Using Control Charts to Evaluate Process Variability Daniel Y. Peng, Ph.D. Quality Assessment Lead Office of Process and Facility (OPF) OPQ/CDER/FDA PQRI 205 Annual Meeting North Bethesda, Maryland October 5, 205
Walter Andrew Shewhart (89-967) A physicist, engineer and statistician Father of statistical quality control Statistical method from the viewpoint of quality control (939) Creator of PDSA (Plan, Do, Study and Act) cycle Creator of control chart Originator of the Chance and Assignable variation concept 2
Uncontrolled variation is the enemy of quality Dr. W. Edwards Deming (900-993) 3
Sources of Variation Variation exists in all processes. Variation can be categorized as either: Chance or Common causes of variation Inherent to a system, random, always present and hence predictable within statistical limits Eliminate inherent variability (noise) is difficult Assignable or Special causes of variation Exterior to a system, non-random, not always present (intermittent) can cause changes in the output level, such as a spike, shift, drift, or non-random distribution of the output. Are usually easier to be detected, controlled or eliminated 4
6.0 Control Chart USL UCL Quality attribute (unit) 5.0 LCL CL LSL 4.0 30 40 50 60 Sample # Definition: a graphical display of a product quality characteristic that has been measured or computed periodically from a process at a defined frequency Every control chart consists of: A set of data A central line (CL) (mean) Two statistical process control limits (UCL and LCL) (Is the process Stable?) Upper and Lower Specification Limits (USL and LSL) Patient s need ( Safety and Efficacy) (Is the process Capable?) 5
Potential Applications To proactively monitor and trend a process To detect the presence of special cause variation To identify continual improvement opportunities To maintain the process in a state of statistical control Using science and risk-based approach Take action in a timely manner 6
Key Considerations for Constructing a Control Chart 7
Choice of Product Quality Characteristics Critical Quality Attributes (CQA) A physical, chemical, biological or microbiological property or characteristic of an output material including finished drug product that should be within an appropriate limit, range, or distribution to ensure the desired product quality (ICH Q8) Identification of CQA: primarily based upon the severity of harm to the patient (safety and efficacy) Critical (input) material attributes and critical process parameters (CMAs/CPPs) Other relevant process characteristics that can assist in process monitoring and controlling 8
Types of Control Chart Variable Control Chart Characteristics that can be measured (continuous numeric data) e.g. Assay, Dissolution, % of Impurity The average and variability charts are usually prepared and analyzed in pairs Average Range chart (Xbar-R chart, subgroup size 2-0) Average Standard Deviation chart (Xbar-S chart, subgroup size >0) Individual Moving Range chart (I-MR chart, n=) Attribute Control Chart Characteristics that have discrete values and can be counted, e.g. % defective, # of failed batches in a month p chart / np Chart: for fraction of occurrence of an event- Binominal distribution e.g. % of unsuccessful batch at a facility every month c chart / nc Chart: for counts of occurrence in a defined time or space increment -Poisson distribution e.g. number of particulate matter in an injection vial Other types of control chart: cumulative sum control chart (CUSUM) exponentially weighted moving average control charts (EWMA) 9
Subgroup Size and Sampling Frequency Subgroup: the observations sampled at a particular time point Subgroup Size and Sampling Frequency (N x K) The number of observations in each subgroup: n the objective of the monitoring (detect large or small shift) how quickly the output responds to upsets consequences of not reacting promptly to a process upset time and cost of an observation Rational Subgroup: Minimize the variation of observations within a subgroup Maximize variation between subgroups 0
