Journal of Bioanalysis & Biomedicine - Open Access Research Article

OPEN ACCESS Freely available online

doi:10.4172/1948-593X.1000026

JBABM/Vol.2 Issue 4

A Note on the Calculation of Intrasubject Coefficient of

Variation in Bioequivalence Trails
Mahmoud Abdel Mohsen
Pharmaceutical Research Unit (PRU), Royal Scientific Society (RSS), P.O. Box 1438, Amman 11941, Jordan

Abstract
Bioequivalence studies are generally performed as crossover studies and, therefore, information in the Intrasubject
Coefficient of Variation (CV) is needed for sample size planning. Recently, a confusing point was noticed in calculating the
Intrasubject Coefficient of Variation (CV) in crossover studies under the additive model (i.e. under normality assumption
and the multiplicative mode (i.e under logarithmic distribution). The aim of this paper is to clarify this confusion. Methods
used in calculating the Intrasubject Coefficient of Variation (CV) are reviewed in this paper from a statistical point of view.

Keywords: Coefficient of variation; Bioequivalence; Crossover The Intrasubject (CV) can be estimated from the analysis of
design; Sample size; Normal and log-normal distribution variance of the parameters: AUC0, Cmax, and Cmax/ AUC0 based on
2

Introduction the residual mean square for error  .

e

Bioavailability (BA) and Bioequivalence (BE) studies investigate Theorem (1): Let X1, X2….Xn be a sample from normal distribution,
and compare pharmacological characteristics of different drug that is X ~ N (2) then the coefficient of variation of X is defined as:
formulations in terms of the rate and the extent of absorption of var iance(X)
active ingredients. Coefficient of Variation (CV) =
mean(X )
The most common pharmacokinetic responses of the rate and the var(X ) 
extent of absorption are: Area under the plasma concentration-time CV=  .
E (X ) 
curve (AUC), maximum plasma concentration (Cmax), and time to reach
2
maximum plasma concentration (tmax). In BE studies    e , μ = μR, where μR is the mean of the reference
formulation. Hence, the Intrasubject Coefficient of Variation (CV)
The distributions of responses such as AUC and Cmax are often
under raw data normality assumption is
positively skewed and exhibit a lack of homogeneity of variances.
e
Therefore, normality assumption on AUC and Cmax may not be CV = ,
suitable. In this case, the assessment of average BE based on raw R
data and normality assumptions may not be appropriate. Logarithmic Which can be estimated by

transformation is the commonly chosen data transformation MSE
CV  (1)
technique for BE studies and is also recommended by the FDA XR
“Guidance on Statistical Procedures for Bioequivalence Studies using Where:
a Standard Two-Treatment Crossover Design (July 1992)”. MSE is the mean square error for within subject variability (not
The most important variation in crossover designs is the for subject within sequence error), X R is the least square (LS) mean for
intrasubject or within-subject variance, which is the variation, the reference formulation.
exhibited by a single person when given the same dose of a drug Multiplication of (1) by 100 gives the percent Intrasubject CV for
over repeated administrations. random variable X (raw data, normality assumption).
The magnitude of the intrasubject variance depends on the Theorem (2): Let X1, X2, …..., Xn, be a sample from log-normal
pharmacokinetics of the drug itself. For instance, drugs with simple distribution; that is Ln(X) ~ N (, ). Then, the Coefficient of
kinetics (e.g. with little metabolism or well absorption) will generally Variation of (X) is defined as:
have estimated intrasubject coefficient of variations (CVs) of less than
20%, whereas drugs with complicated kinetics (e.g. highly variable Cov (X) = exp( 2 )  1
absorption) can have estimated coefficient of variations (CVs) that are
equal to or greater than 40%. Therefore, information on intrasubject
coefficient of variation (CV) is needed for sample size planning of *Corresponding author: Dr. Mahmoud Abdel Mohsen, Division of Biometrics,
future studies. Pharmaceutical Research Unit, Royal Scientific Society, P.O. Box 1438,
Amman 11941, Jordan, Tel: +962-6-5358290; Fax: +962-6-5358261; E-mail:
We believe that (Chow and Liu, 2000) in Chapter (6) of [email protected]
“Transformation and Analysis of Individual Subjects Ratios” had Received July 13, 2010; Accepted August 24, 2010; Published August 24, 2010
erroneously evaluated the Intrasubject Coefficient of Variation (CV) Citation: Mohsen MA (2010) A Note on the Calculation of Intrasubject Coefficient of
using a certain example. In this note, we will suggest a better formula Variation in Bioequivalence Trails. J Bioanal Biomed 2: 075-078. doi:10.4172/1948-
in finding the Intrasubject (CV) under the two models of normality 593X.1000026
assumptions: Copyright: © 2010 Mohsen MA. This is an open-access article distributed under
the terms of the Creative Commons Attribution License, which permits unrestricted
1. Additive model (raw data model).
use, distribution, and reproduction in any medium, provided the original author and
2. Multiplicative model (log-transformed model). source are credited.

