Revisiting The Notion Of Singlet Testing Requirements

In his Feb. 2005 viewpoint article, "In Defense of Singlet Testing," Torbeck (1) draws an important philosophical distinction between "standards" and "specifications." He argues that specifications are criteria selected by manufacturers for statistical control of their products, whereas compendial standards are absolute requirements. This distinction is entirely compatible with modern concepts of statistical process control. US Pharmacopeia (USP) standards provide solid benchmarks for industry and assurance of quality product for patients. We agree that USP standards should (and in many cases do) encourage the design of manufacturing and measuring systems so that the risk that any one lot will fail to meet the standard is acceptably small.

Torbeck also raises other interesting points which merit further discussion. We explain these points and propose an alternative viewpoint.

David LeBlond, Tim Schofield And Stan Altan

Philosophical position on singlet testing

Under the concept of singlet testing (1), every lot of a pharmaceutical product—with possibly millions of dosage units per lot—must be made by processes that guarantee that every dosage unit tested by the filed analytical method at any time before expiry conforms with the stated USP standard. To take the specific example quoted (1), the USP standard for aspirin tablets (2) of not less than 90% and not more than 110% should apply to every USP-grade aspirin tablet.

It is necessary to point out that this view is contradictory to the USP "uniformity of dosage units" standard for aspirin, which allows at least one tablet in a lot to be outside the 85–115% label claim. This contradiction in requirements questions the viewpoint that every dosage unit must meet the 90–110% USP standard. We believe, instead, that this latter standard is intended to apply to the result of the compendial aspirin tablet assay that takes, as a sample, a composite of at least 20 aspirin tablets. Although the USP wording may be open to debate, we agree with the British and European pharmacopeias, whose description of the application of their standard to the result of the assay is quite clear (3, 4):

"Limits of content: Where limits of content are prescribed, they are those determined by the methods described under Assay.

The JapanesePharmacopoeia offers similar wording.

The result taken from a physically blended composite effectively averages out contributions from individual tablets. Thus, we are led to the view that the result of the compendial aspirin tablet assay is meant to provide an estimate of the batch mean and that the stated standard was intended to control the batch mean potency, not the content uniformity of individual dosage units." (5)

We suggest that a "singlet testing" view, in which the stated compendial standard applies to every individual dosage unit, is an unattainable requirement for a manufacturer and is a testing disincentive. It may discourage a full scientific understanding of a manufacturing process and interfere with the objective of providing quality product to the patient. Recognizing the central value estimate (e.g., a mean or median) from a representative sample as a statistic of interest encourages companies to make better decisions that serve the interests of both the customer and the manufacturer.

Standards and specifications

The special concept of a "standard" is useful (1), but we are not sure that a broad dichotomy between standards and specifications is justified. Although standards and specifications may differ in their immediate purpose (e.g., compliance versus process control) and in the rationale for setting the limit (e.g., therapeutic versus statistical), they do not differ in application. In both cases, one compares a statistic with a limit and makes a fixed decision (e.g., lot release or internal action). In both cases, decision gray zones are handled by incorporating staged-testing into the application process. Extending an artificial dichotomy into the area of application does not seem helpful and may inhibit a clear understanding of the operational characteristics of standards.

It is noted that the USP sets process goals (e.g., rigid standards) without specifying how they can be achieved (1). We agree that the strategy for meeting USP standards should be the responsibility of individual manufacturers. Although some general guidelines can be helpful (e.g., ref. 6), it is important that these be understood as optional. It seems more important to ensure that the specific testing methodology associated with each standard be explicitly and unambiguously clear.

Value of statistical considerations

Torbeck states, "if statistical procedures were given in the USP, companies would have little incentive to develop better procedures" (1). It is implied that it is not USP's role to include statistical considerations such as statistical sampling plans and failure probability when issuing standards. This policy seems unwise. Statistical concepts and methods are tools as powerful as many of those used in analytical chemistry.

We offer the following examples in which the application of statistical technology could, or has, positively impacted compendial standards:

• The benefits of statistical sampling plans in ensuring representative and adequate samples are well known. The USP content uniformity, dissolution, and disintegration chapters make use of the efficiency afforded by such multistage testing schemes (7).

• A careful definition of a reportable result can lead to more discriminating tests. Any test is most efficient when the decision criteria can be based on a particular summary measure that most completely captures the information in the test data bearing on the specific quality parameter of interest. Such summary measures might be the mean, median, standard deviation, range, minimum, or maximum, depending on the test objective, data, and assumptions. The lesson here is that certain summary measures may be more informative for a particular compendial purpose, and that statistical concepts can help us identify them.

