OR WAIT null SECS
This article looks at the current good manufacturing practice regulations from a statistical perspective while addressing their requirements and implications and inviting the industry to assess its past performance in meeting the regulations.
The director of quality control walked up to my cubicle at G.D. Searle in Skokie, Illinois, and said, "FDA has just published a revision to the Current Good Manufacturing Practice regulations, and I would like for you to review it for the statistical implications." It was February 1976, I had been with the company less than two months, and so started my 30-year career of applied statistics in the service of the current good manufacturing practice (CGMP) regulations. That internal memo later became a formal presentation at the 20th annual Quality Clinic (1) and I have primarily pursued the same goal ever since.
The 30th anniversary issue of Pharmaceutical Technology now prompts the same question and a related question: How have we, the pharmaceutical industry, done in responding to the statistical implications?
The first set of regulations for finished pharmaceuticals was published in 1963 by the Food and Drug Administration, (FDA) on the basis of standards developed by the industry and by the Pharmaceutical Manufacturers Association in 1961. These regulations were amended in 1965, revised in 1971, amended again February 13, 1976 (2), and published in the Federal Register as 21 CFR 210 and 211.
A reading of 21 CFR 210 and 211 finds statistical topics expressed both explicitly and implicitly. The most frequently occurring topics are representative samples, sampling, and sampling plans. Section 210 (20) finds
Acceptance criteria means the product specifications and acceptance/rejection criteria, such as acceptable quality level and unacceptable quality level, with associated sampling plan, that are necessary for making a decision to accept or reject a lot or batch.
Note that there are three parts to the acceptance criteria: the specifications, the acceptable quality level (AQL) (3), and the unacceptable quality level. The unacceptable quality level is usually known as the limiting quality (LQ). (Note that the definition of AQL changed in the 2003 version of ANSI/ASQ Z1.4 to be the acceptable quality limit.) The acceptable quality limit (AQL) is the percent of defects or defectives that is the worst tolerable process average. The probability of rejection at the AQL is the "producer's risk." A typical sampling plan will reject 5% or less of the lots with a defect level at or less than the AQL. The LQ is the percent of defects or defectives that is unacceptable. The probability of acceptance is the "consumer's risk." A typical sampling plan will accept 5% (or 10%) or more of the lots with this defect level or greater. The implication is that we should look for all three parts anywhere the term acceptance criteria appears. Most standard operating procedures and manufacturing instructions give the specifications and the AQL, but few provide the LQ. Historically, sampling plans were indexed primarily by the AQL with little note of the LQ values. With FDA's new emphasis on risk and risk management, however, the LQ or consumer's risk takes on new importance for compliance.
Manufacturers typically interpret the regulations to mean that they must set their internal product specifications requirements tighter than the regulatory compendial requirements. As stated in USP, "Confusion of compendial standards with release tests and with statistical sampling plans occasionally occurs" (4). Compendial standards define what constitutes an acceptable article and describe test procedures that demonstrate that the article is in compliance. These standards apply at any time in the life of the article from production to consumption. The manufacture's release specifications, and compliance with good manufacturing practices generally, are developed and followed to ensure the article will indeed comply with compendial standards until its expiration date, when stored as directed.
The term representative sample is carefully defined in Section 210 (21) as "a sample ... of units that are drawn based on rational criteria such as random sampling...." Note the use of rational criteria. Sampling is the physical act of selecting the units and the validity of the sample, and its representativeness is directly related to the physical actions used to select it. Once a sample has been taken, it is impossible to determine whether it is actually representative of the lot. Documentation and standard operating procedures (SOPs) are key to compliance.
Although simple random sampling (in which every unit has an equal chance of being selected) is preferred from a theoretical consideration, it is often very difficult or impossible to obtain because of physical limitations. Other statistically valid sampling techniques include systematic sampling with or without a random start and cluster sampling. Nonstatistical sampling includes a judgment sample selected by an "expert" to be representative or a convenience sample taken because it is easy to obtain (also called a grab sample).
Our task is to examine the rational criteria to "assure that the sample accurately portrays the material being sampled," as stated in Section 210 (21), and to observe the act of sampling by the operator. Only by actual observation of the physical act of sampling can we determine validity—reading an SOP is insufficient.
Notice that the objective of the sampling inspection and the acceptance criteria is to make a decision to accept or reject. This must be documented and communicated to management. Sampling is not a substitute for process control and improvement.
The words sampling and representative samples are used in the following sections: 211.80(a); 211.84(a) and (b); 211.110(a); 211.122(a); 211.134(b); 211.160(b)(1), (2), and (3); 211.165(c); and 211.186(b) (9). Sampling plans are statistical descriptions of the procedures used to sample units, including the physical activities to select the units.
