OR WAIT 15 SECS
The author discusses the purpose of analysis and testing and the implications for specifications and their underlying statistical distribution.
Risk-based assessments are required for process validation and manufacturing, computerized system validation, auditing, and supply-chain management, and it would seem logical to include specifications in such approaches. Pharmaceutical regulators, however, are curiously silent on risk-based specifications. The focus of this article is to provide an approach to risk-based product and raw-material specifications that can be applied to the pharmaceutical industry. It is based on an International Organization of Standardization (ISO) approach that is consistent with current regulatory expectations and offers a perspective for analytical results and reportable values.
ADAM GAULT/GETTY IMAGES
Regulatory and compendial considerations
All measurements are subject to error. Traditionally, regulatory and pharmacopoeial specifications for drug substances and drug products have been set with fixed limits without consideration for measurement uncertainty in the analysis and testing procedure itself. This situation exists in the United States and many other parts of the world. For example, in the European Union, the release limits for a drug product as provided in law are (1):
Unless there is appropriate justification, the maximum acceptable deviation in the active-ingredient content of the finished product shall not exceed ± 5% at the time of manufacture.
Within the pharmacopoeias, there is generally no allowance for an appropriate justification, and in general, regulators ignore this caveat. It is clear that the pharmaceutical industry lags behind many other areas of manufacturing with respect to its approaches to quality and risk management.
During the past 10 years, however, there have been global regulatory changes toward risk-based and lifecycle approaches to cGMPs and regulatory issues relating to chemistry, manufacturing, and controls led particularly by FDA. In 2004, FDA proposed that there should be science-based regulation of product quality stating (2):
As pharmaceutical manufacturing evolves from an art to a science and engineering based activity, application of this enhanced science and engineering knowledge in regulatory decision-making, establishment of specifications, and evaluation of manufacturing processes should improve the efficiency and effectiveness of both manufacturing and regulatory decision-making.
As a consequence, FDA developed process analytical technology (PAT) guidance (3). It stated that "product and process specifications are based on a mechanistic understanding of how formulation and process factors affect product performance" (3). It further asserted "the advantage of using these PAT principles and tools during development is to create opportunities to improve the mechanistic basis for establishing regulatory specifications" (3). FDA emphasized that one goal of the PAT guidance was to tailor the agency's usual regulatory scrutiny to meet the needs of PAT-based innovations that improve the scientific basis for establishing regulatory specifications.
FDA's process validation guidance, first published in 1987, required major changes to take into account the new risk-based approaches, and FDA subsequently revised and finalized process validation guidance in 2011 (4). The core principle was maintained that if a process were adequately controlled, the product would consistently meet its predetermined quality attributes. The biggest change was a move from a fixed number of batches constituting validation to placing the emphasis on process design and understanding by the generation of a design space. Within this design space, adjustment of parameters would not constitute a change. Hence, for robust processes, the traditional fixed limit is replaced by a proven acceptable range.
Out-of-specification (OOS) results have always been and continue to be a major regulatory concern. FDA took eight years to finalize its guidance on this topic (5) following the 1993 ruling, regarding OOS results, by Judge Wolin in the United States of America v. Barr Laboratories (6). The arguments regarding the "isolation" of outliers from the specification continues to be contentious not only in the US but in the EU as well. The application of scientifically sound statistical methodologies in conjunction with root-cause investigations is acceptable to the authorities provided that their use and application are predetermined and proceduralized. If the regulators wish to move the industry to risk-based process validation, the industry will require that regulators also move toward risk-based specifications, but this topic has not been adequately addressed.
Public and private specifications
The pharmacopoeias, as the guardians of public standards for pharmaceutical materials and products, have long been the bastions of the rigid specification approach, which has been reinforced by the assignment of a value for a primary reference standard without consideration of its uncertainty. There was a deliberate distancing of the pharmacopoeial reference standards from the activities of the ISO Reference Standards Committee. The introduction to ISO Guide 35:2006 states (7):
Pharmacopoeial standards and substances are established and distributed by pharmacopoeial authorities following the general principles of this Guide. It should be noted, however, that a different approach is used by the pharmacopoeial authorities to give the user the information provided by certificates of analysis and expiration dates. Also, the uncertainty of their assigned values is not stated since it is negligible in relation to the defined limits of the method-specific assays of the pharmacopoeias for which they are used.
