AAPS PharmSciTech ( # 2017)

DOI: 10.1208/s12249-017-0925-8

Brief/Technical Note

A Science and Risk-Based Pragmatic Methodology for Blend and Content

Uniformity Assessment

Naheed Sayeed-Desta,1 Ajay Babu Pazhayattil,1,2 Jordan Collins,1 and Chetan Doshi1

Received 14 October 2017; accepted 21 November 2017

Abstract. This paper describes a pragmatic approach that can be applied in assessing
powder blend and unit dosage uniformity of solid dose products at Process Design, Process
Performance Qualification, and Continued/Ongoing Process Verification stages of the Process
Validation lifecycle. The statistically based sampling, testing, and assessment plan was
developed due to the withdrawal of the FDA draft guidance for industry BPowder Blends and
Finished Dosage Units—Stratified In-Process Dosage Unit Sampling and Assessment.^ This
paper compares the proposed Grouped Area Variance Estimate (GAVE) method with an
alternate approach outlining the practicality and statistical rationalization using traditional
sampling and analytical methods. The approach is designed to fit solid dose processes
assuring high statistical confidence in both powder blend uniformity and dosage unit
uniformity during all three stages of the lifecycle complying with ASTM standards as
recommended by the US FDA.
KEY WORDS: blend uniformity; content uniformity; dosage uniformity; process validation; solid dose;
lifecycle stages.

INTRODUCTION Statistical tools should be used to ascertain sampling locations

Since the withdrawal of the US FDA draft guidance
document for industry, BPowder Blends and Finished Dosage
Units—Stratified In-Process Dosage Unit Sampling and
Assessment^ on August, 2013 (1), the pharmaceutical indus-
try required a science and risk-based approach, justified
sampling, and testing plans to ensure drug product uniformity.
Although a draft, the guidance was extensively used by the
industry and in some circumstances, continued to be used for
lack of an alternate practical approach replacing the guid-
ance. A drug product manufacturer is required to justify their
assessment method for blend and content uniformity. US
FDA has also set an underlying expectation for manufac-
turers to ensure consistent product quality by demonstrating
no significant difference in uniformity between and within
locations. The draft guidance was withdrawn since some
sections did not agree with regulators’ current thinking.
According to the subsequently issued Level 2 guidance:
Question and Answers on Current Good Manufacturing
Practices—Production and Process Controls (2), Sections V
and VII of the guidance were not adequate. US FDA
recommends demonstrating that the variability attributed to
blend sample location is minimal in addition to ensuring that
the sample size is adequate for a valid statistical analysis.
1
Apotex Inc., 150 Signet Drive, Toronto, Ontario M9L 1T9, Canada.
2
To whom correspondence should be addressed.
(e-mail: apazhaya@apotex.com; abapaz@gmail.com)
across the blender and to conduct a valid analysis. The
regulator is concerned about proper sampling of powder
blends since it is supposed to ensure that no differences exist
between locations in a blend that could adversely affect
finished product quality. They acknowledge that traditional
sampling using a powder-thief can have drawbacks such as
causing disturbance to the powder bed, sampling errors, etc.
though it is widely used and provide reliable results. Further,
the agency believes that the procedure and acceptance
criteria in United States Pharmacopeia (USP) General
Chapter <905> Uniformity of Dosage Units are not statisti-
cally sound and hence should not be extrapolated to a larger
sample. US FDA therefore no longer supports its use for
batch release.
One strategy available for blend and content uniformity
assessment was developed by Garcia et al. (3) as part of The
International Society for Pharmaceutical Engineering (ISPE)
sponsored Blend Uniformity and Content Uniformity
(BUCU) Initiative. The teams initial two-part paper (Part 1:
Modifications to Withdrawn FDA Draft Stratified Sampling
Guidance and Part 2: Technical Discussion of Sampling Plans
and Application of ASTM E2709/E2810) defined alternate
approaches for the assessment of blend and content unifor-
mity attempting to address FDA’s primary concerns that led
to the withdrawal of the draft guidance document. The
proposed enhanced sampling, testing, and assessment plan
discussed in this paper was developed as an approach
primarily fit for solid dose manufacturing processes with

