Medicines Compendium

General Information / 10 Assessing Validation Parameters 1

PPAC Resolution: NLT 1.5 for each impurity

10 Assessing Validation Parameters for Reference and Acceptable ProceduresGuideline for Donors/Instructions for Staff
PURPOSE
This document (Document 5) provides information to determine whether Reference Procedures or Acceptable Procedures for the Assay and Impurity tests in the USP Medicines Compendium (MC) are acceptable. The approach is based on a defined set of Procedure Performance Measures (PPMs) and associated Procedure Performance Acceptance Criteria (PPACs) and experiments. Further information is provided in the Appendix. While the concepts in the Guideline are general, it is directed to chromatographic methods for chemical ingredients and products. Adjustments may be needed for non-separation methods and ingredients/ products.

Linearity and Range

Samples: NLT 6 independent replicate Standard solutions over the range the procedure is intended to be used, but NLT at 80%, 90%, 100%, 110%, and 120% Analysis: Analyze the results of each data series of solutions as described in Precision and Accuracy. PPAC: NLT 0.95 for each concentration point

ORGANIC IMPURITY LIMIT PROCEDURE Precision (Repeatability)

Test samples: Six independent samples of the material under test, spiked with reference materials for the specified impurities at the indicated limit PPAC Relative standard deviation: NMT 2C0.1505%. C is the concentration fraction of the impurity [and is calculated by dividing the impurity concentration by the major component concentration. A spiked impurity present at 0.1% of the major peak would have a concentration fraction, C, of (0.1%/100%) = 103, which would lead to an acceptable RSD of 5.7%. Therefore, an alternative impurity procedure would present acceptable precision if it yielded a repeatability of 5.7%]. [NOTEThe Horwitz equation is based on the comparison of a procedures reproducibility as compared to the concentration factor. The repeatability has been reported to be 1/2 of that of the reproducibility. Therefore, the equation based on the repeatability analysis would lead to the equation: Acceptable % RSD = C0.1505. However, for the evaluation of a limit test, the broader reproducibility based equation is used because the precision is less important in this type of test.]

ASSAY Precision and Accuracy

Samples: Six replicate Standard solutions Analysis: Determine the Precision value and Accuracy value. Precision value: Determine the % RSD of the Standard solutions. Accuracy value: Mean of the percentage content of the Standard solutions corrected for the stated content of the Reference material. Analyze results using MS Excel: Result = NORMDIST(Upper Cert, Mean, SD, TRUE) NORMDIST(Lower + Cert, Mean, SD, TRUE)

Limit of Detection
Control sample: A preparation of reference materials for the impurity(ies) of interest at the target concentration(s) Test sample 1: A sample of material under test, spiked with reference materials for the specified impurities at the target concentration, prepared in triplicate Test sample 2: A sample of material under test, spiked with reference materials for the specified impurities 100 (2C0.1505)% of the target concentration for the specified impurity(ies), prepared in triplicate PPAC: Each Test sample 1 provides a peak response equivalent to or greater than that of the Control sample. Test sample 2 must provide a peak response that is less than that of the Control sample. [NOTEThe signal from each sample obtained must show a change from the value obtained compared to a blank determination.] Specificity: The procedure must be able to unequivocally assess, with acceptable Accuracy and Precision (see previous sections), each specified impurity in the presence of components that may be expected to be present, including other specified impurities, the parent compound counter-ions, solvent fronts, and matrix components. Where specified impurities are separated by a resolution of greater than 1.5, then the procedure has acceptable specificity Where specified impurities are not well-resolved, then a spike and recovery study of the closely eluting peaks is completed. If the specified impurity of interest meets

Upper Cert Mean SD TRUE Lower

= upper limit of the Acceptance Range (e.g., 102% in the default case) = measurement uncertainty of the reference standard used (= 0 unless CRM or SRM used) = Accuracy Value = Precision Value = logical operator = lower limit of the Acceptance Range (e.g., 98% in the default case)

PPAC: NLT 0.95

Specificity
Spiked standard solution(s): Contains the measurand at the same concentration as the Standard solution and each specified impurity at 10x the specified value and unspecified impurities at 1.0%. [NOTEExclude all impurities that contribute an acceptable level of assignable bias.] Samples: Standard solution and Spiked standard solution(s) Analysis: Analyze results, and determine the Resolution of each impurity from the primary peak.

