Quality by Design : Design Space

Design Space

Presentation prepared by Drug Regulations – a not for
profit organization. Visit www.drugreulations.org for the
latest in Pharmaceuticals.

www.drugragulations.org 1

Product Profile  Quality Target Product Profile (QTPP)

CQA’s  Determine “potential” critical quality attributes (CQAs)

Risk Assessments  Link raw material attributes and process parameters to
CQAs and perform risk assessment
Design Space  Develop a design space (optional and not required)

Control Strategy  Design and implement a control strategy

Continual  Manage product lifecycle, including continual
Improvement
improvement


Product Profile

CQA’s  This presentation Part IV of the
series “QbD for Beginners” covers
Risk Assessments
basic aspects of
Design Space ◦ Design Space
◦ Design of experiments
◦ Models
Control Strategy

Continual
Improvement


 The relationship between the process inputs
(material attributes and process parameters) and
the critical quality attributes can be described as
the design space.


 The multidimensional combination and interaction of
input variables (e.g., material attributes) and process
parameters that have been demonstrated to provide
assurance of quality.
 Working within the design space is not considered as a
change.
 Movement out of the design space is considered to be a
change and would normally initiate regulatory post
approval change process.
 Design space is proposed by the applicant and is
subject to regulatory assessment and approval (ICH
Q8).


 The Multi-dimensional region which
encompasses the various combinations of
product design, manufacturing process
design, manufacturing process operating
parameters and raw material quality which
produce material of suitable ( defined)
quality.


Critical Quality Attributes

Input Materials Output Materials
Process Step (Product or Intermediate)

Design
Space
Input Measured
Process Parameters
Parameters or Attributes
Process
Control Model
Measurements
and Controls

100.0 2
95.0
90.0 Surface Plot Contour Plot 1.8
Dissolution (%)

1.6
85.0 Dissolution (%)
80.0 1.4
90.0-95.0
75.0

Parameter 2
1.2
70.0 85.0-90.0

65.0 1 80.0-85.0
60.0 75.0-80.0
0.8
55.0 70.0-75.0
0.6
50.0
Design Space
65.0-70.0
2 0.4 60.0-65.0
40
(non-linear)
Pa 2 0.2
ram 50 1
er
ete et Design Space
r1 am 0
60 0 P ar 40 42 44 46 (linear ranges)
48 50 52 54 56 58 60
Parameter 1

• Design space proposed by the applicant
• Design space can be described as a mathematical function or
simple parameter range
• Operation within design space will result in a product meeting the
defined quality attributes


Knowledge Space

Design Space: “Multidimensional
Design Space combination and interaction of input
variables (e.g., material attributes) and
process parameters that have been
NOR
demonstrated to provide assurance of
quality.”

CQA
Knowledge Space: “A
summary of all process
knowledge obtained
during product
development.”


There are no
regulatory
requirements to
have a Design
Space


 Design space can illustrate understanding of
parameter interactions and provides manufacturing
flexibility
 Proven acceptable range alone is not a design
space
 Design space should be verified and operational
at full scale
 No requirement to develop a design space at the
full manufacturing scale


 A design space can be described in terms of ranges
of material attributes and process parameters.
 It can also be described through more complex
mathematical relationships.
 It is possible to describe a design space as a time
dependent function (e.g., temperature and pressure
cycle of a lyophilisation cycle), or
 As a combination of variables such as components
of a multivariate model.


Design Space


 Scaling factors can also be included if the design
space is intended to span multiple operational
scales.
 Analysis of historical data can contribute to the
establishment of a design space.
 Regardless of how a design space is developed, it
is expected that operation within the design
space will result in a product meeting the defined
quality.


 Independent design spaces can be established for one or
more unit operations, or
 Single design space that spans multiple operations can
also be established.
 A separate design space for each unit operation is often
simpler to develop.
 However a design space that spans the entire process can
provide more operational flexibility.
 For example, in the case of a drug product that undergoes
degradation in solution before lyophilisation, the design
space to control the extent of degradation
(e.g., concentration, time, temperature) could be
expressed for each unit operation or as a sum over all
unit operations.


Identify Q T P P

Identify C Q A Risk Assessment

Define product design space

Define process design space Risk Assessment

Refine process design space Process Characterization

Define Control strategy Risk Assessment

Process Validation

Process Monitoring


 Consider QTPP in establishing the Design Space
 Initial determination of CQAs
 Assess prior knowledge to understand variables and
their impact
 Scientific principles & historical experience
 Perform initial risk assessment of manufacturing
process relative to CQAs to identify the high risk
manufacturing steps (->CPPs)
 Conduct Design of Experiments (DoE)
 Evaluate experimental data
 Conduct additional experiments/analyses as needed


 First-principles approach
◦ Combination of experimental data and
mechanistic knowledge of chemistry, physics, and
engineering to model and predict performance
 Statistically designed experiments (DOEs)
◦ Efficient method for determining impact of
multiple parameters and their interactions
 Scale-up correlation
◦ A semi-empirical approach to translate operating
conditions between different scales or pieces of
equipment


 The risk assessment and process development
experiments can lead to an understanding of the
linkage and effect of process parameters and
material attributes on product CQAs and
 Also help identify the variables and their ranges
within which consistent quality can be achieved.
 These process parameters and material attributes
can thus be selected for inclusion in the design
space.


