Journal of Pharmaceutical & Biomedical Analysis 0731-7085/90 $3.00 + 0.

00
Vol. 8, Nos 8-12, pp. 613-618, 1990 ~) 1990 Pergamon Press plc
Printed in Great Britain

A practical approach to method validation in

pharmaceutical analysis*
G.P. C A R R t $ and J.C. W A H L I C H §

$ Medgenix Group, B-6220 Fleurus, Belgium

§ Glaxo Group Research, Ware, UK
Abstract: Guidelines issued by Regulatory Authorities make it clear that validation of analytical methodology is now
widely required in support of registration dossiers. Although some attempts are made at defining terms and some vague
indications are sometimes provided within these guidelines, no clear advice is provided on how validations should be
conducted and what results should be expected. In this paper it is attempted to suggest some practical approaches to
conducting validation and in particular to the determination of accuracy, linearity and limit of detection/quantitation.
Keywords: Analytical validation; accuracy; linearity; limit of detection; limit of quantitation.

Introduction are dependent on: (a) the purpose of the

method, and (b) the sample matrix. With
Analytical method validation has been defined respect to the purpose of the method, one
as a procedure used to prove that a test method should consider for example whether it is to be
consistently yields what it is expected to do applied to a determination of a m a j o r or minor
with adequate accuracy and precision. O v e r c o m p o n e n t , or for evaluation of performance
recent years the Regulatory Authorities have characteristics (e.g. dissolution test for a solid
b e c o m e increasingly more aware of the necess- dosage form) and whether the test will then be
ity of ensuring that the data provided to them used to support a release specification or a
in applications for marketing (and perhaps stability study.
clinical) authorizations have been acquired In this p a p e r it is intended to review and
using validated analytical methodology. This then demonstrate practical approaches by
has resulted in the publication of a series of which accuracy, linearity and limit of de-
requirements and guidelines by various auth- tection/quantitation may be evaluated for
orities [1-4]. various analytical methods.
It is the opinion of the present authors that
the contents of these documents are often very
vague, sometimes quite inaccurate and mis-
Validation Parameters
leading and rarely provide the development
analyst with guidance on what should really be Accuracy
required of a validation exercise. In addition, a A reasonable definition is provided in USP
survey of recent literature on analytical devel- X X I I [2]: "The accuracy of an analytical
o p m e n t reveals very variable standards em- method is the closeness of test results obtained
ployed for validation. by that method to the true value".
P e r f o r m a n c e p a r a m e t e r s which should be F D A Guideline [1] requires that data be
addressed in a validation exercise include: provided " . . . over the range of interest (ca.
accuracy; precision; repeatability and repro- 80-120% of label claim)". Some details of how
ducibility; limit of detection and quantitation; accuracy should be determined are provided in
linearity; selectivity; ruggedness/robustness; USP X X I I and C P M P Guidelines [2, 4], which
and stability of analytical solutions. both contain some useful information, but
T h e p a r a m e t e r s that require validation and there are still a n u m b e r of questions left
the approach adopted for each particular case outstanding, as noted below.

* Presented at the "Second International Symposium on Pharmaceutical and Biomedical Analysis", April 1990, York,
UK.
t A u t h o r to whom correspondence should be addressed.

613
614 G.P. CARR and J.C. WAHLICH

Table 1
Recommended validation ranges for linearity studies
Typical range Recommended validation range
Purpose of analysis (%) (%)

Release specification assay 95 to 105 80 to 120

Check specification assay 90 to 110 80to 120
Content uniformity test 75 to 125 70 to 130
Assay of a preservative in a stability study 50 to 110 40to 120
Determination of a degradant in a stability study 0 to 10 0 to 20