2 10 Assessing Validation Parameters / General Information

the previous Precision requirements while in the presence of interfering peak, then the procedure is acceptable.

Medicines Compendium

APPENDIX: THEORETICAL CONSIDERATIONS Introduction

The principles of validation are provided in United States PharmacopeiaNational Formulary (USPNF) general chapter Validation of Compendial Procedures 1225. Principles of equivalence are presented in the following publication: Acceptable, Equivalent, or Better: Approaches for Alternatives to Official Compendial Procedures. One form of equivalence involves demonstrating that a procedure meets pre-set criteria that indicate acceptability of either a Reference Procedure or Acceptable Procedure in the USP Medicines Compendium (MC). The pre-set criteria consist of the Procedure Performance Measures (PPM) and the Procedure Performance Acceptance Criteria (PPAC) for a specific application. Often the PPM and the PPAC are presented together in a monograph or method-specific general chapter (see Elemental ImpuritiesProcedures 233). The PPAC may also be found in method-specific general chapters (such as Chromatography 621). This Appendix provides a general discussion of PPMs in determining whether Reference Procedures or Acceptable Procedures meet predetermined criteria for acceptable PPACs.

Organic Impurity Quantitative Procedures

ACCURACY Standard solutions: Prepare solutions of the specified impurities at concentrations ranging from 50% to 150% of the limit value for each impurity, using appropriate reference materials or other reliable impurity source. Sample solutions: Prepare solutions of the material under test spiked with appropriate reference materials at concentrations ranging from 50% to 150% of the indicated limit value for each specified impurity. PPAC Spike recovery: = C0.1505% for the mean of three replicate preparations at each concentration. [NOTEThe data obtained for this validation parameter are used for Limit of Quantitation (LOQ), Range, and Linearity in the following section.] PRECISION Repeatability Test samples: Six independent samples of material under test, spiked with appropriate reference materials for the specified impurities at the indicated levels PPAC Relative standard deviation: NMT C0.1505%; where C is the concentration fraction of the impurity RUGGEDNESS The effect of random events on the analytical precision of the method must be established. Acceptable experiments for establishing intermediate precision include performing the Repeatability analysis On different days With different instrumentation, or With different analysts. Note that executing only one of the three experiments listed is required in order to demonstrate intermediate precision. PPAC Relative standard deviation: NMT 2C0.1505%; where C is the concentration fraction of the impurity SPECIFICITY The procedure must be able to unequivocally assess each specified impurity in the presence of components that may be expected to be present, including other specified impurities, the parent compound counter-ions, solvent fronts, and matrix components. Where specified impurities are separated by a resolution of greater than 1.5, then the procedure has acceptable specificity. Where specified impurities are not well-resolved, then a spike and recovery study of the closely eluting peaks is completed. If the specified impurity of interest meets the previous Precision requirements while in the presence of interfering peak, then the procedure is acceptable.

Procedure Performance Measures

General chapter Validation of Compendial Procedures 1225 defines the critical validation parameters (CVP) necessary to validate a procedure. The CVPs can also be considered as the PPMs for the determination of the acceptability of procedures. The methods (chromatographic, spectroscopic, gravimetric, etc) used in a monographs test procedures each have terms, measures, and assumptions that are used to describe critical information about acceptability of the procedure itself. These method-specific terms and measures may not align well with the PPMs necessary to define an acceptable procedure. Additional approaches are thus needed to determine acceptability of a monographs procedure. These approaches are termed PPMs.

Chromatographic Assay
The validation parameters for a chromatographic assay are defined as; accuracy, precision, specificity, linearity, and range. There are thousands of experts practicing chromatography throughout the pharmaceutical industry and these individuals have all come to a general, unwritten agreement as to what an adequate procedure would entail. Most of these individuals would describe the resolution, peak shape, co-elution, baseline drift, dead volume and other chromatographically relevant terms. However, all of these terms are specific measures that describe the features of a separation, but do not parallel the CVPs defined above. For example, the specificity of a procedure can be described in part by the resolution of critical pairs of peaks, and partially by the likelihood of co-elution and also by the number and height of theoretical plates, and several others. Each of these measures help an analyst describe the goodness of their procedure. However, these measures cannot adequately describe an acceptable procedure in isolation. What is needed is an approach that allows an analyst to quickly and decisively determine that their procedure is acceptable. The acceptable values have been included in general chapter Chromatography 621 using the rationale that follows. PRECISION AND ACCURACY

Limit of Quantitation, Range, and Linearity

Demonstrated by meeting the Accuracy requirement.