1530-9932/17/0000-0001/0 # 2017 American Association of Pharmaceutical Scientists


powder blending and dosing operations. The authors applied
well-established principles of pharmaceutical science along
with sound statistical justification in developing the Grouped
Area Variance Estimate (GAVE) method. The blend unifor-
mity and dosage uniformity plan can be practically accom-
modated within the current solid dose manufacturing
operations and laboratory capabilities without confounding
sample logistics.
POWER BLENDS
Bin blending is a batch manufacturing process wherein
all materials are charged before the start of the processing
and discharged at the end of the processing. Blend analysis is
conducted at multiple lifecycle stages as it provides an insight
to the adequacy of mixing and the final dosage unit
uniformity. Blend results are significant in gaining process
knowledge and understanding impact of changes related to
manufacturing process, formulation, raw material, equipment,
and scale up on uniformity. However, it has to be noted that
current blend sampling practice of sampling post blending
operations violates the representative sampling rule of
sampling materials when in motion to determine blend
adequacy (4). Blend and content uniformity data can be
compared but not correlated as powder flow properties differ
between formulations and batch sizes, and may also vary with
change in raw material physical characteristics. Location to
location blend and dosage unit correlation is not applicable
since powder rheology is complex and affected by multiple
product specific factors as understood during the Process
Design stage. For instance, the dependent variables are
different when mixing cohesive materials in comparison with
mixing free flowing materials (5).
It is typical for gravity-fed powder blends to be on dosing
equipment where the blenders are enclosed and the powder
flow patterns during processing are not visible. The variations
in powder blend characteristics can be mitigated with the
application of control strategies based on well-characterized
raw material properties, robust manufacturing processes, and
accurate measurement systems. Blend process parameters are
established during Process Design (Stage 1) blend time
analysis study to assure consistent uniformity. Some of the
main factors systematically studied in establishing uniform
blends are the percentage of active pharmaceutical ingredient
(API) in the formulation, particle size distribution in relation
to other excipients, particle shape, density, batch size and bin
fill, loading and unloading methods, blend time, and blender
speed. The aforementioned factors have not been discussed
further in this paper based on the underlying assumption that
the process and formulation is optimized through extensive
Quality by Design (QbD)-based development studies. Sam-
pling and testing of powder blends and finished dosage units
during manufacturing of development batches may mimic the
sampling and testing plan recommended here for Process
Performance Qualification (Stage 2B) batches.
BLENDERS AND BLEND SAMPLING
Bin blenders are commonly used intermediate bulk
containers for mixing powders and granules during solid dose
manufacturing. This paper further discusses the approach as it
Sayeed-Desta et al.
is applicable to bin blenders though the same methodology
may be applied to other blenders such as V-blenders. Bin
blenders provide good containment of the in-process batch
and easy loading and unloading of the blend. The bins are
also used to store and transport material to the next
processing stage and therefore also referred to as intermedi-
ate bulk containers. Gallay bin blenders are an example of
such blenders. Material is loaded top-to-bottom in bin
blenders and mixing of the powder occurs by tumbling. Low
shear axial and radial mixing is applied when the bin blender
is rotated at a set speed and time on an angled rotational axis
(6). Bin blenders are known to exhibit slower axial mixing
compared to radial mixing (7).
The geometry of the blenders is symmetric in design with
distinct sections being the rectangular and hopper-shaped
section. A numbered sample location is therefore unwar-
ranted as long as representative locations are sampled, that is,
corners of the bin blender versus center of the bin blender. A
newly designed blend sampling plan is therefore proposed to
have five sample locations from the Btop^ area, four locations
from the Bmiddle^ area, and three locations from the
Bbottom^ area of the powder blend. Five locations from the
top assure that the four corners of the bin blender as well as
the center location are sampled. The four locations from the
middle assure that the four corners of the bin blender are
sampled. Three locations from the bottom of the bin blender
hopper section assure that there are a minimum of three
samples from this area for a statistically valid analysis.
Although traditionally 10 locations are deemed adequate to
map tumbling type mixers, 12 locations are recommended
based on the geometry of Gallay bin blenders.
Figure 1 shows the geometrical design of the blenders.
The bin volume is higher in the top rectangular section and
decreases with the hopper-shaped section. Hence, more
sample locations are recommended for the top and the
middle areas of the bin blender than the bottom to
adequately represent the blend as per sectional volumes of
the bin. Table I provides estimated volumes of each of the
sections for the bin blender sizes for comparison.
Based on the estimated volumes, the rectangular section
is ~ 75% and the hopper section is ~ 25% of expected bin fill
volumes. Therefore, from a sampling perspective, there is
