30 THE LABORATORY ANALYSTS ROLE IN THE DALG DEVELOPMENT PROCESS, 15, Foo@hpd Drug Administration, Center for Drug Evaluation and Research, Vali dation Ju Chromatographie Methods, November 1994, 16, G. C-Hokanson, A Life Cycle Approach to Validation, Phatm, Tech, 1994, 18,118 17. Guideline for Submitting Documentation forthe Stability of Htanan Drugs and Bio- logics, Center for Drugs and Biologics, Food and Drag Administration, February 1987 1B, J.T, Carstensen (editor New York, 1990, 5), Drug Stability, Principles and Practices, Marcel Dekker, LABORATORY CONTROLS AND COMPLIANCE Henry Avalione 2.1. INTRODUCTION Throughout history, the U.S. Government has reacted to defective product and fraudulent practices by amending the law, issuing regulations, or issuing guide- lines. Thus, in the late 1980s, when defective product and fraudulent activities reached an unacceptable level, the Food and Drug Administration (FDA) in- troduced the Pre-Approval Inspection (PAT) Program. This program formalized the manner in which FDA inspections of new or pending product applications were conducted. One of the areas where considerable fraudulent activities were identified was the area of laboratory controls. This has resulted in increased FDA serutiny of the laboratory, including more frequent utilization of FDA analysts in the conduct of these inspections. These analysts have brought increased expertise to EDA inspections because they are familiar with the emerging technology and ata handling systems that are utilized in theit own laboratories. Coinciding with the increased scrutiny of laboratory controls by the FDA. has been the advancement of laboratory technology and systems. The latter has resulted in the need for new standards, new methods, increased training, and increased capital expenditures to stay current, The improvements in laboratory cquipment and systems have resulted in increased sensitivity to detect and quantitate substances. This, coupled with the “Analytical Chomisoy a GMP Eneronment. Edited by JM. Mile and. 8 Crowther ISBN G471-31431-3 "2000 John Wily & Sons ne32 LABORATORY CONTROLS AND COMPLIANCE establishment and tightening of impurity specifications or standards, has led to the increase of compliance issues. The impurity profiles and methods detection limits have been particularly critical in obtaining new drug approvals. An even greater challenge has been to bring older products, processes, and methods that fare nearing the end of thet life eycle up to today's standards, Additionally, it hhas been necessary for laboratory records to fully document test procedures, demonstrating that the tests were performed correctly and in accordance with the method, Problems with fraudulent data and many legal cases have led the effort to require increased documentation in order to authenticate the data Failure on the part of industry to comply with GMP requirements has 1e~ sulted in an FDA request or laboratory certification. In at lenst two legal cases [l, 2], laboratory certification by a third party was included in a court order (injunction); in other cases, companies have “voluntarily” submitted to third- party laboratory certification, These “voluntary” cases have been a dircct result of FDA inspections and subsequent agreements with local FDA field (district) management. © This chapter will address some of the compliance issues identified above and provide guidance for maintaining a compliant laboratory In August of 1993 and July of 1994 FDA Injunctions against two drug manufacturers required certfieation of their laboratories [1,2]. The court orders listed basie requirements and the certification of laboratory conformance to Current Good Manufacturing Practice (CGMP)* regulations by a third-party cxpert, At the time of the consent decree (injunction), the specifics of laboratory certification had not been delineated by FDA. Subsequently, an FDA memo it August 1994 [3} provided direction for the certification ofa laboratory, and this document has been used by third-party “experts.” Areas discussed in this docu- ment included the followin; + Management systems + Operating procedures + Personne! training + Data accountability + Method validation + Equipment + Facilities + Certification documentation All of these issues will generally be discussed in this chapter focusing on areas of laboratory mantagement, laboratory controls, and laboratory compli- lance. In reflecting back on the FDA memo, it should be recognized that, while +p the interest of split Ue desinaton “GMP” wil also be red 10 be filly equivalent 19 "CGMP" in this book and is conse 22 LABORATORY MANAGEMENT — 33 appropriate at the time it was written, there were issues addressed that may now be considered outdated by today’s standards, 22. LABORATORY MANAGEMENT 2.2.1. Management Responsibility ‘The basic premise and foundation of the Federal Food, Drug and Cosmetic (FD&C) Act [4| pertains to management accountability, For example, Pream- ble Comment #432 to the 1978 current GMP regulations |S, 6] discussed the need for corporate officers and responsible managers to be aware of objection= able conditions. It states: Moreover, responsible officials already have i legal duties to be aware of, and (0 ake action upon, conditions eontribating to drug adulteration [7 While there is a movement on the part of some companies in the industry toward self-directed work teams, total quality management (TQM), and so on, this logic can be contradictory to FDA philosophy. The FD&C Aet directing accountability and responsibility applies to the management of the laboratory. For example, at a recent training program an FDA manager was asked: “What are FDA thoughts on self-directed work teams?” He replied: “Firms are guired to comply with GMPs and have individuals accountable and responsi- ble. While we have no objection in some environments, in a GMP environment responsibility is @ key issue. For example, in the review of analytical raw data, a manager or supervisor is held accountable for the work performed and review of the analysts’ data. Certainly, a peer or even administrator could check eal- culations. However, a manager needs to review the data.” The atea of resources, particularly regarding competent personnel, often presents a compliance issue, The typical laboratory bas an analyst-to-supervisor ratio of a maximum of eight anaiysts to one supervisor. In an audit, problems ‘can be identified through a review of out of specification (OOS) test results, “deviations, and investigations. For example, incomplete investigations are often. the result of inadequate laboratory resources, For the compliant laboratory, it is important to have data reviewed on a daily basis. Ifa supervisor is missing for not available, even on a limited daily basis, a qualified replacement should be available. As an example, replacement with an “acting” supervisor or a temporary supervisor not only covers the compliance issue, but serves as a developmental opportunity for personnel. In the review of laboratory data, it is acceptable for the peer reviewer to censure that data are labeled, calculations are verified, and data are complete. Mowever, the supervisor must perform the general review of the complete data package. For example, the supervisor reviews raw data, such as chromatograms ‘and their integration, and also evauates the appropriateness of results. The34 LABORATORY CONTROLS AND COMPLIANCE supervisor also needs to conduct reviews and compare assays, content unifor- ity, and dissolution results. While data should be reviewed daily, the complete review of the data package should be performed within a reasonable time, typi- cally five working days. Otherwise, it becomes extremely difficult to conduct any investigation of laboratory data. In a laboratory training program, a senior FDA laboratory director was asked {o comment on the responsibilities of the first line supervisor. He stated that responsibilities include the review of analysis’ work, assurance of adequate procedures, assurance of safety requirements, planning of work, development ‘of analysts, and certain administrative functions. By far, the most important of these is the assurance that analysis are tained and are performing the analyti- cal work correctly From an auditing perspective, if laboratory supervisor's office or work area is notin close proximity to the area where analysts are working, itis usually an indication that the supervisor is not involved with daily luboratary operations. Ifa supervisor is flot aware of the samples that each analyst is working on, it ‘could also be an indication that the laboratory is undermanaged, 2.2.2. Training A major reason for laboratory noncompliance, as detailed in all of the FDA legal actions regarding laboratories, is the failure to have trained analysts. “Training is necessary both from a technical basis and for GMP compliance. Itis expected that a laboratory have a standard operating procedure (SOP), whic provides some overall direction for laboratory training, including basie training ‘on the content of the cGMP regulations, The sections of the regulations that detail laboratory requirements should be highlighted as well have as a discus: sion of basic SOPS, such as data handling and recording data in workbooks and/or on data sheets. Specific training for each type of test to be performed should be reviewed during the technical training. Experienced laboratory directors have com ‘mented that the typical chemist graduating with a degree in chemistry has little practical laboratory experience. Therefore, for general tess, there must be some Jevel of assurance as to the analysts’ competency, It is also expected that each analyst who performs a relatively complex test, such as a high-performance liquid chromatography (HPLC} assay, have method-specific training. For ex- ample, in an audit itis not unusual to identify an analyst having difficulties conducting a specific HPLC assay, and then verify through a review of training records that the analyst has not been trained to conduct the product specific HPLC assay. ‘As new equipment is introduced, analysts should also be trained through a ‘combined use of visual aids and hands-on use. The FDA St. Louis National Laboratory, for example, has a library of tapes for the training provided by the vendor cach time a new piece of equipment is purchased, ‘A major part of the training program is the use of appropriate evaluation 23, LABORATORY GoNTROLS — 35 tools to certify that an analyst is trained. IL is @ relatively easy task to provide training. However, the reat issue centers around the effectiveness ofthe training provided. During a review of laboratory deviations, management can identily analysts who have had problems conducting testing. It is the laboratory man- ager’s responsibility to review analysts’ training records and “certify” that the analyst has been found acceptable to perform the test in question. (By “certify ing” the analyst, the manager acknowledges that the analyst has been evaluated or assessed and that the manager assumes accountability for the competency of the analyst in the area/method/procedure certified.) ‘There are basic documents that should be included in the training file of each analyst. These include: + Analyst training file + Resume or eurriculum vitae (C.V.) + Documentation of SOP training + Documentation of GMP training + Documentation of standard method training + Documentation of product specific training + General skill enbancement training One of the ways in which management commitment can be demonstrated is in the resources allocated to training. Some companies have undertaken on their own a program for the certification of laboratory analysts. For example, Jobnson & Johnson conducts a two-week hands-on laboratory analyst training and certification program. This program is conducted on a bimonthly basis for laboratory analysts during which they receive training in both cormpliance and ‘echnical skills. At the conclusion of the program each analyst is assessed and, if found satisfactory, is issued a certiBcate of satisfactory completion. As “certifi cation" becomes more popular in the industry, it is expected that mote com panies will move toward some type of analyst/laboratory certification. 