2018 Analytical target profile (perfil analitico objetivo control de calidad) PDF

Preview

Summary

perfil analitico objetivo para la validacion de ensayos analiticos para control de calidad de muestras de ingrediente farmaceutico activo de un producto cualquiera...

Description

Journal of Pharmaceutical and Biomedical Analysis 160 (2018) 73–79

Contents lists available at ScienceDirect

Journal of Pharmaceutical and Biomedical Analysis jo u rn a l h om e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / j p b a

Analytical Target Profile: establishmentof precision requirements for assay Joachim Ermer Sanofi-Aventis Deutschland GmbH, Quality Control Services Frankfurt Chemistry, Industriepark Hoechst, Bldg. D711, D-65926 Frankfurt am Main, Germany

a r t i c l e

i n f o

Article history: Received 29 June 2018 Received in revised form 16 July 2018 Accepted18 July 2018 Available online 20 July2018 Keywords: Analytical lifecyclemanagement Analytical TargetProfile Precision Probability distribution Quality-by-design Target measurement uncertainty

a b s t r a c t Approaches are presented to establish precision (or target measurement uncertainty) requirements to drug substance and drug product assays. They are based on the simple and well-known concept of the normal distribution probability around true content values represented either by manufacturing range limits, or by the manufacturing target (usually 100% label claim). A maximum acceptable precision is derived which allows a defined probability of analytical results within the established acceptance limits of the specification and thusan objective and rational establishmentof precision acceptance criteria. By this approach, ␣ or type-I-errors are controlled, i.e. the maximum probability of failure for intrinsically acceptable results islimited.Thecombination of thisnormal distribution probability approach with guard bands allows controlling ß or type-II-errors, i.e. the acceptance of intrinsically not conforming results is limited. Here, no assumptionsconcerning the manufacturing range are needed; therefore this approach can also be applied for quantitation of impurities. The guard band approach allows the highest level of control, but requires in turn high demands on the precision. Therefore, it should be restricted to drug product assays or impurity determinations with larger risks, i.e. justified by a corresponding clinical relevance, such as narrow therapeutic ranges or substantialtoxicity. © 2018 Elsevier B.V. All rights reserved.

1. Introduction An analytical procedure must be demonstrated to be fit for its intended purpose, which applies to its entire lifecycle. To achieve this goal, a quality-by-designapproach has beenproposed for pharmaceutical analyses which includes the three stages Procedure Design and Development, Procedure Performance Qualification, and Continued Procedure Performance Verification [1,2], in alignment to manufacturingprocess validation [3] (see Fig. 1). A fundamental component of this approach is having a predefined objective that stipulates the performance requirements for the analytical procedure, i.e. the Analytical Target Profile (ATP). “The ATP states the required quality of the reportable value produced by an analytical procedure in terms of the target measurement uncertainty (TMU).”[2] The performance requirements, i.e. accuracy and precision (or TMU)in case of quantitative assay procedures should be defined by the measurement objectives of the given test(Quality Attribute) which is linked to the product control strategy, such as water

Abbreviations: ATP, Analytical Target Profile; TMU, targetmeasurement uncertainty; GB, guard band(s); SL, specification limit(s); QL, quantitation limit. E-mail address: [email protected] https://doi.org/10.1016/j.jpba.2018.07.035 0731-7085/© 2018 Elsevier B.V. All rights reserved.

content in a drug substance, assay of active in drug substance or drug product, content of impurities, etc. As based on the measurement objective, the TMU should be (as far as possible) not directly linked to a given analytical method or technique. If the TMU can be established unambiguously, any method conforming to the ATP requirements can be applied. Such a concept is already applied in compendia for determination of elemental impurities [4,5]. All other performance attributes are method-specific, and eventually consolidate in either accuracy (bias), or precision, for example linearity (justification of the calibration model), specificity, or quantitation limit. But how can an acceptable precision or TMU be derived in an objective way? Often,these acceptance limits have been definedfrom the capability of the given analytical technique, for example 2.0% or 3.0% [6] relative standard deviation (RSD) of intermediate precision for HPLC assay. However, such a capability-approach (“what can be achieved”) lacks scrutinization versus the requirements represented by the acceptance limits of the specification (SL) (“what must be achieved”). Even 2.0% intermediate precision are obviously not suitable for a drug substance assay with the usual SL of 98.0 to 102.0%. Although these generic SLs (as well as 95.0 to 105.0% for drug products) are also based on historical experience with no direct (patient) safety link, they havebecome ingrained in

