dx.doi.org/10.

14227/DT120305P19

USP Dissolution Test with Pooled Samples

Statistical Analysis of the Third Stage
Nora S. Meneces1, Carlos D. Saccone1,2 and Julio Tessore1
email: csaccone@fing.edu.uy

Abstract
Dissolution test as defined by USP is an important determination in assessing the quality of different pharmaceutical
forms. Some years ago, the pooled samples test was introduced. This test is carried out in three stages like the test for indi-
vidual samples, but it uses different acceptance criteria. Some doubts have been raised regarding calculation of the third
stage average. In this paper, the statistical precision of an equation for this calculation is discussed. Average dissolution val-
ues used to assess compliance are affected by several sources of variability: measurement, dissolved amount unit to unit
differences, etc. These sources of variability were studied and preferred zones are described for different values of those
components, to calculate third stage averages. The results obtained were verified with a simulation study. It was demon-
strated that it is possible to reduce the number of units tested in the third stage without introducing a noticeable increase
of the risks in the acceptance/rejection decisions.

Introduction • Product variability: variability between units of the prod-

I
n the third stage of the USP Dissolution Test with pooled uct refers to the dissolved amount in the dissolution test
samples (1),(criteria shown in Table 1),the average value (i.e. variability of the concentration of analyte present in
of the lot submitted to inspection is compared with the the vessel at the end of the dissolution test). This compo-
value of the parameter Q specified in the monograph. When nent was described using the RSD (relative standard
the third stage is reached,the calculation of the average of deviation). A RSD range of 1-10% covering the usual val-
the 3 values obtained in the three stages,can originate the ues (2) was studied.
question if the value of the third stage must be weighted • Variability (or uncertainty) due to measurements, includ-
double taking in account the fact that they are twelve units ing:
instead of six. The following aspects are investigated in this • Variability of aliquot volumes used to make the pooled
paper: sample. This variability has been modeled using CVp
(coefficient of variation of volume delivered); its usual
a. Which is the best estimate of the mean value (weighted value was considered to be 0.01 (1%).
vs. non-weighted estimations), • Variability related to the operation of adjusting pooled
b. Which are the consequences of applying an expression volume. This variability has been modeled using CVe
that differs from the optimum for estimation of the (coefficient of variation of the filling of the volumetric
mean, flask); its usual value was considered to be 0.005
c. Which would be the consequences of taking a sample (0.5%).
of only six units also in the third stage instead of twelve. • Variability related to the measurement of the concen-
tration of the analyte in the pooled solution. This has
Discussion been modeled using CVa (coefficient of variation of
To perform this work a statistical model of the inspection analysis); figures taken were between 0.005 and 0.03
has been built,including the following sources of variability (0.5-3%) in order to consider the different analytical
in the results of each stage: procedures that might be used.

Performing the usual approximation, where:

any and any , variability of
results of each stage is obtained:

(1)

1University of the Republic of Uruguay, Faculty of Engineering, Industrial 2 Correspondingauthor, Facultad de Ingeniería, Instituto de Ingeniería
Production Department, Montevideo, Uruguay Mecánica y Producción Industrial, Julio Herrera y Reissig 565, Montevideo,
Uruguay

18 Dissolution Technologies | AUGUST 2005

 Figure 1. Differences [CV 2m (w = 1/4)− CV2m (w = 1/3)] % as a function of
CVa for different RSD.

(2)

Where:
R1, R2, R3 are the pooled results obtained in stage 1, 2 and 3.
CVR1 is coefficient of variation of result in stage 1, CVR2 is
assumed to be the same as CVR1.
CVR3 is coefficient of variation of result in stage 3, and
RSD is expressed as a percentage.
With the results of stages 1, 2 and 3, the mean value of
the lot is estimated (m) and the decision of acceptance or
rejection is taken by comparing this value with Q.
The method established by USP gives equal weights to
the three results despite the fact that they have been
obtained with different numbers of dosage units.
Figure 2. Weighting factors preferred zones (CVp = 0.01 and CVe = 0.005).
In a general way, weighting of the results can be
expressed as follows:
The partial derivative of equation 4 with respect to w is:
(3)
In this equation the following considerations have been (5)
assumed:
Then it can be obtained the value of w, which makes
• Same weighting is given to results 1 and 2 because they minimum the variability of the calculated average value m,
have similar variability, and therefore the most efficient estimator of the mean (3).
• Result for stage 3 is weighted by a factor (1-2w), which The expression is as follows:
assures that the three factors add to 1 and m is an aver-
age of the results obtained. (6)

