Using Process Capability to Enhance Product Quality Daniel Y. Peng, Ph.D. Senior Product Quality Reviewer Office of Process and Facility (OPF) OPQ/CDER/FDA IFPAC 2015 Annual Meeting Arlington, Virginia January 27, 2015 1
Vision for 21 st Century Manufacturing A maximally efficient, agile, flexible pharmaceutical manufacturing sector that reliably produces high quality drug products without extensive regulatory oversight. - J. Woodcock, M.D. CDER/FDA - 2005 and 2012 3
Are We There Yet? CMC Supplements Total Product Recall Lawrence Yu, NIPTE meeting, 2013 June 3 4
Lagging and Leading Indicator Lagging indicator An indicator that follows an event. Output oriented, easy to measure, but hard to influence Reactive in nature Leading indicator An indicator that predicts future events and tends to change ahead of that event. Typically input oriented, measuring activity or behavior Proactive in nature, focusing on continual improvement and failure prevention 1. Calories taken in 2. Calories burned It is recommended to use a combination of Lagging and Leading Indicators. 4
Concept of Process Capability First introduced in Statistical Quality Control Handbook by the Western Electric Company (1956). process capability is defined as the natural or undisturbed performance after extraneous influences are eliminated. This is determined by plotting data on a control chart. ISO, AIAG, ASQ, ASTM.. published their guideline or manual on process capability index calculation. 5
Four indices: Nomenclature C p : process capability index C pk : minimum process capability index P p : process performance index P pk : minimum process performance index ASTM E2281: Standard Practice for Process and Measurement Capability Indices 6
Calculation Formula C p ( USL LSL) = 6σ Mean LSL Cpkl = 3σ ( USL LSL) P p = 6SD Mean LSL Ppkl = 3SD Cpku USL Mean = 3σ Ppku USL Mean = 3SD C pk = min (Cpkl, Cpku) P pk = min (Ppkl, Ppku) USL: upper specification limit; LSL: lower specification limit; Mean: grand average of all the data Sigma hat: estimated inherent variability (noise) of a stable process SD: overall variability 7
A Perfectly Centered Process LSL LSL USL USL For this case: USL= +4σ LSL = -4σ USL-LSL= 8σ Cp= 1.333 Cpku=1.333 Cpkl=1.333 Cpk=1.333-5 -4-3 -2-1 0 1 2 3 4 5 Mean (μ ), Sigma (σ) 8
Process Mean is not Centered For this case: USL= +4σ LSL = -4σ USL-LSL= 8σ Cp= 1.333 Cpkl = 1.667 Cpku = 1.0 Cpk= 1.0 When the process is not centered, or deliberately run off-center for economic reasons, or only a single specification limit is involved, Cpk should be used. 9
Cpk, Sigma Value, and PPM Cpk Value Sigma Value Area under normal distribution curve (%)* Non conforming parts per million (ppm) Unilateral specification Bilateral specification* Capability Rating** 0.333 1 68.27 158650 317300 Terrible 0.667 2 95.45 22750 45500 Poor 1.0 3 99.73 1350 2700 Marginally capable 1.333 4 99.993636 32 64 Capable 1.667 5 99.999942 0.29 0.58 Good 2.0 6 99.9999998 0.001 0.002 Excellent *Process mean is centered at middle of the specification limits and has normal distribution **Bothe, D. R., Measuring Process Capability, Cedarburg, W.I., Landmark Publishing Inc., 2001 10
Denominator Difference between C pk and P pk Inherent variability Overall variability σ = R d or MR d or S 2 2 c4 SD = N ( X i X ) N 1 i= 1 2 SD: standard deviation of all individual (observed) values, which accounts for both common cause variability (noise) and special cause variability. It is often referred to as overall variability. σ : the inherent variability (noise) due to common cause of a stable process. It is often estimated by using within subgroup variability which is linked to the use of control charts. 11
Difference between C pk and P pk Cpk represents the potential process capability (i.e. how well a given process could perform when all special causes have been eliminated). Ppk addresses how the process has performed without the demonstration of the process to be stable. Predict future batch failure rate Cpk (Yes) Ppk (No) 12