Statistical Process Control Limits UCL and LCL: the thresholds at which the process output is considered statistically unlikely typically, ±3 SD (Shewhart limits) Rationale: to balance the two risks: Failing to signal the presence of a special cause when one occurs; False alarm of an out-of-control signal when the process is actually in a state of statistical control
How out-of-control points are identified? Rule No. any point falls outside UCL/LCL Other Rules certain nonrandom patterns of the plotted data Use it judiciously Risk of false alarm 8 Western Electric Rules 2
Over-Reaction vs. No-Reaction Procedures should describe how trending and calculations are to be performed and should guard against overreaction to individual events as well as against failure to detect unintended process variability (20 FDA Process Validation Guidance) Control chart and process capability analysis often go hand-in-hand 3
Illustrative Examples 4
Within Batch Variability Example Xbar-R Chart of Disso@240min Sample Mean 45 40 35 30 25 054 233 309 338 356 429 Time 452 55 539 602 625 UCL=3.75 _ X=29.25 LCL=26.75 Not Stable & Not Capable 0.0 UCL=0.36 Sample Range 7.5 5.0 2.5 _ R=5.7 0.0 LCL=0 054 233 309 338 356 429 Time 452 55 539 602 625 ER coated beads, mixed with extra-granular cushioning excipients and compressed into tablets Compression: ~ 5h, sample frequency: every 8-0 min (total 33 subgroups), subgroup size= 6 5
Between Batch Variability Example Subgroup Mean Process Capability Analysis of Tablet Assay (first 25 batches, subgroup size =3) 02 00 98 3 5 7 9 Xbar Chart 3 5 Batch No. 7 9 2 23 25 UCL=02.08 _ X=00.287 LCL=98.466 LSL 96 Capability Histogram 98 00 02 USL 04 Specifications LSL 96 USL 04 USP: 90-0 Cpk: 2.95 Subgroup Range 4 2 0 3 5 7 9 R Chart 3 5 Batch No. 7 9 2 23 25 UCL=4.582 _ R=.78 LCL=0 96 Normal Prob Plot A D: 0.636, P: 0.094 99 02 05 Stable & Capable Assay (%) 04 02 00 98 96 5 Run Chart 0 5 Batch No. 20 25 Within StDev.05 Cp.27 Cpk.8 PPM 229.4 Capability Plot Within Overall Specs Overall StDev.079 Pp.24 Ppk.5 Cpm * PPM 323.5 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), 202, pp. 98-5 6
Between Batch Variability Example Process Capability Analysis of Tablet X Content Uniformity (AV) Individual Value Moving Range 5.0 2.5 0.0 3.0.5 28 3 34 I Chart 37 40 43 46 49 Moving Range Chart 52 55 UCL=5.558 _ X=3.37 LCL=0.76 UCL=2.974 MR=0.90 2 Capability Histogram 4 6 8 0 2 4 USL Normal Prob Plot AD: 0.637, P: 0.088 Specifications USL 5 Not Stable but Capable 0.0 28 3 34 37 40 43 46 49 52 55 LCL=0 0 2 4 6 AV 6 4 2 28 3 34 Last 30 Observations 37 40 43 46 49 Batch No. 52 55 Within StDev 0.8070 C p * C pk 4.90 PPM 0.00 Capability Plot Within Overall Specs O v erall StDev 0.9460 Pp * Ppk 4.8 C pm * PPM 0.00 Tablet content uniformity (AV) of last 30 commercial batches of Tablet X manufactured by Firm Y (subgroup size =, I-MR chart) 7
Site Performance Monitoring Example % of unsuccessful batch /month at Site A (# of lots attempted: 20-30/month) Binomial Process Capability Analysis of Unsuccess Batch Proportion 0.5 0.0 0.05 0.00 3 5 7 9 P Chart 3 5 Month 7 9 2 23 25 UC L=0.809 _ P=0.0437 LC L=0 % Unsuccess Rate 20 0 Unsuccess Rate 0 20 25 30 T otal Batch Manufactured/Month Stable but Not Capable Tests performed w ith unequal sample sizes Cumulative Unsuccess Rate 6 5 4 3 2 Cumulative Unsuccess Rate 5 0 5 20 Month 25 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: 2797 Upper C I: 6489 Process Z:.7090 Low er C I:.550 Upper C I:.923 Frequency 0.0 7.5 5.0 2.5 0.0 Histogram Tar 0 3 6 9 2 % Unsuccess Rate Binomial process capability index: 0.569 8
Paradigm Shift Culture of Quality 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 9
Summary Brief introduction of control chart: history, definition, types Key considerations for constructing a control chart: Choice of drug product quality characteristics Subgroup size and sampling frequency Statistical process control limits (UCL and LCL) Illustrative examples for process monitoring and control: Within batch variability Between batch variability Site performance monitoring Control Chart can be a valuable tool to: Proactively monitor and trend a process Detect the presence of special cause variation Identify continual improvement opportunities Maintain the process in a state of statistical control 20
Acknowledgements Dr. Christine Moore Dr. Naiqi Ya Dr. Ubrani Venkataram 2