Particle Size Limits to Meet USP Content Uniformity

Criteria for Tablets and Capsules

BRIAN R. ROHRS, GREGORY E. AMIDON, RICHARD H. MEURY, PAMELA J. SECREAST,
HARRY M. KING, CONNIE J. SKOUG
Pharmaceutical Sciences, Pfizer, 7000 Portage Rd, Kalamazoo, Michigan 49001

Received 19 September 2005; revised 2 December 2005; accepted 6 January 2006

Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/jps.20587

ABSTRACT: Content uniformity (CU) of pharmaceutical dosage units can be affected by

active pharmaceutical ingredient (API) particle size and size distribution. Previous
authors have estimated this impact but use of different particle size descriptors led to
confusion and difficulty in applying the theoretical models developed. We show that when
the same descriptors for particle size and distribution are used (i.e., median diameter on a
weight basis (d50) and geometric standard deviation (sg)), previously published models
are consistent. The approach of Yalkowsky and Bolton4 [Pharm Res 7:962–966, 1990] is
modified to use these descriptors and updated for current USP28/NF23 CU criteria. A
nomograph is provided to allow easy estimation of an acceptable d50 for a given dose and
sg. To test the model’s validity, tablets were manufactured over a wide range of doses and
assayed for CU. As predicted, %relative standard deviation (RSD) increased as dose
decreased. However, for APIs that deviate significantly from the assumed log-normal
distribution, sg is more appropriately described by the upper region of the API size distri-
bution, presumably because large particles have greatest influence on CU. At very low
doses, CU values deviate significantly from normality, consistent with the presence of
single large API particles causing super-potent dosage units. ß 2006 Wiley-Liss, Inc. and
the American Pharmacists Association J Pharm Sci 95:1049–1059, 2006
Keywords: content uniformity; particle size; tablet; capsule

INTRODUCTION regarding particle size need to be made. CU, in

Particle size is an important consideration in the
development of solid dosage forms since it can
have a significant impact on such critical product
characteristics as dissolution rate and dosage unit
content uniformity (CU). In the early stages of
development, the active pharmaceutical ingredi-
ent (API) supply may be limited, yet decisions
Brian R. Rohrs’s and Harry M. King’s present address is
Bausch & Lomb, 1400 North Goodman Street, Rochester, New
York 14609.
Richard H. Meury’s present address is Analytical Sciences
Research and Development, Eli Lilly and Co., Indianapolis,
Indiana 46285.
Correspondence to: Brian R. Rohrs (Telephone: 585-338-
5394; Fax: 585-338-8019;
E-mail: Brian_R_Rohrs@bausch.com)
Journal of Pharmaceutical Sciences, Vol. 95, 1049–1059 (2006)
ß 2006 Wiley-Liss, Inc. and the American Pharmacists Association
particular, can be a problem for low dose drugs
and appropriate selection of particle size for early
clinical supplies is needed, yet a convenient way of
estimating particle size requirements to achieve
CU is not yet available.
A review of the literature reveals several
significant publications that address particle size
and CU from a theoretical and statistical view-
point. Johnson1 derived an equation that calcu-
lates the expected variation in a unit dose when the
particle size distribution is known. This equation
allows one to move forward in a predictive sense
from a characterized powder to an estimate of CU.
Validation of the conceptual approach has been
demonstrated through measurement of CU of
well-mixed powders.2,3 Johnson’s method is useful
in the development process as a troubleshooting
device. If, for example, a CU problem arises with a

