US FDA - EU Process Validation : Determination of Number of PPQ Batches

This presentation is compiled by “ Drug Regulations” a
non profit organization which provides free online
resource to the Pharmaceutical Professional.
Visit http://www.drugregulations.org for latest
information from the world of Pharmaceuticals.
4/28/2016 1

 This presentation is compiled from freely
available resources like the websites of FDA and
ISPE specifically the Discussion Paper of ISPE on
the subject
 “Drug Regulations” is a non profit organization
which provides free online resource to the
Pharmaceutical Professional.
 Visit http://www.drugregulations.org for latest
4/28/2016 2
Drug Regulations : Online
Resource for Latest Information

4/28/2016 3
Drug Regulations : Online Resource for Latest
Information

 We published a presentation on previous
ISPE discussion papers which can be
accessed below :
 US FDA Stage 2 Process Validation
 US FDA Stage 3 Process Validation
4/28/2016 4
Information

 This presentation now deals with use of
Statistical Techniques to determine number of
Process Performance Qualification Batches to
demonstrate
◦ A high degree of assurance in the manufacturing
process
◦ That the control strategy is sufficiently robust to
support commercial release of the product
4/28/2016 5
Information

 The presentation
◦ Proposes 4 statistical tools to determine the PPQ batches.
◦ Discusses their assumptions and limitations
◦ Presents simulated examples
 The ISPE paper on which this presentation is based
calls for feedback on this approach
 The paper may be modified and expanded further
4/28/2016 6
Information

 US FDA PV guidance, New Annex 15 and EMA
regulations require high degree of assurance in
the
◦ Product Manufacturing Process
◦ Control Strategy
 Once this assurance is obtained then only a
product commercialization decision can be
supported
4/28/2016 7
Information

 The science and risk based approach to
determine number of batches is preferred
 The earlier 3 batch approach is no longer
preferred
 There are differences of opinion on how to
translate Risk Based approach to number of
batches
4/28/2016 8
Information

 Earlier ISPE papers featured 3 approaches
 See here and here
 These papers provoked a lot of scientific
discussions
 Additional approaches have been developed
which are given in the current paper and
this presentation.
4/28/2016 9
Information

 To determine number of initial PV batches
or Process Performance Qualification
Batches using statistical tools.
 Each statistical tool is intertwined with risk
assessment approaches.
 Risk assessment is as per ICH Q 9
4/28/2016 10
Information

 Strategizes presented here may be modified
to suit
◦ Established systems
◦ Procedures / polices
 The approach is not industry consensus
 There are no preferences of one approach
over the other presented here
4/28/2016 11
Information

 Earlier ISPE paper presented three Risk
based approaches which are summarized in
subsequent slides
4/28/2016 12
Information

 Approach # 1 :
 Based on Rationales and Experience
 Appears quite reasonable
 Does not use statistics in the determination
of number of batches
 Uses rationales based on historical
information and process knowledge
4/28/2016 13
Information

 Approach # 2 :
 Target Process Confidence & Target Process
Capability
 Uses statistics
 Applicable only to normally distributed data
& processes under statistical control
4/28/2016 14
Information

 Approach # 2 :
 Transformation from normal to non normal
distribution is possible , this is matter of
trial and error
 User need to draw conclusions so as to
apply them to the original process that
generated non normal data
4/28/2016 15
Information

 Approach # 2 :
 Recommends implementing improvements
to processes if process capability index is
not met
4/28/2016 16
Information

 Approach # 3 :
 Expected Coverage
 Non parametric approach
 Based on order statistics
◦ Minimum & Maximum observed results are referred to
as x(1) , x(n)
◦ And properties of intervals based on these order
statistics.
4/28/2016 17
Information

 Approach # 3 :
 The expected probability of a future
observation being within the range defined by
◦ { x(1), x(n) } is ( n -1)/(n+1)
 Users of this approach are required to select
appropriate level of coverage most suitable for their
processes.
4/28/2016 18
Information

4/28/2016 19
Information

4/28/2016 20
Information

 FDA guidance stated that
◦ Number of batches should be adequate to provide
sufficient statistical confidence of quality both within a
batch and between batches
 This prompts two questions
◦ What is sufficient statistical confidence
◦ How will the between batch quality be measured
4/28/2016 21
Information