• An argument for considering the mean, rather than individual dosage units, is that individual doses are not tested before they are administered to subjects during clinical development. In these cases, only the relationship of the mean dose measure to the mean clinical outcome is relevant. Thus, the translation of development experience to quality control is best preserved by holding the lot mean to the standard or specification. In cases in which the objective of a particular standard may be to govern the properties of a population of doses, rather than of individual doses, the mean is a more discriminating statistic and also argues in favor of applying compendial limits to the mean, not to individual results. This criterion does not preclude monitoring consistency of dosage units through a parameter such as the standard deviation or range, or as routinely practiced through dose uniformity testing.

• Variability in reported values can arise from many sources (e.g., analytical, manufacturing process, interlaboratory, and distribution-chain variability). As Torbeck recognizes (1), the tolerances and limits in compendial monographs allow for certain sources of variability. In making these allowances, it is important to establish a proper "error budget" for all sources (8). The application of statistical concepts and methods to better understand the sources of variation should therefore be an important consideration in the development of standards.

• Proactive consideration of the statistical operating characteristics (e.g., discriminatory power) of tests gives important insight into the material ultimately released and should be (and often are) considered when different compendial tests are compared. Discriminatory power and efficiency have implications for efficacy, availability, and cost of the drug products available to patients. In choosing between two alternative tests of compliance to a given standard, the test that is most statistically discriminating and efficient is preferred.

We should avail ourselves of technology in all areas. If statistical considerations are helpful in guaranteeing quality product to the patient, we feel regulators and industry alike should have the option of using such tools to ensure the development and application of appropriate standards.

Alternative viewpoint: standards based on appropriate statistics

We suggest that the notion of a standard can only be properly understood in the specific context of a well-defined analytical method. A standard definition must include a clear, unambiguous description of the sample being analyzed, the characteristic of interest being measured, and the final decision rule and statistic against which the standard limits are to be compared.

Standards and specifications differ in their purpose and objective but are similar in their application. If statistical concepts can be useful in developing and applying specifications (e.g., in the quality control world), then we should not preclude the use of these tools in connection with standards any more than we would avoid the use of chemical or physical concepts and tools. Statistical concepts and tools traditionally associated with quality control specifications can be applied to ensure that standards have maximum discriminatory power and are properly linked with the true quality of the material being tested.

We recognize that standard limits are fixed and should allow for certain sources of variation and that it is the responsibility of individual manufacturers to develop processes and methodologies to meet those limits. At the same time, we believe that compendial standards should be flexible to the limitations of current technology, the value of new technology in all applicable fields, and sources of error that are inherent in the product.

Everyone (USP, manufacturers, regulators, and the public) who has input into specific monographs or other USP methods or standards may benefit from the use of statistical concepts or tools. Such tools can help foster the use of optimal standards that encourage good science and facilitate the availability of life-saving products.

Ackowledgment

The authors gratefully acknowledge helpful discussions with William Porter of Abbott.

References

1. L.D. Torbeck, "In Defense of USP Singlet Testing," Pharm. Technol. 28 (2), 105–106 (2005).

2. "Aspirin Tablets Monograph," USP 28–NF 23 (US Pharmocopeial Convention [USP], Rockville, MD 2005), p. 183.

3. General Notices, Part III, Section 1.4, "Limits of Content," British Pharmacopeia 2004 (Medicines Commission, London, UK, Vol. 3, 2004), p. 2090.

4. General Notices 1.4, "Limits of Content," European Pharmacopoiea (Council of Europe, Strasbourg, France, 4 ed., 2002), p. 7.

5. General Notices 27, Japanese Pharmacopoeia (Society of Japanese Pharmacopoeia, Tokyo, Japan, 14th ed., English ed., 2001), p.3

6. General Chapter ‹1010› "Analytical Data: Interpretation and Treatment," USP 28–NF 23 (USP, Rockville, MD 2005), p. 2516.

7. General Chapters‹905› "Uniformity of Dosage Units,"‹701›"Disintegration," ‹711› "Dissolution," and ‹724› "Drug Release," USP 28–NF 23 (USP, Rockville, MD 2005), pp. 2503, 2411, 2412, 2415.

8. EAL Task Force, European co-operation for Accreditation, "Expression of the Uncertianty of Measurement in Calibration," EA–4/02 (Dec. 1999), http://www.european-accreditation.org.

David LeBlond is a senior statistician of nonclinical statistics at Abbott, Tim Schofield, is a director at Merck Laboratories, and Stan Altan is a senior research fellow at R.W. Johnson Pharmaceutical R&D, LLC, OMP Admin Building, 1000 Route 202 S. Room G-038, Raritan, NJ 08869, Tel. 908.704.4083, saltan@prdus.jnj.com.

*To whom all correspondence should be addressed.