This position, however, changed dramatically in November 2011 when the United States Pharmacopeial Convention was granted accreditation (8) to ISO Guide 34 (9) and started specifying measurement uncertainty of assigned values for its chemical reference standards. The European Directorate for the Quality of Medicines & Healthcare, which is responsible for the provision of reference standards for the European Pharmacopeia (10), announced that it, too, is moving toward ISO Guide 34 accreditation, which will probably be accomplished in the next two years. From a statistical perspective, this means that in the future, the assigned value of the primary reference standard is no longer a point value but a confidence interval that contains the true value. Therefore, fixed-limit specifications will no longer be scientifically justified on the basis that the primary standard itself has measurement uncertainty.
The cGMPs are quite explicit on the requirement for scientific soundness of specifications as 21 CFR 211.160(b) states:
Laboratory controls shall include the establishment of scientifically sound and appropriate specifications, standards, sampling plans, and test procedures designed to assure that components, drug product containers, closures, in-process materials, labeling, and drug products conform to appropriate standards of identity, strength, quality, and purity.
It is essential to consider analytical procedures as processes to quantify sources of variation and their effect on the uncertainty of an analytical result or reportable value. It is illogical to assign any fixed limits without taking these factors into account.
The nature of testing and measuring processes
The basics of process understanding is well understood. A generic process flow is shown in Figure 1. This model is widely applicable to synthetic API production, drug-product manufacture as well as analysis and testing methods and procedures. This generic model has been adapted by Burgess and McDowall for analytical procedures (see Figure 2) (11). The intent of the analytical testing of a pharmaceutical substance or product is to be able, based upon the testing of a sample, to predict the properties of the population (i.e., batch) from which it was taken and to assign an unbiased best estimate of the property being determined as predefined in the procedure. This unbiased best estimate of the property being determined as predefined in the procedure is called the reportable value or reportable result, which is compared with the registered specification. This approach is the traditional method of determining if a product meets specification.
Figure 1: Generic process flow. (ALL FIGURES ARE COURTESY OF THE AUTHOR)
This approach, however, is flawed because of the inherent variabilities of the measurement/analysis process, which includes the variability of the reference standard. Any analytical or testing result is subject to the influences of the manufacturing process, the appropriateness of the sampling scheme to ensure representivity, and the test method or procedure. It is important to quantify the capability of the test method or procedure to make an informed decision regarding the analytical outcome with respect to a specification. The capability of the manufacturing and sampling processes should be known.
Figure 2: The analytical "factory," the process of turning samples into data and information (Ref. 11).
The nature of specifications
The nature of a specification depends upon the type of data and their inherent mathematical distribution. In this article, the discussion is confined to continuous data where normal distribution may be assumed. This assumption has been shown to be well founded for traditional chemical analytical testing (12).
Figure 3: Conventional specification from a customer perspective. OOS refers to out of specification. LSL refers to lower specification limit. USL refers to upper specification limit.
The conventional view of fixed-limit specification is generally known as the "voice of the customer," or in the case of pharmaceuticals, the "voice of the regulator" (see Figure 3); this approach is from Western countries. In Japan, particularly in the engineering and electronics industries, the Taguchi approach to quality loss (13) is often adopted, whereby only at the target is the loss zero (see Figure 4). This approach is widely used in Six Sigma and Lean Sigma quality initiatives. The "voice of the process," on the other hand, is for the majority of continuous analytical measurements described by normal distribution. Two parameters define this distribution: namely a measure of its location denoted by the arithmetic mean (i.e., the average) and a measure of its dispersion (i.e., the standard deviation).
Figure 4: Taguchi quality loss function. LSL refers to lower specification limit. USL refers to upper specification limit.