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006 1049

1050 ROHRS ET AL.
product, one can use the equation to evaluate
whether the particle size distribution could be
responsible for the unit dose variation. If not,
particle size is eliminated as a potential cause and
one can focus efforts on other aspects of the
manufacturing process that can be a cause of poor
CU such as inadequate mixing or segregation.
Although Johnson’s approach is useful if one
has a specifically characterized powder distribu-
tion, it would also be useful to have a more general
method to allow one to move backwards from a
desired limit on unit dose variation to an appro-
priate particle size distribution. In this way, one
can a priori set particle size limits for a given dose
and eliminate altogether problems of CU asso-
ciated with particle size. In a 1990 paper, Yalk-
owsky and Bolton4 provided the basis for such
a method. They used the predicted variation
in CU to calculate the probability of passing USP
CU criteria. Since 1990, however, the USP CU
criteria have changed, and the acceptance criteria
now also differ between tablets and capsules.
In addition, after the Yalkowsky–Bolton paper
was published, H. Egermann5, in a Pharmaceu-
tical Research Letter to the Editor, stated that
Yalkowsky and Bolton’s approach is ‘‘inaccurate
and misleading’’ and that Johnson’s method
should be used. In fact, this misunderstanding is
related to the differences in the particle size
distribution descriptor-basis from which the two
methods are derived. The Yalkowsky–Bolton
method reports limits as arithmetic means on a
number basis, whereas Johnson’s method is
derived from a volume basis. We demonstrate
(Appendix I) that in fact, if a log-normal distribu-
tion is assumed, the methods are identical when
expressed with the same particle size distribution
descriptors.
The objective of this paper is to provide a
reliable method of estimating particle size needs
based on mean particle size, particle distribution
width, and dose. In this report, the Yalkowsky–
Bolton approach is used to estimate appropriate
particle size limits, but takes into account current
USP28/NF23 criteria for tablet and capsule CU.
We also convert the suggested limits to metrics
commonly used to describe a particle size distribu-
tion, the median diameter on a weight basis (d50),
and the geometric standard deviation (sg) as the
descriptor of the distribution width. CU data on
tablet lots manufactured from two-particle size
distributions are used to test the limits and provide
guidance on the selection of the width parameter
from the actual distribution.
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006
For the benefit of readers who do not wish to
be buried in the theoretical and calculation
details, a practical product of this report is
the nomograph provided in Figure 3, which allows
the reader to easily estimate particle size (d50)
distribution width (sg ) and dose (D) needed to
have a 99% probability of passing USP Stage I
CU criteria.
MATERIALS AND METHODS
Materials
Microcrystalline cellulose, NF (PH 101, FMC
Biopolymer, Philadelphia, PA), lactose monohy-
drate-modified spray process standard, NF (315,
Foremost Farms, Inc., Baraboo, WI), pregelati-
nized starch NF (Starch 1500, Colorcon, West
Point, PA), croscarmellose sodium, NF (FMC
Biopolymer, Ac-Di-Sol), colloidal silicon dioxide
NF (Cab-O-Sil, Cabot, Tuscola, IL), and magne-
sium stearate NF powder food grade (Mallinck-
rodt Chemicals, Hazelwood, MD) were used as
received from the supplier. The API was milled
using a Bepex PCS-10 Pulvercron mill (Bepex
International LLC, Minneapolis, MN) prior to
use.
Particle Size Measurement
A representative test sample of API was obtained
by randomly selecting multiple times from the
bulk API container until a total of 150–200 mg
API was obtained. This sample was dispersed as
primary particles in 20 mL of 1,2-dichloroethane.
The particle size distribution was measured with
a Sympatec HELOS light diffraction particle-sizer
(Sympatec GmbH, Clausthal-Zellerfeld, Germany),