What tests need to be validated for accuracy? the equivalent standards. Normally for im-
If one accepts the USP XXII definition, it is purities for which the levels of interest are
arguable that virtually every physico-chemical around 0 . 1 - 1 . 0 % , a variance of not greater
test method will require some form of valid- than + 5 % may be considered acceptable.
ation for accuracy.
(e) Bulk drug assays. The approach adopted
Design of accuracy determination for differ- depends on the assay method to be adopted.
ent types of analytical method. For titrations the expected equivalence point
(a) Identity tests. It should be demonstrated may be calculated on theoretical grounds,
that a positive result is obtained if the analyte is taking into account the number of titratable
present and a negative result if it is absent. functions in the analyte and the MW, but it is
recommended that this should then be verified
(b) Physico-chemical characteristics (e.g. MP, by carrying out the intended titration pro-
optical rotation). It should be demonstrated cedure on a well characterized reference
that the result is not influenced by other standard.
components which may be present in the For light absorption (UV) assays based on
sample matrix. specific absorbance values, i.e. A ( l % , 1 cm),
validation would require that this value be well
(c) TLC tests for related substances. It should selected. It should be noted here that these
be demonstrated that the impurity of interest values for closely related compounds may vary
in the presence of the principal component considerably, so it would also be appropriate to
provides a response to the method of detection demonstrate that this is not significantly in-
which is comparable to that of a standard fluenced by the presence of likely impurities at
application of the same quantity of the im- their intended maximum limits.
purity. In addition, it is a popular approach in For chromatographic assays, the mass bal-
such T L C tests to compare the responses of ance approach of Kirschbaum et al. is recom-
impurity zones in sample applications with mended [5]. This is based on the comparison of
those of the principal zones in standard appli- total peak response with and without the
cations which are prepared by dilution of chromatographic column in place; a variance
sample solutions. This approach is widely of not more than + 2 % should normally be
adopted in pharmacopoeial monographs and achieved.
should be justified by demonstrating equival-
ence of response between the principal com- Design of accuracy determination for differ-
ponent and its main impurities. ent sample matrices. H e r e it is intended to
discuss additional complications which arise
(d) HPLC/GC tests for related substances. The when the procedure includes extraction of the
approach described above of spiking the analyte from a sample matrix prior to carrying
sample with the impurity at around the level of out a measurement. So it is now necessary to
interest would be applicable here also. In this demonstrate extraction efficiency in addition
case, however, test results will probably be to the above discussed parameters. This type of
calculated by electronic integration of detector determination is generally referred to as "re-
responses and based on peak heights or areas. covery" and is the aspect of accuracy validation
The validation should then be evaluated by which has been best recognized by the Regu-
comparing the integrator values obtained in latory Authorities, some of whom refer to the
the same way for spiked sample with those of "technique of analysis of spiked, active-free
M E T H O D V A L I D A T I O N IN P H A R M A C E U T I C A L ANALYSIS 615