 Prior knowledge may include :
 Internal knowledge from development and
manufacturing
 External knowledge: scientific and technical
publications (including literature and peer-reviewed
publications)
 Citation in filing: regulatory filings, internal company
report or notebook, literature reference
 No citation necessary if well known and accepted by
scientific community


 Risk assessment is based on prior knowledge and
relevant experience for the product and
manufacturing process
 Gaps in knowledge could be addressed by further
experimentation
 Assignments of risk level must be appropriately
justified
 Risk assessments/control will iterate as relevant new
information becomes available
 Final iteration shows control of risks to an
acceptable level


◦ Design space could include critical and non-critical
parameters
 Critical parameter ranges/model are considered a regulatory
commitment and non-critical parameter ranges support the
review of the filing
 Critical parameter changes within design space are handled by
the Quality System and changes outside the design space need
appropriate regulatory notification
◦ Non-critical parameters would be managed by Quality
System


 DOE useful tool in development of a DS but
not the only one
◦ 1st Principles models
 DS may cover one, or multiple unit
operation(needs to be clear in the dossier)
 Not all unit operations must have a DS
 Unit operations without a DS will obviously
not achieve the regulatory benefits (ie, ability
to move within DS)


 DS is usually developed at lab scale
 There is no need to perform full DOEs at full scale to
confirm the DS at full scale.
 Good understanding of scale up phenomena is
needed, some parameters may be scale independent
(needs to be justified)
 Scale up factors could be used to reduce concern about
moving within DS at scale
 Experiments within the DS at full scale could also be used
to reduce the same concern.
 Another option is have additional monitoring controls
applied when there is a change within the DS to ensure
that the DS is still valid and then relaxation to a less
stringent control strategy.

 DS needs to be complemented by an appropriate
control strategy
 Critical process parameters remain critical even if
 controlled,
 CQAs: appropriate specs need to be set, even if
not tested routinely
 Release based on CQAs and control of process
parameters is possible if satisfactorily
demonstrated ( e.g. dissolution release based on
particle size control, and disintegration test)


 One-factor-at-a-time (the classical approach)
 Designed experiments (DOE)


 One-factor-at-a-time
◦ Procedure (2 level example)
 Run all factors at one condition
 Repeat, changing condition of one factor
 Continuing to hold that factor at that condition, rerun
with another factor at its second condition
 Repeat until all factors at their optimum conditions
◦ Slow, expensive: many tests
◦ Can miss interactions!


Process: Yield = f(temperature, pressure)

50% yield

30% yield

Max yield: 50% at 78 C, 130 psi?


A better view of the maximum yield!

Optimized yield is over 85%

Process: Yield = f(temperature, pressure)


 Multiple-Factors-at-a-Time, DOE
◦ Full Factorials
◦ Fractional Factorials
◦ Plackett – Burman designs
◦ Central Composite designs


 DOE is defined as “a structured analysis
wherein inputs are changed and differences
or variations in outputs are measured to
determine the magnitude of the effect of each
of the inputs or combination of inputs.”
 Full factorial example:
Dependent
Independent Variable Variable
(Controlling Factors) (Response)
Run Factor X1 Factor X2 Factor Y1
1 High High Output1
2 Low High Output2
3 High Low Output3
4 Low Low Output4

(1) Choose experimental design
(e.g., full factorial, d-optimal) (2) Conduct randomized
experiments
Experiment Factor A Factor B Factor C

1 + - -
A
2 - + -
3 + + +
B
C 4 + - +
(3) Analyze data
(4) Create multidimensional
surface model
(for optimization or control)

www.minitab.com
www.drugragulations.org
33

 Several families
 n = number of Factor tested and L : level/factor
 Semi Factorial Design : the lowest number of experiments required : 2n-k
 Used for a first screening of mains factors and at least single interactions
 Used for demonstration of a Proven Acceptable Range (PAR) or Design Space
◦ Don‟t be afraid by the number of factor.
 Factorial Design : higher number of experiments : 2n
 For both Design, only two levels (L = 2) + eventual central point(s), Models will always
be linear.
 Response Surface Model : higher number of experiments : Ln. Non linear models.
The number of experiments can be decreased by historical methods or by computer
optimisation (D Optimal).
◦ Used for optimisation/ modeling of a process
◦ Used for searching the „‟ edge of failure‟‟
 Mixture : RSM + constraint (sum of component = fixed value)
 Used in chemistry, formulation,…
 Combined : Mixture + (semi) Factorial or RSM
 Used for combines mixture/process such as formulation (excipents) and freeze drying
conditions.

34

 The kind of question to answer must be
understood :
◦ Critical parameters
◦ Interactions
◦ Optimisation
◦ Demonstration of Proven Acceptable Range
◦ Modeling
 The experiments are planned before starting
 Apparently a high number of experiments, more
work, more time, more money.
 In reality, far less experiments (semi factorial or
reduction for RSM) to obtain far less valuables
results. Allow a better planning of experiments
including Analytical.

35

 Randomization, blocking and replication are the three
basic principles of statistical experimental design.
 By properly randomizing the experiment, the effects of
uncontrollable factors that may be present can be
“averaged out”.
 Blocking is the arrangement of experimental units into
groups (blocks) that are similar to one another.
 Blocking reduces known but irrelevant sources of variation
between groups and thus allows greater precision in the
estimation of the source of variation under study.
 Replication allows the estimation of the pure experimental
error for determining whether observed differences in the
data are really statistically different


 ANOVA results should accompany all DOE data
analysis, especially if conclusions concerning the significance of
the model terms are discussed.
 For all DOE data analysis, the commonly used alpha of 0.05 is
chosen to differentiate between significant and non significant
factors.
 It is important that any experimental design has sufficient power
to ensure that the conclusions drawn are meaningful.
 Power can be estimated by calculating the signal to noise ratio.
 If the power is lower than the desired level, some remedies can
be employed to increase the power.
 For example, by adding more runs, increasing the signal or
decreasing the system noise.
 ICH Points to Consider document for guidance on the level of
DOE documentation recommended for regulatory submissions.