The necessity to control both parameters is recognized in FDA's OOS guidance (5), whereby the reportable result or value is controlled as well as the range of individuals as a measure of dispersion. The analytical reportable value is represented by a distribution around the best estimate found (see Figure 5). There is a fundamental incompatibility of the voice of the customer and the voice of the process. The problem with near-the-limit OOS results is indicated in Figure 6.
Figure 5: Reportable value (RV) and a normal distribution.
In Figure 6, there are two analytical reportable values lying close to the lower specification limit. The reportable value (indicated by the red "X") is conventionally OOS, and the reportable value (indicated by the green "X") is not. From a statistical viewpoint, given measurement uncertainty in the reportable value, it is likely that both are within the known variability of the test procedure. FDA, in its final OOS guidance, says, in such cases (5):
Figure 6: Conventional voice of the customer issues with out-of-specification (OOS) results. LSL is lower specification limit; USL is upper specification limit.
...where a series of assay results (to produce a single reportable result) are required by the test procedure and some of the individual results are OOS, some are within specification, and all are within the known variability of the method, the passing results are no more likely to represent the true value for the sample than the OOS results. For this reason, a firm should err on the side of caution and treat the reportable average of these values as an OOS result, even if that average is within specification.
Confidence in analytical data
As previously discussed, the outcome of an analytical procedure is not an exact value but a predicted range at a predetermined level of confidence. The current accepted practice within the pharmaceutical industry is to validate analytical procedures in accordance with ICH Q2(R1), last revised in 1995 (14). This approach also is adopted by the US Pharmacopeia (USP) General Chapter <1225>. In this approach, the conventional analytical chemistry methodology of separating the measure of location (there called accuracy) and the measure of dispersion (there called precision) is adopted. In addition, ICH Q2(R1) describes three different precision levels according to the number of process factor influences. As ICH Q2(R1) focuses on the methodology, it omits the fourth and lowest level, which is system or instrument precision. The hierarchy of the four precision levels in ascending order is system (i.e., instrument) precision, repeatability, intermediate precision, and reproducibility (15). From a quality-control perspective, the most important precision level is intermediate precision as this level represents the day-to-day capability of a given procedure within a particular laboratory.
The relationship between accuracy and precision in conventional analytical chemistry methodology is illustrated by the target model (see Figure 7[(a]). The definition of accuracy in Validation of Compendial Procedures, USP General Chapter <1225>, and in ICH Q2(R1) corresponds to unbiasedness only. In the International Vocabulary of Metrology (VIM) and ISO documents, accuracy has a different meaning. ISO defined a new term, "trueness," to mean the closeness of agreement between an average value obtained from a large series of measurements and an accepted reference value. In other words, trueness implies a lack of bias. The concept of trueness in ISO leads directly to the idea of measurement uncertainty, which combines the variabilities associated with both accuracy and precision. The ISO approach is illustrated by an adapted target model in Figure 7(b). The United States Pharmacopeial Convention has updated USP General Chapter <1010> on the interpretation of analytical data to include a comparison of the ISO measurement uncertainty (MU) approach with the conventional confidence interval approach.
Figure 7: The relationship between accuracy and precision: (a) conventional analytical chemistry methodology and (b) ISO approach.
ISO recognized the incompatibility of the concept of (MU) of the testing or measuring process and fixed-limit specifications, particularly in the engineering and medical-device industries. The result was the publication of a standard (ISO 14253-1:1998) on decision rules for proving conformance or nonconformance with specifications, which took into account measurement uncertainty of the testing/metrology process (16). The current version of ICH Q2(R1) (14) on the validation of analytical methods was developed almost 20 years ago, and the parent guideline was published in 1994. It is long overdue for revision.
ISO approach to conformance with specification
ISO 14253-1 addresses the problem arising when a measurement result falls close to or on a specification limit. In this instance, it is not possible to prove conformance or nonconformance with specification because the measurement result plus or minus the expanded uncertainty of measurement includes the limit itself. The estimated uncertainty of measurement has to be taken into account when providing evidence for conformance or nonconformance with specification. This ISO guide led to the development of more detailed and recent guidance from the American Society of Mechanical Engineers (17, 18) although the basis of this approach has been available in the literature for more than 60 years (19, 20). The overall ISO risk-based approach may be summarized in terms of conformance versus specification zones and a complete statement of a result of measurement of a reportable value. This complete statement of a result of measurement is determined by a full execution of the analytical procedure, including the uncertainty range, 2U.