The precision of a procedure affects the acceptability of all of the CVPs. As the precision of a procedure increases, any

Medicines Compendium

General Information / 10 Assessing Validation Parameters 3

Figure 1. Bias%CV Tradeoff, 98%102% Limits, True Value = 100, Probability Passing 0.95. bias inherent in the system becomes more apparent, the linearity and range become better defined, the ability to determine an error caused by a lack of specificity becomes more apparent, and the likelihood of a material meeting the Assay acceptance criteria receiving a passing result is improved. Of these relationships, Precision and Accuracy are the most closely linked. Accuracy: Accuracy of a liquid chromatographic procedure is linked to the accuracy of the detector, detector sampling rate, and the consistency and quality of the separation. Because chromatography is a relative technique, the results of an unknown must be compared to those from a known standard to obtain a quantitative result. The true value or concentration of the known (Standard) solution must be determined using an orthogonal procedure or through the use of a standardized material. To empirically determine the accuracy of a CVP, multiple (6) weighings of a standard material would be used to produce independent solutions having the same concentration. Each of the solutions would be chromatographed, and corrected for weighing differences. The difference between the average of the calculated content of Standard solutions (obtained by using the calibration curve developed during the validation of the procedure) and the known content of the Standard solution would represent the procedure bias (accuracy). Precision: Like accuracy, the precision of a quantitative chromatographic procedure is directly linked to the precision of the detector used. The precision of a measurement is also dependent upon the quality of the separation. Precision can be determined empirically using the data obtained for the Accuracy evaluation. The data obtained and corrected are then used to calculate the relative standard deviation or %CV. RELATIONSHIP BETWEEN ACCURACY AND PRECISION When making decisions about the best method for a particular analysis, there are a number of factors aside from the CVPs. These factors include: length of analysis, cost of solvent use and disposal, column life, solvent-column compatibility, gradient lengths, equilibration times, pump capacity, etc. Therefore, it is often necessary to weigh Precision and Accuracy requirements with these outside concerns. Obviously a method with 0 RSD and bias would be the best available technique, but where it requires solvents and column to be maintained at 50 C or that requires 2 hours to equilibrate between each analysis, it is unlikely to be selected. Therefore, the definition of an acceptable procedure must provide flexibility to the user, while clearly defining the boundary condition between acceptable and unacceptable chromatography. This boundary is defined statistically as a simple normal distribution that describes the likelihood (at a 95% confidence level) of a passing value to demonstrate compliance to an Assay acceptance criterion of 98%102% (see Figure 1). For the 98%120% case, a user could simply plot the values of Bias and %CV. If the point described by a procedures accuracy and precision falls in the shaded area of Figure 1, then the procedure will (with a 95% probability) provide a passing value when the true value passes. Because this is a statistical evaluation and not a chemical determination, the results are conclusive. Although the Assay acceptance criteria of 98%102% is the most prevalent in USPNF, there are many other criteria in use. For the CVP approach to determining the acceptability of a procedures accuracy and precision to be applied to cases other than the default, it is necessary to calculate a new boundary line for each case. The development of a new boundary line is not difficult, but is time consuming. However, there is another approach. Because the boundary line is based on the statistical evaluation of a normal distribution, it is possible to solve for the probability of the precision and accuracy values providing the correct answer. When the calculated probability is less than 0.95, then the precision and accuracy are acceptable. One way to speed the calculation is to use a program such as Microsoft Excel. When using Excel, the probability would be calculated using the following equations: 1. Where the acceptance criteria are symmetric around 100%, then calculate: Result = NORMDIST(Upper Cert, Mean, SD, TRUE) NORMDIST(Lower + Cert, Mean, SD, TRUE)

Upper Cert

Mean SD TRUE Lower

= upper limit of the Acceptance Range = measurement uncertainty of the Standard used (= 0 for Standards with unknown measurement uncertainty) = accuracy value (e.g., 100% in the default case) = standard deviation = logical operator = lower limit of the Acceptance Range