good representation of the samples from each of the sections
relative to the volumes where 75% of the samples (9/12) are
obtained from the Btop^ and Bmiddle^ areas of the rectangu-
lar section and 25% of the samples (3/12) are obtained from
the Bbottom^ area of the hopper section. The samples from
the 12 predetermined locations of the bin blenders are
obtained using qualified powder-thief sampling methods.
The second sample from each location is required in case
there are issues with first samples (e.g., spilled or missed
sample). The blend sample size is set at 1×–3× dosage units
unless otherwise justified during Stage 1.
Even though one sample is taken from each location,
there are sufficient samples to determine within- and
between-location variability. This is accomplished by group-
ing of locations within an area. Samples obtained from each
of the Btop,^ Bmiddle,^ and Bbottom^ areas of the bin are
considered as one group for the purpose of statistical analysis.
Analysis of within-location (within-group) variability for each
of the three areas is conducted by analyzing each of the
A Science and Risk-Based Pragmatic Methodology
Fig. 1. Schematic of bin blender areas (T, M, B) with sampling
locations
samples grouped by location area. Therefore, in this ap-
proach, replicate sampling is not recommended as there are
≥ 3 samples obtained from each of the three areas. Statistical
justification of using area grouped samples versus single
location replicate is provided later in this paper. The
statistical criteria for blend uniformity is stringent in this
method, i.e., the overall blend uniformity standard deviation
(SD) for the batch should be ≤ 3.0% in order to proceed with
Tier 1 dosage uniformity (DU) assessment which includes
stratified sampling and testing of three unit doses from each
of the 20 locations sampled during the dosing process. If the
blend uniformity SD is > 3.0%, an extended Tier 2 DU
assessment is applied wherein seven units are tested from the
20 locations sampled.
DOSING AND STRATIFIED SAMPLING
Once the blend uniformity has been verified, the unit dosages
produced from the blend are tested for uniformity as well. This is to
ensure that the end unit dosage given to patients has uniform active
content/s as per its approved label claim. Since the internal flow
dynamics of the powder/granule systems are not completely known
for each of the formulation, it is important to ensure that the
powder flow from blenders and subsequent dosing process does
Table I. Comparison of Rectangular and Hopper-Shaped Sections of Bin Blender
Bin size (cu.ft.) Volume (L) Rectangular section
L (in.) W (in.)
2 56 15.1 17.9
4 113.3 18.8 22.2
11.3 320 24 30
80 2266 44.3 52.2
not contribute to any potential segregation issues resulting in
variable dosage uniformity. If powder segregation occurs during the
process, it will be more likely identified through a stratified
sampling and testing of the dosage units produced from each blend
batch.
In a typical setup, bin blenders are staged over dosing
equipment, and blends are discharged from the bottom of the
blender into lay flats that transfer the blend into hoppers. Powder
blends are either compressed or encapsulated into unit doses.
Commercial dosing operations are typically semi-continuous
processes where material is simultaneously charged and discharged
and completed once the entire blend batch is consumed. Tableting
presses and encapsulators are set up so that each unit dose
consistently meets the required output of the controlled process (8).
This is verified by frequent sampling and testing for in-process
physical quality attributes such as weight, hardness, thickness,
friability, and disintegration time. In addition to equipment control
capabilities and verification of the physical attributes, the dosage
uniformity of the units is also verified.
Samples for dosage uniformity testing are taken throughout
the semi-continuous dosing process. A stratified sampling plan is
followed during a dosing operation. The first sample is taken at the
start and remaining samples are taken at equal intervals until the
end of the process with sampling points at no more than 5% of the
batch. This results in a total of 20 strata samples of dosages
representing the entire dosing run. Stratified sampling increases
confidence of the uniformity of the batch as sample locations target
problematic areas prone to potential segregation. A risk-based
sampling plan that includes additional sampling points at the
beginning and the end of the run may be applied based on product/
process knowledge. Random variation is expected throughout a
population; therefore DU results are expected to be a normal
distribution with no special cause variation. A minimum of seven
units from each of the 20 locations are required to be sampled for
testing. For statistical rigor, 20 sample locations are sufficient. Three
random samples are tested from each of the 20 strata resulting in 60
units from each blend for Tier 1 testing and a total of 140 (20 × 7)
units for Tier 2 dosage uniformity testing.
BLEND UNIFORMITY AND CONTENT UNIFORMITY
ASSESSMENT
It should be noted that the measure of blend uniformity
is designed to capture homogeneity of the blend batch. The
measure of homogeneity is determined by the dispersion
measured in a blend sample, and not the location. Further-
more, given that mean and individual results may be affected
by bias introduced due to sampling, a measure of dispersion
within areas is scientifically and statistically appropriate than
Hopper section
H (in.) Volume (L) R (in.) H (in.) Volume (L)
10.6 47 8.9 12.5 9
13.8 94 11.1 14.9 19
20.5 242 15 22.7 78
47.8 1811 26.1 37.1 454
 Sayeed-Desta et al.