Figure 8 shows that the reportable value (RV) is the best estimate of the true value, but at 95% confidence, this value can fall within the range 2U. Consideration of how to determine U is discussed later in this article. In the ISO approach, the conformance zone(s) is defined as the specification zone reduced by the expanded uncertainty of measurement, U. A recent article on using target measurement uncertainty to determine "fitness for purpose" for analytical procedures discusses the guard-band principle (21).
Figure 8: Pictorial representation of the reportable value (RV) and its measurment uncertainty. U is uncertainty of measurement.
The essential features of the ISO approach are illustrated in Figure 9. Because of U, there are three distinct zones rather than the normal two. The nonconformance zone(s) lies outside the specification zone extended by the expanded uncertainty of measurement, U. These two zones are separated by the uncertainty range equal to 2U. The nonconformance zone(s) are OOS. The variance contributions that define U are only due to the analytical procedure itself and includes the uncertainty in the reference standards and must not include any manufacturing or sampling variances. The uncertainty range surrounds the specification limit(s), where neither conformance nor nonconformance can be proved taking into account the uncertainty of measurement (2U).
Figure 9: ISO 14253-1:1998 approach to conformance with specification. LSL is lower specification limit; USL is upper specification limit. U is uncertainty of measurement. OOS is out of specification.
Calculating the value of U
The evaluation of uncertainty requires consideration of all the possible sources of variability affecting the analytical reportable value. Although a detailed study of this kind may require considerable effort, it is important that the effort expended should not be disproportionate. In practice, a preliminary study will identify the most significant sources of uncertainty, and the value obtained for the combined uncertainty is almost entirely controlled by the major contributions. A good estimate of uncertainty can be made by concentrating effort on the largest contributions.
This measurement uncertainty is arrived at by means of an error budget that enables the calculation of uc, the standard deviation of the procedure and is described in detail in a EURACHEM/CITAC Guide QUAM:2012.P1 document (22). An error budget is used to:
Uncertainty of measurement is estimated and evaluated according to the Guide to the Expression of Uncertainty in Measurement (GUM) (23) and is consequently expressed as the expanded uncertainty, U, where U = k*uc with a default coverage factor (k) of 2. This coverage factor (k) of 2 corresponds to a probability of 95.45% (Note if k = 1.96, the probability would be 95.0%). The prerequisite for adopting these approaches is that both the manufacturing and measurement processes are shown to be under statistical control and are statistically capable.
Consequences for pharmaceutical specifications
The primary purpose of quality control is to assure quality of the manufactured product by testing of samples in accordance with registered methods and specifications contained in a marketing authorization in the EU or in a new drug application or abbreviated new drug application in the US. Safety and efficacy are assessed and assured during development prior to the granting of such authorization. Impurities, however, are always a safety concern. As D. Jacobson-Kram and T. McGovern of FDA note (24):
While the use of pharmaceuticals is always a balance of risks and benefits, the same is not true for impurities in pharmaceuticals; impurities convey only risk. While impurities should always be reduced to the lowest levels that are reasonably practical, it is acknowledged that impurities cannot be reduced to zero and specifications for impurities need to be established.
A risk-based approach to the specification of physicochemical quality parameters, such as assay and in-process controls, where a small difference in value is highly unlikely to have any adverse effect on the patient, should be different from that for impurities that might have an adverse effect.
A guard band, G, is defined as the magnitude of the offset (U) from the specification limit to the acceptance or rejection zone boundary (17). For nonclinical critical specifications, it is reasonable to adopt a risk-based relaxed specification zone that allows for the uncertainty of the method or procedure itself (see Figure 10). In this approach, the author proposes that a reportable value does not become OOS until it exceeds the guard band. It is not expected to occur frequently, so reportable values falling in this guard zone should be treated as out-of-expectation (OOE) results and investigated to assure analytical integrity. Impurities, on the other hand, which are regarded as clinically critical, should be subject to a stringent specification zone (see Figure 11).