J Bioanal Biomed
Volume 2(4) : 075-078 (2010) - 075
ISSN:1948-593X JBABM, an open access journal
Citation: Mohsen MA (2010) A Note on the Calculation of Intrasubject Coefficient of Variation in Bioequivalence Trails. J Bioanal Biomed 2: 075-078.
doi:10.4172/1948-593X.1000026

Proof: Now let X1, X2….., Xn be a sample from the log-normal is the (1 - ) level confidence interval for the intrasubject coefficient
distribution then Y1= Ln(X1), …..Yn= Ln(Xn) is a sample from a N of variation (CV).
() distribution.
The measure of sensitivity
Straight forward computation yield:
The index of measure of sensitivity is defined as the ratio of the
2 width of the average BE interval (2 = U – L), to standard error
E (X) = exp (   ) and
2 of the difference in (XT  X R ) least square means for the Test and the
Var (X) = exp (2 ) exp (() – 1),

Reference formulations, that is:
Then by using Theorem (1), the Coefficient of Variation (CV) of X 2

is defined as:
 e21 1
(  )
var iance (X ) 2 n1 n2
Coefficient of Variation (CV) = 
E (X ) 1. Under raw data, normality assumption, and according to the
± 20 rule,   20% , therefore,  can be written in terms of
2

exp ( + ) exp (( 2 ) - 1) R
2  exp ( 2 ) - 1 
CV  e as follows:
2 R
exp ( + )
2 2
R 40
In BE studies,  2   e2 , where:  e2 is within subject variability.   (4)
 e2 1 1 1 1 1
Hence, the Intrasubject CV under log-transformed data, normality (  ) CV (  )
2 R2 n1 n2 2 n1 n2
assumption is
CV  exp( e2 )  1 , (See for example in (Chow and Liu, 2000).

This can be estimated by 2. Under Log transformed data, normality assumption, and
 according to the equivalence limit = (Ln (0.8), Ln (1.25)),  can
CV  exp(MSE )  1 (2)
be written in terms of CV  exp( e2 )  1 as follows:
Where: MSE is the mean square error for within subject variability
(not for subject within sequence). 2Ln(1.25) 0.44629
  (5)
2
Ln(CV  1) 1 1 Ln(CV 2  1) 1 1
Multiplication of (2) by 100 gives the percent Intrasubject CV for (  ) (  )
a random variable X (log-transformed data, normality assumption). 2 n1 n2 2 n1 n2

Remark In the two cases, this index () is a function that depends on the
Now we can distinguish between (formula 1) and (formula 2) in Coefficient of Variation (CV) and the sample sizes; n1 and n2 used
finding the Coefficient of Variation (CV) for a given random variable. in BE studies.

Under raw data, (formula 1) depends on two unknown parameters (Schuirmann, 1987) used this index to compare the size of the
 e2and R. On the other hand, under log-transformed data, (formula two one-sided test procedures to the power approach.
2) depends on one unknown parameter  e2 .
Total Sample CV
Consequently, in order to build a good BE study, we recommend Size (n1+n2) 10% 15% 20% 25% 30% 35% 40%
using the log-transformed data transformation in crossover designs. 8 8.00 5.33 4.00 3.20 2.67 2.29 2.00
10 8.94 5.96 4.47 3.58 2.98 2.56 2.34
If one is interested in finding the (1 - ) level confidence interval 12 9.80 6.53 4.90 3.92 3.27 2.80 2.45
for the intrasubject coefficient of variation (CV) under log-normal 14 10.58 7.06 5.29 4.23 3.53 3.02 2.65
distribution in a 2x2 crossover design, one can use the following: 16 11.31 7.54 5.66 4.53 3.77 3.23 2.83
18 12.00 8.00 6.00 4.80 4.00 3.43 3.00
(Theorem 3): Assuming a log-normal distribution and a 2x2 20 12.65 8.43 6.32 5.06 4.22 3.61 3.16
crossover design, it is known that a (1 - ) level confidence interval 22 13.27 8.84 6.63 5.31 4.42 3.79 3.32
for  e2 is [ Le, Ue ]. 24 13.86 9.24 6.93 5.54 4.62 3.96 3.46

Table 1: Values of the Index of Sensitivity.

