分析方法的确认(6) 生物分析方法的实验设计

JOURNAL OF CHROMATOGRAPHY A ELSEVIER Journal of Chromatography A, 730 (1996) 381-394 Rational experimental design for validation bioanalytical methods Illustration using an assay method for total captopril in plasma J. Wieling*, G. Hendriks, W.J. Tamminga, J. Hempenius, C.K. Mensink, B. Oosterhuis, J.H.G. Jonkman Pharma Bio-Research International BV, Science Park, NL-9471 GP Zuidlaren, Netherlands Abstract Generally, bioanalytical chromatographic methods are validated according to a predefined programme and distinguish a pre-validation phase, a main validation phase and a follow-up validation phase. In this paper, a rational, total performance evaluation programme for chromatographic methods is presented. The design was developed in particular for the pre-validation and main validation phases. The entire experimental design can be performed within six analytical runs. The first run (pre-validation phase) is used to assess the validity of the expected concentration-response relationship (lack of fit, goodness of fit), to assess the specificity of the method and to assess the stability of processed samples in the autosampler for 30 h (benchtop stability). The latter experiment is performed to justify overnight analyses. Following approval of the method after the pre-validation phase, the next five runs (main validation phase) are performed to evaluate method precision and accuracy, recovery, freezing and thawing stability and over-curve control/dilution. The design is nested, i.e., many experimental results are used for the evaluation of several performance characteristics. Analysis of variance (ANOVA) is used for the evaluation of lack of fit and goodness of fit, precision and accuracy, freezing and thawing stability and over-curve control/dilution. Regression analysis is used to evaluate benchtop stability. For over-curve control/dilution, additional to ANOVA, also a paired comparison is applied. As a consequence, the recommended design combines the performance of as few independent validation experiments as possible with modern statistical methods, resulting in optimum use of information. A demonstration of the entire validation programme is given for an HPLC method for the determination of total captopril in human plasma. Keywords: Validation; Experimental design; Pharmaceutical analysis; Chemometrics; Captopril; Paracetamol 1. Introduction Bioanalytical methods must be validated if the results are used to support the registration of a new drug or the reformulation of an existing one. The validation is required to demonstrate the performance of the method and the reliability of * Corresponding author. Present address: Bio-Intermediair Europe BV, P.O. Box 454, NL-9700 AL Groningen, Nether- lands. analytical results. It is essential to use well characterized and fully validated methods to yield reliable results that can be interpreted 0021.9673196/$15.00 0 1996 Elsevier Science B.V. All rights reserved PIZ SOO21-9673(96)00006-4 382 .I. Wieling et al. I J. Chromatogr. A 730 (1996) 381-394 satisfactorily. Many papers have appeared re- cently on the validation of bioanalytical methods. The most relevant of these are referred to here [l-5]. The conference report by Shah et al. [4] (the Washington Conference Report) is generally accepted as one of the most important guidelines for bioanalytical methods validation. The confer- ence was attended by many workers in the field from pharmaceutical industries, regulatory bodies, contract research organizations and uni- versities and many of them contributed to the report. Nevertheless, some critical notes to cer- tain aspects of the guidelines in the Washington Report have been made by Hartmann et al. [5]. They have extracted some recommendations after analysis of the Washington criteria, namely that it would be preferable for the terminology to be consistent with the existing guidelines in other fields of chemical analysis. More importantly, the authors drew attention to statistical considera- tions and to the nature of experimental errors and the separation of the total measurement error into its constant (bias) and its random (precision) components. An excellent paper on the practical issues of bioanalytical methods validation was published by Dagdar et al. [6]. They discussed all aspects of bioanalytical methods validation, from prereq- uisites to method validation to revalidation and cross-validation. Useful procedures are described for the long-term stability validation of biological samples. Interesting comments on other stability questions are included and ideas on ruggedness testing and internal standard selection, For research and development laboratories within a service, the validation of a method should be cost-effective and results should be available as soon as possible. Therefore, the experimental set-up