Figure 10: A risk-based relaxed specification zone for nonclinical critical specifications. LSL is lower specification limit. USL is upper specification limit. G is guard band. OOS is out of specification.
This approach has already been applied to regulated pesticide residue analyses in the EU (25), and the approach is more fully described by Ellison and Williams (26). This document referenced and built on the American Society of Mechanical Engineers' approach from 2001 (17).
Figure 11: A risk-based stringent specification zone for clinically critical specifications for impurities. USL is upper specification limit. G is guard band. OOS is out of specification.
A key point for regulatory concern using this approach is the magnitude of U and hence the guard band. The size of the guard band must be strictly controlled and limited to prevent abuse. For example, it has been suggested for physical metrology (17) that the size of the guard band should not exceed one-eighth of the specification range. For analytical methods, the size of the guard band should be based on the measured process capability index Cpk and an uncertainty budget.
Risk-based approach to shelf-life estimation
While a risk-based approach is not currently used in product- or material-release specifications, it is accepted as the standard within ICH Q1E on evaluation of stability data for establishing or confirming a registered shelf life (27):
Regression analysis is considered an appropriate approach to evaluating the stability data for a quantitative attribute and establishing a retest period or shelf life.
An appropriate approach to retest period or shelf life estimation is to analyze a quantitative attribute (e.g., assay, degradation products) by determining the earliest time at which the 95% confidence limit for the mean intersects the proposed acceptance criterion.
For an attribute known to decrease with time, the lower one-sided 95% confidence limit should be compared to the acceptance criterion.
The ICH Q1E approach is illustrated in Figure 12. The example is taken for a product with a registered shelf life of 24 months. The regression line is derived from long-term product stability data of the API expressed as the percentage of labeled claim with respect to time generated under the appropriate storage condition. As the API degrades with time, calculation of the lower 95% confidence interval is appropriate and is shown as the blue dashed line. This confidence interval intersects the registered regulatory end of life limit at a time beyond 24 months, thereby confirming the shelf life. This use of this confidence interval approach is equivalent to the guard-band proposal for a stringent specification requirement shown in Figure 11.
Figure 12: Illustration of product shelf-life confirmation based upon a confidence Interval equivalent to a stringent specification requirement.
The adoption of an ISO risk-based specification approach to the control of drug substances and drug products using the guard-band principle and the clinical relevance of the specification is proposed. This approach is summarized in Figure 13. Suppose that the reportable value and its measurement uncertainty are taken from Figure 8 and combined with the four illustrative examples with respect to a specification limit.
Figure 13: The proposed guard-band principle and analytical process capability. G is guard band. U is uncertainty of measurement. RV is reportable value.
From a compliance perspective, options Figure 13 (a) and Figure 13 (d) present no difficulty of interpretation as they are demonstrably OOS and in specification respectively. Figure 13 (b) and Figure 13 (c), however, are more difficult to interpret in a conventional approach whereas this interpretation is easier using the proposed approach. For a clinically relevant specification, example Figure 13(b) would be OOS, but only OOE for a nonclinically relevant specification. Figure 12(c) is OOE for a clinically relevant specification but in specification for a nonclinically relevant specification.
The argument has been made that fixed-limit specifications for pharmaceuticals (i.e., the voice of the customer or regulator) are fundamentally incompatible with the voice of the process. This has been recognized for many years in industries other than the pharmaceutical industry. An internationally recognized ISO standard is available and a proposal to apply it to risk-based specifications pharmaceutical products and processes using the guard-band principle is presented.
The author wishes to thank Dr. R.D. McDowall, principal McDowall Consulting, and R.M. Bonner, chairman of the European Compliance Academy, for helpful discussions during the development of this article.