 A design space can be updated over the lifecycle as
additional knowledge is gained.
 Risk assessments, as part of the risk management process,
help steer the focus of development studies and define the
design space.
 Operating within the design space is part of the control
strategy.
 The design space associated with the control strategy
ensures that the manufacturing process produces a product
that meets
◦ The Quality Target Product Profile (QTPP) and
◦ Critical Quality Attributes (CQAs).


 Since design spaces are typically developed at small
scale, an effective control strategy helps manage
potential residual risk after development and
implementation.
 When developing a design space for a single-unit
operation, the context of the overall manufacturing
process can be considered, particularly immediate
upstream and downstream steps that could interact
with that unit operation.
 Potential linkages to CQAs should be evaluated in
design space development.


 In developing design spaces for existing
products, multivariate models can be used for
retrospective evaluation of historical production
data.
 The level of variability present in the historical
data will influence the ability to develop a design
space, and additional studies might be
appropriate.


 Design spaces can be based on
◦ scientific first principles and/or
◦ empirical models.
 An appropriate statistical design of experiments
incorporates a level of confidence that applies to
the entire design space, including the edges of an
approved design space.


 However, when operating the process near the
edges of the design space, the risk of excursions
from the design space could be higher because of
normal process variation (common cause
variation).


 The control strategy helps manage residual risk
associated with the chosen point of operation
within the design space.
 When changes are made
(e.g., process, equipment, raw material
suppliers), results of risk review can provide
information regarding additional studies and/or
testing that might verify the continued
applicability of the design space and associated
manufacturing steps after the change.


 Capturing development knowledge and
understanding contributes to design space
implementation and continual improvement.
 Different approaches can be considered when
implementing a design space (e.g., process
ranges, mathematical expressions, or feedback
controls to adjust parameters during processing
(see also Figure 1d in ICH Q8(R2)).
 The chosen approach would be reflected in the
control strategy to assure the inputs and process
stay within the design space.


 Although the entire design space does not have
to be reestablished (e.g., DoE) at commercial
scale, design spaces should be initially verified as
suitable prior to commercial manufacturing.
 Design space verification should not be confused
with process validation.
 However, it might be possible to conduct
verification studies of the performance of the
design space scale-dependent parameters as part
of process validation.


 Design space verification includes monitoring or
testing of CQAs that are influenced by scale-
dependent parameters.
 Additional verification of a design space might be
triggered by changes (e.g., site, scale, or
equipment).
 Additional verification is typically guided by the
results of risk assessments of the potential
impacts of the change(s) on design space.


 A risk-based approach can be applied to
determine the design of any appropriate studies
for assessment of the suitability of a design space
across different scales.
 Prior knowledge and first principles, including
simulation models and equipment scale-up
factors, can be used to predict scale-independent
parameters.
 Experimental studies could help verify these
predictions.


 Some aspects of the design space that could be
considered for inclusion in the regulatory submission:
 The design space description, including critical and
other relevant parameters.
 The design space can be presented as ranges of
material inputs and process parameters, graphical
representations, or through more complex
mathematical relationships.
 The relationship between the inputs (e.g., material
attributes and/or process parameters) and the CQAs,
including an understanding of the interactions among
the variables.


 Data supporting the design space, such as prior
knowledge, conclusions from risk assessments as part
of QRM, and experimental studies with supporting
data, design assumptions, data analysis, and models.
 The relationship between the proposed design space
and other unit operations or process steps.
 Results and conclusions of the studies, if any, of a
design space across different scales.
 Justification that the control strategy ensures that the
manufacturing process is maintained within the
boundaries defined by the design space.


 The control strategy used for implementation
of a design space in production depends on
the capabilities of the manufacturing site.
 The batch records reflect the control strategy
used.
 For example, if a mathematical expression is
used for determining a process parameter or
a CQA, the batch record would include the
input values for variables and the calculated
result.


 As part of the technology transfer of a design space
to a site and throughout the lifecycle, it is important
to share the knowledge gained during development
and implementation that is relevant for using that
design space both on the manufacturing floor and
under the PQS of the company or site.
 This knowledge can include results of risk
assessments, assumptions based on prior knowledge,
and statistical design considerations.
 Linkages among the design space, control strategy,
CQA, and QTPP are an important part of this shared
knowledge.


 Each company can decide on the approach
used to capture design space information and
movements within the design space under the
applicable PQS, including additional data
gained through manufacturing experience
with the design space.
 In the case of changes to an approved design
space, appropriate filings should be made to
meet regional regulatory requirements.


 Movement within the approved design
space, as defined in the ICH Q8(R2)
glossary, does not call for a regulatory filing.
 For movement outside the design space, the
use of risk assessment could be helpful in
determining the impact of the change on
quality, safety, and efficacy and the
appropriate regulatory filing strategy, in
accordance with regional requirements.


 A model is a simplified representation of a
system using mathematical terms.
 Models can enhance scientific understanding
and
 Possibly predict the behavior of a system
under a set of conditions.
 Mathematical models can be used at every
stage of development and manufacturing.