samples". There are still a number of practical available, and then to examine for the presence
issues to address, as noted below: of activity in the excipient residue after ex-
traction.
(a) Spiking range. This will be dependent on
the intended purpose of the method, but for a Linearity
procedure intended for a specification assay, a A reasonable definition is provided in CPMP
range of 80-120% is normally considered Guidelines [4]: "The linearity of a test pro-
appropriate. This may also be suitable for an cedure is its ability (within a given range) to
assay to support a stability study, but if produce results which are directly proportional
degradation below 80% is anticipated this to the concentration of analyte in the sample".
should be taken into consideration. A suitable All guidelines recognize that linearity is a
approach is to spike at five levels within this parameter which should be determined, but a
range and for each level the variance should number of issues are not addressed, including
normally be no greater than +2% of the the following.
theoretical value.
What tests need to be validated for linearity?
(b) Method of spiking. To provide an appro- This is only required for quantitative methods
priate model, consideration should be given to but not necessarily in all cases. It is not
how the active will be introduced into the normally necessary to demonstrate linearity for
mixture of excipients. For example, in the case a titration based on a well established equiv-
of a tablet, manufacture includes a com- alence factor.
pression stage and this should be taken into Linearity is usually conducted to justify
account. single point standardization, i.e. the analyst is
Possibilities which have been used include assuming a linear response with zero intercept.
addition of drug into excipient mixture as a In some methods, e.g. fluorescence, or atomic
solution in a volatile solvent and then drying spectroscopy, this assumption is not made and
under vacuum to produce more intimate mix- calibration curves are used. Then it is not
ing of drug with excipient. necessary to demonstrate linearity, but it
Another possibility is a "recovery efficiency should be shown that the calibration curve has
experiment"; this is most appropriate for a satisfactory slope sensitivity over the range of
H P L C and GC methods and requires an interest.
internal standard. The sample is extracted with
solvent as intended in the final procedure, but Design of linearity determination. The FD A
with the addition of an internal standard. After Guideline [1] recommends a concentration
centrifuging or filtering, about 75% of the range equivalent to 80-120% of the theoretical
supernatant solution is taken, subjected to the content of active. In the literature it is often
remaining procedure and the ratio of analyte seen that a range of perhaps 0-200% is
response/internal standard response noted. examined. In practice the study should be
The sample residue (including about 25% of designed to be appropriate for the intended
supernatant from the first extract) is then re- analytical method, as shown in Table 1. The
extracted with a further volume of solvent range selected for validation should not be
without internal standard, centrifuged or fil- unrealistically wide, as this may lead to rejec-
tered, subjected to the remaining procedure tion of a method which is really quite suitable
and the ratio of analyte response/internal for the intended purpose.
standard response is again noted. The variance
between the two ratios should not be greater
than + 2 % . This approach has the advantage Evaluation of results. Very variable stan-
that it can be conducted on "real samples" and dards are observed in the literature. Authors
does not require specially prepared "validation may sometimes simply claim that the pro-
samples". However, it should be noted that cedure was linear, with the additional mention
this technique will not detect inefficiency of of a "near zero intercept". More often, data
extraction due to some irreversible binding of are processed by linear least-squares regression
drug to excipient. and authors quote values obtained for the
Another approach would be to use a radio- regression coefficients "a" and "b" of linear
labelled analogue of the drug, if this were equation y = ax + b, together with a corre-
616 G.P. CARR and J.C. WAHLICH

(a)

SampLe
/ chromatogram

BLank
_/ chromartogram I
ill
20w, or 20w2

(b)

/
A
e

Figure 1
Estimation of chromatographic baseline noise. (a) Noise is measured over the region of a blank chromatogram
corresponding to 20 times the width of the analyte peak and situated around the region where the analyte peak would be
located in a sample chromatogram. (b) Noise magnitude is determined from the largest peak-to-peak fluctuation (No p)
or from the largest positive or negative deviation from the mean (Np).

lation coefficient. This is inappropriate be- (c) T h e value o f the correlation coefficient as
cause: a criterion of linearity has b e e n criticized in the
(a) T h e value o f a does not provide any literature [6, 7], but nonetheless authors of
i n f o r m a t i o n on the linearity of the p r o c e d u r e analytical d e v e l o p m e n t papers continue to
or on the g o o d n e s s of fit to a regression line. a d o p t it, or its squared product.
(b) T h e value of b is very i m p o r t a n t as it is T h e correlation coefficient (r) was devel-
the value of the intercept of the regression line. o p e d to d e m o n s t r a t e w h e t h e r or not any
U n f o r t u n a t e l y this is often presented as the relationship exists b e t w e e n two sets of data. If
n u m b e r c o m p u t e d in the regression calculation n o n e , r = 0; if a linear relationship exists, r =
and will not have any m e a n i n g to the reader + 1 or - 1 d e p e n d i n g on the slope.
without some further information, e.g. a b o u t F o r validation the question is not w h e t h e r a
the a b s o r b a n c e range of a detector. This is relationship exists; it is normally expected to
easily r e m e d i e d by expressing b as a per- be linear. A n o t h e r a p p r o a c h which has b e e n
c e n t a g e of the value of analytical response at suggested is based on curve fitting [7, 8].
the 100% analyte level. It is then suggested R e s p o n s e versus c o n c e n t r a t i o n data are fitted
that this should normally fall within the range to e0uations of the type: y = ax" + b; y = a x z
- 2 . 0 to + 2 . 0 % for the validation to be + b x + c; y = a ( e x p ) b x ; or y = a + b x + c / x ,
considered satisfactory. to establish the best fit.
METHOD VALIDATIONIN PHARMACEUTICALANALYSIS 617