fitted with a 200-mm lens providing a sizing
range of 0.5–350 microns and analyzed using the
Fraunhofer diffraction model. The particle size
was considered large enough that the Fraunhofer
diffraction model would yield satisfactory results
with the low angle scatter instrument used in this
study. Results obtained by light diffraction were
qualitatively consistent with those obtained by
optical microscopy. Particle size distributions for
lots A and B are shown in Figure 1 with d50’s
determined to be 22 mm and 29 mm, respectively.
Tablet Materials and Manufacture
The API and 230 g pregelatinized starch were
preblended by hand in a polyethylene bag for
DOI 10.1002/jps
 PARTICLE SIZE LIMITS TO MEET USP CONTENT UNIFORMITY CRITERIA
Figure 1. Particle size distributions of API lots used
in tablet manufacture.
approximately 1 min to disperse the API. This
premix and other excipients except for magne-
sium stearate were screened using a Comil model
197S (Quadro Engineering, Milburn, NJ) equip-
ped with a 2A045R031/37 screen and 2A1601-
173 impeller with a 0.12500 spacer at 2000 rpm.
The screened materials were placed into a one
cubic foot V blender, and mixed for 15 min at
26 rpm. A portion of the blend was removed from
the blender and preblended with magnesium
stearate in a polyethylene bag by hand for
approximately 1 min, returned to the blender
and mixed for an additional 2 min. Tablets were
compressed on a Picolla rotary tablet machine
(Riva SA, Buenos Aires, Argentina, Model B,
10 Station) equipped with a force feeder and using
0.304800
0.300200 pentagonal tooling. Tablets
were compressed to a theoretical weight of
150 mg (3%) using approximately 6.7 kN of
compression force at a press speed of 18 rpm.
About 150 consecutive tablets were collected for
assay once the established tableting parameters
were reached and the equipment was running
smoothly. Consecutive tablets were collected to
minimize variance in CU that might be attributed
to segregation.
Analytical Assay
The amount of API in individual tablets was
quantified using HPLC (Agilent Technologies,
Palo Alto, CA, PH 1100) with UV detection at
280 nm.
The relative standard deviation of n ¼ 5
repeated injections for the lowest concentration
standard (0.236 mcg/mL) was 0.2%. Therefore,
analytical error contributed little to the variation
in CU results.
DOI 10.1002/jps
1051
THEORETICAL TREATMENT
To estimate the influence of particle size on
the probability of passing CU criteria, two
components are needed: the ability to predict the
variation in CU from the particle size distribu-
tion, and a way to take that variation in CU
and predict the probability of passing the USP
criteria.
Following the lead of Yalkowsky and Bolton’s
derivation,4 we first describe the relationship
between variation in CU and probability of meet-
ing the current USP CU criteria. After selecting
the appropriate population relative standard
deviation (RSD) to pass at Stage I, we then
used the Yalkowsky and Bolton method to predict,
as a function of dose, the desired particle size
distribution using conventional weight-based
descriptors.
Probabilities for Passing USP Content
Uniformity Criteria
The goal for this section is to find the standard
deviation of the CU distribution that enables USP
criteria to be met with a certain acceptable
probability (say p ¼ 99%). The population of
individual dosage units is assumed to be distrib-
uted normally about the mean with a standard
deviation, scu. If we assume the mean to be 100,
scu becomes numerically equivalent to the RSD
expressed as percent.
USP CU criteria differ between tablets and
capsules, and are listed in Table 1.
The following probabilities are defined:
P1L ¼ Probability for one unit that
85% 
 115%
P2L ¼ Probability for one unit that
75% 
< 85% OR 115 <
 125%
P1S ¼ Probability that RSD on 10 units  6:0%
P2S ¼ Probability that RSD on 30 units  7:8%
The probability of passing Stage I for tablets is:
P1S  P10
1L
This expression is translated as, the RSD is
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006
1052 ROHRS ET AL.