2.3. LABORATORY CONTROLS 23.1. Laboratory Records Issues relating to data integrity and authenticity have raised the bar of labora- tory records beyond what some consider “good science.” Some analysts, and particularly research analysts, have a difficult time with regard to compliance with data handling requirements of the regulations. The CGMP Regulations, Section 211.194(a) (6, state: Labotatory records shall include complete data devived from al tests necessary 10 assure compliance with established specifications and standards, including exam nations and assays, as follows:36 LABORATORY CONTROLS AND COMPLIANCE ‘Sample description ‘Method used Sample weight ‘Compete records, in insteamentation Record of all calculations 6. Statement of results and comparison to specifications 7. The initials or signature of @ second person showing that the original records dhave been reviewed for aceuracy, completeness, and compliance with estas lished standards uding all graphs, charts tnd spectra from laboratory In order to assure compliance, i is important to have a SOP that specifically defines what raw data are, not what they can be, When dealing with compu- terized data hand systems and electronic data, its important to define raw data up front. Generally, its both the electron signal and the hard copy that are generated and reviewed. As such, both are considered to be raw data, With movement toward paperless systems, there may be some consideration for paperless laboratory data. Howover, the volume of data and chromatograms senerated inthe typical laboratory is 100 area to affect review on a screen. For example, in an audit of a rescarch laboratory that was considering the uiliza- tion of a paperless system, a manager replied that she was unaware of any supervisor or reviewer who would sitin front of a sreen and review chroma tograms and data. Furthermore, there have been some questions about the resolution of graphs displayed on a terminal or personal computer. Thus, at this time most laboratories continue to utilize both the electronic signal and hard copy. It is also recognized that some electronic data in the laboratory may. be considered to be transient eleetroie data. For example, data from some pH meters could be considered transient data as included inthe Eleetronie Signa- ture Regulations (Part 11) 7]. While itis widely recognized that some clavif- cation with regard to electronic data may be needed, it behooves the laboratory to have detailed procedures regarding the composition of raw data, (Further discussion of the definition of raw data can be found in Chapter Id). Improvements in technology have also resulted in development of more complex HPLC gradient methods, particularly for quantifying impurities. Many of the newer, more specifie methods have made it necessary to reprocess chromatograms. While the electronic signal is not changed, the parameters around integration may be changed to improve the clarity of the peak. In the assessment of data, it is important to recognize that specific results must be reproducible, In these situations, SOPs are needed ineluding direction for the reprocessing of data SOPs should also define the data systems: notebooks, data sheets, electronic, and combinations. Current requirements for prenumbered pages and data sheets reflect the concern for data integrity. If results re derived via the treat rent of raw data, this should also be defined. The processing of data by a 29 LABORATORY CONTROLS 37 computer using changeable software, such as a spreadshest, requires SOPs as Well as validation of the spreadsheet, 2.3.1.1. Computerized Data Acquisition Systems. ‘The use and validation ‘of computerized data acquisition systems and specifically computerized liquid chromatography (LC) systems is 1 major compliance issue. FDA policy on the validation of computerized LLC systems was published in 1994 [8]. This docu ment was written by FDA laboratory managers and circulated as an FDA policy document, The holistic approach presented in the document represents a practical application to validation and is the same philosophy expressed in the United States Pharmacopeia (USP). In the section on chromatography the USP. slates: “To ascertain the effectiveness of the final operating systems, it should bbe subjected to a suitability test prior to use. The essence of such a testis the concept that the electronics, the equipment, the specimens, and the analytical aperations constitute a single analytical system, whichis amenable to an overall lest of system function” 9} Years ago, with the advent of the Good Laboratory Practice (GLP) Regu- lations and increased FDA interest in nonclinical testing, the FDA provided {guidance on computerized data acquisition systems that continues to represent basic policy. Dr. George James, FDA pharmacologist, FDA's Center for Drugs, commented on security and authenticity issues: In ceference to the growing trend towanls computerized data acquisition systems, the Agency has provided the following guidaace for computerized systems 1. Provision must be made so that only authorized individuals ent make data 2. Data entries may not be deleted. Changes must be made in the form of amendments 3, The database must be made as temperproof as possible 4, The Standatd Operating Procedures must deseribe the procedures for ensuring he validity of the data, Additionally, » basic aspect of validation of computerized data acquisition systems for 2 laboratory provides far a comparison of data from the specific instrument with that data electronically transmitted through the system and emanating on a printer. Once this functional testing has been accomplished ‘vera period of time, periodic data comparisons would be suficient. The other basic aspect of validation is the qualification of equipment. In a laboratory, and particularly for an automated system, i is important that the test equipment, such as the HPLC unit and its components (pump, detector, injectors, ete), be qualified (sce Chapter 15). Generally, the laboratory equip. rent supplier will provide the aspocts that need to be evaluated in the qualif cation exercise, Improvements in LC technology increased interest in unknown peaks and sensitivity adjustments atthe noise level and have made it cificuit to resort 1038 LABORATORY CONTROLS AND COMPLANCE sess LC systems In a numberof cass, eegulaors have asked for doc mentation of data integration and reprocessing. In some cases data ate retained foreach eprcesing, whl in others ony the final data ae eaned, Because it isa requirement thatthe lboritory supervisor review all data, it becomes a challenge for this supervisor to tata computer work station and review all chromatograms, along withthe reprocesing dala foreach chromatogram. In fay event, would sem appropiate for some documentation each tine data se reprocessed 2.3.2, Out of Specification/Trend (00S/00T) Perhaps the area that has received the most regulatory attention is that of retesting. FDA comments and observations concerning the practice of retesting ‘without appropriate investigation and justification have been a major compli- ance issue for thg last two decades. In the United States of America vs, Bare Laboratories court decision [10}, Judge Wotin clarified the requirements. ‘The gules that comes rom the Won deion mgpe tat every OOS result must be investigated. There are two mechanisms for investigation: infor: mal and formal, Informal insestigaiions are appropriate for single-event OOS results. The informal investigation generally takes place at the departmental level between the analyst and the first-line supervisor. The investigation should be documented and follow a written procedure that directs the review of test procedures, caleulations, equipment performance, and the data record. Typi- cally, an informal investigation for an assay or impurity failure ineludes the following: + Acreview of all lab data + A review of calculations to ensure they are accurate + A review of test procedure utilized + A review of glassware utilized for the test; this includes pipettes + Acreview of equipment, columns, charts and previous analyses of samples of the same product + Reanalysis of dilutions + Acomplete investigation and evaluation of initial results prior to a retest If this investigation identifies a problem that resulted in the OOS result, the initial OOS result is invalidated and the test is repeated. In the context of this document, this repeat testis not considered a retest. It is documented in the laboratory deviation/nonconformance system (see below). I the initial results not invalidated, a laboratory manager or director usually becomes involved and a formal investigation is initiated. At this point, retests may be performed to either invalidate or tonfirm the initial results(s) wilizing preestablished procedures. Typically, two analysts perform two weighings and uplicate injections, thus generating eight results. It is then up (o the responsi- 23 LABORATORY CONTROLS 38 ble manager to review the laboratory and manufacturing data and history and either invalidate or confirm the initial results, If the analysis is of tables, the investigation includes a retest of the same mix or grind. In some eases, tablets are not ground, but are placed in a flask, weighed, and diluted to volume where inital extraction is performed. For retest purposes, this equates to the same mix or grind. In the event that the initial tablet extraction flask is unstable. @ new sample can be prepared from the initial composite sample of the batch. I the analysis is of a solution, a retest may be conducted of the same sample con- tainer. It should be pointed out that the retest procedure should be specific and include the number of sample weighings and number of duplicate tests When reviewing lab data associated with an QOS/OOT incident, there are some basic procedures that should be followed. This includes weighing every sample tested, even when content uniformity and dissolution testing bas been conducted, Additionally, it would be difficult to eonduct a meaningful invese ligation without duplicate injections of each sample preparation to rule out instrument/test system error. There should be acceptance criteria established to demonstrate the reproducibility of test results; for example, assay values should be within 2-3%, ‘At Some point after the informal investigation is complete and the decision is ‘made to embark on the formal retest mode, other issues have to he addressed and documented. These include the Following + A review of production and controi records. + A review of the product history * A documentation of the investigation and the logic and recommendation of disposition of the bateh by the manager/director. If the conclusion is that the value is an aberrant value, it should be reviewed by the laboratory director Retests for content uniformity and dissotution features are the second (S2) and third (S3) stages (for dissolution) testing, as provided for in the USP gen- eral test procedures. FDA policy has been that retesting is the second (or third) Stage unless orginal results can be invalidated as described above. The second and third stage of testing follow the same formal investigation procedures. All 53 testing should include « review of production and control records and his- torical data, 2.3.3, Laboratory DeviationsiNonconformances, Another system associated with OOS/OOT investigations is thet of laboratory deviations or nonconformances. They can be defined as laboratory situations, including deviations agreed upon prior to any analysis being performed, it which a departure has occurred from an SOP, method, specification, or proto- col. These incidents include, but may not be limited to, equipment failures,40. LABORATORY CONTROLS AND COMPLIANCE Table 21. List of Laboratory Deviation System suitability failure Failure to use bracketed standard when required Reference standard not run ‘Wrong sample sianderd dilution Degraded column Changed injection volume Tcorrect wavelength Deseribed resolution not attained Tnsuficient mobile phase Described RSD (eften <2.0%) not attained Significant pressure drop Deseribed cosflcient of correlation not attained Pressure increase (shut awn) Change in programming time for wavelength change during HPLC run ‘pH of bullery/Soluttons used for HPLC analyses cuanged Degradation compound not available or expired Incorrect filter used Incorrect sample preparation Low theoretical plates Injector malfunetion Detector variability Change in flo rate (Change in peak shape Retention time shi, equipment calibration failures, documentation errors, and procedural deviae lions. Examples of typical deviations are included in Table 2.1. Examples that, would not be classified as deviations are in Table 2.2. The identification and tracking of deviations is an important management tool to determine overall laboratory compliance, to identily training needs, to track trends and highlight method problems and to specifically evalvate analyst performance. Because both the laboratory deviation and laboratory OOS/OOT invest gation procedures are significant laboratory management tools, itis important for management to account for and to periodically discuss and review these investigations In some cases a company may merge the laboratory deviation system into the OOS/OOT system. While from a compliance perspective it is acceptable, it involves considerably more work, particularly when the laboratory deficiency is obvious and documented and any result (if one is generated) can be easily invalidated. Alll laboratories must have written procedures that describe how errors and failures are handled, investigated, and resolved. Itis expected that failure in- Table 22. Oceurrences That Are Not Deviations Power faire Primer erorjiam Calculation error Sample