a measure within an individual location. Since blend uniformity is
of interest as a precursor to homogeneity in the dosage units
themselves, the statistical criteria for the BU results should not be
designed such that the batch could be questionable solely upon its
basis. The statistical criteria proposed for the blend uniformity is
appropriate as it focuses on measures of variability and not an
acceptance criterion with individual results within ± 10% of the
mean. The criteria put forth in this paper serve the purpose of
dictating whether extended testing should occur at the DU stage of
testing. Apart from being applied to determine whether extended
DU testing is required, it is proposed that the results be further
analyzed utilizing Variance Component Analysis (VCA) when
needed, to increase knowledge of the product and as an aide in
Stage 3 continuous improvement efforts (9).
The proposed GAVE method for blend and dosage
uniformity during solid dose manufacturing processes at the
Process Performance Qualification stage is provided in Fig. 2.
A tiered approach is applied to the testing of dosage
uniformity using ASTM E2709/E2810 methodology sampling
plans (10).

The sampling, testing plan, and acceptance criteria for

Stage 2B is summarized below.

Stage 2 PPQ Blend Sampling:

Assay 1 blend sample from each of the 12 locations
1. If the overall SD ≤ 3.0% of target, proceed to Tier 1 Dosage Unit Testing
2. If the overall SD > 3.0% of target, proceed to Tier 2 Dosage Unit Testing
Stage 2 PPQ Dosage Unit Sampling:
Sample at least 7 in-process dosage units from 20 approximately equally spaced
predetermined locations through the batch
Tier 1 Testing: Tier 2 Testing:
Assay 3 dosage units per location from all 20 locations (n = 60) Assay an additional 4 dosage units per location from
all 20 locations (n = 140)
Acceptance criteria:
All individual values within 75.0–125.0% (as-is) and data complies with the ASTM
E2709/E2810 acceptance criteria standards
If either is not met, proceed to Tier 2 dosage unit testing If either is not met, dosage units are not uniform

The proposed assessment of blend and dosage unifor- Continued/Ongoing Process Verification stage is provided in
mity during solid dose manufacturing processes at the Fig. 3.

The sampling, testing plan, and acceptance criteria for

Stage 3A (11) and 3B is summarized below.

Stage 3A CPV Blend Uniformity Sampling:

Assay 1 blend sample from each of the 12 locations
1. If the overall SD ≤ 3.0% of target, proceed to Tier 1 Dosage Unit Testing
2. If the overall SD > 3.0% of target, proceed to Tier 2 Dosage Unit Testing
Stage 3A CPV Dosage Unit Sampling:
Sample at least 7 in-process dosage units from 10 approximately equally spaced
predetermined locations through the batch
Tier 1 Testing: Tier 2 Testing:
Assay 3 dosage units per location from all 10 locations (n = 30) Assay an additional 4 dosage units per location from
all 10 locations (n = 70)
Acceptance criteria:
All individual values within 75.0–125.0% (as-is) and data complies with the ASTM
E2709/E2810 acceptance criteria standards.
If either is not met, proceed to Tier 2 dosage unit testing If either is not met, dosage units are not uniform