of the validation study should be efficient and sensible. This paper deals with the design and the experimental set-up of the validation of bioanalytical methods within our laboratory. The design was originally applied to chromatographic assays, but may, after a few adaptations, also be used for other analytical techniques, such as immunoassays. Particular attention will be given to statistical analysis of validation results. This paper will not emphasize items such as the preparation of standards for calibration, quality control and validation or strategies for method optimization. Other work- ers have highlighted these subjects in the context of bioanalytical methods validation [l-3]. Defini- tions of the required performance characteristics (validation criteria) are also not included; for descriptions we refer to other publications [l-5]. These papers also discuss in detail the back- ground, the meaning and the rationale for the characteristics to be validated. 2. Validation study protocol All analytical methods developed in our lab- oratories are validated comprehensively, and all aspects with regard to specificity, sensitivity, calibration model, recovery, accuracy, precision, stability and overcurve control of samples are covered. When bioanalytical studies are per- formed, the relevant international guidelines, recommendations and requirements are taken into account as comprehensively as possible. For the development and validation of the assay method, this concerns the Note for Guidance on Analytical Validation [7] and the Conference Report on Analytical Methods Validation (Shah et al. [4]). Development and validation of assay methods are conducted in accordance with cur- rent good laboratory practice standards [8,9]. Proper interpretation of these guidelines and regulations directs investigators towards the de- sign of protocols for all studies to be performed in support of drug registration. Studies also include methods validation, hence a signed protocol for assay method validation should be in place prior to the conduct of a validation study. In general, the protocol should be followed. Scientifically justified changes can be made, if the changes are documented and authorized by the study director. Usually, in protocols for assay method valida- tion we include items such as identification of the analyte(s), including sample matrix and concen- tration range to be applied, the preparation of calibration, quality control and validation sam- ples, the performance characteristics to be evalu- J. Wieling et al. I J. Chromatogr. A 730 (1996) 381-394 383 ated (including the limits to be maintained), the procedures for evaluation (number of replicates, number of batches/runs, statistics), the concen- tration range to be evaluated, data filing and reporting. 3. Validation programme 3.1. Analytical method We demonstrate here our validation pro- gramme by means of a recently developed meth- od for the determination of total captopril con- centrations in human plasma. Captopril is a potent and selective inhibitor of angiotensin-con- verting enzyme (kininase II). Oral captopril administration lowers blood pressure in hyper- tensive humans. Captopril is readily converted into its disulphide dimer and forms disulphide conjugates with endogenous thiol compounds [lo]. Only the free captopril is pharmacologically active; however, the formation of the inactive disulphides is reversible; subsequently, they may act as a reservoir of free captopril and contribute to a longer duration of action than predicted by the blood concentrations of free captopril [ll]. As a consequence, the total captopril plasma concentration is probably an important parame- ter in relation to therapeutic effect. To measure total captopril (free captopril + captopril disul- phides), all captopril disulphide conjugates have to be reduced to free captopril. Immediately after this reduction, a chemical stabilizer must be added to the biological samples, to prevent the re-formation of disulphides. In the method for the quantitative determination of total captopril in human plasma presented in this report, valida- tion samples were used, which had been pre- pared by spiking blank human plasma with known amounts of captopril-captopril disul- phide. No internal standard was used. The re- quirements for the method had been docu- mented in a Study Protocol for Assay Method [121. Captopril disulphides in plasma were reduced to free captopril with tris(2-carboxyethyl)phos- phine (TCEP). The free captopril was captured with N-pyrenylmaleimide (NPM), in order to protect the free thiol group from the re-forma- tion of disulphides. After the TCEP reduction and after NPM treatment, the reaction mixture was washed to remove interferences. Further sample clean-up was performed by means of liquid-liquid extraction of the NPM adduct of captopril. Separation was performed by reversed- phase high-performance liquid chromatography and quantification with fluorescence detection. The detection limit of the method was approxi- mately 5 ng ml-’ using 500 ~1 of plasma. The lower and upper limits of quantification were 10.0 and 1005 ng ml-‘, respectively. 