4 10 Assessing Validation Parameters / General Information

Medicines Compendium

Figure 2. Guard Band for a CRM Labeled as 100% 0.06% Applied to Figure 1. 2. Where the acceptance criteria are asymmetric around 100%, but the entirety of the range is between 97% and 103% (includes about 85%* of monographs in USPNF) then: The same equation is used, but the Lower and/or Upper value would be changed to describe the smallest symmetric range having a mean of 100%. 3. Where the acceptance criteria are asymmetric mean other than 100%, and the value of 100% is not included in the range, then calculate: Result = NORMDIST(Upper Cert, Mean, SD, TRUE) NORMDIST(Lower + Cert, Mean, SD, TRUE) sion will need to be adjusted. This adjustment takes the form of a guard band in the distribution. The guard band effectively tightens the acceptance criteria by the stated uncertainty of the standard. An example of a guard band for a certified reference material labeled as 100% 0.06% is illustrated in Figure 2. This offset is accomplished by a small change in the equation used to calculate acceptance probability. The change is accomplished by adjusting the boundary conditions of the acceptance criteria using the measurement uncertainty (Cert) in cases 13 in the previous subsection. SPECIFICITY The ability to describe an adequate separation of a critical analyte from interferences is currently measured using several different approaches including: resolution, separation factors, capacity factors, peak shape (tailing and fronting), height and number of theoretical plates, retention times, and several others. All of these separation measures assume that the detection system being employed is capable of measuring the analyte and other materials likely to be present. Resolution is the most effective measure of a separation. Generally speaking, a resolution value of 2 would be described as being baseline separated and a resolution of 0 would indicate no separation. Complete (baseline) separation is preferred, but may represent much longer chromatographic run times, difficult chromatographic conditions, or may be unattainable. Therefore, it is necessary to determine what resolution is necessary to provide an adequate separation of two close eluting peaks. One way to quantify the effect of resolution upon a chromatographic procedure is by the evaluation of the error caused by a co-eluting or overlapping peak. SPECIFICITY ERROR The signal (peak height or peak area) used to measure chromatographic response is susceptible to error due to peak height or area shifts caused by co-elution of analytes. Most detectors do not allow the differentiation of chemical entities present in a measured peak. The exceptions are multi-dimensional detectors with deconvolution algorithms. Where peaks overlap, there are many different ways to estimate the approximate individual component concentrations and each has a related error.

Upper Cert

Mean SD TRUE Lower

= upper limit of the Acceptance Range = measurement uncertainty of the Standard used (= 0 for Standards with unknown measurement uncertainty) = target content = (Upper + Lower)/2 = standard deviation = logical operator = lower limit of the Acceptance Range

4. 4.Where the range is larger than 97%103% and is not centered on 100%, the determination of the intended mean and range need to be considered prior to calculating the acceptable value. The details of these very rare cases are still being considered. The problem of intent can be illustrated using an example acceptance criteria of 85%103%. This criteria could be considered in two different ways: A symmetric range around 94% or An asymmetric range around 100% Assuming an SD of 4, case 1 would yield a probability of 0.97 and would be considered acceptable, while case 2 would yield a probability of 0.54 which would not be acceptable. Further discussion on this topic is being considered. GUARD BANDS When a reference standard is used that includes a measurement uncertainty value, such as those found in CRMs or SRMs, the results of the determination of Accuracy and Preci-

Medicines Compendium

General Information / 10 Assessing Validation Parameters 5

Figure 3. Relationship Between Error and Resolution. From a purely theoretical perspective, it is possible to calculate the percentage error caused by overlapping peaks by evaluating the effect of secondary Gaussian peak with differing sizes on a primary peak. The error is directly related to the size of the secondary peak, the width of the peaks, and the distance between their peak values (retention time). Because peak width and retention times are the critical variables in calculation of resolution, then resolution and error are also directly linked. That relationship is described in Figure 3. The values in Figure 3 are the area counts obtained by the traditional extending the sides of the peak to baseline approach described in general chapter Chromatography 621 with and without overlapping peaks. The size of the minor overlapping peaks range from 0.1% to 5% of the primary peak. The generally accepted rule of thumb for chromatographers is that a resolution of 1.25 is sufficient to adequately quantify overlapping peaks but that 1.5 is preferred. Based on Figure 3, the rule of thumb appears to be correct. A resolution of 1.25 contributes significantly more error to a determination than a resolution of 1.5, but with more sophisticated analytics (i.e., better peak area approximations) it is possible to reduce that error further. It is also clear that below a certain level the presence of a small impurity will contribute an insignificant error in the quantification of the primary component. The size of an insignificant impurity is directly linked to the Precision and Accuracy evaluation discussed in the previous section. Please note that the presence of impurities in a drug substance is not desirable, but for the purposes of assaying the content of the active ingredient, small impurities or undetectable impurities may not need consideration. The determination of these impurities will be considered separately. ASSIGNABLE BIAS 1. If an impurity that co-elutes with the principal peak is considered a component of the Bias (assignable bias) in the analysis, then the likelihood of the presence of that impurity impeding the quantification of the primary component can be calculated. The assignable bias is the known concentration of the impurity presented as a percentage of the total primary component concentration. The calculation is the same as given previously, but with the assignable bias added to the accuracy value. Result = NORMDIST(Upper Cert, Mean, SD, TRUE) NORMDIST(Lower + Cert, Mean, SD, TRUE)