 They can be derived from
◦ first principles reflecting physical laws (such as
mass balance, energy balance, and heat transfer
relations), or
◦ From data, or
◦ From a combination of the two.


 There are many types of models.
 The selected one will depend on
◦ The existing knowledge about the system,
◦ The data available, and
◦ The objective of the study.


 Models can be categorized in multiple ways.
 The categorization approaches are intended
to facilitate the use of models across the
lifecycle, including
◦ Development,
◦ Manufacturing,
◦ Control, and
◦ Regulatory processes.


 For the purposes of regulatory submissions,
an important factor to consider is the model‟s
contribution in assuring the quality of the
product.
 The level of oversight should be
commensurate with the level of risk
associated with the use of the specific
model.


 Low-Impact Models:
These models are typically used to support
◦ Product and/or
◦ Process development
◦ (e.g., formulation optimization).


 Medium-Impact Models:
Such models can be useful in assuring quality
of the product.
 However these models are not the sole
indicators of product quality
 (e.g., most design space models, many in-
process controls).


 High-Impact Models:
A model can be considered high impact if
prediction from the model is a significant
indicator of quality of the product.
 (e.g., a chemometric model for product assay,
a surrogate model for dissolution).


 For the purpose of implementation, models
can also be categorized on the basis of the
intended outcome of the model.
 Within each of these categories, models can
be further classified as
◦ Low,
◦ Medium or
◦ High,
 Classification based on their impact in
assuring product quality.


 Models for supporting process design:
This category of models includes (but is not
limited to) models for
◦ Formulation optimization,
◦ Process optimization
 (e.g., reaction kinetics model),
◦ Design space determination, and
◦ Scale-up.


 Models for supporting process design:
 Models within this category can have different
levels of impact.
 For example, a model for design space
determination would generally be considered
a medium-impact model,
 While a model for formulation optimization
would be considered a low-impact model.


 Models for supporting analytical procedures:
In general, this category includes empirical
(i.e., chemometric) models based on data
generated by various Process Analytical
Technology (PAT)-based methods.


 Models for supporting analytical procedures:
 A calibration model associated with a near
infrared (NIR)-based method.
 Models for supporting analytical procedures
can have various impacts depending on the
use of the analytical method.
 For example, if the method is used for release
testing, then the model should be high-
impact.


 Models for process monitoring and control:
◦ Univariate Statistical Process Control (SPC) or
◦ Multivariate Statistical Process Control (MSPC)-
based models:
 These models are used to detect special
cause variability;
 The model is usually derived and the limits
are determined using batches manufactured
within the target conditions.


 If an MSPC model is used for continuous
process verification along with a traditional
method for release testing, then the MSPC
model would likely be classified as a
medium-impact model.


 However, if an MSPC model is used to support
a surrogate for a traditional release testing
method in an RTRT approach, then the model
would likely be classified as a high-impact
model.


 Models used for process control (e.g., feed
forward or feedback).
 Data-driven models should be developed
through appropriately designed experiments.
 These models are typically medium-impact or
high-impact.
 For example, a feed forward model to adjust
compression parameters on the basis of
incoming material attributes could be
classified as a medium-impact model.


 Sequential steps
 Steps can be repeated to impart an iterative
nature to this process.
 Overall steps are given in following slides:


1. Defining the purpose of the model.
2. Deciding on the type of modeling approach.
◦ (e.g. mechanistic or empirical) and
◦ Possible experimental/sampling methodology to
be used to support the model development.


3. Selecting variables for the model; this is
typically based on
◦ Risk assessment,
◦ Underlying physicochemical phenomena,
◦ Inherent process knowledge, and
◦ Prior experience.


4. Understanding the limitations of the model
assumptions to:
◦ Correctly design any appropriate experiments;
◦ Interpret the model results; and
◦ Include appropriate risk-reduction strategies.


5. Collecting experimental data to support
model development.
◦ These data can be collected at
 Laboratory,
 Pilot, or
 Commercial scale, (depending on the nature of the model. )

◦ It is important to ensure that variable ranges
evaluated during model development are
representative of conditions that would be
expected during operation.


6. Developing model equations estimating
parameters, based on a scientific
understanding of the process and collected
experimental data.


7. Validating the model, as appropriate.
8. In certain cases, evaluating the impact of
uncertainty in model prediction on product
quality.
◦ If appropriate, defining an approach to reduce
associated residual risk
 (e.g., by incorporating appropriate control strategies
(this can apply to high-impact and medium-impact
models)).


9. Documenting the outcome of model.
◦ Development
◦ Assumptions
 Developing plans for verification and update
of the model throughout the lifecycle of the
product.
 The level of documentation would be
dependent on the impact of the model


 Model validation is an essential part of model
development and implementation.
 Once a model is developed and
implemented, verification continues
throughout the lifecycle of the product.


 In the case of well-established first principles-
driven models, prior knowledge can be leveraged
to support model validation and verification, if
applicable.
 The following elements can be considered for
model validation and verification and generally
are appropriate for high-impact models
 The applicability of the elements listed below for
medium-impact or low-impact models can be
considered on a case-by-case basis.


 Acceptance criteria relevant to the purpose and
to its expected performance.
 In setting the acceptance criteria, variability in
sampling procedure (e.g., for blending) could
also be considered.
 In situations where the model is to be used to
support a surrogate for a traditional release
testing method, the accuracy of the model
performance versus the reference method could
be considered.