Stage 3B CPV Dosage Unit Sampling:

Tier 1 Testing: Tier 2 Testing:
Assay 1 dosage units per location from all 10 locations (n = 10) Assay an additional 2 dosage units per location from
all 10 locations (n = 30)
Acceptance criteria:
All individual values within 75.0–125.0% (as-is) and data complies with the ASTM
E2709/E2810 acceptance criteria standards
If either is not met, proceed to Tier 2 Dosage Unit Testing If either is not met, dosage units are not uniform
A Science and Risk-Based Pragmatic Methodology

FOR BLEND UNIFORMITY (BU): Collect duplicate samples from 12 locations (2 x 12)

Assay 1 sample per location (1 x 12)

SD > 3.0 % SD < 3.0 %

BU Statistical DU Tier 1
DU Tier 2
Criteria

Sample at least 7 dosage units Sample at least 7 dosage units

from at least 20 locations from at least 20 locations
(7 x 20) (7 x 20)

Assay 7 units per location Assay 3 units per location

(7 x 20) (3 x 20)

DU Acceptance Criteria:
Fail All individual ‘as-is’
Assay remaining 4 units per
results within 75.0 –
location (7 x 20)
125.0% and mean
results pass ASTM std.

Pass
DU Acceptance Criteria:
All individual ‘as-is’ Pass
results within 75.0 – BU & DU Acceptable
125.0% and mean
results pass ASTM std.

Conduct Variance
Component Analysis (VCA)

Fail

BU & DU Not
Acceptable

Fig. 2. Process flow diagram for assessment of blend and dosage uniformity for Process Performance Qualification batches (Stage 2B)

If the variability in any Stage 3A blend batches is > 3.0%,
VCA is recommended to understand the contributing sources
of variation and determine a path forward. Firms may decide
to continue with the heightened sampling and testing in Stage
3B to gain more product understanding. Note that the ASTM
E2709/E2810 acceptance criteria are known to provide a
conservative approach for DU acceptance limits (12), and
therefore, using these acceptance criteria provides great
assurance of product quality. The GAVE method may be
enhanced based on product formulation and process (ex. low
active content drug products).
VARIANCE COMPONENT ANALYSIS
VCA allows for the quantification of sources of variability
across different levels of units from a sampling scheme. VCA
analysis is conducted to determine the potential source of
variability: either within area and or between areas. High
 Sayeed-Desta et al.

FOR BLEND UNIFORMITY (BU): Collect duplicate samples from 12 locations (2 x 12)

Assay 1 sample per location (1 x 12)

SD > 3.0 % SD < 3.0 %

BU Statistical DU Tier 1
DU Tier 2
Criteria

Sample at least 7 dosage units Sample at least 7 dosage units

from at least 20 locations from at least 20 locations
(7 x 20) (7 x 20)

Assay 7 units per location Assay 3 units per location

(7 x 20) (3 x 20)

Assay remaining 4 units per
location (7 x 20)
DU Acceptance Criteria:
Fail All individual ‘as-is’
results within 75.0 –
125.0% and mean
results pass ASTM std.

Pass
DU Acceptance Criteria:
All individual ‘as-is’ Pass
results within 75.0 – BU & DU Acceptable
125.0% and mean
results pass ASTM std.

Conduct Variance
Component Analysis (VCA)