Table 1. USP28/NF23 Content Uniformity Acceptance Criteria for Tablets

and Capsules

Testing Tablets Capsules

Stage I Pass if: Pass if:
Test 10 None outside 85–115% and NMT 1 outside 85–115%
RSD  6.0% None outside 75–125% and
RSD  6.0%
Stage II Pass if, for all 30 units: Pass if, for all 30 units:
Test 20 NMT 1 outside 85–115% NMT 3 outside 85–115%
None outside 75–125% and None outside 75–125% and
RSD  7.8% RSD  7.8%

acceptable at Stage I and 10 of 10 units are within Functions exist within Microsoft Excel to calcu-
the 85%–115% range. late the probabilities. P1L and P2L are calculated
For Stage II tablets, the probability of passing via the cumulative normal distribution function
is: (NORMDIST in Excel). Therefore,


P2S  10  P29
1L  P2L þ ð1  P1S Þ  P1L 
10
P1L ¼ NORMDISTð115; 100; scu ; TRUEÞ
  NORMDISTð85; 100; scu ; TRUEÞ
1L þ 20  P1L  P2L Þ
P20 19

P2L ¼ NORMDISTð125; 100; scu ; TRUEÞ

The two terms inside the bracket derive from the
fact that there are two ways Stage I could have  NORMDISTð75; 100; scu ; TRUEÞ  P1L
failed, either by missing the 85–115 limit or by
To calculate P1S and P2S, we recognize that the
missing the 6% RSD criteria. For the first term, 1
ratio of (n-1)s2 to s2 is a random variable
of 10 tablets at Stage I fell outside the 85–115
distributed as w2 with (n-1) degrees of freedom:
range but inside the 75–85:115–125 limit. This
could have been any of the first 10 tablets, and ðn  1Þ  s^ 2
¼ w2ðn1Þ
since there are 10 ways to arrive at P91L  P2L , the s2
coefficient 10 is needed. At Stage II, 20 of 20
where s2 is the square of the sample standard
tablets have to be within the 85–115 range, so
deviation.6 To determine the probability of pas-
the first term is multiplied by P20 1L to yield sing Stage I or II RSD limits, the one-tailed
10  P91L  P2L  P20
1L or 10  P 29
1L  P 2L .
probability of the chi-squared distribution is
In the second term, 10 of 10 tablets could have
subtracted from unity. The Excel statistical
met the 85–115 limit, but failed the RSD criteria,
function CHIDIST is used, where the function
hence the factor ð1  P1S Þ  P10 1L . Moving to Stage II, CHIDIST(x,DOF) is the probability of w2ðDOFÞ > x
either 20 of 20 tablets could be within 85–115, or
with DOF degrees of freedom.
there are 20 ways that 19 of 20 tablets fall within
Therefore:
85–115 and 1 of 20 falls within 75–85:115–125.
 
Both the first and second terms are then multiplied 9  62
by the probability of having acceptable RSD on P1S ¼ 1  CHIDIST ; 9
s2cu
30 units, P2S.
Similar arguments can be made for capsules. and
The probability of passing Stage I is:  
29  7:82
  P2S ¼ 1  CHIDIST ; 29
P1S  P101L þ 10  P1L  P2L
9 s2cu

and Stage II is: for any scu.

 
1L  P2L
þ 1020  P1L  P2L
45  P28 2 27 3

P2S 
þð1  P1S Þ  P1L þ 30  P1L  P2L þ 390  P28
30 29
1L  P2L þ 3040  P1L  P2L
2 27 3

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006 DOI 10.1002/jps
 PARTICLE SIZE LIMITS TO MEET USP CONTENT UNIFORMITY CRITERIA 1053

The probability of passing USP Stage I and II Table 2. Population Relative Standard Deviation (%)
CU criteria versus the unit dose population to Pass USP CU Criteria at Various p-Values
relative standard deviation is illustrated in
Tablets Capsules
Figure 2A and B for tablets and capsules, Probability
respectively. The RSD values that give rise to Pass CU Stage I Stage II Stage I Stage II
various probabilities of passing CU are listed in
Table 2. 0.900 4.64 6.33 4.70 6.86
So, for example, to pass Stage I CU for tablets 0.950 4.33 6.03 4.38 6.59
0.990 3.84 5.46 3.87 6.12
with p  0.99, the relative standard deviation on the
0.999 3.39 4.84 3.41 5.65
entire population would have to be 3.84% or less.

where,
Modification of the Yalkowsky–Bolton Equation
Cv ¼ Coefficient of variation ðRSDÞ of the
Yalkowsky and Bolton4 derived an expression
for the variance of drug content based on dose ð%Þ
assumptions of homogeneous mixing and the D ¼ DoseðmgÞ
particle size distribution described by a log-
¼ Density of the drug ðg=ccÞ
normal function. That relationship, expressed in dm ¼ Arithmetic mean particle diameter on a
terms of the coefficient of variation (or RSD) is:
number basis ðmmÞ
p  r
1=2  12
1=2 9=
Cv ¼ 100   d3m  1 þ C2 10 2 C ¼ Coefficient of variation ðRSDÞ of the
6D
particle diameter distribution ðfractionÞ