3.2. Validation experimental design In principle, a common validation study within our laboratories consists of six analytical runs (during a period of at least 72 h), optionally extended by additional analytical runs for the validation of less common performance charac- teristics. Here, an analytical run is defined as a batch of samples consisting of a test sample for system suitability, calibration samples (>8), quality control samples (26) a blank sample and clinical study samples (20) which can be ana- lysed within a predefined period of time (com- monly 24 h). A diagram of the entire procedure for the bioanalytical methods validation is given in Fig. 1. For each validation study, the pre-validation phase (run 0) is used to assess the validity of the expected concentration-response relationship (lack of fit, goodness of fit), to assess the spe- cificity of the method and to assess the benchtop stability of processed samples. The latter experi- ment is performed to justify overnight analyses. After approval of the suitability of the method after the pre-validation phase, the main valida- tion phase (the next five runs) is performed to evaluate method precision and accuracy, re- covery, freezing/thawing stability and over-curve control/dilution. An efficient and rational design has been developed for this main validation phase. The design is nested, i.e., many ex- perimental results are used for the evaluation of several performance characteristics. Below, de- 384 .I. Wieling et al. I J. Chromatogr. A 730 (1996) 381-394 I 3loanalytkal v~lldauon - Common Validation - PfeValldatlon SplflW - Main Vdldatlon - Preclslon Wlthlll+Un Between-fun 1 A-J=Y RSCOV~ --/ OVer-cuwe conlrol/dlluUon 4 Fmezlng and thawing atablllty - Follow-up Vdldatlon ?? -( Long-term !Habllny Rugged- wng --j llwmloatabllily Photostablllty 1 C.SWpOVW ??Not dIscussed In paper Fig. 1. Diagram of the entire procedure for bioanalytical methods validation. tails of the entire programme are given. The complete validation study is presented in a scheme in Table 1. If applicable and relevant, follow-up validation runs may be carried out. Here, follow-up validation will not be discussed. Pre-validation In run 0 the following samples were analysed: (i) 24 calibration samples; eight concentra- tions (10.0, 25.1, 50.2, 100, 251, 502, 754, 1005 ng mll’) distributed over the concentration range with increasing intervals and analysed in trip- licate (entire procedure). (ii) samples for the assessment of the stability during storage in the sample compartment (benchtop stability); pooled extracts of spiked plasma samples at two concentrations (approxi- mately 25.0 and 750 ng ml-‘) were injected every 2 h for a total period of 30 h, during which the extracts were kept in the sample compartment of the injector (protected from light and at a tem- perature of 1oOC). The specificity of the assay method was checked by analysing at least six independent blank plasma samples. The chromatograms of these blank plasma samples were compared with chromatograms obtained by analysing test solu- tions of the pure compound treated with NPM. The specificity of the method was also checked for other compounds, namely drugs (and their metabolites, if applicable) that were used as co- medications during drug interaction studies. The above-mentioned experiments for pre-validation are general experiments, especially those for specificity testing. Additional or modified experi- ments may be performed for reasons such as the inavailability of reference materials for metabo- lites. Suggestions for solving such difficulties have been given by Dagdar et al. [6] in their prerequisites to methods validation. Main validation In runs l-5 the following samples were ana- lysed: (i) eight calibration samples containing eight concentrations (10.0, 25.1, 50.2, 100, 251, 502, 754, 1005 ng mll’) distributed over the con- centration range with increasing intervals; (ii) twelve precision and accuracy samples: four concentrations (10.0, 25.1, 251 and 754 ng mll’) in triplicate. (iii) three over-curve control samples; one concentration (2008 ng ml-‘) diluted five times in triplicate; (iv) six samples for stability assessment after repeated freezing and thawing; two concentra- tions (25.1 and 754 ng ml-‘) in triplicate [sub- J. Wieling et al. I J. Chromatogr. A 730 (1996) 381-394 Table 1 Scheme of the validation study: number of assays per analytical run for validation 385 Run Linearity Specificity Stability (30 h) 0 24 calibration samples 6 independent blank plasma 32 