4867 due to proprietary software Use side-by side to compare 2 different DP sites 6

7 Visual Inspection- Example Defect Trend I Chart of Major Defects Orencia SC 125 mg/syringe % % % % % Statistical Alert Simple but powerful tool to reduce complaint/quality risk Lot defect rates compared to in-process control limits (not specs) Further defect pareto can be explored in regions of concern (e.g. scratch, product on stopper) Side by side control charts used for comparing performance of 2 different DP sites. 7 AAD2732 AAD4848 AAD4849 AAD7169 AAE0418 AAE0418 AAE3662 AAE3746 AAE3749 AAE4762 AAE4762 AAE7586 AAE7879 AAE8686 AAE8707 AAE8714 AAE8724 AAF5108 AAF7888 AAF0208 AAD2076R AAF2676 AAE8730 AAF4881 AAF8393 AAF8392 AAD4849R AAF7195 AAE3749 AAE3752 AAE3750 AAF3932 AAE3757 AAG9564 AAG7286 AAG7289 AAG7290 AAH2522 AAG7291 AAE3747 AAE3747R AAE3753 AAE3748 AAE3751 AAE3754 AAE3756 AAE3754R AAH5395 AAH6736 AAH5393 AAH9195 AAH7734 AAH5394 AAJ1246 AAH5397 AAG7293 AAG7293R AAJ5501 AAH7744 AAH9193 AAH9194 Individual Value UCL= % _ NMT=1.0% X= % LCL= % Lots

8 Example Product Related Impurity Variability.. Orencia SC Genealogy: Method or Deamidation Process or Both? 6.0 USL=6.0 CpK = 1.0 DP DS Deamidation (%) Impurity AAA5066 AAB6158 AAC2885 AAD2076 To Be Continued. AAE8707 AAE8730 AAF8393 AAH6736 AAH5397 AAH9193 8

9 Example Process PpK Heat Map Parameter Ppk per Manufacturing Stage Drug Substance Formulation Filling Capping / Thawing / Mixing Sealing Pooling 100% Visual Inspection Finished DP Protein concentration Purity ph SEC Moisture Reconstitution Time Fill Dose Machine Speed Plunger Pressure Height Adjustment Critical Defects Major Defects Total Defects Process Improvement Initiatives Focused Here 9

10 Drug Product CPV In Practice - 3 Levels Product A Product B ACME ACME CMO % Cpk>1.3 80% % Cpk > Quarterly Portfolio- Sr. Management Scorecard comparing Products and Drug Substance and Drug Product Sites Rejection Rate 2/50 0/20 Yield 2. End to End Product Reviews --Cross functional team across Drug Substance, Drug Product, Analytical, Stability and Quality --Deeper Statistical Analysis applied to Hot Spots Attribute X Drug Product Manufacturing Site -Approximately Monthly interbatch data reviews -Voice of the Process at the Shop Floor -Site Focused with MS&T participation 10 BMS Proprietary 0 and Confidential 10

11 End to End Product Reviews Each Cross-Functional Product team presents data package to technical management, approximately 3-4X per year Drug Substance Drug Product Analytical and Stability End to End view provides powerful insights Most CQA s are common from DS to DP Analyses such as DS-to-DP genealogy assessments and Assay Variation Ratio provide insights to source of variation (Assay or Process?) Selective multivariate analysis used as needed Input into Annual Product Quality Review 11

12 DS to DP Batch Genealogy-Product Related Impurity Both DP result and Input DS result(s) plotted against DP lot number Orencia SC Genealogy: Deamidation 6.0 Batch of interest DP DS USL=6.0 Deamidation (%) Impurity Variability not attributable to lab. Lower purity DS results in lower purity DP 2. DP mfg process contribution should be explored: shorter Time out of Refrigeration or light exposure? 3. Consider multivariate analysis AAA5066 AAB6158 AAC2885 AAD2076 AAE8707 AAE8730 AAF8393 AAH6736 AAH5397 AAH

13 DS to DP Batch Genealogy-Potency Orencia SC Genealogy: B7 Binding DP DS USL= Assay % LSL=70 AAA5066 AAB6158 AAC2885 AAD2076 AAE8707 AAE8730 AAF8393 AAH6736 Random pattern of DS vs. DP result suggests analytical contribution Method precision appears to be increasing with more recent lot history AAH5397 AAH

14 Assay Variance Ratio (AVR) AVR is ratio of analytical system variance (based on reference standard results) to total process variance (DS or DP results) Indicates contribution of assay variability to total variability Cpk > 1.33 & AVR < 0.45 Cpk > 1.33 & AVR 0.45 Cpk < 1.33 & AVR < 0.45 Cpk < 1.33 & AVR 0.45 Guideline No Action Required May need further evaluation of analytical method. Medium risk May need further evaluation of analytical method. Medium risk Needs further evaluation of analytical method. High risk Notes: 0.45 cut off derived on the basis that analytical variability should not contribute more than 50% of the variability compared to the process variability. 14

15 General Takeaways VP-level engagement and broad data transparency down to the shop floor are critical to a successful and sustainable CPV program Centralized Process Analytics team, led by Statistician, helps drive end-to-end product view, and provides second level statistical support (multivariate) When in doubt use time-series plots to visualize the data CpK/PpK heat maps provide a good overview of process risk, and help drive further analyses and corrective actions Investment in automated data capture and analysis platforms (e.g. Discoverant) Provide the following benefits Decreased data lag Less labor Increased data integrity While automation is good, CPV can be resourced and executed with manual data capture when necesary 15

16 Acknowledgements Susan Abu-Absi Peter Millili Syama Adhibhatta Bryan Thurnau Abraham Diaz Questions? 16