The factor of 109/2 has been added to the original

equation to convert units.
The descriptors ‘‘dm’’ and ‘‘C’’ are the arithmetic
mean and arithmetic RSD, respectively, and not
routinely used to describe particle size distribu-
tions. It is useful to convert them to the more
widely used descriptors geometric mean diameter
on a weight basis d0g and geometric standard
deviation sg.
‘‘C’’ is converted as follows. For a log-normal
distribution described as f(d) ¼ LN[m,s], where
"   #
1 1 lnðdÞ   2
LNð; sÞ ¼ pffiffiffiffiffiffi  exp  
d  s  2p 2 s

s and the arithmetic RSD, ‘‘C,’’ are related7 by the

expression
 
1 þ C2 ¼ es
2

The geometric standard deviation ‘‘sg’’ and s are

Figure 2. Fractional probability of passing USP con-
tent uniformity criteria for (A) tablets and (B) capsules
versus relative standard deviation of the dosage unit
population.
related by the expression
sg ¼ es or s ¼ ln sg
Therefore, the relationship between the arith-
metic RSD ‘‘C’’ and the geometric standard devia-
tion ‘‘sg’’ for a log-normal distribution is:
  2
1 þ C2 ¼ es ¼ eðln sg Þðln sg Þ ¼ eln sg
2
Now, to convert from the arithmetic mean on a
number basis (dm) to the geometric mean on a

DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006
1054 ROHRS ET AL.

weight basis (d0g ), we can make use of the Hatch–

Choate equation8 to convert from one to the other:
2
dm ¼ d0g  e2:5 ln sg

Substituting the derived expressions for (1 þ C2)

and dm into the Yalkowsky and Bolton equation
leads to an equation expressed in terms of the
geometric mean on a weight (or volume) basis and
the geometric standard deviation:
p  r
1=2 2 2

1=2 9=
Cv ¼ 100   ðd0g  e2:5 ln g Þ3  ðeln g Þ12 10 2
6D
or
p  r
1=2 2

1=2 9
Cv ¼ 100   ðd0g Þ3  e4:5ln g 10 =2
6D
Since we want to predict d0g from the Cv
determined in the previous section, the equation
is rearranged to:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
   
0 3 6D 4:5ln 2

Cv 2
dg ¼ e g   103
p
100

Note that the geometric mean diameter d0g and

the more commonly used expression for the
volume median diameter d50 are interchangeable
since this equation assumes a log-normal function
that describes the particle size distribution.
RESULTS AND DISCUSSION
Figure 3. Maximum volume median particle dia-
meter (d50) in mm predicted to pass USP Stage I content
uniformity criteria with 99% confidence as a function of
dose (mg) and geometric standard deviation (sg). An
alternative estimation of the particle distribution width
d90/d50 assumes a log-normal distribution and is
calculated by (sg)1.28.

Nomograph
Using the modified Yalkowsky and Bolton equa-
tion, we can estimate the maximum acceptable
geometric mean diameter for a given particle
distribution width (sg), dose D, and assurance of
passing CU. Figure 3 shows the maximum
particle diameter/dose relationship (assuming
r ¼ 1) for passing Stage I tablet criteria with
p ¼ 0.99 for sg ¼ 1.0, 1.5, 2.0, 2.5, and 3.0. Table 2
shows that population RSD values to pass Stage I
are similar for tablets and capsules, and it is only
at Stage II that there is a significant difference.
Therefore, the nomograph in Figure 3 is applic-
able to both tablet and capsule formulations.
The utility of Figure 3 is that it can be used to
quickly estimate the maximum volume median
diameter (d50) on a weight basis. For example, to
pass USP Stage I CU criteria at a p-value 0.99, a
1 mg dose would have an upper limit on d50 of 70 mm
for sg ¼ 2. In fact, a useful feature of Figure 3
is the demonstration that by just reducing the
width of the particle size distribution, one gains a
significant increase in the maximum acceptable
d50. Estimating sg is addressed in a subsequent
section.
Comparison of Predictions to Data
To test the model predictions, mixes were pre-
pared from well-characterized API lots and
standard excipients, tablets compressed, then
assayed. The target doses ranged from 360 mg to
1.3 mg. Assay results are listed in Table 3. As
expected, the variation in CU increased with
decreasing dose.
The %RSD for CU can now be predicted from the
modified Yalkowsky–Bolton equation and com-
pared to the actual data. The descriptor of the
particle size distribution width is the geometric
standard deviation, sg, and its value is typically
estimated by the ratio (d84.1/d15.9)0.5. In this
manner, sg is estimated to be 2.50 and 2.77 for
API lots A and B, respectively. Using these values,
the theoretical %RSDs for CU are shown in

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006 DOI 10.1002/jps
 PARTICLE SIZE LIMITS TO MEET USP CONTENT UNIFORMITY CRITERIA 1055