 For example, a multivariate model (e.g. a
partial least squares (PLS) model), when
appropriate, can be used as a surrogate for
traditional dissolution testing.
 In this case, the PLS model should be
developed in terms of in-process parameters
and material attributes and can be used to
predict dissolution.


 One of the ways to validate and verify model
performance in this case would be to
compare accuracy of prediction of the PLS
model with the reference method (e.g., a
traditional dissolution method).


 Comparison of the accuracy of calibration
versus the accuracy of prediction.
 This can often be approached through
internal cross-validation techniques using the
same data as the calibration data set.


 It can be beneficial to verify the prediction accuracy of the
model by parallel testing with the reference method during
the initial stage of model implementation.
 This testing can be repeated throughout the lifecycle, as
appropriate.
 If models are used to support a design space at
commercial scale or are part of the control strategy, it is
important to verify the model at commercial scale.
◦ If a calibration model associated with an NIR-based method is
developed at the laboratory scale and the method is then
transferred to and used in commercial scale.


 In addition, the data sets used for calibration,
internal validation, and external validation
should take into account the variability
anticipated in future routine production
◦ (e.g., a change in the source of raw material that
might impact NIR prediction).
 Low-impact models typically do not call for
verification.


 Approaches for model verification can be
documented according to the PQS of the
company and can include the following:
◦ A risk-based frequency of comparing the model‟s
prediction with that of the reference method,
◦ Triggers for model updates (e.g., because of changes in
raw materials or equipment),
◦ Procedures for handling model-predicted Out of
Specification (OOS) results,
◦ Periodic evaluations, and approaches to model
recalibration


 The level of detail for describing a model in a
regulatory submission is dependent on the
impact of its implementation in assuring the
quality of the product.
 For the various types of models, the applicant
can consider including:


 Low-Impact Models:
A discussion of how the models were used to
make decisions during process development.


 Medium-Impact Models:
◦ Model assumptions,
◦ A tabular or graphical summary of model inputs and
outputs,
◦ Relevant model equations (e.g., for mechanistic models),
◦ Statistical analysis where appropriate,
◦ a comparison of model prediction with measured data,
and
◦ A discussion of how the other elements in the control
strategy help to mitigate uncertainty in the model, if
appropriate.


 High-Impact Models:
Data and/or prior knowledge (e.g., for established first
principles-driven models) such as
◦ Model assumptions,
◦ Appropriateness of the sample size, number and distribution of
samples,
◦ Data pretreatment,
◦ Justification for variable selection,
◦ Model inputs and outputs,
◦ Model equations,
◦ Statistical analysis of data showing fit and prediction ability,
◦ Rationale for setting of model acceptance criteria,
◦ Model validation (internal and external), and
◦ A general discussion of approaches for model verification during
the lifecycle.


Rittinger’s law: The work required in crushing is
proportional to the new surface created.

Where: P=power required, dm/dt=feed rate to crusher, Dsb =
ave diameter before crushing, DSQ=ave after crushing,
Kr=Rittinger’s coef.

Kick’s law: the work required for crushing a given mass of
material is constant for the same reduction ratio, that is the
ratio of the initial particle size to the finial particle size

Kk=Kick’s coef.

For fine grains, the Characteristic region
boundary between
the characteristic
Blender head space
region and the
remaining powder
bed is parabolic in
shape

n
m o m
The powder bed Vr rV V r 1
below the boundary
r 1
rotates with the
mixer as a solid as fraction mixed
body.
n
f rm rf o
f rm1
r 1

0.40
Avicel® PH-200 compacts

VFS Speed: 200 rpm
0.35 HFS Speed: 30 rpm
Roll Pressure: 6560 lb/in
Slope of NIR Spectrum

0.30

Roll Speed (RPM)
0.25
y = 0.3672x + 0.1754
4 5 6
R2 = 0.9899
7 8 9
0.20
10 11 12

0.15
0.0 0.1 0.2 0.3 0.4 0.5 0.6
20
Force at break/Thickness/Width (N/mm2)
18 Avicel® PH-200 compacts

 The strength is a 16
VFS Speed: 194 - 197 rpm
HFS Speed: 29 - 30 rpm
Roll Gap: 0.031 - 0.038"

linear function of the 14 Roll Pressure: 6551 lb/in
Force at break (N)

12
density which is 10

monitored by NIR 8
y = 21.54e
-0.4493x

 Semi Empirically
6
R2 = 0.9884
4

F=(SNIR-0.17)/0.37 2

0
4 5 6 7 8 9 10 11 12
Roll Speed (RPM)

Avicel® PH-200 Milled Compacts

1000 Increaing Roll Speed
Day1

Day2
800
Particle Size ( m)

d90
600

400
d50

200 d10

0
3 4 5 6 7 8 9 10 11 12 13
Roll Speed (rpm) Avicel® PH-200 Milled Compacts
1200
Increaing Roll Speed

 The particle sizes d90 Day1

of the milled
1000
Day2

material is also 800
Particle Size ( m)

manifest in the 600
d50

slope of the NIR
signal (as
400
d10

predicted) 200

0
2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
1 / Slope NIR Spectrum

Optimum
Conditions

(u,v)= *(u,v)
0
Equilibrium Moisture Content

SIZE
M=M0-Kt

M=M0’exp(-K’t)
18 0.600

GRANULATION TIME
16
0.550
14
Moisture Content (% w/w)

Mean Particle Size (mm)
12 0.500
Moisture Content
Particle Size
10
0.450
8

6 0.400

4
0.350
2

0 0.300
0 20 40 60 80 100
Elapsed Time (min)