inadvertently tested rvige Information omitted on printouts 29, LABORATORY CONTROLS 41 vestigations be specific enough to provide for clear diteetion and responsibilitics for decision making with appropriate sign-off or documentation by manage- ‘ment for investigation of failures 2.3.4, Test Methods/Procedures/Specitications Methods used should be written in company format. Copies of the compendia, Which may allow for interpretations, are not to be used in lieu of a company: written test procedute. Procedures are appropriate for the operation, assembly, cleaning, regenera: tion, and inspection of some equipment. All Inboratory procedures must be readily available to laboratory personnel, Specifications for components, raw material, reagents/solutions, standards, processing intermediatcs/subassemblies, and final produet must be readily available to laboratory personnel. These specifications must be controlled documents, meaning that they should be tniquely identified, they should be subject to document change control policy, and not more than one version of a specification should be in effect at any time, Another issue with regard to method specificity is the utilization of thin-layer chromatography (TLC) methods in lieu of more specific liquid chromatography/ gas chromatography (LC/GC) methods. There should be & program for the review and evaluation of all semiquantitative TLC methods and replacement, if appropriate, by more specific quantitative methods, 2.3. Callbration and Maintenance One area that is frequently cited on FDA 483 observation reports is the failure to calibrate and maintain laboratory equipment. Laboratory equipment must be calibrated. Thore should be a system to indicate calibration due date and to provide traceability to the calibration records. Calibration poliey/procedures, calibration standards, and acceptance criteria must be written in controlled documents, Equipment-specific logbooks detailing the use, cleaning, regeneration, cali: bration, repair, and so on, are required for complex equipment used for multi- ple product evaluations or used by multiple technicians Each laboratory should have a preventative maintenance program with re sponsibilities. Because maintenance records, particularly from outside vendors, are utilized in laboratory audits, there should be a system for senior laboratory ‘management to be informed of any unacceptable calibration/maintenance. AAs discussed in the next section, itis not unusual for the auditor to identify instances of maintenance work on equipment, and then review raw data gen- erated during the maintenance period to ensure that there is correlation. From an auditing perspective, there can also be some review of correlation with lab- oratory deviations.42 LABOAATORY CONTROLS AND COMPLIANCE 2.4. LABORATORY COMPLIANCE 2.4.1. General Notices The USP [11] contains considerable information that clarifies compliance re- quitements. The General Notices section discusses tolerances, as well as the basis for their establishment, including overages for analytical error, manu facturing variability, and deterioration, Furthermore, the preface points out that the in-house quality release tests specified by a manufacturer should not be confused with official compendial standards. Basically, the trend among man- ufueturers has been to have tighter release specifications forall tests except for the sterility and identity tests, Tests that should have tighter in-house release limits include assay, impurities, content uniformity, dissolution, endotoxins, pacticulate matter, microbial limits, and pH. For example, if the dissolution specification is set too tight by regulatory authorities, then fhere will be an excessive number of nonconformances and retests. Some regulatory managers believe that there should be @ percentage of batches having S2 {second stage} level testing. Unfortunately, they fail to rec- ognize that dissolution testing cannot be repeated if the OOS result cannot be clearly and directly attributed (o analytical error. Having a low percentage of tablets below Q (the means of the dissolution values) increases the probability of having a dissolution failure that may indicate a product adulteration, ‘The USP also defines information regarding added substances. It points out that added! substances cannot interfere with tests and assays. Because official (compendia! and filed) tests are regulatory tests, they should be able to be con ducted by a regulatory body on a stand-alone basis. This precludes the use of a placebo to blank out interfering substances. ‘The USP also provides information with regard to parametric release (re- lease based on processing results, without testing). It points out that itis not a requirement to perform every compendial test to assure compliance. It states the following: ‘Data derived from manufacturing process validation studies and from in-process controls may provide greater assurance that & batch meels a particular mono- raph requirement than analytical data derived from an examination of finished luits drawn from that batch. On the basis of such assurances, the analytical pro- ‘cedures in the monograph may be omitted by the manufacturer in judging com- plianee of the batch with the pharmacopeial standards [12) Depending upon the wording in an application, one could interpret the above to mean that fled tests do not have to be performed as long as compli- tance with the particular specification is assured. For example, it is not unusual for a company to delete performance of filéd identity tests when there are other ‘means to assure product identity. However, it would be difficult to make & case for deleting the very basic assay and impurity tests. These tests serve as a 24 LABORATORY COMPLIANCE 43 check on the process and operator performance and could identify foreign contaminants 2.4.2. Method Development The FDA PAL Program initiated in 1990 inereased focus on the development of processes and the documentation of this development. The 1994 FDA Guide to Inspections of Oral Solid Dosage Forms {13] discussed product development reports. The Guide points out that while there ae no statutes or regulations that specifically require a development report, companies are required to produce scientific data that justify processes. Cleary, the FDA. could support an analo- gous interpretation for methods and the need for scientific data to justify methods, This same FDA Guide addreses the laboratory and points out that the teansfer of laboratory methods and technology from the Research and) Devel- ‘opment Department to the Quality Control Department should be reviewed during inspections. Any review of method validation and method development could include a review ofthe development trials and experiments that result in a laboratory method. From a compliance perspective, there are some very basic issues that show bbe addressed in the development of a method. Ideally the drug substance and drug product should have the same stabilty-indicating methods for assay and itmputity testing, For example, itis not unusual to have a foreign contaminant appear in an assay or purity test, One of the sources of the forcign contami= nant could be a foreign contaminant in the drug substanes. Having the same method hips to identify the contaminant source. Ifthe tet methods are dif ferent, the dosage form manufacturer should look at the drug substance vith the dosage form method Another general issue would inchude partnering with formulators to ensure that there will be no excipient interference. As discussed previously, itis un- acceptable to utilize a placebo blank for a regulatory method. With regard to method development, good practices would include a proce- ‘ute for developing methods and a protocol detailing acceptance criteri forthe method. The procedure should also include provisions for the referencing of raw data in the method development report In order to demonstrate robusiness of a method, some basic areas that should be addressed in method development and ineluded in the method de- velopment report areas follows: + Stability of reagents, mobile phases, standards, sample solutions + Temperature variations + Mobile-phase composition variations + Chromatographic gradient profiles and mixing systems + Flow rate (pressure) variations44 LABORATORY CONTROLS AND COMPLIANCE + Multiple fots of columns + Alternate columns, + Instrument parameters (wavelength, bandwidth) + Sources of reagents + Limits of detection with different instruments 2.4.3. Method Validation ‘The GMP regulations require that test methods must meet proper standards of accutacy and reliability. Users of analytical methods described in the USP and in the National Formulary (NF) are not required to validate accuracy and reliability of these methods, but merely verify their suitability under actual conditions of use. However, for demonstrating and testing for stability, other noncompendial methods may be used in addition to, or in lieu of, compendial ‘methods. Such meyhods do require validation, The USP provides direction for parameters needed for the validation of test ‘methods. Parameters include accuracy, precision, specificity, limit of detection, limit of quantitation, linearity, range, and ruggedness. The USP also defines and provides direction for the application of these parameters to the specific type of test procedute being validated {14} In the validation of methods, the USP points out that method specificity is ‘a requirement. Its defined as the ability to measure the analyte in the presence of components. Therefore, methods must be able to separate impurities and dogradents from the active drug substances; components of the formulation ‘must similarly be separated ‘A major aspect of method validation, which has generated some discussion, is the establishment of the actual limit of detection (LOD) and the limit of quantitation (LOQ). In the development and validation of an analytical test ‘method for impurities or related compounds it is important to establish the lowest limit of detection and quantitation. In a typical audit or FDA inspec tion, the review of chromatograms will include the determination of total im- purities by adding together the area under all of the peaks. From a compliance perspective, it is important to quantitate any impurity peak above the noise level. However, from a practical aspect for dosage forms, itis very difficult and laborious to quantitate any impurity peaks at_a concentration of less than 0.03%. For the most part, the Intemational Conference on Harmonization (ICH) and previous references point out the need to quantitate impurities at levels of 0.1% or greater. In order to establish 0.1% as threshold, LOQ values of less than 0.1% are needed. As technology advances, the true LOQ for many substances is becom. ing less than 0.1%. In some cases, FDA reviewing chemists have been looking for impurities of less than 0.1% (0.03%). Thus, itis becoming increasingly im- portant for analysts to quantitate impurities down to levels of 0.03% in purity ‘methods that they use. It then becomes the responsibility of the supervisor to review the results, evaluate their significance, and report them as appropriate. 