Section 211.84 notes that "The number of containers to be sampled ... shall be based upon appropriate criteria such as statistical criteria for component variability, confidence levels and degree of precision desired, the past quality history of the supplier...." Section 211.165(d) goes further by stating, "The statistical quality control criteria shall include appropriate acceptance levels and/or appropriate rejection levels."
Sampling plans have historically been either Military Standard 105E for attribute sampling or Military Standard 414 for variables. These documents describe a set of sampling plans that have been used in the industry since 1942. In fact, we find this quote in an article by Olson and Lee in 1966 (5):
For practical purposes the need to design a sampling plan has been eliminated by a series of government sponsored sampling plans, two of which are MIL STD 105D for attribute single, double, and multiple sampling plans; and MIL STD 414 for variables sampling plans. These books have gained acceptance throughout most of the United States industry in a manner much like the USP and NF. Government contracts for the purchase of pharmaceuticals usually refer to one or both of these books. The obvious advantage of selecting plans from either of these books is communicability and acceptance throughout industry. Hence, there is little or no advantage to specially designed sampling plans.
However, Military Standard 105E was discontinued by the US government in February of 1995. No reason was given, but cost may have been a factor as the government tried to downsize. Thus, technically Military Standard 105E is not available any more. However, a variation of it is available from ANSI/ASQ as Z1.4 (3) for 105E and Z1.9 (6) for 414.
Other statistical issues in CGMPs
The CGMPs also directly discuss or are related to process control, setting specifications, stability testing method validation, and out-of-specifications.
Process control is required in sections 211.100 and 211.110(a). The latter requires us to"monitor the output and to validate the performance of those manufacturing processes that may be responsible for causing variability...." Moreover, section 211.165(d) says, "to assure that batches ... meet ... appropriate statistical quality control criteria ...." Thus, we should look for data and techniques used to control the process. These could include statistical control charts and process capability studies. Ideally, critical sources of variation would be identified and controlled.
Setting specifications is addressed in section 211.110(b), where it states:
Valid in-process specifications for such characteristics shall be consistent with drug product final specifications and shall be derived from previous acceptable process average and process variability estimates where possible and determined by the application of suitable statistical procedures where appropriate.
Clearly, we should look for evidence that the specifications were set using real data, averages, and standard deviations and not just wishful thinking.
Labeling discrepancies, 211.125(c) says, are to be compared with "narrow preset limits based on historical operating data." This would imply trending and summary statistics to set those limits.
A statistical basis for stability is required in section 211.166(a)(1), where "sample size and test intervals based on statistical criteria..." are to be in the written program. Furthermore, sections 211.137(a) and 211.137(b) expect expiry dates to be determined by "... appropriate stability testing..." and "... related to any storage conditions...." ICH Q1 addresses stability in considerable detail (7).
Method validation, of course, is required in section 211.165(e) where "The accuracy, sensitivity, specificity, and reproducibility of test methods ... shall be established and documented." Section 211.166(a)(3) requires "reliable, meaningful, and specific test methods." Note that there have never been operational definitions in the literature for reliable and meaningful in this context. Section 211.194(a)(2) also requires "... that the methods used in the testing of the sample meet proper standards of accuracy and reliability..." and that "the suitability of all testing methods uses shall be verified under actual condition of use." The standards for method validation are given in USP <1225> and the ICH Q2 documents (8, 9).
Out-of-specifications investigations are required in section 211.192:
Any unexplained discrepancy (including a percentage of theoretical yield exceeding the maximum or minimum percentages established in master production and control records) or the failure of a batch or any of its components to meet any of its specifications shall be thoroughly investigated, whether or not the batch has already been distributed. The investigation shall extend to other batches of the same drug product and other drug products that may have been associated with the specific failure or discrepancy.
Based on this, FDA issued in final form its Guidance for Industry: Investigating Out-of-Specification (OOS) Test Results for Pharmaceutical Production (10). Many of the issues in OOS are statistical, including defining the reportable results, retesting, resampling, sample size, averaging, outliers, testing into compliance, and specification limits.
Finally, it should be emphasized that these statistical topics are of a specialized nature and require that personnel be qualified to perform these tasks. As stated in section 211.25:
Each person engaged in the manufacture, processing, packing, or holding of a drug product shall have the education, training, and experience, or any combination thereof, to enable that person to perform the assigned functions. Training shall be in particular operations that the employee performs ....
Another paragraph requires the same for supervisors. Clearly, personnel must be trained in the statistical topics and techniques and must have practical experience in their application.