Funicular Modeling
Wet
Granulation
Pendular

Over
Wetting

Droplet
Capillary
Drying

0.0010 X1=110 g
H13 (1 min) (=X3)
H15 (3.5 min) X2=255 rpm
NIR Treated Response

H14 (6 min)
0.0008
610 m
410 m
Slope

320 m
0.0006

0.0004

MIXING SPRAYING WET MASSING
0.0002
0 100 200 300 400 500 600

Process time (s)

[Kunii and Levenspiel, Fluidization Engineeri
[Kunii and Levenspiel, Fluidization Engineering, Pub. Krieger, pg. 424-428,1977]

180 65

Evaporative

Moisture Content

Temperature
63
160
Q Qo Kt
61

Critical
140 moisture

Temperature (°C)
59
MM55 Reading

T 57
120

55

100
53

80 Diffusive 51

MM55 Q Q Q'ok EXP( k' t) 49

60
47

40 45
0 5 10 15 20 25 30

Drying Time (min)

K.R. Morris, S.L. Nail, G.E. Peck, S.R. Byrn, U.J. Griesser, J.G. Stowell, S.-J. Hwang, K. Park Pharm Sci Tech Today 1 6 235–245 (1998).

235
NIR Monitor (Arbitrary Values)

215

195

175

155

135

115 Conventional Drying

Fast Drying
95

75
0.00 5.00 10.00 15.00 20.00 25.00
Time (min)

Morris et.al., Drug Dev. Ind. Pharm., 26 (9):985-

60.00
240.0
Average Exhaust Temp
55.00
220.0
MM55 Gauge Reading

200.0 50.00

180.0 45.00

160.0
40.00
140.0
Active Melting Temp 35.00
120.0
30.00
100.0

80.0 25.00

60.0 20.00
0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0 40.0
Elapsed Drying Time (min)
120
Batch 018 sub1 Batch 018 sub2 Batch 019 sub1 Batch 019 sub2 Batch 020 sub1 Batch 021 sub1 Batch 021 sub2

Average % Release 100

80

60

40
Baseline Data
20
Temp Excursion
0
0 20 40 60 80 100 120
Time (min)

WHOLE TABS HALF TABS QUARTER TABS
Active 1 Active 2 Active 1 Active 2 Active 1 Active 2
MEAN 101.9 100.9 101.8 99.6 102.1 100.5
SD 0.7 1.6 1.4 2.8 2.4 5.1
CV (%) 0.7 1.6 1.3 2.8 2.3 5.1

CU for constant size portions of tablets must be larger than for the
whole, so in spec using real time monitoring of “part” of the tablets
means in spec for the whole tablet
CVP CVT
T. Li, et. al., in press Pharm. Res.
BioMed Anal.

HPMC and Sulfanilamide Calculations (Peak Height)
0.4

0.35
Absorbance (log(1/R))

0.3
HPMC
Sulfanilamide
0.25

0.2

0.15
0 30 60 90 120 150
Elapsed Time (min)

200 700

100
500
Sulfanilamide Gauge

0

M oisture Gauge
Sulfanilam ide 300
Moisture
-100

100
-200

-100
-300

-400 -300
0 20 40 60 80 100 120 140 160
Elapsed Time (min)

 These principles and techniques are applicable to
batch and continuous processing and may be
linked by multi-variate (chemometric) methods
after univariate conformation.
 Ultimately this give us the ability to understand
how development variables interact to influence the
final product and to design in the quality

10

Quality by Design for ANDAs:
An Example for Immediate-Release
Dosage Forms
Published by FDA


Aqueous 0.1 N HCL 0.015 mg/ml
solubility as a
pH 4.5 buffer 0.015 mg/ml
function of
pH:
pH 6.8 buffer 0.015 mg/ml
Hyroscopicity Acetriptan Form III is non-hygroscopic and requires no special protection
from humidity during handling, shipping or storage
Density (Bulk, • Bulk density: 0.27 g/cc
Tapped, and • Tapped density: 0.39 g/cc
True) and • True density: 0.55 g/cc
Flowability: • The flow function coefficient (ffc) was 2.95 and the Hausner ratio was
1.44 which both indicate poor flow properties.
Chemical • pKa: Acetriptan is a weak base with a pKa of 9.2.
properties • Overall, acetriptan is susceptible to dry heat, UV light and oxidative
degradation.

Biological • Partition coefficient: Log P 3.55 (25 °C, pH 6.8)
properties • Caco-2 permeability: 34 × 10-6 cm/s. Therefore, acetriptan is highly
permeable.
• BCS Class II compound (low solubility and high permeability)


Drug Substance Attributes
Drug Solid PSD Hygrosc Solubil Mois Residual Process Chemi Flow
Product State opicity ity ture Solvent Impurit cal prop
Cont
CQA Form ies stabili
ent
ty
Assay Low Med Low Low Low Low Low High Med
CU Low High Low Low Low Low Low Low High
Dissolution High High Low High Low Low Low Low Low

Degradation Med Low Low Low Low Low Low High Low
products


Component Function Unit Unit
( mg/tablet) ( % W/W)
Acetriptan, USP Active 20 10
Lactose Monohydrate, NF Filler 64-86 32-43
Microcrystalline Cellulose Filler 72-92 36-46
(MCC), NF
Croscarmellose Sodium Disintegrant 2-10 1-5
(CCS), NF
Magnesium Stearate, NF* Lubricant 2-6 1-3
Talc, NF Glidant/Lubricant 1-10 0.5-5
Total tablet weight 200 100