24 (ABORATORY COMPUANCE 45 As stated earlier, one of the major issues in the industry isthe compliance of older products under current standards and utilizing curfent technology. With improvements in analytical equipment, such as detectors, columns, and so on, previously unidentified impurities that are close to the level of 0.1% are being detected. For dosage forms where the daily dose is 10 mg to 2 g, impurities present at around 0.2% or more should be identified. In the review of stability data, this can present problems when the level of the unidentified impurity exceeds 0.2%, Thus, itis important to identify unknowns at levels of less than %. Industry needs to be proactive in the quantification and even the identi- fication of impurities by having a plan for corrective action in the event that ‘otal impurities found in product and/or drug substance approach limits, The obvious corrective actions include improving the manufacturing process and/or changing the limi ‘When addressing method validation, the titration method is a method that is frequently overlooked. The USP 24 states in Section <1225) [14] that general assays and tests (tittametric methods) shoiild also be validated to verify their accuracy when used for a new product and raw material, From a compliance perspective, the use of titration methods, for the assay of drug substances, should be discouraged because the method is nonspecific and not stability indicating. Furthermore, some impurities may not be discernible or may not absorb when chromatographic methods are used to detect impurities and de- {gtadants; the use of a nonspecific assay method may be of value, Additionally, methods and specifications may not be married, While the USP states that the identity, strength, quality, and purity are determined by tests and assays relating to the article (USP 24, p. 7) [li], it also states that stability testing (Section <1191}) [15] for both the drug substance and drug product should be performed by validated stability-indicating test methods, For drug products USP 24 states [16] that monograph assays may be used for stability testing if they are stabilty-indicating (capable of differentiating intact ‘drug molecule from degradants). There have been examples in which drug substances have been tested by the FDA utilizing specific LC methods, and not utilizing a nonspecific official titration method. In a few cases the FDA has utilized the assay limits in the ‘monograph purity rubric statement as the specification for the substance, re- gardless of the method specified in the monograph. Such situations have re- sulted in improvements in the drug substance manufacturing process, methods, and specifications Other aspects of method validation, which have presented compliance prob- lems, include robustness and ruggedness. In order to determine the ruggedness of a method, different laboratories would need to be used. One of the compli- lance activities associated with the introduction of a new product is the transfer of the method from the developing laboratory to the operational laboratory. A. successful transfer demonstrates the ruggedness of a method. It is not unusual for a method-developing laboratory to issue a method validation report without demonstrating ruggedness (Which includes testing at another laboratory). From46 LABORATORY CONTROLS AND COMPLIANCE 4 compliance perspective, it could be concluded that the method has not been validated because it hasn't been transferred or performed by another labora- tory. (Refer to Chapter 15 for further discussion on method transfer.) ‘With regard to robustness, it is a measure of the capacity of an analytical procedure to remain unaffected by small but deliberate variations in method parameters. Issucs that pertain to robustness are addressed in the development of ‘method. (Refer to Chapter 12 for further discussion on method development.) 2.4.4. Method Transfer Prior to performing any method transfer for drug substance or drug product, there should be a protocol in place. Some compliance requirements to be addressed in each transfer protocol are as follows: + The transfer should be performed on two batches, with three sample preparations for each batch. Ideally, the batches are not marketed and contain some level of impurities. + Mean assay values between laboratories should be within 2% for drug substances and 3% for drug product. + Precision should comply with suitability), + For known impurity determinations, the absolute difference between results should be no greater than 0.2%, Precision (system suitability) should be within 10-15% for the knowa impurities tralaboratory requirements (system + For unknown impurities, there should be qualitative comparability + For oral solid dosage forms, mean dissolution values should be within 6¥% and mean content uniformity values should be within 5%, * Linearity is performed during method validation and is not a transfer requirement + The LOD/LOQ provided for the method should be verified by utilizing standards at these levels, + For compendial methods, the laboratory should verify that the method can be performed and that there are no interfering substances, 2.48. Auditing the Laboratory ‘As discussed in the introduction to this chapter, the major reason for conducting. aan audit of a laboratory is to ensure that data and results are not fraudulent. Also, audits are conducted to ensure that GMPs and commitment to regulatory filings are being followed, ‘There are some very general techniques that are utilized when auditing the laboratory. Basic to any auidit isthe observation of analysts, All of the arcas of laboratory compliance can be evaluated by actual observation, rerenences 47 Other techniques, including having the responsible manager open drawers in the areas of analysts not present, can also be employed to identify data not reported or samples not teste. A good auidit will also include the review of maintenance reports and sub: sequent comparison to data in analyst logbooks or data sheets. If problems with instrumentation are identified during analyses, there should be some rec« cognition of ths in the analyst logbook or data sheets. Generally, auditors look for missing electronic files, out-of-sequence data, and cut chromatograms as possible indicators of fraudulent data, {is important to identify laboratory personnel and the laboratory manage ‘ment chain early in the audit. This will enable the auditor to identify respon- sible personnel and personnel who are outstanding performers. Typically, the best analyst will perform testing on the more difficult and/or problematic tests A good laboratory audit will generally include a review of a representative rumiber of laboratory deviations and OOS/OOT investigations. 2.486. Use of Outside Testing Laboratories A laboratory is responsible for the quality of the outside testing facilities that it uses, Laboratory management is responsible for assuring the quality of the work performed by these service laboratories; the use of audits and sending spiked) known samples, and so on, may be appropriate. Additionally, there should be some type of written agreement regarding the responsibilities, reporting of data, tnd supply of copies of raw data, 2.6. CONCLUSION A number of issues central to laboratory compliance have been discussed in this chapter. These include laboratory certification and management, records, ‘method validation, retesting, training, computerized data acquisition systems, calibration, and auditing. Certainly, there are other issues that need to be addressed, and this discussion was not intended to be all-inclusive, Although the establishment and monitoring of systems designed to support laboratory compliance may seem @ daunting task, the company with a true commitment to (quality will find a way to incorporate all issues in an effective quality plan. REFERENCES 1. United States of America vs. Warner Larshert Company, Cisil No, 93-3828, Aves 16, 1993, 2, United States of Armerica rs, Bioerat Laboratories, Ine, Cixil No, 94-3499, July 28 1994 3. HLL. Avallone, Laboratory Certification, August 24, 1994,48 1, 4, R. LABORATORY CONTROLS AND COMPLIANCE: Title 21, U.S, Code Faderal Food, Drug and Cosmetic Ae. ‘Current Good Manufacturing Practice Regulations Federal Register, September 29, 1978, pp. 45067. Current Good Manufacturing Practice Regulations, Federal Reystr, September 29, 1978, p. 45086, 21. CRF, Flectonic Signature Records Part I: Bleetronie Signatures, Part Hl, August 20, 1997 . Laylolf and W. Forman, AOAC Int, 1994, 77, 1514. United Stares Pharmacopeia, th edition, (621) US. Pharmacopeial Convention, Ine., Rockville, MD, 2000, p. 1923, United States of Americas. Barr Laboratories, Civil No, 92-1744, February 4, 1993 United States Pharmacopeta, 24h edition, 2000, p.7. Gurrent Good Manufacturing Practice Regulation, Federal Register, Soptember 29, 1978, p, 45086, Food and Drugedministration, Guide to Inspections af Oral Solid Dosage Forms, Jamwary 1994. United Sites Pharmacopeia, 24th edition, (1225) 2000, United States Pharmacopela, 24th edition, (1191) 2000, p. 2128, United Stares Pharmacopeia, 24th edition, <1151) 2000, p. 2106 THE USP, ICH, AND OTHER COMPENDIAL METHODS Jennifer G. Feldman 3.1. INTRODUCTION In order to ensure the quality and uniformity of pharmaceutical substances throughout established geographical regions, several compendia were exeated These lengthy documents, established separately in the United States, Great Britain, Europe, and Japan, provide standards and specifications for all facets of pharmaceutical materials, including thei testing, packaging, and storage With globalization of the pharmaceutical industry @ need arose to harmonize these separate standards, and the International Conference on Harmonisation (ICH) was formed, This committee has the task of harmonizing the requize iments for pharmaceutical products among the participating members: The United States, Europe, and Japan. While the prooess has Been a slow and arduous one, much progress has been made. This chapter will discuss the indi- vidual pharmacopeia, which are still in effect, and briefly summarize the eon- tent of the various guidelines 3.2. USPINF 3.2.1. Introduction ‘The United States Pharmacopeia and National Formulary (USP/NF) is the Jatgest and most comprehensive of the national compendia, whieh also include Analytical Chay m 2 GMP Enrvrmen. Bed by JM. Miller and J. B. Crowther ISBN 0-1 SEATS C2000 Fo Wiey f Sons, ie_4 STATISTICS IN THE PHARMACEUTICAL ANALYSIS LABORATORY Alvin J. Melveger When You Measure You Know. “—Lowd Kelvin ‘The essence of obtaining knowledge about our universe, environment, or ana- |ytical samples depends upon making reliable measurements. The measurements usually involve the generation of data expressed as numbers. The knowledge of a sample that we ate examining follows from our ability to record and analyze the numbers that we generate. Unless these data are treated in a consistent and correct manner, the knowledge we are seeking will be fawed and perhaps incorrect. Handling data, and making decisions about their “correctness,” is usually accomplished with statistical methods. In the pharmaceutical industry, statistics are applied throughout—-in the design of manufacturing processes, in sampling, in the evaluation and control of processes, in setting proxiuct specifications and shelf life, and in the analytical laboratory where methods are developed and evaluated and products are analyzed The use of statistical techniques is quite pervasive in the pharmaccutical industry. The United States Pharmacopeia (USP} has addressed the issue as nasil Chemtary na GMP Environment. Biel by J. M, Mille and J. 8, Crowther ISBN 0471-31431 “© 2000 Johm Wiley & Sons, Ine7B STATISTICS IN THE PHARMACEUTICAL ANALYSIS LABORATORY related to biological assays as in the USP general chapter, “Design and Analy- sis of Biological Assays” [1, Section <111)] For statistical treatment of analytical method development and data derived from these methods, the USP has published general information chapter, “Validation of Compendial Methods” [I, Section 1225). Although this chapter outlines the statistical tests necessary, it does not present the method- logy to be employed. The USP is presently considering more formal use of statistical methods ap- plied to interpretation of analytical ata (2. This isin response to judge Wolin’s Aecision [3] im the United States vs. Barr pharmaceutical litigation. The judge clearly asked for compendial guidance for understanding analytical data and ‘out of specification (0S) results. Statistical methods are useful in determining the best method of sampling incoming raw materials, analysis of data for accuracy and conformance to 4 specification, and comparing the measured values of product to established specifications [4 ‘When developing new analytical procedures, it is necessary to evaluate the accuracy, precision, tineary, specificity, limit of detection (LOD), and limit of ‘quantitation (LOQ) of the method and to compare the method to alternative available methods [5) Determining the shelf life of a pharmaceutical product enables the assign- iment of expiry dating. Inciuded in the process is determining stability under accelerated conditions and applying statistical and chemical kinetic methods to extrapolate to actual shelf stability under ambient conditions This chapter will focus on the basic concepts regarding the generation and treatment of data obtained from analyzing samples in the pharmaceutical lab- oratory. The treatments described may not be completely rigorous but should be sufficient for most purposes. These will include a review of basic statistical approaches for handling and evaluating dat, the variability of testing method, and the confidence placed on the results, A list of general references is provided at the end of the chapter for more complete discussions of the subjecs. 4.1. ERRORS ASSOCIATED WITH MAKING MEASUREMENTS. 1 should be recognized that variability is inherent in every physical process or ‘measurement. Some factors may be carefully controlled, but even well-designed and controlled laboratory experiments will exhibit variability that may origi- nate from slight fluctuations in environmental conditions such as temperature, pressure, lhumidity, cosmic radiation, and so on. Statistical methods are aimed at providing objective, quantitative, and reproducible ways of assessing the cifects of variability There are two types of variability or error, systematic and random, The total error is the sum of the two, but systematic errors can be, and should be, found and eliminated. 