Fail

BU & DU Not Continue with routine IPDU

Acceptable sampling and testing for
Stage 3B

Fig. 3. Process flow diagram for assessment of blend and dosage uniformity for Continued/Ongoing Process
Verification (Stage 3A)

between-area variance could indicate poor mixing, resulting in non-
uniformity within the blender and potential for segregation. High
within-area variance could indicate sampling bias or analytical
errors (13). In the sampling plan proposed, single testing per
location is performed with multiple tests per area, hence VCA is
possible. Adequate mixing is directly assessed since it is more likely
that mixing issues will appear as a difference between top, middle,
and bottom areas of the blender than between locations lying
potentially in the same area. Figure 4 provides representative
examples of products where BU results were out of specification;
however, dosage uniformity results of the batches met the
acceptance criteria. The results indicate that the variability in BU
is not related to mixing issues. Analyzing the within-area and
between-area variance components supports this. In both exam-
ples, the between-area SD is 1.80 and 1.02 compared to the within-
area SD which is higher at 3.82 and 2.91.
If all dosing batches tested meet Tier 2 dosage uniformity
(n = 140) acceptance criteria, it also indicates that the within-
area blend variability is not carried over to the dosage
uniformity and hence is non-process/product related. There-
fore, it should have no impact on the dosage unit uniformity
of the batches. When between-areas variability is observed, it
indicates uniformity issues related to process/product. The
identified potential source of variation observed in blend
batches may be further evaluated and implement preventa-
tive actions if required. For further assessment, additional
data is required which can be gathered from heightened
sampling and testing of Stage 3 batches.
STATISTICAL CONSTRUCT FOR THE BLEND
SAMPLING PLAN DESIGN
To elucidate the benefits of the proposed methodol-
ogy, it is compared to one of the recent well developed
plan (Garcia, 2015). The value of taking three BU
samples per location and testing as opposed to grouped
BU samples by area is unclear. Although further investi-
gation is required, it appears that the extra cost of testing
A Science and Risk-Based Pragmatic Methodology

Fig. 4. BU variance component analysis examples

three samples does not outweigh the benefits gained from testing of BU at 12 locations with a sample each will
the data analysis. The plans are compared here with a allow for an analysis of variance via VCA comparing
view of practical implementation at brand, generic, and within-group (within-area) and between-group (between-
contract manufacturing pharmaceutical organizations. The area) variability when grouping the 12 locations according

Table II. Review of Assessment Conditions Across Various Methodologies

Methodologies

Conditions Draft Garcia et

GAVE
Guidance al

Within and between group variability is evaluated

Statistical tools used to justify number of sampling locations

Statistically valid blend uniformity criteria

Minimal blend disturbance with powder-thief sampling

Minimal potential for sample management errors

Application of VCA for enhanced product knowledge

Minimal sampling and testing cost.

Application of ASTM statistical acceptance criteria

Cells in gray represent most fitting for the listed conditions

GAVE Grouped Area Variance Estimate, VCA Variance Component Analysis, ASTM American Society for Testing and Materials