74

J. Ermer / Journalof Pharmaceutical and Biomedical Analysis 160 (2018) 73–79

Fig.1. Analytical performance (validation) lifecycleapproach.

regulatory expectation and wouldneed hard justification tochange [28,29]. Anotherproposal hasbeen to use a defined fraction of the specification range, for example 60% resulting in a TMU of 3.0% for content limits of± 5.0%, i.e. 95.0–105.0% [7], which might be considered as somewhat arbitrary, too. However, in that paper the precision and accuracy requirements for the ATP have been further modelled using a two-sided beta-content tolerance interval approach thus providing the link tothe requirements. The disadvantage of thisapproach is that an evaluationis only possible with the experimentalprecision results obtained with the final analytical procedure. In the CITAC-Guideline [8],an approach is describedto calculate the targetstandard uncertainty by dividing the specification range (“compliance interval”) by a factor of 16. This includes a coverage factor of2, which gives a level of confidence of approximately 95%. As often the underlyingmanufacturing processes arenot symmetrical to the specification ranges anda shiftof the process mean can be expected, in the following the literature approaches to define TMU will be adjusted to a one-sided calculation. For the TMU calculation, always the tighter range between SL and process mean or true assayvalue is relevant. Such a one-sided approach can also be directly applied to one-sided specification limits. Thus, in the CITAC-approach, the one-sided specification range is divided by a factor of8 to obtain the TMU. In analogy to the six sigma approach to process capability [9], the method capability is calculated by dividing the one-sided specification range by 3 times the (actual) standard deviation (SD). The method capability should be at least 1.0, but usually a “safety margin”is applied and1.33 or 2.0 is recommended.In theprecisionto-tolerance ratio approach (PTOL), even an index of 3.33–10 should be applied [10]. Transformingthis calculation, the TMU corresponds to the one-sided range divided by afactor of4–6 or even 10–30. Note that by this transformation,not the actual analytical variability is addressed (i.e. capability-based), but the maximum one allowed conforming to the SL (i.e.requirement-based). The toleranceinterval can be defined to contain a defined fraction (e.g. 90%) of futureresults with a defined confidence (e.g. 95%) [11] with tabulated tolerance factors for the number of determinations used to obtain the standard deviation. Alternatively, the Student-t-factor can be used for the defined levelof statistical confidence andthe degreesof freedom inthe respective precision study [12]. Generaltolerance or coverage factors,e.g. 2 (corresponding to 95% levelof confidence) or 3 (corresponding to 99% level of confidence) havebeen proposed to be used directly as a division factor to obtain the TMU [13]. For volumetric titrations, an acceptable repeatability is proposed as one third of the one-sided content limits, for example 0.33% RSD in case of acid/base titrations with content limits of ±1.0% [14]. In order to take intermediate precision into account, a division factor of 4 can be assumed for the TMU.