42, SIGNIFIGAWT FIGURES AND ROUNDING 79 4.1.1. Systematic Error Errors that could be avoided but creep into measurements at times unknown to the analyst are called systematic ertors or determinate etrors. They may or may not be proportional to the concentration of analyte and may bias repeated or replicate measurements equally. These errors may be unpredictable or constant, Examples of such errors and possible remedies are given in Table 4.1. In general, statistieal methods cannot be used to analyze such errors since they are unpredictable. As a result, they may seriously affect assay resulls. The good ‘ews is that once recognized, these errors may be addressed and rectified, as suggested in Table 4.1 4.4.2. Random Error Random errors are inherent in the measurement system and may be due to small ‘changes in environmental factors such as temperature, atmospheric pressure, humidity, cosmic radiation, and so on. Statistics may be applied to these ran. dom errors to determine their magnitude in a series of replicate measurements. Averaging of replicate measurements ensures that random error will be mini. ‘mized and will end toward zero. Statistics allows us to analyze replicate mea surements fo assess the variability or uncertainty in the analysis. A fow of these Statistical tools will be examined in this chapter, but belore describing the treatment of data let us discuss the essence of data which includes treatment of ‘numbers, significant figures, and rounding. 42, SIGNIFICANT FIGURES AND ROUNDING Any number generated in a laboratory measurement has only a finite number of digits associated with the measurement. These digits are determined by the analyst's ability to read a scale or an instrument's electronic or visual output. The number of significant figures associated with a measurement depends on the capability of making such a measurement. For example, an analyst using 4@ laboratory balance that reads out to 0.01 g should not report a result to 0.0001 g. 4.2.4, Number of Significant Figures ‘The number of signifieant figures is the sum ofall those digits you can read with certainty from your measuring scale plus one extra digit that is estimated, The estimated digit isa careful estimation from the scale. I is therefore the size and ature of the scale and only the ability to read it wiich determines the nurnber of significant figures Consider the following example, shown in Figure 4.1, which is associated with reading the same point using two separate scales having differing numbers80 STATISTICS INTHE PHARUACEUTICAL ANALYSIS LABORATORY ‘Table 4.1. Systematic Errors Type Observed Resulis™ Gonaaaion Sere ny have rin tl eda Aes ths ipury cred ond ty be sy ceed tome yo thos teamen Caltrarion Errors Incorrect standards, outdated, 1,2 standards, or cheek oos solutions. Instruments not propery calibrated Lasses(Degradation® Samples andor sample ' solution could be degrading during storage or analysis, Heat, light, or humidity may be ut fake Unvepresentative Sampling AA specific analytical sample 12 oes not represent the bulk of the materia Cnsuitable Merhods A particular method may 12 introduce bias, Incorrect columns in chromatography ‘may nat give proper elution of analytes, Incompetence Improperly rained analysts 12 utilizing instrumensation or methods inypropery. 1, Low assay 2, high asay; O05, out of spietion. Possible Remedies Reject samples that may appear discolored or have foreign inclusion, Utlize traceable standards and follow established procednres ‘Make sure thatthe standards are ceriied and not outdated Minimize exposure of samples to sources of degradation, Examine results of stability ‘studies to understand environmental factors. Metis development and validation activities should suggest sample handling conditions and time Limitations, Set up a sampling pla involving statistical sampling so thatthe collection of samples will represent the actual bulk ‘material Use only validated methods, Ensure that analyst has trained ‘on method and type of samples being assayed. 42, SIGNIFICANT FIGURES AND ROUNDING Bt — 4000 = 400 _ 3766 3800 3800 2000 3000 Figure 4.1. Etect of scale divslons onthe urbe of signiteant gues that canbe read (of graduation marks, Note that the let scale only allows fortwo digits (ie, 37) {o be read with certainty, and a third is estimated (37.7), This reading has three significant figures, The right scale, which has more graduation lines, allows for three digit (.., 37.6) tobe read with certainty, and a fourth is estimated (37.66) This reading has four significant figures {In digital instrumentation, typically the last digit is assumed to be estimated. For instruments such as balances, pH meters, and calculators, the same rules apply. Only report significant figures that are known with certainty based on 4 known standard, for example, Then provide an additional figure from the readout as an estimate, Traceable standards must be used to verify the in. Strument's capability or displaying the certain and the estimated digits. When using volumetric glassware or weights, itis necessary to know the classification and tolerances of these so that the correct number of significant figures for Weights and volumes may be reported. Tolerances for weight and volume82 STATISTICS IN THE PHARMACEUTICAL ANALYSIS LABORATORY standards are reported in ASTM standard specifications E 617-78 [6] and E. 694-79 [7], and some NIST volume tolerances are included in the next chapter (Table 5.2) Icisimportant to understand the number of significant figures the instrument or device is capable of generating and not to express more significant figures than this number. Specifications for materials must be consistent with the ‘capabilities of the instrument as (o the number of significant figures that may be reported, 4.2.1.1. Signiticant Figure Conventions. There are a number of conven- tions to be followed in determining the amount of significant figures in a reported number: 1, The digits 1 through 9 ae significant. 2, Zeros betweEh nonzero digits are significant; for example, 1.084 has four significant gures 3. Zeros to the left of nonzero digits are not significant. These zeros only show the position of the decimal point; for example, 0.084 has two nificant figures. This could also have been expressed as 8.4 x 10"? or 84 x 10- or 0.00084 x 102. 4, Zeros that fall at the end of a number are not significant unless they are ‘marked as significant by having a line placed over them or by some other convention, 1600 has two significant figures and 1600 bas four. 5. Zeros to the right of the decimal point are always significant. 25.00 has four significant figures but 25. only has two, ‘When reporting @ value, do not automatically add zeros after the decimal ppoint unless the actual measurement has given those figures as significant. If an instrument is incapable of giving more than a certain number of significant figures, do not assume that addition of zeros after a decimal point is acceptable. Use correct judgment in reporting data calculated with an electronic calculator, it probably has more significant figures than is justified. The correct number of significant figures should be specified in the analytical method being used. 4.2.2, Rounding Afler making a laboratory measurement consistent with the rules for significant figures, we may then be asked to employ numbers in calculations. The question arises as to methods for handling the significant figures in our calculations and in the reported final result, There is a tendency to report more significant figures ‘han is actually generated in the measurement. For example, if we were asked to determine the density of a metal by measuring its weight and the volume of liquid it displaced, we would proceed to substitute our measurements in the 42, SIQNIFGANT FIGURES AND ROUNONG 83 formula _Weight (g) Volume (tt) Density = (aay For a measured weight of 15.125 g and displaced volume of 6.1 mL. the com: uted density read from a calculator is 15.125/6.1 or 2.4795081 g/mL. Clearly this computed result has many more significant figures compared to the ind vidual weight or volume measurements, The question is, How do we make the calculated result consistent with the measurements, and at what stage of the calculation do we make an adjustment? The process of adjusting the result to the correct number of significant fig- lures is called rounding. The following rules are followed in the rounding pro. cess. There are other conventions used in the rounding procedure, but we will limit ourselves to the following: - | To round a number, look at all digits beyond the last decimal place desired and raise this lst place value by one if the number following the last place is $ or greater (see example below). 2. For whole numbers, if the series of numbers after the desired place is exactly halfway up (5, $00, 500,000, tc), then raise the desired digit by The following examples demonstrate the rounding process. In the previous calculation of density, the computed value was 2.4795081. The measurement of volume had only two significant figures (one after the decimal). The final reported density may therefore not have more than two significant figures! Tio ‘una! we 100k atthe computed value and examine the numbers after the fist fecimal place. For 2.4795081 the value 795,081 is more than halfway 10 '-000,000; therefore the required rounding is to report the resull as 2. (add 1 to the 4), consistent with the number of significant figures of the least known measurement (volume), leis important to know at what point in our calculation to round. Consi the following guidelines ° Constr |. In addition or subtraction, the result is rounded off to the smallest rnuraber of decimal places to be found in the numbers that were added or subtracted. The result ean have no more decimal places than the number With the least such places. 2. In multiplication or division the sesult is rounded off to the smallest ‘number of significant figures to be found in the numbers that were mult- plied or divided. 3. In a chain of calculations, rounding off is not done until the final result hhas been obtained. The number of significant figures in the final result is84 STATISTICS IN THE PHARMACEUTICAL ANALYSIS LABORATORY Table 42. Some Statistical Deinitons Name ‘Symbol 1 Mean (overage) x 2. Average mean deviation D 3. % Relative mean deviation - 4, Standard deviation ° 5. Variance a 6. Relative standard deviation RSD Coefficient of variation wv 7, Standard error © SE Standard deviavion of the mean om 8, Confidence iit cL the same as the least number of significant figures found in the numbers that were used in the chain of calculations. 4. Ac least one more place beyond the last figure required in a specification ‘will be cartied through the calculations to determine the reported values. 5. Rounding will not be performed until the average final value is calculated 6. Numerical results will be reported to the same number of Higures shown for the test requirement in the specification. 7. The number of figures shown in a specification requirement must be con sistent with the number of significant figures the vest method or equip- ‘ment can generate. 4.3. SOME DEFINITIONS* Before proceeding to a discussion of the uses of statistics in handling laboratory data, it is necessary to summarize the terms and symbols to be used. The most ‘important ones are listed in Table 4.2 [8-10), 43.1. Accuracy Accuracy describes how close « measurement is to the correct or true value, tis not unusual for the accuracy to be indeterminate because an incoming sample is * See Chapter 3 Appeodis I for ICH detuitins 43. SOME DEFINITIONS — 85 ‘not identified with a true value. In fact the purpose ofa lab analysis of a sample is to determine what that value is and if it agrees with a specification, Accuracy may be assessed from running traceable standards such as from the National Institute of Standards and Technology (NIST), USP, oF internally generated samples that have been well-characterized. Standards are available from USP and NIST and ate provided as certified standards with the appro- priate assay or other parameter given. In developing new methods, accuracy is assessed by determining the agreement of the analytical procedure with the known standard value. For new drug substances, one may use internally de- veloped standards that have been well-characterized by appropriate techniques {or purity levels and assay values 4.3.2. Precision Precision is 2 measure of the reproducibility of a measurement, It expresses in 4 quantitative fashion the degree of scatter of tepettive measurements on the same sample. Good precision implies that the analyst is producing a consistent ‘result each time the same measurement is made on the same sample, Precision does not address how close this measurement is to the true value. It is possible to have high precision but poor accuracy, This may come about, for example. in a situation where there is a systematic error causing a bias, If a sample degrades 50% during a measurement, the reported result could show very litle scaiter in the data but would yield an assay only one-half the true value, An incorrectly calibrated instrument could ako give rise to a situation of poor accuracy but good precision, Poor