With 3 samples tested; SD = 0.76
Scenario 1
With 3 samples tested; SD = 3.72
Scenario 2
Fig. 5. Representing four scenarios
to top, middle, and bottom areas of the bin blender. It is
acknowledged that with only three areas, the estimate of
between-area variability will not be as precise as the plan
with 10 locations. If a more precise estimate is required,
the recommendation for performing VCA with 12 location
triplicates can be followed (14). However, given that the
between- and within-location partitions are used as a
rough guide for further learning and investigation, precise
estimates are not necessary; one is looking here for overt
deviations from what is expected, not subtle differences
dependent upon the most precise estimates possible for
blend uniformity.
In the proposed sampling plan, total variation in the
blend can be measured as the sum of the variability
between areas and variability within areas, due to random
unknown sources. The latter comprises variability due to
sampling, equipment, personnel, etc. Attributable to the
unbalanced nature of the sampling plan, total variation is
estimated from an REML estimate of the variance
components. It is possible to devise balanced sampling
plans with an equal number of replicates per area; in this
case an ANOVA table breakdown of the sources of
variation into the variance components between areas
Sayeed-Desta et al.
With 1 sample tested; SD = 0.64
With 1 sample tested; SD = 3.80
versus within areas can be used. For this proposed
sampling plan, the analysis of total variation is based
upon the following model of blend uniformity:
8
i ¼ 1; 2; 3
9
>
> >
>
j ¼ 1; 2; 3; 4; 5 for i ¼ 1
< =
Yij ¼ a þ Li þ EjðiÞ
> j ¼ 1; 2; 3; 4; for i ¼ 2 >
>
:
j ¼ 1; 2; 3 for i ¼ 3
>
;
Where Yij is the jth blend uniformity measure at the
th
i area, a is the overall average blend uniformity values,
Li is the random effect attributable to the ith area where
Li is normally distributed with mean 0 and variance σL,i2,
and Ej(i) is the random error which is assumed to be
normally distributed with mean 0 and variance σ2.
The purpose of this breakdown is to estimate the total
variance and to have an estimate of the variability between
areas. Since the measure of the blend is meant as an indicator
of potential segregation, and determining if there is a need for
extended dosage unit testing requirement, rough estimates is
all that is required here; the cost of extended error prone
blend sampling and testing for the sake of purported
A Science and Risk-Based Pragmatic Methodology
With 3 samples tested; SD = 4.16
Scenario 3
With 3 samples tested; SD = 5.16
Scenario 4
Fig. 5. (continued)
improvements in statistical measures of homogeneity are not
warranted. The proposed sampling plan includes grouped
testing of BU samples by area. Grouped area BU testing is
sufficient for providing an estimate of variance within an area.
The difference in performance between single and triplicate
testing for 12 locations has been thoroughly examined (15)
demonstrating that grouped BU testing provides clearer dis-
crimination versus triplicate location testing. It is generally well
known that size of group-level variables is more important than
individual-level variables (16–18). In fact, based on slightly
different criteria, Madsen (15) simulated OC curves for the
probability of passing as a function of standard deviation with 12
locations (4 from the top, 4 from the middle, and 4 from the
bottom) with single samples tested per location versus OC
curves for triplicate samples tested. It is remarked that the
relation between total variation and probability of passing is less
clear with three replicate samples per location than under the
sampling plan with grouped area observations. The study
therefore shows that grouped area testing is in fact more
appropriate than triplicate per location testing when the goal is
to assess the total variability of the blend. Extensive testing is
appropriate during Stage 1, Stage 2B, and Stage 3A, and not for
Stage 3B if it can be justified with the substantial data generated
as part of the previous stages (Table II).
With 1 sample tested; SD = 3.56
With 1 sample tested; SD = 5.46
COMPARATIVE SIMULATION STUDIES
The following (Fig. 5) graphical representations of below
scenarios highlight the way in which the proposed approach
will be different. For the sake of comparability, it is assumed
that there are 12 locations with three samples tested and 12
locations with one sample tested as proposed.
In Scenario 1, the data is generated such that there is a
small between-location and small within-location variability,
SD < 3.0%. The practical outcome of these BU results in both
proposals would be to continue with Tier 1 DU testing. In
Scenario 2, there is a large between-location and small within-
location variability. In Scenario 3, there is small between-
location and large within-location variability. In Scenarios 2
and 3, the SD < 5.0%. In Scenario 4, there is a large between-
location and large within-location variability, SD > 5.0%. The
practical outcome of these BU results in Scenarios 2, 3, and 4
would be to continue with Tier 2 DU testing. As a result of
the non-exhaustive list of scenario simulations, it is apparent
from a practical perspective that the proposed sampling/
testing plans do not differ much in the outcome. The plan
proposed in this article requires a greater number of total
tests, but the emphasis is on extended DU testing, as opposed
to BU testing.