All these approaches are related to probability distributions, but the division factors larger than 2 (95% confidence) or 3 (99% confidence) try to include additional or unknown (long-term) uncertainty sources, i.e. they add up worst-case scenarios. This often results in unfeasible requirements, in particular for HPLC drug substance assay. With SL of 98.0–102.0%, a TMU of 0.25%, 0.50%, or 1.0% would result for the CITAC-approach, the methodcapability-approach, and the95% tolerance approach, respectively. Ifthe tighter TMU requirements would be reallynecessary, usually obtained intermediate precisionsfor drug substance assays(pooled averages 0.80% up to 1.4% [15,16] would result in a much larger frequency of out-of-specification resultsdue to random variability thanobserved in Quality Control(QC) practice. Although these simple division factors are also related to a probability distribution, they do not consider the possibility that the distribution may overlap with both specification limits, which would alter the probability of results outside the limits. Therefore, thisarticleproposes approachesto establish TMUcriteria for assays of active in drug substances and drug products directly based on calculations of the normal distribution probability. 2. Fundamental assumptions The startingpoint to derive the TMU criteria are theestablished acceptance limits ofthe specification, asrepresentation ofthe measurement requirements for the given Quality Attribute. Usually, thesecontent limits follow inpharmaceutical QC traditional regulatory expectations, i.e. 98.0–102.0 or 95.0–105.0 for drug substance anddrug product, respectively. As acceptance ranges ofthespecification must include both manufacturing and analytical variability, assumptions for the manufacturing part of the specification range (i.e. the rangeof the true content of manufacturedbatches) have to be made to establish the TMU. In principle, accuracy (bias) and precision can be evaluated simultaneously, or separately. The former would allowa “trading” between bias and precision, i.e. a larger bias might be acceptable in case of high precision (or vice versa). However, a combined experimental approach for drug product assay necessitates that spiked samples which are usually needed for accuracy are representative enough to allow a routine sample preparation. The latter is crucial to obtain the precision ofthe reportable value, i.e. that of the routine application of the analytical procedure. The main objective of Stage 1, analytical procedure design, is the elimination of bias (systematic errors).When this is achieved, the main focus can be directed to the investigation and control of precision (random variability). Statistically, a true bias of zero is assumed, with an acceptance criterion for theobserved bias in the ATP based on the expected range of random experimental variability. The derivation of the TMU for assay from SL is based on the assumptionof anormal distribution, which is almost impossible to prove, but can be expected with good reasonfor physico-chemical assay techniques usually appliedin pharmaceutical analysis. Although strictly the thus obtained TMU represents only the allowed random variability, it should also include any systematic bias which cannot be specified. However, the preferred approach should be to determinethe bias andeliminateor correct it. Any shift which may develop over time, such as stability-relevant degradation will reduce the specification range available for theanalytical procedure and has to be added to the manufacturing range. 3. Establishment of TMU for drug substance The maximum allowed manufacturing range for anactive ingredient in drug substance is defined bythe allowed sum of impurities

75

J.Ermer / Journal of Pharmaceutical and Biomedical Analysis 160 (2018) 73–79

due to degradation (increase of the sum of degradation products) during shelf-lifemust be considered andincluded inthe worst case true assay value. For example, a shelf-life limit for the sum of impurities of1.0% would lower the worst-case truecontent to 99.0% and would require aTMU SD of 0.78%. Of course (if it can be sufficiently justified), the sum of impurities could also be considered in the establishmentof the lower SL (i.e. 97.5% in the example), which would leave ± 2.0%for the TMU calculation. TheSD obtained represents the maximum allowed true TMU. It has also to include the experimental uncertainty components of analytical procedure steps, e.g. the calculation on an anhydrous and solvent free basis. In case of low SL for water and residual solvents, this errorpropagation can usually be considered to be negligible. However, these steps must be included in the experimental validation of the reportable value precision in Stage 2, Procedure Performance Qualification [7,18]. 4. Establishmentof TMU for drug product 4.1. Worst-case scenario Fig.2. Normal probability distribution witha true mean of 99.5% and 90.0% results inside the specification limits.

(as far as assay-specific and for the assay on anhydrous and solventfree basis). Assuming a specification range of 98.0–102.0%, and a maximum sum ofimpurities of0.50%, themanufacturinglimits (i.e. the minimum and maximum limits for the true assay values) will be 99.50% and100.00%, respectively. Now, the TMU can be defined as the maximum uncertainty which would ensure an acceptable probability of results within the specification limits. Assuming a normal distribution, the probability within and outside SL can be calculated [17]. As the analytical uncertainty range at the lower specification side isthe tighter one (1.5% vs. 2.0% towards the upper specification limit), the probabilities are calculated for a true value of 99.50%. However, this represents a worst-case scenario, and for a vast majority of batches, the sum of impurities will be less than 0.50%. Therefore, for this worst-caseapproach, a probability of 90% of results within the specification is recommended, i.e. 90% of all resultswould conform to thespecification if the trueimpurity content wouldbe 0.50%. For atotal impurity content ofless than 0.50%, the probability of resultswithin therequired range would be larger. In case of the possibility of a two-sided limit-violation (see Fig. 2), the probabilities bothoutside the lower and upper SL have to be considered (see Eq. (1)). Using Eq. (1), e.g. in anexcel spreadsheet, the entry for the true SD (sigma) is varied until the added probability within the SL is 90%.