Where:
 
(n1  n2  2) 2 (n1  n2  2) 2 Total Sample CV
Le  e and U e  e Size (n1+n2)
2  2  10% 15% 20% 25% 30% 35% 40%
 (n1  n2  2,1  )  (n1  n2  2, ) 8 8.95 5.98 4.51 3.63 3.04 2.63 2.31
2 2 10 10.00 6.69 5.04 4.05 3.39 2.94 2.59
Where X2(n1 + n2 − 2) denotes the cumulative distribution function 12 10.96 7.33 5.52 4.44 3.72 3.22 2.84
14 11.84 7.92 5.96 4.79 4.02 3.47 3.06
16 12.65 8.46 6.37 5.13 4.30 3.71 3.28
18 13.42 8.98 6.76 5.44 4.56 3.94 3.48
of a central chi-squared distribution with (n1+n2 − 2) degrees of 20 14.15 9.46 7.13 5.73 4.81 4.15 3.66
freedom. Thus 22 14.84 9.92 7.47 6.01 5.04 4.35 3.84
24 15.49 10.36 7.81 6.28 5.27 4.55 4.01
( exp(Le )  1, exp(U e )  1) (3) Table 2: Values of the Index of Sensitivity.

J Bioanal Biomed
Volume 2(4) : 075-078 (2010) - 076
ISSN:1948-593X JBABM, an open access journal
Journal of Bioanalysis & Biomedicine - Open Access Research Article
OPEN ACCESS Freely available online
doi:10.4172/1948-593X.1000026

JBABM/Vol.2 Issue 4

Period I Period II sequences from (Chow and Liu, 2000).

Sequence Subject
AUC Cmax AUC Cmax
Nine subjects were randomized to sequence (TR) (1-9), while the
1 2.52 1.04 5.47 1.58
2 8.87 3.80 4.48 1.35 remaining nine subjects were randomized to sequence (RT) (10-18). A
3 0.79 0.12 2.25 0.77 one-week washout period separated the two periods. Blood samples
4 1.68 0.62 1.82 1.23 were taken just before drug administration and then at: 0.50, 1.00,
TR 5 6.95 2.92 7.87 2.25 1.50, 2.00, 3.00, 4.00, 6.00, and 8.00 hours post-dosing. The primary
6 1.05 0.16 3.25 0.85
7 0.99 0.28 12.39 3.00
summary variables of interest, were AUC(0-8) and Cmax. The data are
8 5.60 2.62 4.77 2.05 given in the tables, Bioequivalence data set from (Clayton and Leslie,
9 3.16 1.51 1.88 0.60 1981).
10 4.98 1.61 3.19 1.37
11 7.14 1.98 9.83 3.74 The results of the Analysis of Variance (ANOVA) on log-transformed
12 1.81 0.56 2.91 1.16 and untransformed data are given in (Tables 4, 5, 6, and 7) for the two
13 7.34 2.40 4.58 1.79 parameters; AUC(0-8) and Cmax.
RT 14 4.25 1.04 7.05 2.51
15 6.66 2.67 3.41 1.55 Comments on the results obtained by (Chow and Liu, 2000)
16 4.76 1.47 2.49 0.82
17 7.16 1.90 6.18 2.57 concerning the analysis of variance of the log-transformed
18 5.52 2.47 2.85 1.07 data for AUC(0-8)
Table 3:
(Chow and Liu, 2000) concluded the following statement: “The
Source D.F SS MS F-value p-value estimated CV is about 36.55%, which is high, but not unusual in
Subjects 17 9.64872 0.56757 1.8368 0.11543 bioavailability and bioequivalence studies. The index of sensitivity
Sequence 1 1.30529 1.30529 2.5031 0.13319 defined in (Equation 5.3.10) is estimated to be about 3.28. Therefore,
Subject (sequence) 16 8.34343 0.52146 1.6876 0.15275
as discussed in (section 5.3.2), this study has little power to conclude
Periods 1 0.19586 0.19586 0.6338 0.43762
Treatments 1 1.04699 1.04699 3.3883 0.08428
bioequivalence if (Schuirmann, 1987) two one sided test procedures
Residual 16 4.94398 0.30899 - - and the ± 20% rule are applied.”
Total 35 15.83555
(Chow and Liu, 2000) assumed that the results obtained from
Table 4: Analysis of Variance for in (AUC(0-8)).
the analysis of variance for log-transformed data are as the results
Source D.F SS MS F-value p-value obtained from the analysis of variance of raw data. Accordingly,
Subjects 17 149.64430 8.80260 1.4690 0.22358 they computed the intrasubject coefficient of variation (CV) and the
Sequence 1 7.07560 7.07560 0.7941 0.38607 measure of sensitivity using (Equations 1 and 4):
Subject (sequence) 16 142.56870 8.91054 1.4870 0.21808
Periods 1 0.93444 0.93444 0.1559 0.69817  MSE 0.30899
Treatments 1 11.17788 11.17788 1.8653 0.19090 % CV   100%   100%  36.55%
Residual 16 95.87808 5.99238 - - XR 1.5209
Total 35 257.63470
40 40
Table 5: Analysis of Variance for AUC(0-8).    3.28
1 1 1 1 1 1
Under raw data, normality assumption, (Chow and Liu, 2000) CV (  ) 36.55 (  )
2 n1 n2 2 9 9
calculated values of  for different combinations of CVs, n1, and n 2
as shown in (Table 1). Source D.F SS MS F-value p-value
Under log-transformed data, normality assumption, we Subjects 17 13.59333 0.79961 1.6484 0.16186
Sequence 1 1.85384 1.85384 2.5266 0.13150
recalculated values of  for different combinations of CVs, n1, and
Subject (sequence) 16 11.73949 0.73372 1.5125 0.20841
n2 as shown in (Table 2). Periods 1 0.60542 0.60542 1.2481 0.28042
Treatments 1 0.54057 0.60542 1.1144 0.30681
From the two (Tables 1 and 2), the results indicate that  Residual 16 7.76141 0.48509 - -
increases as sample size increases and as CV decreases. Total 35 22.50073