The variability of the calculated mean value (m) is given by:

A value of w = 1/3, indicates that the best estimation is
(4) obtained giving to the three results the same weighting,

Dissolution Technologies | AUGUST 2005

19
 while a value of w = 1/4 indicates that in order to obtain
minimum variability in m (estimation of mean lot value),
double weighting must be given to the stage 3 result.
In order to quantify the results, w has been studied as a
function of the analytical parameters and the variability of
the lot expressed as RSD. Data are shown in Table 2.
Values obtained show that:
• There is not a unique optimum value of w in the range of
parameters considered,
• When the variability of m is dominated by measurement
(low RSD, high CVa) the optimum w value approaches 1/3,
but when product variability dominates (high RSD, low
CVa), optimum value of w approaches 1/4. If measurement
variability were negligible, optimum value of w will be 1/4,
which means double weighting for the stage 3 value. is preferred. This curve was obtained equating variability
calculated with w = 1/3 and w = 1/4. Points on the curve
To assess the loss of efficiency caused by using a fixed have the same variability.
value of w, lot average variabilities obtained with an opti- It can be seen in Table 3 that total variability coefficient is
mum value of w were compared with those obtained using a function of RSD and CVa.
values of 1/3 and 1/4 (data are shown in Table 3). Due to the analysis variability contribution, an increase in
From the analysis of the table it is shown that: sample size (6 to 12) reduces variability of the stage 3 aver-
age less than could be expected if analysis variability con-
• Variability of measurements is key to determining which tribution didn’t exist.
weighting is more convenient. A more detailed analysis In order to evaluate the improvement due to additional
is shown in Figures 1and 2. samples, Table 4 has been prepared. In these calculations,
• The loss of efficiency when w = 1/3 is used (compared the same values of CVe and CVp were used and a weighting
with the optimum w value) is not significant; in the worst factor of w = 1/3 was applied.
case, variability increases 4.8%. The reduction of variability due to additional samples is
less than 9,1% in all the ranges of parameters studied.
Figure 1 shows differences in variability due to weighting
factor as a function of RSD (lot variability) and CVa (analysis Verification
variability). The difference [CV2m (w = 1/4)- CV2m (w = 1/3)] In order to verify the theoretical analysis, simulations
% is plotted as a function of CVa for different RSD (CVp = were performed taking into account the statistical model
0.01 and CVe =0.005). Above the x-axis, CVm (w = 1/4) > outlined.
CVm (w = 1/3) and, therefore, w = 1/3 should be preferred. Each value shown in Table 5 is the average of 1000 repe-
Below the x-axis, w = 1/4 is preferred. titions and the resultant confidence interval (95%) is
Figure 2 shows preferred zones as a function of RSD and approximately 0.06.
CVa. In the zone above (below) the curve w = 1/4 (w = 1/3) The results of the simulation are in agreement with cal-

20 Dissolution Technologies | AUGUST 2005

culations within the confidence intervals. The maximum
observed difference (a = 1/4, RSD = 2 and Cva = 0.005) is
5.9%.
The Dissolution
Conclusions
The optimum weighting factor is variable. Depending on
Solutions Network
the relative values of the sources of variability, equal
weighting (as established by USP 27) or double weighting
third stage, provides estimators of lot mean with less vari-
ability. In any case, loss of efficiency (increase in variability)
due to the use of a fixed, equal weighting to calculate aver-
age is not significant. Third stage testing could be reduced
to six units without introducing a noticeable increase of
risks in the acceptance/rejection decisions since there is lit-
tle increase in the variability of the estimated average.

Acknowledgement
The authors wish to thank Ms. Elena Saccone for her lan-
guage revisions of the manuscript.

References
1. THE UNITED STATED PHARMACOPOEIA 27, The U. S.
Pharmacopoeia Convention, Inc., Board of Trustees,
Webcom Limited, Toronto, Ontario, 2303-2304, 2004.
2. Hofer, Jeffrey D., Gray Vivian A,“Examination of
Selection of Immediate Release Dissolution
Acceptance Criteria”, Dissolution Technologies, 10 (1,
February), 16–20, 2003.
3. Montgomery, D.,“Introduction to statistical quality con-
trol”, 4th ed., John Wiley & Sons, Inc., New York, NY, 91,
2001.

Dissolution Technologies | AUGUST 2005

21