Formulation Variables
Drug product DS PSD MCC/ CCS Level Talc Level Mag Stearate
CQA Lactose Level
ratios
Assay Medium Medium Low Low Low
Content High High Low Low Low
Uniformity
Dissolution High Medium High Low High

Degradation Low Low Low Low Medium
Products


 Formulation development focused on evaluation of
the high risk formulation variables as identified in the
initial risk assessment shown earlier.
 The development was conducted in two stages.
 The first formulation study evaluated the impact of
the drug substance particle size distribution, the
MCC/Lactose ratio and the disintegrant level on the
drug product CQAs.
 The second formulation study was conducted to
understand the impact of extragranular magnesium
stearate and talc level in the formulation on product
quality and manufacturability.
 Formulation development studies were conducted at
laboratory scale (1.0 kg, 5,000 units).

 Goal of Formulation Development Study #1
 Select the MCC/Lactose ratio and
 Disintegrant level and
 To understand if there was any interaction of
these variables with drug substance particle size
distribution.
 This study also sought to establish the
robustness of the proposed formulation.
 A 2³ full factorial Design of Experiments (DOE)
with three center points was used to study the
impact of these three formulation factors on the
response variables.


Process step Equipment
Pre-Roller Compaction 4 qt V-blender
Blending and Lubrication o 250 revolutions for blending (10 min at 25 rpm)
Alexanderwerk10 WP120 with 25 mm roller width and 120
mm roller diameter
o Roller surface: Knurled
Roller Compaction and o Roller pressure: 50 bar
Integrated Milling o Roller gap: 2 mm
o Roller speed: 8 rpm
o Mill speed: 60 rpm
o Coarse screen orifice size: 2.0 mm
o Mill screen orifice size: 1.0 mm
Final Blending and 4 qt V-blender
Lubrication o 100 revolutions for granule and talc blending (4 min at 25
rpm)
o 75 revolutions for lubrication (3 min at 25 rpm)
16-station rotary press (2 stations used)
o 8 mm standard round concave tools
Tablet Compression o Press speed: 20 rpm
o Compression force: 5-15 kN
o Pre-compression force: 1 kN


Factors : Formulation Variables Levels
-1 0 +1
A Drug substance PSD (d90, μm) 10 20 30

B Disintegrant (%) 1 3 5

C % MCC in MCC/Lactose combination 33.3 50 66.7


Responses Goal Acceptable
Range
Y1 Dissolution at 30 min (%) (with hardness of 12.0 kP) Maximize ≥ 80%

Y2 Disintegration time (min) (with hardness of 12.0 kP) Minimize < 5 min

Y3 Tablet content uniformity (% RSD) Minimize % RSD < 5%

Y4 Assay (% w/w) Target at 100 % 95.0 to 105.0 w/w
Y5 Powder blend flow function coefficient ( ffc) Maximize >6

Y6 Tablet Hardness @ 5 kN ( kP ) Maximize > 5 kP

Y7 Tablet Hardness @ 10 kN ( kP ) Maximize > 9 kP

Y8 Tablet Hardness @ 15 kN ( kp ) Maximize > 12 kP

Y9 Friability@ 5 kN ( kp ) Maximize <1%

Y10 Friability@ 10 kN ( kp ) Maximize <1%

Y11 friability@ 15 kN ( kp ) Maximize <1%

Y12 Degradation products (%) (observed at 3 months, 40 Minimize ACE12345: NMT 0.5%
°C/75% RH) Any unknown impurity:
NMT 0.2%
Total impurities: NMT
1.0%


A B C Y1 Y3 Y5 Y7
Batch DS PSD Disintegra % MCC Dissolution Content Ffc Tablet
No nt level in MCC/ in Uniformity Value Hardness
Lactose 30 min @ 10 kN
Mix
(d90, μm) ( %) (%) (%) ( % RSD ) -- (kP)

1 30 1 66.7 76.0 3.8 7.56 12.5

2 30 5 66.7 84.0 4.0 7.25 13.2

3 20 3 50.0 91.0 4.0 6.62 10.6

4 20 3 50.0 89.4 3.9 6.66 10.9

5 30 1 33.3 77.0 2.9 8.46 8.3

6 10 5 66.7 99.0 5.1 4.77 12.9

7 10 1 66.7 99.0 5.0 4.97 13.5

8 20 3 50.0 92.0 4.1 6.46 11.3

9 30 5 33.3 86.0 3.2 8.46 8.6

10 10 1 33.3 99.5 4.1 6.16 9.1

11 10 5 33.3 98.7 4.0 6.09 9.1


 Initially, dissolution was tested using the FDA-
recommended method.
 All batches exhibited rapid and comparable
dissolution (> 90% dissolved in 30 min) to the RLD.
 All batches were then retested using the in-house
dissolution method .
 Results are presented in earlier table.
 Since center points were included in the DOE, the
significance of the curvature effect was tested
using an adjusted model.
 The Analysis of Variance (ANOVA) results are
presented in next table


Source Sum of df Mean F value P value Comme
squares square nts
Model 742.19 3 247.40 242.94 < 0.0001 Significant

A- Drug Substance PSD (d90, μm) 699.8 1 699.78 657.72 < 0.0001 Significant

B- Disintegrant ( % ) 32.81 1 32.81 32.21 0.0013 Significant

AB – Interaction 39.61 1 39.61 38.89 0.0008 Significant

Curvature 1.77 1 1.77 1.74 0.2358 Not
Significant

Residual 6.11 6 1.02 --- ----- ----

Lack of fit 2.67 4 0.67 0.39 0.8090 Not
Significant

Pure error 3.44 2 1.72 ---- ---- ------

Total 750.07 10 --- ---- ----- -----


 The curvature effect was not significant for
dissolution;
 Therefore, the factorial model coefficients were
fit using all of the data (including center points).
 As shown in ANOVA results of the unadjusted
model (next slide), the significant factors
affecting tablet dissolution were
 A (drug substance PSD),
 B (disintegrant level) and
 AB (an interaction between drug substance PSD
and the intragranular disintegrant level).