injections of pooled (8 concentration levels samples; test samples; sample extracts (every 2 h, in triplicate) co-medication samples at two concentrations) Total 62 injections Run Calibration Validation samples samples (8 calibration LLQ X, x, X, Total levels) P&A” P&A” F/T Rech P&A” Rech Dil P&A” F/T Ret?’ 1 (8) 3 3 3’ 3‘ 3 3’ 3’ 3 3’ 3’ 38 2 (8) 3 3 3’ 3’ 3 3’ 3’ 3 3’ 3’ 38 3 (8) 3 3 3’ 3’ 3 3’ 3‘ 3 3’ 3’ 38 4 (8) 3 3 3’ 3’ 3 3’ 3’ 3 3’ 3’ 38 5 (8) 3 3 3’ 3’ 3 3’ 3’ 3 3’ 3’ 38 Total 190 injections a P&A = precision and accuracy determinations. b Experiments are direct injections with concentrations equal to 100% recovery in extracts. ’ Results are used to calculate the performance characteristic using the precision and accuracy data of the same analytical run at the same concentration level. samples were taken from samples which were prepared from the validation pools and which were frozen and thawed before each next freez- ing and thawing cycle (run)]; (v) no recovery measurements were per- formed for total captopril analyses, since a re- covery experiment would not only include the liquid-liquid extraction procedure, but also the reduction of disulphides and the derivatization with NPM. Therefore, for simplicity reasons, a demonstration of recovery measurements and data processing with the same validation ex- perimental design is shown for paracetamol (acetaminophen) in plasma here: nine direct injections of test solutions for recovery determi- nation were performed; these direct injections contained such amounts of the analytes as would correspond with 100% recovery from validation samples at three different concentrations (1 .OO, 20.0 and 35.0 pg ml-‘) in triplicate. The recovery of theophylline (internal standard for paraceta- mol) was evaluated at the concentration used during the actual analysis of the plasma samples (40.0 pg mll’). 4. Data processing Calculations for the determination of the vali- dation parameters were performed using spread- sheets programmed in Lotus 123 on IBM-com- patible computers. These spreadsheets hold the analysis of variance tables for the determination of precision, accuracy, goodness of fit, lack of fit, freezing/thawing stability and over-curve con- trol/dilution. The peak height of the captopril-NPM adduct was taken as the response for a given sample. Calibration graphs were calculated by weighted linear regression (W= X-‘) on the responses of a series of calibration samples versus the corre- sponding nominal concentrations. The measured concentration in a sample was calculated by substituting the response for that sample in the equation of the corresponding calibration graph. The calibration data from runs 1-5 were subjected to the following acceptance criteria. A calibration point was rejected as an outlier if the back-calculated concentration for a calibration sample (on the basis of the equation of the 386 J. Wieling et al. I J. Chromatogr. A 730 (1996) 381-394 corresponding calibration graph) deviated more than 15% from the nominal value for the two lowest concentrations and more than 10% for the other concentrations. A calibration graph was accepted unless there were more than two out- liers, or if there were two outliers on adjacent concentrations. The validation data from runs O-5 were sub- jected to the Grubbs test [13] for the detection of outliers. Outliers, if any, were excluded from the calculation of performance characteristics. The measured concentrations were rounded to three significant digits. 4.1. Pre-validation Benchtop stability The measured peak heights for the assessment of the stability in the sample compartment were plotted versus time. The data were used for regression analysis to estimate an increase or a decrease in the measured peak heights. A de- crease or an increase of 10% in the measured peak height (based on regression analysis) is the limit we suggest. Crossing these limits is a warn- ing to improve benchtop stability, for example by decreasing the autosampler temperature or by changing the reconstitution medium. Choice of calibration model The responses as obtained for the 24 cali- bration samples were used to establish a relation- ship between the concentration and response and to evaluate the goodness of fit and the lack of fit by means of analysis of variance. If a significant lack of fit is observed, measures should be taken, e.g., selection of an alternative model, applica- tion of a detector with a better performance or the application of an alternative extraction pro- cedure. 4.2. Main validation Precision and accuracy The 15 measured concentrations per concen- tration level (triplicates from five runs) as ob- tained by analysing the validation samples were subjected to analysis of variance (ANOVA) to estimate the within-run precision and the be- tween-run precision. The accuracy of the method was dete