Upper Cert

Mean SD TRUE Lower

= upper limit of the Acceptance Range = measurement uncertainty of the Standard used (= 0 for Standards with unknown measurement uncertainty) = accuracy value + assignable bias (%) = standard deviation = logical operator = lower limit of the Acceptance Range

2. If the Result is less than 0.95, then the impurity is inconsequential to the Assay. Where the Result is less than 0.95, then additional separation is necessary to assure the ability of the procedure to assay the primary component. Figure 3 also provides insight into both the source of variability in the analysis and the level of separation necessary to achieve an acceptable procedure. If an analyst knows the

6 10 Assessing Validation Parameters / General Information

Medicines Compendium

Figure 4. Effect of Random Error on CVP Linearity. level of error (accuracy value + assignable bias) it would be possible to approximate the necessary resolution using Figure 2. However, due to the significant convergence of the curves, it is safe to say that a resolution of 1.5 will provide sufficient separation for all but very large impurities. LINEARITY AND RANGE The linearity and range of a liquid chromatographic procedure are largely controlled by the detector sensitivity and by column capacity and retention time. Both the linearity and range can be empirically determined by evaluating Standard solutions ranging in concentration from 80% to 120% of the target concentration of the analyte. There should be five separate concentrations with each of the solutions being prepared in triplicate. The results are then evaluated for adequate precision and accuracy as well as the function of a linear regression of the calculated concentration versus known concentration plot. The measurement of the linearity and range are coupled for this discussion because, like precision and accuracy, the experiments and outcomes are directly linked. Up to this point, all of the CVPs have been focused on the ability of the procedure to determine the true value of a sample when present at a target concentration. However, not all samples will be present at 100.0%. Therefore, it is necessary to ensure that a given procedure will yield acceptable results over a range of likely concentrations. This range is described as 80%120% of the target concentration in the previous CVP section. Range: The acceptability of a range is evaluated by calculating the precision and accuracy values for each of the prepared solution concentrations described previously. Although these solutions are prepared in triplicate and the acceptance values calculated in the Precision and Accuracy section are based on 6 measurements, the calculations provide an adequate level of confidence in the acceptability of the measurements. If all of the five solution triplicates yield a value of 0.95 or greater, then the range of the instrument is adequate for the measurement. Linearity: Typically linearity is evaluated by comparing the signal to the concentration. However, the slope and intercept of the linear regression of that line is of little consequence as long as it is not too vertical or too horizontal. In addition, there are a number of detection systems in development and use that provide unique capabilities but require non-linear relationships between signal and concentration that can be modeled. Therefore, the best manner to assess the linearity of a system is not to evaluate the signal-response curve, but instead to evaluate the calculated concentration-known concentration curve. This approach provides a means to assess the entire system including sample preparation, injection, separation, detection, and post-processing. The calculated versus know concentration curve should be very linear, and pass through the origin. However, as random error is introduced into the system, there is a direct effect on the R2 value of the linear regression. This effect is illustrated in Figure 4. This plot represents the results of a linear regression on data that ranged from 80% to 120% at known values with a random noise of a known percentage added. As the amount of noise is increased, the variability of the R2 value increases and the centers of the ranges decrease. Viewed another way, the error captured as the % RSD of the data when compared to the square of the residuals indicates a strong relationship (see Figure 5). This plot suggests that a R2 value of greater than 0.994 and 0.990 would describe an acceptable procedure for 98%102% and 97%103% acceptance ranges, respectively. In summary, it is clear that the precision and accuracy of an analytical procedure are pivotal to defining the acceptability of a procedure. The specificity, linearity, and range are also necessary that ensure the results of a procedure having acceptable precision and accuracy will provide a meaningful approximation of the true value of an unknown sample. When taken together, it is possible to describe an acceptable procedure using measures that are independent of chromatographic figures of merit typically used to describe a procedure.