precision is influenced by random errors that are part of any measuree ‘ment, Running replicate analyses will minimize the effect of these random ‘errors. A distinction may also be made for within-ran precision and between ‘run precision. The repeatability, or within-run precision, is a measure of the precision of replicate runs determined for the same sample preparations and lassware within a short time span. The reproducibility or between-run precie sion is determined for replicated runs at diferent times, This precision is gen erally not as good as the repeatability 4.3.3. Absolute Error When the true valuc of a measurement is known, then the absolute error is expressed as Absolute error = Measured value — True value (42) This is usually expressed as a positive number having the units of the measured Property. For example, ifa standard has a known value of 24.2 g ofa particular Constituent and we measure it in our laboratory and get the value as 24.5 g, the absolute error of the measurement is 0.3 g,85 STATISTICS iN THE PHARMACEUTICAL ANALYSIS LABORATORY 4.3.4, Relative Error Relative error compares the magnitude of the absolute error to the size of a ‘measurement and is usually expressed ns a percent, Its defined as Nolte eto Relative exor = (ASME STEN) 5 10 3) In the example just given, the relative error is (0.3/24.2)100 = 1.2%. In this ease the relative error is a measure of the accuracy of our measurement because We ate comparing it to a known standard value, AC times an analyst may be biasing a reading, thereby introducing error. For example, in reading a burette it is found that no matter what volume the ‘meniscus is at, an analyst cannot read the scale to better than 0.02 mL. This absolute error of reading occurs at any point along the burette scale. ‘The relative uncertainty of this measurement is very dependent on the mag nitude of the voluihe being read. Therefore for a burette reading of 30 mL the relative error is (0.02/30) x 100 = 0.07%, If the bureite reading had been at 10 il, the relative error would be (0.02/10) x 100 = 0.2%. 11 ean be scen that by reducing the scale reading from 30 to 10 we have increased the relative error in reading the burette a factor of almost threefold. In making laboratory measurements it is therefore important to choose sample sizes that allow for minimizing relative error. For example, do not per form a titration in which only very small volumes are dispensed. In chroma tography or absorption spectroscopy, make sure that area counts or integrated band absorptions are large enough s0 that noise levels only contribute a small relative error to the measurement, 43.8. Mean The average of any number of repeated measurements is given by (4a) where P= mean, X;= value of single measurement, and n = total number of measurements. This is simply the sum of each mensurement divided by the number of measurements. Consider the following separate measurement values for an assay: 11, 12,10, 8, 9, The mean is calculated as +12 +10 3 43) 4.36. Average Deviation ‘The average deviation is a measure of how close each measurement is to the ‘mean (use positive or absolute numbers). Deviation is defined as the difference between any given measurement and the mean of the measurements, 43, SOME DEFIIMONS 87 Fer the previous example in which the mean was 10 we have the following: ‘Measurement Value () m1 L 12 2 10 0 2 1 6 Deviation (X; — 8 9 ‘The average or mean deviation is calculated as [4G -¥|_ 142404241 paEWMiM 14740471 1, iy where | | represents the absolute value, A convenient alu. A convenient way to expres this pre- cision isto use the % relative mean deviation mene ‘Average mean deviation’ Relative mean deviation = x or pan 100 (4.7) For the example given, the relative mean deviation is 1296. 43.7, Standard Deviation Another important measure of the precision of a series of measurements is the standard deviation, «, The meaning of this will be discussed in the next section, but the conventional formula for a series of measurements is (a8) (49)(8B STATISTICS IN THE PHARMACEUTICAL ANALYSIS LABORATORY ‘The precision of the set of measurements may then be expressed as the average + an uncertainty range. For example: Fac (4.10) Or, as in the case of our example, 10 + 1.6, The + indicates that the deviation is above and below the mean and equally spaced. The actual range is therefore 84-116, 4.3.8. Relative Standard Deviation ‘Usualy the standard deviation is exposed eative tothe mean and dined as fi deviation % Relnive stand deviation (Rb) = (SHMBEESEWIANER) «199 (4.11) In our example the 96RSD is 16%. The RSD is also referred to asthe coefficient of variation (CV). During the validation process of an analytical method, limits are placed on the magnitude of the allowed RSD or CV, Requirements for good precision will permit RSD of only a few percent. For example, the USP requires an RSD of 2% or less in any assay analysis, 4.3.9. Comparison of Precision and Accuracy Tt should be stressed that precision and accuracy may fall in opposite directions for a series of measurements. It is possible to have high or low precision cou- pled with high or low accuracy. Ideally we strive to make our measurements highly precise and highly accurate. To illustrate, consider the following four sets of replicate measurements of a standard having a true value of 20. These dllferent sets of data could have been generated by different analysts or on dif ferent instruments, 15, 25, 10, 30 19, 20, 21, 22 25, 24, 26, 23 25, 30, 35, 45 pap> We can determine the accuracy and precision of these sets of measurements using the calculated values of relative error of the mean compared to the known true value (accuracy) and the relative mean deviation (precision). (See Table 4.3.) These calculations illustrate that the four sets of measurements may be categorized as to their precision and accuracy: 49. SOME DEFINITIONS 89 Accuracy Mean Deviation % Relative Mean Relative from True Value Deviation from Average Mean Mean x 20) True Value Deviation, Deviation AD ° 0 15 a5 Bo 2 1 5 10 4a cus 4s 2s Lo 4 DBR 58 @ 64 189 A. Poor precision and good accuracy B. Good precision and good accuracy . Good precision and poor accuracy - « D. Poor precision and poor accuracy Hone were to visualize the results as shots at a target, the results would look like those depicted in Figure 4.2. Obviously, set B is the best and is always our objective. Set C could represent a situation in which there is a determinate ot systematic error; for example, the given sight could be misaligned, causing the projectile to miss the bull's eye, In such a situation, the cause of the determinant errors should be found and corrected. Notice that sample set A represents ‘measurements that are not precise but are aceurate, This is not a desirable si. uation, The high accuracy may be a fortuitous occurrence in a set of measure ‘ments showing poor reproducibility. The poor precision informs us that the ‘measurement is not under proper control. Clearly, set D is not desitable and indicates serious problems with the method. The analyst must pay attention to how the measurements are being made to evaluate situations of poor accuracy and precision, Attention must be paid to Proper instrumental calibration, correct preparation of standards and working Solutions, and noninterference from impurities or excipients, 4.3.40. Standard Error Statistical theory also predicts that the uncertainty of a series of measurements will decrease a8 the number of measurements increases, The uncertainty de- creases in proportion to 4 va where ris the number of measurements. For an infinite number of measure- ‘ments the uncettainty tends to zero, (4.12)80 STATISTICS IN THE PHAAVACEUTICAL ANALYSIS LABORATORY a Number of Measurements Number of ‘Measurements © Number of Measurements D ‘Number of| ‘Measurements ‘Conceneation Figure 42. itustiatons of the fou posse combinations of presen and accuracy. T equals target value ard M equals mean valve ‘The standard error is an additional estimate of precision, and for large sample sizes it is referred to as the standard error of the mean, This standard error is calculated by dividing the standard deviation by the square root of the number of replicate measurements: (4.13) ‘This measure of precision implies that increasing the number of replicates will decrease the error in a square toot fashion. Assuming a standard deviation of L |U4, NORWAL DISTRIBUTION OF REPEATED MEASUREMENTS 91 for a series of measurements, the change in standard error for 4, 16, and 64 Ineasurements would be proportionate as the ratios 1/2, 1/4, 1/8, or 4:2:1 ‘Therefore 64 measurements would only give usa fourfold decrease in uncertainty compared to 4 measurements. The extra work in running such a large number ‘of replicates may not be worth the effort in terms of decreasing the uncertainty. 4.4, NORMAL DISTRIBUTION OF REPEATED MEASUREMENTS. We have discussed some methods for estimating the precision of repeated or replicate measurements. The standard deviation, o, is an important indicator of precision or repeatability. Ths parameter is derived from statistical theory and has the following implications. ‘When we perform a chemical avalysis, perhaps on a standard, random error will eause repeated measurement to give sattered results around the true value (assuming no systematic error is present). If wé were to repeat the same mea Surement many times the measured values of X would take on the appearance of the curves in Figure 4.3 and Figure 4.4, where the Y axis represents the numberof times a particular valve of is found {811 This distribution curve comes about de to random errors associated in making the repeated measurements and is symmetrical around a velue of X. labeled , whichis the tre valve ofthe standard or the mean valuo for any se ob repeated measurements. The more measurements that are made, the closer the value isto the mean or average value, This curve is called a normal or Gaussian distribution and is igorously correct for an infinite number of re peated measurements, In practise one can develop such distribution curves with ‘many les than an infinite number of measurements Siatistical probability theory uses this normal distsibtion to prodiet the probability that a given measurement wil fll into a certain range of X values. The standard deviation becomes the messuring yardstick of data spreading as ilstrated in Figure 44, Its predicted that for repeated measurements on the same specimen, any measurement of X would have a value inthe range of (ie 0) 68.3% ofthe time, (ut 2n) 95.8% of the time, and (1+ 3) 99.7% of the time [11]. As the measured points deviate more and more from the rue value (large spread of the curve), the probability of finding w measurement ‘within tis lneger range inereases, This implis that the uncertainty of finding & ‘alu in this expanded range goes down. The standard deviation, c, defines the aarrowness of the distribution curve. For low standard deviation the curve narrows andthe repeated measurements cluster about the average value ‘We expres th confidence limits ofa series of measurements in tems ofthe standard deviation, Thus a 68% confidence interval refers to measurements that fallin the range of the mean + le. The 95% confidence interval includes the mean + 27 and the 99.7% confidence limit includes the range ofthe mean + 3 The greater the confidence interval, the greater the expectation (probability) that a value will fll into the range982 STATISTICS IV THE PHARIACEUTICAL ANALYSIS LABORATORY y a Fraction of peak height Figure 4:3. A normal dlerbuton of roped measuromonts. (From H. M. MeNait and J. M. Miler, Basic Gas Chromatography, Wiey, New York, 1898, p. 27, Reprinted by permission of stn Way & Sons, ne) 45. STUDENT f TEST There is a useful statistical tool for analyzing data which allows us to deal with a limited number of replicate measurements rather than infinite numbers. This tool was elucidated by an Irish chemist, W. Gosset, working at the Guinness Brewery in Dublin in the early 1900s. The statistic is referred to as the Student ¢ ‘or more usually just 1, This tool allows us to determine the uncertainty interval of the mean for @ series of replicates. The calculation may be performed at any Aesired level of confidence. The t test enables calculation of the agreement of a seties of measurements compared to a knolwn standard (accuracy) and allows us to compare the results of different analytical methods, instruments, and ‘operators for the level of agreement. 