Table 3. Content Uniformity Assay Results, n ¼ 30 estimate of sg be made. Conceptually, the presence
of a large API particle has a much greater impact
API Lot Target (mg) Mean (mg) %RSD
on CU than that of a small one. Estimating sg from
A 360 343.1 1.27 the d84.1/d15.9 ratio for the API lots used to
A 90 85.30 2.18 manufacture the tablets puts a greater emphasis
A 90 86.20 2.29 on the smaller particles because of the negatively
A 33 31.40 3.39 skewed distributions. To better evaluate the
A 8.2 7.70 4.75 theoretical predictions, sg should be estimated
A 3.6 3.50 12.27 from the portion of the distribution containing the
A 3.6 3.55 6.60
large particles.
B 15 14.22 2.35
The width descriptor sg can be estimated from
B 15 14.33 3.25
B 5 4.92 5.23 any part of the curve by the equation
B 1.3 1.55 30.54
 z z
1
da a b
sg ¼
db
Figure 4 as solid lines. Comparing the theoretical
lines with the plotted data from Table 3, it is where da and db are the particle size, and za and zb
apparent that the theory significantly overesti- are the z-values of the standard normal distribu-
mates the %RSD. tion at the ath and bth percentiles. (Note that if the
To determine if this is a problem with the particle size distribution is described perfectly by
theory, we must critically evaluate the assump- a log-normal function, the same value of sg would
tions made. The theory assumes that the particle be obtained regardless of the choice of ‘‘a’’ and ‘‘b’’).
size distribution is log-normal. If the actual Table 4 contains estimates of sg for API lots A and
particle size distribution were log-normal, the B using different regions of the particle distribu-
curves in Figure 1 would appear symmetrical. tion curves. The values obtained are significantly
Indeed, they are not, and in fact they are both lower as more emphasis is placed on the larger
significantly skewed to the negative side. This is a particles, a consequence of the negative skew of
natural consequence of the milling process that the the distributions. Using sg estimated from the
API underwent whereby any large particles are d97.7/d50 ratio, the theoretical %RSD is again
broken up into fines. plotted in Figure 4 and displayed as dashed lines.
The modified Yalkowsky–Bolton equation for These predictions are much closer to the mea-
%RSD is inherently sensitive to the value chosen sured values and demonstrate that the theory is
for the geometric standard deviation, sg. There- reasonably accurate. However, an important
fore, to be able to compare theoretical predictions point is illustrated: when translating the esti-
with actual data, it is imperative that a good mated d50 from the nomograph in Figure 3 to a
real formulation effort, it is important to utilize
the upper portion of the particle size distribution
when estimating sg.

Assumptions and Limitations

Log-Normal Particle Size Distribution
There are several assumptions built into this
model. The first is that the particle size distribu-
tion can be described by a log-normal function.
Table 4. Estimation of sg from Particle Size
Distributions for API Lots

Figure 4. Relative standard deviation of content Estimator for sg Lot A Lot B

uniformity versus dose. Symbols are measured values
(d84.1/d15.9)0.5 2.50 2.77
for tablet lots from API Lots A * and B &. Solid lines are
(d84.1/d50)1.0 2.05 2.28
calculated using a particle size distribution width for
(d90/d50)0.78 1.97 2.19
Lots A and B estimated by sg ¼ (d84.1/d15.9)0.5, dashed
(d97.7./d50)0.5 1.83 1.97
lines are calculated from sg ¼ (d97.7/d50)0.5.

DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006
1056 ROHRS ET AL.