Control Chart Central line (CL): the grand average Statistical process control limits (UCL and LCL): Typically: ±3 sigma from CL Should not be confused with upper and lower specification limits (USL and LSL) To evaluate if a process is in a state of statistical control Western Electric Rules Two Types of Control Chart Variable control chart: continuous numeric measurements (e.g. Xbar- Range chart, Moving Range chart, Std. deviation chart) Attribute control chart: discrete data (pass or fail, or counts of defects) (e.g. p chart, np chart, c chart, u chart) ASTM E2587- Standard Practice for Use of Control Charts in Statistical Process Control 13
Product level: CQA Monitoring and Trending Subgroup Mean Process Capability Analysis of Tablet Assay (first 25 batches, subgroup size =3) 102 100 98 1 3 5 7 9 Xbar Chart 11 13 15 Batch No. 17 19 21 23 25 UCL=102.108 _ X=100.287 LCL=98.466 LSL 96 Capability Histogram 98 100 102 USL 104 Specifications LSL 96 USL 104 USP: 90-110 Cpk: 2.95 Subgroup Range 4 2 0 1 3 5 7 9 R Chart 11 13 15 Batch No. 17 19 21 23 25 UCL=4.582 _ R=1.78 LCL=0 96 Normal Prob Plot A D: 0.636, P: 0.094 99 102 105 Assay (%) 104 102 100 98 96 5 Run Chart 10 15 Batch No. 20 25 Within StDev 1.051 Cp 1.27 Cpk 1.18 PPM 229.14 Capability Plot Within Overall Specs Overall StDev 1.079 Pp 1.24 Ppk 1.15 Cpm * PPM 323.15 Cpk: 1.18 Data source: Chopra, V., Bairagi, M., Trivedi, P., et al., A case study: application of statistical process control tool for determining process capability and sigma level, PDA J Pharm Sci and Tech, 66 (2), 2012, pp. 98-115 14
Understand the Science and Risk Product and process understanding Identification of the critical aspects of drug substance, excipients, formulation variables, process variables, container closure systems, in process control Establish appropriate control strategy & risk mitigation Product and process monitoring and trending Listen to the voice of the process Learn what s normal and observe if different Continual improvement Detect early and take actions Transform from reactive trouble shooting to proactive failure reduction or prevention 15
Site Level: Performance Monitoring % of rejected batch /month at Site A (# of lots rejected/# of lots attempted 100%) Binomial Process Capability Analysis of Unsuccess Batch P Chart Unsuccess Rate Proportion 0.15 0.10 0.05 0.00 Cumulative Unsuccess Rate 6 5 4 3 2 1 3 5 7 9 11 13 15 17 19 21 23 Month Tests performed w ith unequal sample sizes Cumulative Unsuccess Rate 5 10 15 20 Month 25 25 UC L=0.1809 _ P=0.0437 LC L=0 Summary Stats (95.0% confidence) % Defectiv e: 4.37 Low er C I: 2.79 Upper C I: 6.49 Target: 0.00 PPM Def: 43726 Low er C I: 27917 Upper C I: 64891 Process Z: 1.7090 Low er C I: 1.5150 Upper C I: 1.9123 Frequency % Unsuccess Rate 20 10 0 20 25 30 T otal Batch Manufactured/Month Histogram Tar 10.0 7.5 5.0 2.5 0.0 0 3 6 9 12 % Unsuccess Rate % rejected batch: 4.37% Binomial process capability index: 0.569 (Cpk <<1) 16
Correlation coefficients based on data samples from 14 production sites Katy George, McKinsey & Company, Brookings Institute, Washington DC (May 1, 2014) 17
Paradigm Shift Culture of Quality Pharmaceutical Quality System (PQS) obviously important, measurement of PQS maturity/effectiveness not easy Manufacturers take full responsibility for quality of their products Focus on meeting patients expectations Regulators expectations considered minimal approach Strive for continual improvement Management and organizational commitment to prioritizing quality Each person in organization understands and embraces their role in quality 18
Summary: Process Capability Indices Patient first: clinical relevant specification based on safety and efficacy Consider not only process mean & variability but also in relation to the specification Process capability index (Cpk): in control and capable Quantitative and action enabling Applicable for cross sectors (brand, generic, OTC and biotech) No additional testing is required since batch release data is available per current regulation A valuable tool to transform from the reactive trouble shooting paradigm to a proactive failure reduction or prevention paradigm 19
Acknowledgements Lawrence Yu Christine Moore Naiqi Ya Ubrani Venkataram 20