Chromatographic Analysis of Impurities

Organic impurities in drug substances are chemical species that should not be present. Their presence may pose a health risk or may simply indicate a lack of quality. The acceptance criteria for individual impurities are evaluated by the regulatory agencies prior to the acceptance of a drug product for use in humans. Impurities are evaluated for safety and those that show toxicity or have the potential to be a health risk are termed toxic impurities and are specifically limited to low levels by the regulatory agency. Those organic impurities that are not toxic at expected levels represent quality issues alone. International standard setting groups have determined that non-toxic impurities present at less than 0.10% need not be quantified and those of less than 0.05% may be ignored. Individual regulatory agencies may lower these limits to better represent the drug substance being approved, but for a public standard, the levels selected by the international body form an acceptable impu-

Medicines Compendium

General Information / 10 Assessing Validation Parameters 7

Figure 5. Relationship Between % RSD and R2. rity level. The acceptability of a procedure to evaluate nontoxic impurities in a drug substance is directly linked to the CVPs defined by general chapter Validation of Compendial Procedures 1225. The validation chapter subdivides impurity procedures into those used in a semi-quantitative limit test and those intended to be quantitative. The requirements for each are significantly different; however, the requirements described for a limit test are insufficient for the determination of acceptable limit tests. The means to demonstrate the acceptable limit test are included in the following Chromatographic Limit Tests subsection. Fortunately, chromatography is typically used in a quantitative mode, and the validation parameters described in the validation chapter are a good indicator of those necessary to determine an acceptable quantitative impurity procedure. The parameters and the performance characteristics necessary to demonstrate acceptability are also included in the following subsection. CHROMATOGRAPHIC LIMIT TESTS The CVP necessary to validate a limit test procedure include precision, specificity, and limit of detection (LOD). The PPM for LOD also provides information about that acceptability of the accuracy of the procedure for a limit test. The CVP for a limit test must answer the following questions: (1) Will this procedure consistently provide the correct pass/fail decision for the specific impurity being evaluated? and (2) What is the smallest change in concentration between a passing and a failing result? The answer to the first question requires knowledge of the precision and specificity. The answer to the second question requires knowledge of the LOD and accuracy of the measurement. However, the accuracy of a limit test and the LOD can be determined using the same PPM and therefore is only described as the LOD for the purposes of this discussion. Precision: Unlike the analysis of an analyte that is a major constituent in a drug substance, the evaluation of minor components of indeterminate concentration creates complications for determining the acceptability of a procedure. The major complication concerns the ability of the detection systems to differentiate a measurand from background noise. The ability to differentiate is determined through measurement of the LOD, but the conclusion that a procedure with sufficient LOD has the differentiating power capable of presenting a true result is unfounded. Instead, it is important to examine the precision of the measurement. The relationship between the analytical concentration and the acceptable level of variability was extensively explored by William Horowitz in the 1980s. He found that as the content of the measurand decreases, the amount of variability from measurement to measurement increases, regardless of the procedure used. Through the examination of a large body of data, he developed an empirical relationship for laboratory-to-laboratory reproducibility. The most common form of that relationship is: Predicted % RSD = 2
(1 0.5 log C)