Since the model will be used in a prospective

manner, that is, at the beginning of a formulation
effort ask, ‘‘What particle size is appropriate for
my dose?’’ we have to assume some distribution. It
is generally accepted that the log-normal function
is a reasonable approximation to the actual
particle size distribution in many instances.
However, as illustrated by the data herein,
emphasis must be placed on the upper region of
the particle size distribution if trying to estimate
%RSD of CU from a measured distribution. In
fact, for retrospective analyses, the method
derived by Johnson1 is more expedient.
Number of Particles Described by Figure 5. Estimated number of particles in a single
Poisson Distribution tablet versus dose.
Both the Johnson and Yalkowsky and Bolton
methods begin with the assumption that out of a good assumption up to about 1% drug load if the
sample of particles constituting the dosage unit, sizes of the drug and excipient particles do not
the Poisson distribution governs the number of differ too widely (fivefold). Above 1% and below
drug particles found within that sample. Two 10%, the %RSD estimate for CU should be multi-
attributes of the Poisson function are that the plied by the excipient load (i.e., 0.9 for a 10% drug
mean and the variance are equal, and that at load) to get a more realistic value. The consequence
large n-values, the distribution approaches nor- is that %RSD is reduced at higher drug loads and a
mality. In the Yalkowsky and Bolton derivation, higher d50 is acceptable. Therefore, estimates for
the mean number of particles ‘‘
’’ in a single d50 based on Figure 3 will be conservative as drug
dosage unit is estimated to be load increases.
D

¼ Normal Distribution of Content Uniformity Values
p 2

  6 d350  e4:5ln sg
A third assumption is that the CU values follow a
but the estimate is highly sensitive to sg, the normal distribution. This is an important
distribution width metric. Since for the present assumption to test because it is the basis for the
case the particle size distribution is known, a estimates of the population RSD needed to pass
direct estimate of the total number of particles USP CU criteria. Figure 6 contains the measured
can be made. The number of particles in any CU values for each dose of the test lots. Super-
weight fraction is equal to the mass of that imposed are the normal distributions calculated
fraction divided by the mass of one particle, from each lot mean and standard deviation.
therefore, the total number of particles in a dose is Figure 7 shows the calculated skewness of the
CU distributions. The two lowest doses have
D X fi
nTotal ¼   106 significant positive skew, and normality may not

 6 i d3i be a good assumption at these low doses. From a
practical viewpoint, given the particle size dis-
Estimates for the mean number of particles in the
tributions used in the test lots, the doses predicted
doses studied are illustrated in Figure 5. Even
to have RSD ¼ 3.84% (pass Stage I with 99%
at the lowest dose tested, there are far too many
probability) are about 30 mg and 100 mg for the two
particles to count to experimentally test the
API lots. These are well above the dose where the
Poisson assumption.
skewness becomes large in Figure 7. Normality,
At what point does the assumption begin to
therefore, appears to be a reasonable assumption
break down? The Poisson distribution describes
for the dose range of interest.
the probabilities of random occurrences. Intui-
tively, one would expect that as long as drug
Single Particle Impact on Content Uniformity
particles are independent of one another (i.e., low
drug load), the Poisson distribution is a reasonable Most of the 30 values of the CU distribution for
approximation. Johnson1 reasoned that this is a the 1.3 mg tablets (Fig. 8) are clustered between

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006 DOI 10.1002/jps
 PARTICLE SIZE LIMITS TO MEET USP CONTENT UNIFORMITY CRITERIA 1057

Figure 6. Content uniformity distributions for manufactured tablet lots overlaid with
normal distribution function based on lot mean and standard deviation. Units on the
x- and y-axis are dose (mg) and frequency, respectively.

1.2 and 1.8 mg, with two tablets at 2.2 and 3.8 mg.
Figure 9 illustrates the upper region of the
measured particle size distribution converted
from weight to number fraction, and the corre-
sponding mass of a single particle having the
indicated diameter. Although there are very few
large particles in the API, the impact on the 1.3 mg
targeted dose is enormous if even one large
particle is present. Therefore, at low doses, the
reason for the positive skewness in the CU
distribution is most likely due to the presence of
single API particles.

CONCLUSIONS

Figure 7. Skewness versus dose for tablet content The nomograph in Figure 3 is useful to estimate
uniformity distributions in Figure 6. a priori the necessary particle size to ensure CU

DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006
1058 ROHRS ET AL.

In order to use Figure 3 effectively, it is critical

to know the basis by which the particle size
distribution is described. This work utilizes med-
ian diameter and geometric standard deviation on
a weight (or volume) basis. If the descriptors are
different from those used here, they should be
transformed, otherwise the conclusions from
Figure 3 will be misleading.