P (%)inside = NORM.DIST SLupper ;␮; ; TRUE − NORM.DIST (SLlower ;␮; ; TRUE)

With respectto drug product, itis morechallenging toestimate the range of manufacturingvariability compared to drug substance, because it cannot be derived from other Quality Attribute specifications. Therefore, it has to be assumed, for example based on prior knowledge from comparable formulations, or postulating as a fraction of the specification, e.g. 50%. Then, the TMU is calculated according toEq. (1) for the analytically available range between the SL and a true content atthe limits of the manufacturing range, as worst case. Theacceptable probability can be defined taking the reliability of the manufacturing variability estimate into account. For example, if the manufacturing range is well known from measurements (e.g. analysis of variances to separate manufacturing and analytical variability), or historical data from the same type of drug product, alower probability, e.g. 90% as for drug substance can be used, because only a small fraction of drug product batches should be manufactured atthelimits of the manufacturingrange. For allother batches, the probability ofresults within the specification would be higher. Otherwise,a probability of 95% may be used (Table 1). Table 1 Calculationof TMUfor various probabilitiesandmanufacturing variabilities,amanufacturing target of 100.0%, and SLof 95.0–105.0%. Probability within specificationlimits

Manufacturing limit (␮, %)

TMU(%)

95% 90% 95% 90%

98.0 98.0 97.0 97.0

1.82 2.33 1.21 1.56

(1)

With SLupper, lower =upper and lower acceptance limit of the assay specification, ␮ = true assay value for the tightest side of the specification range, ␴ = maximum (true) SD to achieve the defined probability of results inside the specification limits, TRUE =cumulative probability For the given example, the acceptable 90% probability is achieved with a SD of 1.11%. Note that for the sake of simplicity, the SL have been used in the calculation as given, without consideration of the rounding range(e.g. 97.95–102.05 for a formal range of 98.0–102.0). Only the tightest side (in this case the lower end of the specification range) needs to be considered. Using the calculated SD of 1.11% for 99.50% trueassay, the spread sheet can be used to show that the probability for an upper true mean of 100.0% would be 92.8% and thus larger than the desired 90%. Anyshift, for example

In case of a symmetrical range, i.e. with a manufacturingtarget of 100.0%, either the upper or lower manufacturing limit can be used. If unsymmetrical, the limit at the tighter side must be used. Any shift, for example due to degradation during shelf-life must be subtracted from the lower manufacturinglimit. Using Eq. (1) in an excel spreadsheet, the entry for the trueSD (sigma) is varied until the added probability within the SL equals the defined probability (Fig. 3). 4.2. Combined analytical and manufacturing variability Inthis approach, theprobability distribution is estimated for the combined analytical and manufacturingvariability (Eq. (2)). Using Eq. (1) in an excel spreadsheet, the entry for the true SD (sigma) is varied until the added probability within the SL equals thedefined

76

J. Ermer / Journal of Pharmaceutical and Biomedical Analysis 160 (2018) 73–79

Fig.3. Normal probability distribution witha true mean of 98.0% and 95.0% results inside the specification limits.

acceptance criterion. From thisoverall standard deviation, the TMU is calculated by re-arrangingEq. (2) using the assumed manufacturing standard deviation. ␴overall =



␴2

analyt

2 + ␴manuf

(2)

Compared to theworst-case approach (seeSection 4.1), here the probability includes all manufacturedbatches. The manufacturing variability may be used from prior knowledge, e.g. data obtained from comparable drug products. Alternatively, it can be estimated from the assumed manufacturing range. Considering 95% of the range, theSD corresponds to¼ of the manufacturing range, e.g. a manufacturing SD of 1.0 is derived from a manufacturing range of ± 2.0. The true value ␮ corresponds to the manufacturing target, i.e. usually 100.0%. Therequired TMU is less stringent,because the defined probability applies to all manufactured batches (Table 2). Table 2 Calculation of TMU withcombined manufacturingand analytical variability for various probabilities andmanufacturing standard deviations, a manufacturing target of 100.0%, and SLof 95.0–105.0%. Probability withinspecification limits

Manufacturing SD (%)

TMU (%)

95% 90% 95% 90%

1.0 1.0 1.5 1.5

2.35 2.87 2.06 2.64

4.3. Guard bands

Fig. 4. Normal probability distribution with a true value at the specification limit and 5.0% within the guard band (Type-II-error).

procedure is used, these GB can be regardedas method-capability based. This approach, as well as those described in Sections 3, 4.1 and 4.2 take only alpha or type-I-errors into account, i.e. the defined probabilityor confidence outsidethe SL andconcerns the producers riskof a wrong decision (“falsefa...