How well has the industry achieved this task and met this challenge? It is hard to say or to tabulate. The pharmaceutical industry has at times been reluctant to share its status and progress in some areas, statistics being one. Clearly, some of the larger pharmaceutical companies have been implementing statistical techniques for a long time. The 1974 third edition of Juran's Quality Control Handbook had the first and only chapter on "Drug and Allied Industries" written by H. Latham Breunig of Lilly (11). Some companies have sustained very early efforts and others have come to it later under such names as "total quality management (TQM)," "Deming," "zero defects," "quality function deployment, QFD," "quality circles," Taguchi, Ishikawa, Kaizen, "value analysis," "just-in-time, JIT," and Six Sigma/Lean manufacturing.
More specific to pharmaceuticals has been the new USP chapter on statistics <1010> (12) and the ICH documents that contain statistical requirements or implications: stability with Q1, method validation with Q2, setting specifications with Q6 (13), pharmaceutical development, quality by design, and design space with Q8 (14), and risk management with Q9 (15).
Most recently, FDA has been encouraging the industry relative to statistics by issuing guidances related to its process analytical technology (PAT) framework (16) as well as supporting ICH documents by providing presentations on related topics such as quality by design and design space. FDA officials have done the industry a great service by attending and presenting at the many conferences and seminars.
In conclusion, we have made progress in applying statistics under the CGMPs, but it may be spotty with the large companies leading the way. Each company should reevaluate its current position with respect to the statistics in the CGMPs, OOS, ICH, and PAT.
Lynn D. Torbeck is a statistician and consultant at Torbeck and Associates, 2000 Dempster Plaza, Evanston, IL 60202, tel. 847.424.1314, firstname.lastname@example.org.
Mr. Torbeck also is a member of Pharmaceutical Technology's Editorial Advisory Board.
Where were you 30 years ago?
"I was supervisor of quality control and manufacturing applications, a statistical position with G.D. Searle in Skokie, Illinois, supervising two statisticians and a secretary. My group provided statistical consultation and long-range planning to the directors of quality control and manufacturing. We were doing statistical data analysis and designed experiments in support of the two departments and also provided training and support for statistical software."
1. L. Torbeck, "Statistical Implications of the New Current Good Manufacturing Practices Regulations," presented at the 20th Annual Quality Clinic, University of Tennessee, Knoxville, Tennessee, Mar. 8–10, 1979.
2. FDA, "Current Good Manufacturing Practice in Manufacturing, Processing, Packing or Holding of Drugs and Finished Pharmaceuticals," Fed. Regist. Feb. 13, 1976.
3. ANSI/ASQ 1.4, "Sampling Procedures and Tables for Inspection by Attributes" (2003).
4. USP 30–NF 26 (United States Pharmacopeial Convention, Inc. Rockville, MD).
5. T.N. Olson and I. Lee, "Application of Statistical Methodology in Quality Control Function of the Pharmaceutical Industry," J. Pharm. Sci. 55 (1) p. 1 (1966).
6. ANSI/ASQ Z1.9, "Sampling Procedures and Tables for Inspection by Attributes" (1993).
7. ICH Q1 Stability Testing of New Drug Substances and Products (International Conference on Harmonization, Geneva, Switzerland, 2003).
8. Chapter <1225> "Validation of Compendial Procedures," in USP 30–NF 26 (United States Pharmacopeial Convention Rockville, MD).
9. ICH Q2, Validation of Analytical Procedures (International Conference on Harmonization, Geneva, Switzerland, 1994, 1996).
10. FDA, Center for Drug Evaluation and Research, Guidance for Industry, Investigating Out-of-Specification (OOS) Test Results for Pharmaceutical Production (Rockville, MD, Oct. 2006).
11. Quality Control Handbook, J. Juan, Ed. (McGraw Hill, New York, NY, 1974).
12. Chapter <1010> "Analytical Data–Interpretation and Treatment," in USP 30–NF 26 (United States Pharmacopeial Convention, Rockville, MD).
13. ICH Q6A Specification: Test Procedures and Acceptance Criteria for New Drug Substances and New Drug Products: Chemical Substances and ICH Q6B Specifications: Test Procedures and Acceptance Criteria for Biotechnological/Biological Products (International Conference on Harmonization, 1999).
14. ICH Q8 Pharmaceutical Development (International Conference on Harmonization, Geneva, Switzerland, 2006)
15. ICH Q9 Quality Risk Management (International Conference on Harmonization, Geneva, Switzerland, 2006)
16. FDA/CDER/CVM/ORA, Guidance for Industry:PAT—A Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance (Rockville, MD, 2004).