(Chow and Liu, 2000) in (Chapter 6) of “Transformation and Table 6: Analysis of Variance for in Cmax.
Analysis of Individual Subjects Ratios”, considered the following Source D.F SS MS F-value p-value
example from (Clayton and Leslie, 1981). Subjects 17 18.21633 1.07155 1.2875 0.30888
Sequence 1 0.97680 0.97680 0.9066 0.35518
Two Erythromycin formulations were compared using the tow- Subject (sequence) 16 17.23952 1.07747 1.2946 0.30581
square, two-period, two-treatment crossover design involving 18 Periods 1 0.03003 0.03003 0.0397 0.84458
healthy subjects. The two formulations were: Treatments 1 0.00047 0.00047 0.0006 0.98076
Residual 16 13.31608 0.83225 - -
T: Erythromycin stearate [Erythrocin®, 500 mg Ovaloid Total 35 31.56588
Tablets, 6316, Abbott Australasia pty. Ltd.], and Table 7: Analysis of Variance for Cmax.
R: Erythromycin base [Eryc , 2X250 mg Capsules, containing
®
2 
enteric-coated pellets, F.H Faulding & Co. Ltd.]. Parameter XR e % CV
Cmax 1.65444 0.83225 55.14 217.63
The study technique of the analysis of variance was used to test
ln (Cmax) 0.39249 0.48509 79.01 1.92
for treatment differences. However, (Clayton and Leslie, 1981) did AUC(0-8) 5.23111 5.99238 46.80 256.41
not provide the information for the sequence assignment of subjects. ln (AUC(0-8)) 1.52090 0.30899 60.17 2.41
For the purpose of illustration, we will adapt the order of periods and Table 8:

J Bioanal Biomed
Volume 2(4) : 075-078 (2010) - 077
ISSN:1948-593X JBABM, an open access journal
Citation: Mohsen MA (2010) A Note on the Calculation of Intrasubject Coefficient of Variation in Bioequivalence Trails. J Bioanal Biomed 2: 075-078.
doi:10.4172/1948-593X.1000026

On the other hand, different results for CV and  are obtained using Discussion
(Equations 2 and 5):
 In this note, we try to clarify the confusion in finding the
% CV  exp(MSE )  1  100%  60.17% intrasubject coefficient of variation (CV) based on the additive model
(raw data model) and the multiplicative model (log-transformed
0.44629 model). Since the information on intrasubject coefficient of variation
  2.41
is still needed for sample size planning in future studies, one must
ln(CV 2  1) 1 1
(  ) specify the model (raw data or log-transformed data) in the crossover
2 n1 n2 design and then choose the suitable equation for the calculation of
coefficient of variation (CV).
Also in (page 195), (Chow and Liu, 2000) gave summary of points and
interval estimation of intrasubject variability for data sets from the References
Analysis of Variance of ln (AUC(0-8)). That is; for the parameter  e2 , 1. Chow SC, Liu J (2000) Design and Analysis of Bioavailability and Bioequivalence
Studies. (2ndedn), New York, Marcel Dekker.
the point estimate = 0.3090 and 90% confidence interval for  e2 is
2. Clayton D, Leslie A (1981) The bioavailability of erythromycin stearate versus
[Le, Ue] = [0.17, 0.72] if one interested to find the 95% confidence
enteric-coated erythromycin base when taken immediately before and after. J
interval for the parameter CV, the point estimation of CV = 60.17%, Int Med Res 9: 470-477.
then according to (Equation 3), the 95% confidence interval for CV is:
3. Schuirmann DJ (1987) A Comparison of the two one-sided tests procedures and
the power approach for assessing the equivalence of average bioavailability. J
exp(Le )  1, exp(U e )  1 = (43.05%, 102.69%). Pharmacokinet Biopharm 15: 657-680.

J Bioanal Biomed
Volume 2(4) : 075-078 (2010) - 078
ISSN:1948-593X JBABM, an open access journal