45 STUDENT Test 93 | | 2s le # wile wede ee | cassoue | 95.48 percent | —89.73 percent Figure 4.4. tlustaton of standard deviations ona normal eetbuton ‘The confidence interval about the mean of a series of measurements is expressed as (4.14) Where m is the number of replicates, and ¢ is read from a table (see Table 4.4) Examination of Table 4.4 shows that values of rare available at different levels, of confidence and for different numbers of replicates 11, 12}. The degrees of freedom are actually the number of replicates minus 1. Five replicates would hhave four degrees of freedom, 454. Applications of t Test 45.1.1. Confidence Limits. The confidence interval (C.1.) or confidence limit (C.L.) was given in equation 4.14, To see how it is used, consider the fol- Towing example. Caleulate the confidence limits for a set of five measurements ofthe moisture level in @ solid sample with the following individual measurements: 11.6%, 10.9%, 12.0%, 11.7%, 11.5%, cL=¥+ (4.15)94 STATISTICS IV THE PHARMACEUTICAL ANALYSIS LABORATORY ta! 4.4, Table oft Values ‘Confidence Level (%) Degrees of Frealom 50 90=«OSSCORSCSCSSC [000 6318 12706 31821 6.657 12732 6.619 1 2 816 2920 4303 6965 9.925 14089 1.598 3 0765 2353 3.182 4541 Sad) 7459. 12924, 4 O78 2132 2776 4747 doe 5598 R10) 5 0727 2015 2571 3368 4022 4.736.869 6 O78 1983 2447 3143 377 4317 s.959 1 OTL 1895 2365 2998 3300 4029 5.408, 8 0706 1860 2306 285 4358 3832 Si 9 0703 1833 2262 2821 3250 3.490 4781 10 0.700 1812 2228 2764 3160 3381 4sa7 13 0.691 1753 DI 2602 2947 3282 407 20g (0687 1725 285 2528 2845 3.153380 25 0.684 1.708 2068 24852787 3078725, 30 0.683 1.697 2082 2457 2780 3010 3.646 40 0.681 L684 2021-2423 274 = DOT 3st Co 0.679 1.671 2000 239 ©2660 ©2915 3.460 120 0677 1.658 198023582617 2.860 3.379, Calculating the mean gives 116+ 1094 12047415 g - =u (4.16) Calculating the standard deviation gives (OTOH s Os Oi? (COT OFT 4 (4.17) ‘The confidence limits may be calculated for any desired confidence by choosing the appropriate r value from Table 4.4. At 90% confidence, ¢ = 2.132 for four degrees of freedom (five replicates). To calculate the confidence interval at 90% confidence, substitute as shown: (2.132)(04) _ Chass = Us £05 (4.18) Expressed another way, the confidence interval at 90% confidence for our data is 11.0-12.0% water in our sample. . For 50% confidence the confidence interval computes a8 11.5 + 0.1. Please note that at lower confidence levels the uncertainty goes down and the precision 446, PROPAGATION OF UNCERTAINTY (ERRORS) 95. is expecied to be tighter. The probability is thetefore decreased for a given ‘measurement to fall into this narrower interval. In most analytical chemisity ‘work we don’t impose specication requirements below 90-954 confidence 4.5.1.2. Comparison of a Sel of Measurements to a Standard. The results of replicate measurements on a standard may be used to assess the accuracy of the measurement. 10 determine how close our average value is to the knovn value. For a set of measurements on a standard bufler having a pH value of 10.5 the following data were obtained in five replicate measurements: 10.7, 10.5, 10.9, 19.6, 108 Do these data agree with the standard value? We apply the # test and impose the requirement that agreement must be at the 95% confidence level. Substituting into our equation for confidence limits wwe obtain the following: . ¥=107 a=02 122.776 on =107 4 275402) 10.7 40.2 (4.19) ‘The uncertainty interval is therefore the pH range of 10.$ to 10.9. The stan- dard pH value of 10.5 is within our uncertainty interval, and we may conclude that our measurements are accurate at 95% confidence. If we had imposcd icter requirements for agreement such as 68% or 50% confidence, our data ‘would not have been consideved accurate 4.6. PROPAGATION OF UNCERTAINTY (ERRORS) Because we have determined methods for calculating the uncertainty interval for data sets, how do we treat these uncertainties in calculations? ‘The rules liffer when dealing with addition or subtraction and multiplication o¢ division, 48.1. Addition and Subtraction of Uncertainties Ifthe uncertainty of a given mean value is known, then the sum and difference are calculated as shown, Calculate the sum of the mean values obtained for three separate samples and consider the uncertainty98 STATISTICS rv THE PHARMACEUTICAL ANALYSIS LABORATORY Sample ‘Mean Value (X}) Uncertainty (U) T 1.76 £0.03 (Ui) 2 £0.02 (Us) 3 £0.02 (Us) sum £004 (U4) Uncertainties are given as absolute and labeled Us, Us, and Us. The means are treated as in normal addition or subtraction, and in this case the sum is simply 1.76 + 1.89 ~ 0.59 = 3.06 The question is the method for calculating the sum of the uncertainties labeled Us, itis not simply the sum of the uncertainties. The following formula should be applied when adding or subtracting uncertainties: y= FUP + (Uae + OH? (420) For the example given we obtain Us = yf(0.03)? + (0.02)? + (0.02)? = 0.04 (421) ‘The final result is therefore expressed as 3.06 + 0.04, 4.6.2. Multiplication or Division of Uncertainties Yo carry out a series of multiplication or divisions, one must first convert the absolute uncertainties to % relative uncertainties. These % values are then treated as in the previous example, but the end result is an uncertainty expressed as % relative uncertainty that must be reconverted to an absolute uncertainty. Consider the following example. Calculate the valve of the following, including the uncertainty of the calculated result: (1.76 + 0203)(1.89 + 0.02) 4, 0.59 + 0.02 (aa) Excling the weertunts the numbers calculate as (1.76)(1.89) _ , oH ag (423) The uncertainty Us is caeuated as Ue = Oat)? + ebay + CEA? (4.24) ‘The individual % uncertainties are then calculated: 4.7, REJECTION OF OUTLIERS 97 fos ry = 00200. 54 ‘ er) ~ Uy = VIBTF TAT TLS = 3.96 = 4.0% (4.26) ‘The final result is 5.6 + 4.0%, which is now converted to an absolute uncer tainty of 5.64 0.2. Note that the final result can have no mote than two siz- nificant figures because 0.59 used in the calculation has only 19. 47, REJECTION OF OUTLIERS efore going into this section it should be cautioned that no data are to be discarded in any analyses unless there are clear-cut standard operating proce- dures to allow this in your laboratory, ‘The following is presented to make you aware that at times we generate data Points in replicate measurements which do not seem to fit in with the other ‘measurements. If this should occur in series of four or more readings from the same sample, then it is possible using a valid statistical technique to cheek if a particular value is an outlier, In an important court decision [3], the United States v. Barr Laboratories, Inc., the presiding judge, Wolin, decided that outlier tests were permitted in microbiological assays but not in chemical analyses. Its not clear how binding this will be on the pharamaceutical industry in general. Proceed as follows, Check all calculations, instrument settings, and calibra tions to ensure that these were not in error. If a source of error cannot be identified, then a suspected outlying data point may be tested by a statistically valid test. An example of such a test is the Dixon @Q test [13]. This is only applied to a series of four or more replicates where a value appears out of line, A term Q is defined as follows: Gap @ Range (2) where gap ~ difference between a questionable value and its nearest value and range ~ difference between the highest and lowest values. The diferences are expressed as positive numbers. Once the Q is caleulated, it is compared to a table of critical Q values determined for a given confidence level. An example of such is Table 4.5 [14] for 95% confidence, Ifthe calculated Q from our experimental data is greater than the critical Q from the table, then the questionable point may be disearded. Consider the following experimental values determined in a laboratory analysis 12.73, 12.36, 12.95, 12.90, 12.85,98 STATISTICS IN THE PHARMACEUTICAL ANALYSIS LABORATORY Table 45. O-Test Table [Number of Data Points (Q. 95% Confidence on oni7 0.2 0370 0324 0492 0464 Deve fom Reference Th Examination of these data might ead us o suspect the 12.36 value as being out of ine Applying the Qe gves us Gap _ 1273-1236 Range 12.95 — 12.36 e ee) @ Examination of the table of critical Q values for five observations shows the critical value of Q as 0.717. Because our calculated Q is not greater than the Q from the table, the questionable point 12.36 may not be rejected. 4.8. LINEAR REGRESSION ANALYSIS In the analytical laboratory we encounter situations where we generate cali- bration or standard curves that are expected to be linear; that is, the points fall, ‘on a straight Tine whose slope and y intercept may be calculated, This linear curve fiting is encountered in all forms of analysis including spectroscopy and chromatography where @ range of concentrations is used to examine the response of an instrument. The data generated can be tested sta- tistically to examine how closely the data points fit a straight-line relationship or for that matter any other mathematical function, In order to check whether a group of data points ft a straight line we make the assumption that the data fit the general linear relationship: meth (429) where y’is predicted from a known value of x (concentration), m is the slope of the line, and d is the y intercept of the line. ‘When a standard curve is developed, known sample concentrations are made ‘up and the instrument response (area counts, absorption) is measured. The poodness of fit of these data points to a straight line may be estimated by de- 49 QUALITY ASSURANCEICONTROL 99 termining a correlation coeficient, r: The closeness of the r value to | confirms that the assumption of linearity is correct. Usually the data are plotted and a visual or computer estimate of the best straight line is drawn through the data, tis, however, possible to numerically calculate the best straight line from the data, This calculated line is called a regression line and is based upon min: ‘mization of the sum of squares of the residuals or deviations of data poin's from the straight line, It is beyond the scope of this book to go into the actual calculations that are usually generated by a computer program associated with the analytical instrument ‘The value of y may range from —1 to +1. A perfect linear correlation would have an r value of +1. For r equal to ~1 a perfect negative correlation would exist and the slope would be negative, For r equal to zero there is no correlation between the instrument response y and the concentration x, Normally, labora: (ory standardizations give r values between 0.9 and 1.0. We usually don't deal with the negative slope and we report 72 values between 0 and 1. Values of r better than 0.99 are readily obiainable in nornial laboratory operations. Its possible to use statistics (o determine the standard deviation and con‘ dence limits associated with the slope and intercept. When a standard curve is used to cafeulate the concentration of an unknown sample, itis also possible to determine the errors associated with the use of the regression line at any level of| confidence. These calculations are usually included in the statistical software programs supplied with analytical instruments and in common spreadsheet programs. ‘At times the response of an instrument as a function of concentration may only be linear over a limited concentration range. Deviation from linerarity is ‘measured by the r or r2. In many cases these values are high, and trending away from linerarity is masked and not apparent. Another technique for assessing deviations from linerarity include calculat- ing the residuals that are the deviations in ¥ value (signal) of each data point away from the linear regression line. For any measured value of 9, the plotted point will have no residuals if the point is on the line. Residuals are charac- terized as + or ~ deviations fom the line, and the numerical sum will add up to m0, ‘The trending of the residuals may be an indication of the deviation from linearity. For a linear plot, the signs of the residuals (+ or ~) should be ran- domly distributed along the regression line. However, if the residuals appear in groups of + or =, deviations from linearity are indicated. Examples are given in Figure 4.5. 49. QUALITY ASSURANCE/CONTROL Any process may be continuously monitored against specific requirements that ensure that the process is producing materia) according 10 expected require ments. The weight of tablets or the volume of a liquid fll are examples of vari-100 STATISTICS IN THE PHARMACEUTICAL ANALYSIS LABORATORY Zero Residuals, —__] Random | Residuals | Negative Residual Deviation Concentration(mg!mg) Figure 4.5, Tending of escuals na near plot, 48, QUALITY ASSURANCE:CONTROL 101 Upper action ine Upper warming ine va ° wl ° Target ele ne 2 ne _ Lower suming line g rr aston ine 86 Time Figure 4.6. 