APPENDIX I

EQUIVALENCE OF JOHNSON
AND YALKOWSKY–BOLTON EQUATIONS

Here we demonstrate that if the particle size

Figure 8. Content uniformity distribution for 1.3 mg
dose tablet.
criteria are met. Larger particles impact CU to a
12 X
much greater extent than smaller ones. Therefore
when using the nomograph in a quantitative
sense, the estimation of the distribution width
metric should weight the upper region of the
particle size distribution, especially as the dis-
tribution shape deviates from log-normal. This
also has implications for particle size reduction
strategies. Merely eliminating large particles can
significantly alter the required particle size. For
example, by narrowing the distribution width
d90/d50 from 4 to 2, the acceptable d50 increases
about fourfold.
distribution is assumed to be log-normal, then the
Johnson method and the Yalkowsky–Bolton
method are equivalent. We start first with
Johnson’s equation, which expresses the coeffi-
cient of variation in terms of weight fraction and
diameter:1
!12
p  r
Cv ¼ 100   fi  d3i
6D i
In an extension of Johnson’s work, Egermann3
showed that
X
fi  d3i ¼ d3v
i
where dv corresponds to the volume-weighted/
volume-number mean diameter. Egermann9 had
previously shown that dv could be defined using
the Hatch–Choate equations in terms of the
geometric mean diameter on a number basis dg
and the geometric standard deviation sg as:

ln dv ¼ ln dg þ 4:5 ln2 g

For a log-normal distribution f(d) ¼ LN(m,s), the
geometric mean diameter on a number basis
dg ¼ em and the geometric standard deviation
sg ¼ es. Therefore,

ln dv ¼  þ 4:5  2 leading to dv ¼ eþ4:5 and

2

d3v ¼ e3þ13:5s
2

The expression for the coefficient of variation Cv

in terms of m and s then becomes
Figure 9. Upper tail of API lot B particle size  

12
1
distribution with corresponding spherical single particle Cv ¼ 100   e3þ13:5s
2 2

mass (density ¼ 1.4 g/cm3). 6D

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006 DOI 10.1002/jps
 PARTICLE SIZE LIMITS TO MEET USP CONTENT UNIFORMITY CRITERIA 1059

Yalkowsky–Bolton’s expression for the coefficient REFERENCES

of variation is:4
 

12 1. Johnson MCR. 1972. Particle size distribution of the
 12
12
Cv ¼ 100   d3m  1 þ C2 active ingredient for solid dosage forms of low
6D dosage. Pharm Acta Helv 47:546–559.
‘‘dm’’ is the arithmetic mean diameter on a 2. Johnson MCR. 1975. The effect of particle size
upon mixture homogeneity. Pharm Acta Helv 50:
number basis. We can use one of the Hatch–
60–63.
Choate equations8 to convert to geometric mean 3. Egermann H, Kemptner I, Pichler E. 1985.
and standard deviation on a number basis Effects of interparticulate interactions on mixing
homogeneity. Drug Dev Ind Pharm 11:663–
ln dm ¼ ln dg þ 0:5 ln2 g ; or dm ¼ emþ0:5s
2

676.
For a log-normal distribution f(d) ¼ LN(m,s), the 4. Yalkowsky SH, Bolton S. 1990. Particle size
arithmetic standard deviation ‘‘C’’ is related to s and content uniformity. Pharm Res 7:962–966.
5. Egermann H. 1991. Comments on ‘‘Particle size and
through the equation:
content uniformity,’’ by Yalkowsky and Bolton.
 
1 þ C2 ¼ es
2
Pharm Res 8:1076–1077.
6. Hays WL, Winkler RL. 1971. Statistics. New York:
Substituting into the expression for Cv gives Holt, Rinehart and Winston. p 356.
7. Limpert E, Stahel WA, Abbt M. 2001. Log-normal
 

12 
3 2
12 12 distributions across the sciences: Keys and clues.
þ0:52
Cv ¼ 100   e  e BioScience 51:341–352.
6D 8. Martin A. 1993. Physical Pharmacy, 4th edition.
 

12
1
2 2 Baltimore: Williams & Wilkins. p 430.
¼ 100   e3þ13:5 9. Egermann H. 1982. Definition and conversion
6D
of the mean particle diameter referring to
which is identical to that obtained for Johnson’s mixing homogeneity. Powder Technology 31:231–
expression. 232.

DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 5, MAY 2006