These approaches may be of interest if the the impurity of interest to establish the smallest
objective were to establish calibration curves, quantity which can be reliably detected. The
but in practice it is intended to use these data following points should be noted.
to justify the linear model: y = ax. This would (a) The weight of sample and volumes of
then allow the conduct of analyses using single- sample solutions applied to the plate should
point standardization. For this to apply, a remain constant and be the same as intended in
satisfactory intercept value must be demon- the analytical procedure.
strated as described above. (b) After development, plates should be
In addition, goodness of fit of data to the examined by all intended detection methods,
regression line may be evaluated by a pro- e.g. UV radiation at 254 nm, UV radiation at
cedure based on the residual sum of squares. 366 nm and any spray reagents. LOD values
Taking the regression line as the mean, a should then be separately reported for each
relative standard deviation (RSD) is calculated method of detection.
for the data; normally this value should not be (c) Results are best expressed as a weight
greater than 2.0%, but when evaluating this and also as the equivalent percentage (or
determination, the results of precision deter- relative scale of units) in relation to the drug of
minations should also be taken into account. interest e.g. x (txg) and equivalent to y (%) or z
(ppm).
Limit of detection (LOD)/quantitation (LOQ) USP XXII [2] also includes a proposal for
Limit of detection (LOD) is a parameter of the determination of LOQ for non-instrumen-
limit tests and may be defined as the smallest tal methods, but in the opinion of the authors
quantity of analyte which may be expected to this is not appropriate.
produce a response which is significantly differ-
ent from that of a blank. Determination of LOD and LOQ for in-
Limit of quantitation (LOQ) is a parameter strumental methods. According to IUPAC [9]
of determination tests for minor components for spectrochemical methods these values may
and may be defined as the smallest quantity of be determined for the smallest concentration
analyte which can be determined with accept- (Cl) or amount (ql) from: Cl ( q l ) = ksa/S,
able accuracy and precision. where k is a constant, sB is the standard
deviation of analytical blank signal, and S is the
What tests need to be validated for LOD/ slope of response versus concentration curve
LOQ? These are essentially the parameters of (slope sensitivity). Recommended values for k
"trace analysis methods", i.e. procedures that are: for LOD, k -- 3; and for LOQ, k -- 10.
require methods to be operating at or close to Justification of these values has been pro-
maximum attainable sensitivity. It may require vided by Long and Winefordner [10] and by
judgment from the development analyst the Analytical Methods Committee [11].
whether this validation is appropriate, but in For this to be used in practice then, requires
general it may be assumed that it will be determination of the value of sB, which is
required for tests for impurities including: TLC related to baseline noise.
and HPLC tests for impurities and related For spectrochemical methods this is not a
substances; GC tests for residual solvents; problem as the analyst can conduct a series of
atomic spectroscopy tests for metals, e.g. absorbance determinations with blank sol-
catalysts. utions, using instrumental conditions intended
However this validation may also be re- for sample determinations. The standard
quired to support some assay methods such as deviation may then be calculated.
content uniformity tests for dosage forms For chromatographic methods, the method
containing very small amounts of active per is not so obvious, but an approach based on
unit dose, or assays of dissolution test samples. one proposed by Foley and Dorsey is recom-
mended [12]. This has now been introduced
Determinations for non-instrumental into the European and British Pharmacopoeia
methods. This would probably most commonly monographs on gas chromatography and liquid
apply to TLC tests for related substances. In chromatography as the basis of the method for
practice LOD determination may be achieved determining signal-to-noise ratios [13, 14].
by chromatographing samples of analyte which The value of Sa is estimated from the
have been spiked with decreasing quantities of magnitude of noise in a blank injection over a
618 G.P. CARR and J.C. WAHLICH