4 typical (Shewhan) contol chart. (Reprinted wth poemssion from J. ©. Miler and JN. Miler, Stasts or Anavtcal Chemisty, vil edn, Prentce-Ha!, Upoor Saddle ve, NU, 1998, gure 42), ables that could be monitored. These parameters, when monitored on a con: tinuous basis, allow for examination of drifting or trending and for corrective action to be taken before the process gives out of specification product. Such data charts are called control charts and may take on different formats. ‘A convenient control chart model (Shewhart) is shown in Figure 4.6. The value of a particular attribute is plotted on the y axis as a function of time. ‘Typically we establish a target value for a process; for example, the fill volume ‘may be targeted at 10 mL. The actual value may vary for each bottle fill due to random error and perhaps errors inthe fling operation, The control chart is set Up so that the average fil volume is monitored with time over many runs. We expect the measured fill volume (sampled by QA) to cluster about the target value (which we call 4) and have a standard deviation of 2. According to our previous discussion, itis expected that the 95% uncertainty range will be rep- resented by ro (430) ‘The 99.7% uncertainty range will be (4.31)102 STATISTICS IN THE PHARMACEUTICAL ANALYSIS LABORATORY ‘The lines on the chart represent these uncertainty limits. In praetice the fill volume would be monitored to cusure that the value hovers around the target value within its standard deviation ‘The 95% control line may be designated as an alarm line indicating that data outside this limit indicate that the process may be drifting out of control. The 99.7% uncertainty limit is usually considered an action limit that requires cxamination of the process and that an appropriate adjustment be made, ‘There are other forms of control charts, and one can use any uncertainty level for alarm and action limits. In some cases it might be necessary to estab= lish an action limit at 95% confidence, The control line may be calculated at any level of confidence according to the methods described in this chapter using the 1 test. These control charts provide a visual record of the important attributes. A continuous drift upward ‘or downward might indicate the influence of systematic error. Machinery parts could be wearing, jncoming materials may have different levels of impurities, and so on. 4.10. CONCLUSION ‘This chapter has dealt with basic concepts relating to the treatment of numbers and data relating to working in an analytical laboratory. Many of the topics hhave not been treated in depth; should you desire a fuller treatment, you may consult some of the references listed below particularly reference 8. Remember that there is no substitute for doing accurate and precise work. Statistics may Lell you how you are doing but will not carreet for sloppy laboratory practives REFERENCES 1. United States Pharmacopeia Convention, United Stotes Pharnacopeta 24iNF 19, Rockaille, MD, 2000, 2. RG, Whitheld, D. W. Hughes, T. P. Layofl, R. R. Cox, G. E. Gressett, P. J Jimener, J. Andersen, R. R. Peck, and S. Schneipp, Interpretation and Treatment of Analytical Data, Phormacopeial Foruys 1988, 24, 7051. 3. United States w. Barr Laboratories, ne, 812 Federal Supp. 458(1), NJ, 1993. 4. Z. Depl leitor), Quality Control in Pharmaceutieat Analysis: Separation Methods Elsevier, New York, 1997 5. Christopher M. Roley and Thomas W. Rosanshe (editor), Development and Vale dation of Analytical Methods Progress in Pharmaceutical and Biomedical Analysis, ‘Vol. 3, Pergamon Pres, Elmsford, NY, 1996 6. ASTM, Vol, 14.02 7. ASTM, Vol. 14.02. « 8.1.C Miller and J. N, Milles, Statstes for Analyrea! Chemistry, 3d edition, Elis Horwood PTR Prentice-Hall, Englewood Cliffs, NJ, 1993, REFERENCES — 103, 9. F, Mosteller, S. E, Flenberg. R. E. K. Rourke, Beginning Staisttes with Data Analysis, Addison-Wesley, Reading, MA, 1983 10, J.T. McClave and FH. Dietrich H, A First Course in Statistics, 4th edition, Macmillan, New York, 1992. 11, W. J, Youden, Experimentation and Measurement, NIST Special Publication 672, 1991, 12, Stondard Guide for Staistical Procedures to Use in Developing and Applying ASTM Test Methods, ASTM E 1488.92, Vol. 1402. 13, Standard Practice for Dealing with Outlying Obsercations, ASTM E 178.94, Vol. 14.02, 14, P. King. J. Am, Stat, Asse. 1958, 48, 31472 GLOSSARY OF TERNS USED IN ICH DOCUMENTS chemical entity defined as the drug substance or an excipient in the drug product. Impurity profile: A description of the identified and unidentified impurities resent in a drug substance or product, New drug substance: The designated therapeutic moiety which has not been previously registered in a region or a member state. It may be a complex, ester, or salt of a previously approved drug (also called a new molecular entity or new chemical entity). Potential impurity or degradation product: An impurity which, from theoretical considerations may arise during or after manufacture or storage of the drug product. It may or may not actually appear in the drug substance or drug product. ‘Qualification: The process of acquiring and evaluating data which establish the biological safety of an individual impurity or a given impurity profile at the level(s) specified Reaction produet: Product arising from the reaction of drug substance with an ‘excipient in the drug product or immediate container/closure system. Safety information: The body of information that establishes the biological safety of an individual impurity or a given impurity profile at the level(s) specified. Specified degradation product: Identified or unidentified degradation product that is selected for inclusion in the new drug product specification and is individually listed and limited in order to ensue the safety and quality of the new drug product. Toxic impurity: An impurity having significant undesirable biological activity. lentified degradation product: An impurity that is defined solely by quali- tative analytical properties such as chromatographic retention time. ‘Unspecified degradation product: A degradation product that is not recurring from batch to batch. APPENDIX | | UNIVERSAL TESTS, DOSAGE-FORM-SPECIFIC TESTS, AND ACCEPTANCE CRITERIA Universal Tests and Acceptance Criteria Area of Application Tesi/Critera New drug substances 1. Description: qualitative statement about sate and colon 2. Identification: specific and disriminate between ‘compounds of closely related structure whieh are likely to be present; include tests for ions ia salt; enantiomers, 3. Assay: specific and stability indicating: may use same method for assay and impurities (ff nonspecific assy method, include suitable impurity tes), 4, Impurities: include organic, inorganic, and residual solvents 5. Physicochemical propertis (pH, melting pont, reftactve index, et); particle size and where applicable ize distribution 6, For solid-state forms, include polymorphs or solvates where bioavailability, stability or dissolution could be affected 7 Chiral impurities shouldbe treated according tothe principles in ICH impurity guidelines (enantioeeective assay and specific idemiy ts), 8, Water content is important i drug substance is hygroscopic, degraded by moisture oF a stoichiometric hydra, 9, Microbial limits may include total aerobic counts and absence of objectionable organisms, Compendial methods, sterility, and endotoxin testing should be used 18 appropriate. Now drug products 1. Description: qualitative statement of unit of use (9. size, shape, and colo).474 UNIVERSAL TESTS, DOSAGE FORM-SPECIFIC TESTS, AND ACCEPTANCE GAITERIA UNIVERSAL TESTS, DOSAGE FORMSPEGIFG TESTS, AND ACCEPTANCE CrIrERA 475 ‘Area of Application ‘Tesy Criteria Dosage Form “Tests and Criteria New drug products 2. Identification: establish identity oF a new drag substance (Oral liquid products “4 Preservative effectiveness (Contnued) ina new drug product by a specif technique (Continued) + Base specifications on level needed to mainain 3. Assay: specie and sability-adicating; may use same method for asay and impurities (if nonspecific assay method, include suitable impurity tex), 4, Impurities: inelude organio, inorganic, and rescal solvents, particularly degradation products; state limits {or individual identified, unidemtied, and total impurities; reduce or eliminate testing when drug substance does not degrade in given formulation, Dosage-Form-Specile: Tests and Acceptance Criteria Dosage Form ‘Tests and Criteria Solid oral products 1. Dissolution/disintegration + Disintegration my be appropriate for rapidly Aissolving products + Single-point measurements normally suitable for immediate release products + Use multiple time points for extended release products + Tworstaze testing may be appropriate for delayed release products 2, Hardness/iailty + Normally an in-process control not included in specifications 3. Uniformity of dosage units + Includes uniformity of content and mass + Use compencial methods + May be performed in-process + Included in regulatory specications 4, Water content + Method specific for water prefered 5. Microbial limits + Test drug product unless components tested and ‘manufacturing proces carries no significant risk of contamination Oral liquid products 1. Uniformity of dosage units + May apply to single and muliple dose packages + Weight variation or fl may sulice * Uniformity of mass testing generally acceptable for dry powders for reconstitution Parenteral products 2. pl * Provide test afd specification 3. Microbial limits + As with solid orals product quality + Support with preservative efficacy testing + In-process testing may sufi 4. Antioxidant preservative effectiveness * Release testing normally performed + May be necessary on stability + In-process testing may sufice + Include in specifications 5. Extracables Testing generally not required if development and stability data show no sigificam extractables 6. Alcobol Content * May berassyed or calculated where declared ‘quantitatively on the label 7, Dissolution * May be appropriate for oral suspensions an dry powder for resuspension + Use compendia! procedures if possible 8, Particle size distribution + May be tequized for oral suspensions + Describe particle size distribution + May replace dissolution test + Different release and shelf ie limits may be required + Investigate potential for particle growth 9, Redispersbility + May be appropriate for suspensions which settle on ing to resuspend 10, Rheological propeties + Speclyacceplable viconity rage where appropriate + May conduct skip fot testing or eliminate test based on developmental dae 11, Specie easily + Establish lest method and specification as approptate 12, Reconstitution ime + Should be provide for dry powders which requte 1. Unity of dosage units + As with oral quis 2 pH As with oral liquids476 UNIVERSAL TESTS, DOSAGE-FORM-SPECIFIC TESTS, AND ACCEPTANCE CRITERIA Dosage Form “Tests and Criteria Parenteral products & Sterility (Continued) + Should have speciation + Compendial procedures preferred + Parametric release may be proposed when justified 4. Endotoxins/pyrogens + Perform endotoxins testing when required + Pyrogen testing may be proposed as alternative to endotoxin waen justified 5. Particulate matter * Should have appropriate specification, which may include limits for visible particulates andjor clarity of solution 6, Antimicrobial preservative effectiveness = As with oal liquids 7. Antionidant preservative effectiveness + As with oral liquids 8, Extracables + As with oral liquids 9, Punctionalty testing + Por items such as prefilled syringes, autoinjctor sartidges, ete + May include pressure, eal inearty, ipfeap removal fore, etc, a8 appropriate 10, Osmoiatty * Control when tonicity of product is delated on labeling * Development data may justify in-process, or skip- lot testing or ditet calelation 11. Particle size distribution + May be appropriate for injectable suspensions + Development data should be used to select between dissolution and particle size distebution + Investigate the potential for particle growth vpon storage 12, Redispersability As with ora iquids 13, Reconstitution time + As with oral liquids “Goneepis described shove should aso be considered for inhalation dosage forms (acoso, sl ions, sprays), opie ormultion erm, einmsits, gls), and transdermel systems, appenoixl V USP CHROMATOGRAPHIC PHASES Stationary Phases for GC USP Designation ~~ (Commerical Equivalent GI Dimethylpolysitoxane oil ov-ior G2__Dimethylpolysitoxane gum ov-l G3 50% Pheny!-30% methylpolysloxane ov.i7 G4 _Diethylene glyco! sucinate polyester DEGS: GS Cynnopropsipalyilonane ov-los G6 Triluoropropylmethyipolysloxane ov-202, ov-210 G1 50% 3-Cyanopropy-S0%e phenylmethylsilcone OV-225. DB.225 G8 90% 3-Cyanopropy!-10% phenylmethylsiicone OV-2330, DB-23 69 MethyWvinylpolysloxane SES2 G10 Polyamide Polyamide GI Bis2-thythexyi}scbacate polyester — G12 _Phenyldiethanolamsine succinate polyester G13 Sormito! GI4 Polyethylene glycol (MW 950-1050) Carbowax 1000 GIS Polyethylene glyeol (MW 3700) Carbowax 4000 G16 Polyethylene lyeo! compound (MW 15000) Carbowax 20M. 75% Pheny1-25% methylpolysiloxane ovas Polyalkplene glycol = 25% Phenyl-25% cyanopropylmethyliicone — — Polyethylene glycol (MW 380-420) Carbowax 400 ‘Neopentyl glycol sucinate as Bis(2-ethyThexy) phthalate por Polyethylene glyco! adipate EGA Diisodecy! phthalate ss G25 Polyethylene glycol compound TPA ‘Carbowax 20M-TPA G26 28% 2-Cyanoethyl-78% methylpolyslloxane — — G27 _ 5% Phenyl-95% methylpolysiloxane = G28_ 25% Pheny!-75%% methylpolysiloxane = '-Thiodipropionitle G30 Tetraethylene glyeol dimethy! ether G3I_ Nonylphenoxypoty(ethvleneoxy}ethanot Nonoxynol 30, an