where C is the concentration as a dimensionless fraction (see Table 1). When this equation is extended to concentrations of 100 ppb (100 ng/g) or lower, the relationship becomes constant. This observation was published by Thompson, et.al. , leading to the Horowitz-Thompson equation used by the Association of Official Analytical Chemists (AOAC) and by the Codex Alimentarius to describe acceptable procedures. This relationship is typically called the Horowitz ratio or HORRAT in these publications. Subsequently, Massart showed that the Horowitz ratio appears to be approximately twice that found in the evaluation of repeatability. The Horowitz relationship is plotted in Figure 6 for reproducibility and repeatability. The recommended procedure for specificity is the evaluation of replicates of spiked standards. This is considered a repeatability study, but for the purpose of defining the PPAC for a limit test, the use of the Horowitz-Thompson Reproducibility factor is more appropriate as it allows a larger degree of variability. For ease of use, Table 1 includes common concentrations, associated fractions, the HorowitzMassart repeatability value, and the Horowitz-Thompson reproducibility value. Specificity: The specificity of a limit procedure is best stated in general chapter Validation of Compendial Procedures 1225 as the ability to assess unequivocally the analyte in the presence of components that may be expected to be present, including other specified impurities, the parent compound, counter ions, solvent fronts, and matrix components. For chromatographic procedures, such as for assays, the resolution provides the best approach PPM for specificity. The rationale for the specificity PPAC is described in the Assay section. However, the basis for the PPAC is illustrated in Figure 3, but the figure presents the error in the calculation of major component caused by co-elution with a minor component. In the case of Impurities testing, co-eluting species will often be of equal or greater concentration than the impurity of interest (measurand), therefore causing an unresolved peak to be over-estimated. However, from an acceptability perspective a false-positive result is preferable to a false negative. Therefore, the PPAC of a resolution of 1. 5 is acceptable. However, there are times that a resolution of 1.5 cannot be achieved. In these cases, a complete spike and recovery study or an orthogonal procedure will need to be completed or used, respectively. The PPAC for these studies is the requirement to meet the previously described precision requirements of each of the specified impurities.

8 10 Assessing Validation Parameters / General Information

Medicines Compendium

Figure 6.

Table 1. Horowitz-Thompson Values Horowitz-Massart Repeatability Value (RSD) 1% 2% 2.8% 3% 4% 5.7% 8% 11% 11% HorowitzThompson Reproducibility Value (RSD) 2% 4% 5.7% 6% 8% 11% 16% 22% 22%

Concentration of Analyte (%) 100% 1% 0.1% 0.05% 0.01% 0.001% 0.0001% 0.00001% <0.00001%

Concentration of Analyte (ppm) 10,000 ppm 1000 ppm 500 ppm 100 ppm 10 ppm 1 ppm 100 ppb <100 ppb

Concentration with w/w Units 1000 g/g 10 mg/g 1 mg/g 500 g/g 100 g/g 10 g/g 1 g/g 0.1 g/g <0.1 g/g

Concentration Fraction 1.0 0.01 0.001 0.0005 0.0001 0.00001 0.000001 0.0000001 <0.0000001

Limit of detection: The LOD is a PPM that is intended to assure that a procedure is capable of detecting an impurity if it is present. There is agreement in the literature that the best way to determine the LOD is through evaluating serial dilutions until a peak is no longer detected. This is often a long and tedious process that yields a definitive answer. However, too often analysts fall back on the 3 times noise rule of thumb that often obtains results similar to the serial dilution approach. However, for a limit test, neither of these approaches is optimal. Instead it is important to know that a procedure can differentiate between a passing and a failing result. The interval between passing and failing must be linked to the capabilities of the procedure. The best measure of the capability is the precision of the measure. The PPAC for precision has been defined by the Horowitz equation. To determine if the procedure has sufficient resolving power to determine whether a procedures passes or fails, it is necessary to first demonstrate that the peak obtained from a spiked measurand yields a peak area equal to or greater than the Standard solution at the limit. Then a solution that has an x% lower than the limit is measured. The x% is defined by the Horowitz equation. This solution must give a passing result, thereby demonstrating an acceptable level of detection and an acceptable accuracy of that determination.

CHROMATOGRAPHIC QUANTITATIVE TESTS Many of the considerations for the CVP described in the Chromatographic Limit Tests section also apply to the quantitative test. However, due to the need to provide a specific numerical content of the impurity (measurand) in an article (drug substance or product), a more exhaustive evaluation of accuracy, precision, and linearity is necessary. Accuracy: The accuracy of a procedure is defined as the ability to obtain the true value for a known measurement. In the case of an impurity measurement, like that for a content assay, the precision and accuracy are very closely related. As the concentration of the measurand decreases in relation to the concentration of the major component, the ability to accurately and precisely measure the impurity also decreases as described by the Horowitz equation. Generally speaking, the accuracy of a procedure is linked to systematic error of a measurement. The systematic error is independent of the number of analyses and is generally fixed or varies in a predictable manner. The systematic error so described is also called the systematic bias. The evaluation of bias is best completed using spike recovery studies over the range of values that a measurand is expected to be present. The accuracy or bias of a measurement should not be described by the precision of that measurement, but should be negligible. However, the term negligible has little meaning without a frame of reference. For the determination of