representative section of baseline which is 20 in particular for accuracy, linearity and L O D /

times the peak width of the analyte, as illus- L O Q . Proposals for ensuring that methods
trated in Fig. 1. There is no internationally continue to maintain their valid status during
recognized approach to conducting this deter- routine use (system suitability tests) are pre-
mination and some workers use 20 times the sented elsewhere [16].
baseline peak width (Wz) [12], while others use In considering these validation procedures,
20 times the peak width at half height (wl) [13- the authors have attempted to recommend
15]. approaches which will challenge the method
The noise magnitude may then be deter- being examined to demonstrate that it is
mined from the largest peak-to-peak fluctu- capable of providing the desired information.
ation (Np_p) [12], when Sa = Np_p/5, or from
the largest deviation from the mean (either
positive or negative) (Np) [13-15], when SB =
References
Np/2.
Whichever approach is adopted it must be [1] Guideline for Submitting Samples and Analytical
noted that the results depend on predictions Data for Methods Validation. US Department of
Health and Human Services, Food and Drug Ad-
made with analytical blank solutions. In ad- ministration, Maryland, USA, Feb. (1987).
dition, estimations of noise intensity are based [2] United States Pharmacopeia XXII, Section 1225,
on the heights of noise fluctuations, whereas 1710-1712. United States Pharmacopeia Convention
Inc,, Rockville, MD (1990).
the analytical method being validated may well [3] Department of Health Medicines Act 1968, Guidance
be based on peak area measurements. Notes on Applications of Product Licences (MAL2)
Thus, it is essential that the development A20-7 and A2E-2. HMSO (1987).
[4] Committee for Proprietary Medicinal Products Guid-
analyst verifies the predicted results by ance Note on Analytical Validation, Final Draft,
examining sample solutions containing the Document No. 111/844/87-EN,Aug. (1989).
computed concentrations of analyte to ensure [5] J. Kirschbaum, S. Perlman, J. Joseph and J. Adamo-
vics, J. Chromatogr. Sei. 22, 27-30 (1984).
that the values are indeed realistic. [6] M.F. Delaney, Liq. Chromatogr. 3, 264-268 (1985).
[7] Analytical Methods Committee, Analyst 113, 1469-
1471 (1988).
Conclusions [8] E. Debesis, J.P. Boehlert, T.E. Givand and J.C.
Sheriden, Pharmaceut. Technol. 120-137 (1982).
It is clear from the various guidelines issued [9] Nomenclature, Symbols, Units and their Usage in
by Regulatory Authorities that analytical Spectrochemical Analysis II, Spectrochim Acta B 33,
242 (1978).
methodology should be thoroughly validated. [10] G.L. Long and J.N. Winefordner, Anal. Chem. 55,
Although the guidelines provide indications of 712A-724A (1983).
the validation required, they do not normally [11] AnalyticalMethods Committee, Analyst 112, 119-204
(1987).
include much information on acceptable [12] J.P. Foley and J.G. Dorsey, Chromatographia 18,
approaches to conducting these studies, or on 503-511 (1984).
the kind of results which should be considered [13] European Pharmacopoeia (2nd edn), 11th Fascicule,
V.6.20.3 and V.6.20.4, Maisonneuve SA, Sainte-
acceptable. This is not very surprising, as such Ruffine, France (1987).
detail can be very dependent on the nature of [14] British Pharmacopoeia, Vol. II, A82 and A84.
the sample, the type of analytical methodology HMSO, London (1988).
[15] J.E. Knoll, J. Chromatogr. Sci. 23, 422-425 (1985).
and the purpose of carrying out the test. [16] J.C. Wahlich and G.P. Carr, J. Pharm. Biomed.
Some indications are provided in this presen- Anal. 8, 619-623 (1990).
tation of approaches which may be adopted for [Received for review 5 April 1990;
validating various types of analytical method, revised manuscript received I June 1990]