SUMMARY
The new GAVE method adequately addresses the
concerns highlighted by the US FDA while withdrawing
the draft guidance. The proposed approach requires a
minimum of 12 blend samples and 20 × 7 dosage unifor-
mity samples for the purpose of uniformity testing and
assessment. The recommended blend and dosage sampling
plan discussed here is fit for implementation at high
volume, multi-product manufacturing sites. It minimizes
blend disturbance and potential operational interruptions.
The risk-based sampling and testing approach applies
tighter acceptance criteria for uniformity assessment. This
paper is intended to firmly rationalize the use of an
enhanced approach for solid dose manufacturing. It
provides drug product manufacturers with a pragmatic
blend and dosage unit sampling, testing, and assessment
plan based on sound science and statistical constructs that
can be applied to all stages of the lifecycle.
COMPLIANCE WITH ETHICAL STANDARDS
Conflict of Interest The authors declare that they have no
competing interests.
REFERENCES
1. Guidance for Industry, BPowder Blends and Finished Dosage
Units—Stratified In-Process Dosage Unit Sampling and
Assessment^. U.S. Department of Health and Human Services,
Food and Drug Administration, Center for Drug Evaluation
and Research (CDER), October 2003, Pharmaceutical CGMPs.
2. Question and Answers on Current Good Manufacturing
Practices, good guidance practices, level 2
guidance—Production and Process Controls, CDER/OC Office
of Manufacturing and Product Quality: CGMP Subject Matter
Contacts, 6 August 2013. https://www.fda.gov/Drugs/
GuidanceComplianceRegulatoryInformation/Guidances/
ucm124782.htm#18.
3. Garcia T, Bergum J, Prescott J, Tejwani R, Parks T, Clark J,
et al. Recommendations for the assessment of blend and content
uniformity: modifications to withdrawn FDA draft stratified
Sayeed-Desta et al.
sampling guidance. J Pharm Innov. 2015;10(1):76–83. https://
doi.org/10.1007/s12247-014-9207-0.
4. Brittain HG. Particle size distribution II: the problem of
sampling powdered solids. Pharm Technol. 2002;26(7):67–73.
5. Alexander A, Sudah O, Arratia P, Duong N-H., Reynolds S,
Muzzio F. Characterization of the performance of bin blenders
part 3 of 3: cohesive powders. Pharm Technol. 2004;28(9):54–74.
6. Alexander A, Arratia P, Goodridge C, Sudah O, Brone D,
Muzzio F. Characterization of the performance of bin blenders
part 1 of 3: methodology. Pharm Technol. 2004;13(5):70–86.
7. Alexander A, Sudah O, Arratia P, Goodridge C, Alani L,
Muzzio F. Characterization of the performance of bin blenders
part 2 of 3: free flowing mixtures. Pharm Technol.
2004;28(7):56–7.
8. Sayeed-Desta N, Pazhayattil A, Chowdari S. BDetermining
minimum batch size,^ APIs, excipients, and manufacturing
supplement to. Pharm Technol. 2016;40:s16–9.
9. Collins J, Sayeed-Desta N, Pazhayattil AB, Doshi C. A novel
metric for continuous improvement during stage three.
BioPharm Int. 2017;30(6):32–5.
10. Bergum J, Parks T, Prescott J, Tejwani R, Clark J, Brown W,
et al. Assessment of blend and content uniformity. Technical
discussion of sampling plans and application of ASTM E2709/
E2810. J Pharm Innov. 2015;10(1):84–97. https://doi.org/10.1007/
s12247-014-9208-z.
11. Sayeed-Desta N, Pazhayattil AB, Collins J, Chen S, Ingram M,
Spes J. Assessment methodology for process validation lifecycle
stage 3A. AAPS PharmSciTech. 2017;18(5):1881–6. https://
doi.org/10.1208/s12249-016-0641-9.
12. Lewis RA, Fan A. Improved acceptance limits for ASTM
standard E2810. Stat Biopharm Res. 2016;8(1):40–8. https://
doi.org/10.1080/19466315.2015.1093959.
13. Nunnally B. Variance component analysis to determine sources
of variation for vaccine drug product assays. J Validation
Technol. 2009;15(3):78–88.
14. Gelman A. Prior distributions for variance parameters in
hierarchical models. Bayesian Anal. 2006;1(3):515–33. https://
doi.org/10.1214/06-BA117A.
15. Madsen C. Statistical Methods for Assessment of Blend
Homogeneity. PhD Thesis for the department of Informatics
and Mathematical Modelling, Technical University of
Denmark;2002.
16. Snijders TAB. Power and sample size in multilevel linear
models. In: Everitt BS, Howell DC, editors. Encyclopedia of
statistics in behavioral science, vol. 3. Chicester: Wiley; 2005. p.
1570–3. https://doi.org/10.1002/0470013192.bsa492.
17. Maas CJM, Hox JJ. Sufficient sample sizes for multilevel
modeling. Methodology. 2005;1(3):86–92.
18. Bell BA, Ferron JM, Kromrey J.‘Cluster size in multilevel
models: the impact of sparse data structures on point and
interval estimates in two-level models’ Joint Statistical
Meetings—Section on Survey Research Methods;2008.