Medicines Compendium

General Information / 10 Assessing Validation Parameters 9

ble of determining an accurate and precise quantity for the measurand over the range that a typical unknown measurement will be made. The accuracy and precision requirements described for a quantitative impurity test also provides this assurance. By determining the precision at several points across the analytical concentration range appropriate to the measurand, the need for a separate linearity requirement is eliminated. Limit of quantification and range: The limit of quantification and the range are values that are evaluated to ensure that the analytical procedure is appropriate for the evaluation of a measurand at an unknown value. In the case of a compendial procedure, the value of the measurand is known before a procedure is validated. Typically, an impurity procedure will define the maximum concentration that an impurity may be present in a sample. By completing the accuracy requirement, the analyst knows that the procedure is capable of quantifying the measurand at half of the maximum acceptable value for the impurity. The accuracy requirement will also ensure that a procedure will provide acceptable results at twice the maximum acceptable concentration of the impurity. Successful completion of the accuracy requirements will provide sufficient confidence in the procedures adequacy, and therefore no additional evaluation of these characteristics is required to define an acceptable procedure.

an acceptable procedure, negligible can be defined as having a bias that is smaller than the acceptable precision of a measurement. Once again, the Horowitz/Massart repeatability value describes the acceptable precision relative to the relative concentration of the measurand. Precision: The precision of a measurement can be split into three different types: repeatability, intermediate precision, and reproducibility. The first evaluates instrumental and sample preparation sources of variability, the second examines variability caused by deliberate changes to the measurement environment, and the third examines the variability typically found through interlaboratory studies. In the determination of an acceptable procedure an understanding of all sources of variability does not need to be individually determined. Instead it is possible to model the types of variability likely to be encountered in reproducibility through the deliberate changes from the intermediate precision studies. RepeatabilityThe repeatability portion of a quantitative procedure validation is conducted by preparing six spiked sample solutions, each containing one or more of the impurities of interest at the analytical significant concentration. The relative standard deviation of these measurements is then compared to the Horowitz-Massart value appropriate for the concentration of the measurand(s). Where it is not possible to spike the analyte with a standard for the measurand, then the use of a retention time and relative response factor can be used. Although this approach is not optimal, it may be the only way to evaluate unstable or transient degradation products. Alternatively, a reference standard with the critical impurities already present of developed in situ could be used. The knowledge of the actual concentration of these impurities is impaired, but the materials could be used to complete the validation of the procedure. Wherever possible, a pure standard of the degradation product or process impurity is preferred. Intermediate precisionThe evaluation of the intermediate precision may be determined by repeating the Repeatability measurements previously described after a controlled deliberate change is made to the testing environment. The types of changes that are most useful for the determination of an acceptable procedure include different days, different instrumentation, different analysts, and using different instrumental conditions. The first three are particularly useful because they link the variability to the Horowitz-Thompson equation. The evaluation of variability caused by all four specified conditions is not necessary because the information provided by more than one condition does not generally provide any additional information. Therefore, the analyst should determine which conditions to be pursued. The result of the two sets of data (this should be 12 individual measurements) should be compared to the Horowitz ratio appropriate for the concentration of the measurement(s). Specificity: The requirements of the specificity in a quantitative procedure are the same as those described above for a limit test. Linearity: The linearity of the procedure is intended to assure the analyst that the procedure being validated is capa-

The Meaning of Acceptability

Up to this point, this document has focused on describing an acceptable procedure from a purely empirical and approachable manner. To that end, the CVPs have been defined and the performance criteria for each of these CVPs has been described. However, the fundamental question of the adequacy of the CDP has yet to be discussed. Comprehensive descriptions of the types of equivalence should be considered when comparing an analytical procedure with a compendial procedure. The first, and in many ways most appropriate, approach is that of an acceptable procedure. This is a procedure that with pre-defined performance criteria that describe a procedure adequate to achieve an approximation of the true value of an unknown without previously defining the procedure itself. Looking at the criteria previously described, it should be clear to a chromatographer that any procedure meeting the PC described herein will provide adequate assurance that the procedure under question is appropriate. Therefore, any procedure that meets these criteria should be considered to be equivalent or better than the Standard procedure. Where a procedure is equivalent to the Standard procedure (or acceptable), then according to the General Notices of USPNF, that procedure may be used as an alternative procedure without the need to actively compare these procedures.