Source Sum of df Mean F value P value Comme
squares square nts
Model 742.19 3 247.40 219.84 < 0.0001 Significant

A- Drug Substance PSD (d90, μm) 699.8 1 699.78 595.19 < 0.0001 Significant

B- Disintegrant ( % ) 32.81 1 32.81 29.15 0.0010 Significant

AB – Interaction 39.61 1 39.61 35.19 0.0006 Significant

Residual 7.88 7 1.13 --- ----- ----

Lack of fit 4.44 5 0.89 0.52 0.7618 Not
Significant

Pure error 3.44 2 1.72 ---- ---- ------

Total 750.07 10 --- ---- ----- -----


Under Quality by Design, establishing a design
space or using real-time release testing is not
necessarily expected (ICH Q8(R2)).

12

 It is not necessary to study multivariate
interactions of all parameters to develop a
design space.
 The applicant should justify the choice of
material attributes and parameters for
multivariate experimentation based on risk
assessment and desired operational
flexibility.

12

 When appropriately justified design space can be
applicable to scale-up.
 Design space can be applicable to a site change.
 It is possible to justify a site change using a site
independent design space based on a
demonstrated understanding of the robustness
of the process and an in depth consideration of
site specific factors (e.g., equipment, personnel,
utilities, manufacturing environment, and
equipment).

12

 There are region specific regulatory
requirements associated with site changes
that need to be followed.
 Design space can be developed for a single
unit operations or across a series of unit
operations.

12

 It is possible to develop a design space for
existing products.
 Manufacturing data and process knowledge
can be used to support a design space for
existing products.
 Relevant information should be utilized from
◦ Commercial scale manufacturing,
◦ Process improvement,
◦ Corrective and preventive action (CAPA), and
◦ Development data

12

 For manufacturing operations run under narrow
operational ranges in fixed equipment, an expanded
region of operation and an understanding of multi
parameter interactions may not be achievable from
existing manufacturing data alone.
 Additional studies may provide the information to
develop a design space.
 Sufficient knowledge should be demonstrated, and
the design space should be supported experimentally
to investigate interactions and establish
parameter/attribute ranges.

12

 There is no regulatory expectation to develop a
design space for an existing product.
 Development of design space for existing
products is not necessary unless the applicant
has a specific need and
 Desires to use a design space as a means to
achieve a higher degree of product and process
understanding.
 This may increase manufacturing flexibility
and/or robustness.

12

 Design space can be applicable to formulations.
 It may be possible to develop formulation (not
component but rather composition) design space
consisting of the
◦ ranges of excipient amount and
◦ its physicochemical properties (e.g., particle size
distribution, substitution degree of polymer)
 Based on an enhanced knowledge over a wider
range of material attributes.

12

 The applicant should justify the rationale for
establishing the design space with respect to
quality attributes such as
◦ bioequivalence,
◦ stability,
◦ Manufacturing
◦ robustness etc.
 Formulation adjustment within the design space
depending on material attributes does not need a
submission in a regulatory postapproval change.

12

 A set of proven acceptable ranges alone does not
constitute a design space.
 A combination of proven acceptable ranges
(PARs) developed from univariate
experimentation does not constitute a design
space
 Proven acceptable ranges from only univariate
experimentation may lack an understanding of
interactions between the process parameters
and/or material attributes.

12

 However proven acceptable ranges continue
to be acceptable from the regulatory
perspective but are not considered a design
space.
 The applicant may elect to use proven
acceptable ranges or design space for
 different aspects of the manufacturing
process

13

 Outer limits of the design space need not be
evaluated during process validation studies at
the commercial scale.
 There is no need to run the qualification
batches at the outer limits of the design
space during process validation studies at
commercial scale.
 The design space should be sufficiently
explored earlier during development studies.

13

 “If the experimental design is poorly chosen, so
that the resultant data do not contain much
information, not much can be extracted, no
matter how thorough or sophisticated the
analysis.
 On the other hand, if the experimental design
is wisely chosen, a great deal of information in
readily extractable form is usually available,
and no elaborate analysis may be necessary.
 In fact, in many happy situations all the
important conclusions are evident from visual
examination of the data.”


Product Profile  Quality Target Product Profile (QTPP)

CQA’s  Determine “potential” critical quality attributes (CQAs)

Risk Assessments  Link raw material attributes and process parameters to
CQAs and perform risk assessment
Design Space  Develop a design space (optional and not required)

Control Strategy  Design and implement a control strategy

Continual  Manage product lifecycle, including continual
Improvement
improvement


Quality by Design : Design Space

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (20)

En vedette

En vedette (6)

Similaire à Quality by Design : Design Space

Similaire à Quality by Design : Design Space (20)

Plus de GMP EDUCATION : Not for Profit Organization

Plus de GMP EDUCATION : Not for Profit Organization (20)

Dernier

Dernier (20)

Quality by Design : Design Space