Christopher Burgess is managing director of Burgess Analytical Consultancy Limited, member of the European Qualified Person Association Advisory Board, member of the USP Council of Experts 2010 to 2015, chairman of the ECA Analytical Quality Control Group, member of the ECA Executive Committee, and a visiting professor, University of Strathclyde's School of Pharmacy and Biomedical Sciences, tel: +44 1833 637 446; firstname.lastname@example.org
1. Article 3, Council Directive 83/570/EEC Amending Directives 65/65/EEC, 75/318/EEC and 75/319/EEC on the Approximation of Provisions Laid Down by Law, Regulation or Administrative Action Relating to Proprietary Medicinal Products.
2. FDA, Pharmaceutical cGMPs for the 21st Century–A Risk-based Approach Final Report (Rockville, MD, Sept. 2004).
3. FDA, Guidance for Industry–PAT–A Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance (Rockville, MD, Sept. 2004).
4. FDA, Guidance for Industry–Process Validation: General Principles and Practices (Rockville, MD, Jan. 2011).
5. FDA, Guidance for Industry–Investigating Out-of-Specification (OOS) Test Results for Pharmaceutical Production (Rockville, MD, Oct. 2006).
6. United States of America vs. Barr Laboratories, US District Court for the District of NJ, Civil Action 92-1774, Opinion, A.M. Wolin (Feb. 4, 1993).
7. ISO Guide 35:2006 Certification of Reference Materials-General and Statistical Principles (Geneva, 2006).
8. USP, Scope of Accreditation to ISO Guide 34:2009,
, accessed June 4, 2013.
9. ISO Guide 34: 2009 General Requirements for the Competence of Reference Material Producers (Geneva, 2009).
10. EDQM, "Quality, Safety Environment," webpage, www.edqm.eu/en/EDQM-quality-safety-76.html, accessed June 4, 2013.
11. C. Burgess and.R.D. McDowall, "Introduction to Raw Data, Results and Reportable Values," presented at the European Compliance Academy (2011).
12. M. Thompson and R J. Howarth, Analyst 105 (1257, 1188-1195 (1980).
13. G. Taguchi, "System of Experimental Design, Engineering Methods to Optimize Quality and Minimize Costs" (Quality Resources and American Supplier Institute, 1991).
14. ICH, Q2(R1) Validation of Analytical Procedures: Text and Methodology (1995).
15. J. Ermer and J.H. McB. Miller, Eds., Method Validation in Pharmaceutical Analysis: A Guide to Best Practice (Wiley-VCH, Weinheim, Germany, 2005), p. 30.
16. ISO 14253-1:1998 Geometrical Product Specifications (GPS)—Inspection by Measurement of Work Pieces and Measuring Equipment—Part 1: Decision Rules for Proving Conformance or Non-conformance with Specifications (Geneva, 1998).
17. ASME, B18.104.22.168-2001, Guidelines For Decision Rules: Considering Measurement Uncertainty In Determining Conformance To Specifications (2001).
18. ASME, B22.214.171.124-2007 Guidelines for the Evaluation of Dimensional Measurement Uncertainty (2007).
19. A. E. Eagle, Ind. Qual. Control 10 (3) 10-15 (1954).
20. F. E. Grubbs and H.J. Coon, Ind. Qual. Control 10 (3) 15-20 (1954).
21. M.L.J. Weitzel and W.M. Johnson, Accredit. Qual. Assur. 17 491-495 (2012).
22. EURACHEM/CITAC Guide CG 4: Quantifying Uncertainty in Analytical Measurement (3rd Ed., 2012), QUAM:2012.P1
23. ISO/IEC/BIPM Guide to the Expression of Uncertainty in Measurement (GUM) (1993), also ANCSL 2540-2-1997 in the US.
24. D. Jacobson-Kram and T. McGovern, Adv. Drug Delivery Rev., 59 (1) 38-42 (2007).
25. EC, 2002/657/EC, Commission Decision: Implementing Council Directive 96/23/EC Concerning the Performance of Analytical Methods and the Interpretation of Results (Aug. 12, 2002).
26. EURACHEM/CITAC Guide: Use of Uncertainty Information in Compliance Assessment (1st Ed, 2007).
27. ICH, Q1E, Evaluation of Stability (2003).