 Both questions can be addressed by using
tolerance intervals as PPQ protocol acceptance
criteria for critical quality attributes.
 Acceptance criteria in the PPQ protocol could
be that the tolerance interval must be within
the routine release criteria for the CQA
 This may be the specification or a tighter
internal requirement.
4/28/2016 22
Information

 Tolerance Intervals
◦ Intervals that cover fixed proportion of population (p)
with stated confidence (γ).
◦ A γ/p tolerance interval provides limits
 within which at least specified proportion (p, also known
as coverage) of future individual observations from a
population
 Is predicted to fall
 With a specified confidence level (γ).
4/28/2016 23
Information

 Tolerance Intervals : Example
◦ A 95%/99% Tolerance Interval of 98.5 to 101.7
means that there is at least 95 % confidence that
at least 99 % of future results will be between
98.5 to 101.7.
4/28/2016 24
Information

 Tolerance Intervals : Assumptions
◦ Continuous normally distributed data
◦ Two sided specifications
◦ k values approximated using Howe method
◦ Process is in a state of statistical control
4/28/2016 25
Information

 Tolerance Intervals : Risk Assessment
◦ Confidence & coverage levels should be based on
Risk Assessment
◦ lower Risk Level :Lower confidence & converge
levels
◦ Higher Risk Level : Higher confidence & converge
levels
4/28/2016 26
Information

 Tolerance Intervals : Risk Assessment
◦ Confidence & converge levels are arbitrary
◦ A minimum acceptable level should be stated
◦ A 50 % level would be too low
◦ A floor of 80 % is suggested
◦ This may be too low for certain companies or
applications
4/28/2016 27
Information

 Example given in following table
4/28/2016 28
Information

 Tolerance Intervals : Other Assignment Levels
 Other assignment levels are possible
 Key feature : Levels be non decreasing with risk
 Confidence & Coverage levels should be fixed
in advance
 This way focus then can be uniquely on risk
assessment
4/28/2016 29
Information

 To determine the number of batches ( N)
needed for PPQ one additional value should
be set.
4/28/2016 30
Information

4/28/2016 31
Information

4/28/2016 32
Information

 Choices for k are extremely intuitive
 They can be narrowed down based on common
relationships
 If k = 3 , then TI is the width of the control
chart limits ( ± 3σ )
 If k < 3 , then TI would not cover expected
common cause variability.
 Therefore k should be at least be 3.
4/28/2016 33
Information

 A k = 4.0 Process Performance index (Ppk)
of 1.3
 A k = 6 Process Performance index (Ppk) of
2.0
 (Ppk) = 2.0 = a 6 σ process
 Though a laudable goal, commercial
specifications would be closer to a 4 σ
4/28/2016 34
Information

 Smaller k = more conservative the final
sample size for validation.
 No absolute answer for k value
 Reasonable that k is 3.0 to 4.0
 In this presentation examples are given with
k = 3.5
4/28/2016 35
Information

4/28/2016 36
Information

 Minimum number of PPQ batches refers to the
total number of unique batch results included
in the PPQ final report TI calculations.
 These results are not limited to to batches run
as part of PPQ, but can include previously
identified stage 1 batches form comparable
processes.
4/28/2016 37
Information

 Validation of New Drug Substance
 K = 3.5
 SME perform Risk Assessment & determine the
Residual risk as moderate
 Number of PPQ batches required = 10
 Five earlier batches , stability = 3 and clinical =
2 with comparable processes can used for PPQ
 Additional batches required for PPQ = 5
4/28/2016 38
Information

 Criteria in the PPQ protocol for each
quantitative CQA will be 95 %/95 %
Tolerance Interval on Release results
 All 10 batches must fall within the
acceptance criteria for the CQA
4/28/2016 39
Information

 If the 95 % / 95 % tolerance Interval is
within the acceptance criteria for the CQA
i.e. readily pass then there is high
confidence that the process can
reproducibly comply with the CQA.
 Sampling and testing may be adjusted to
routine level in stage 3.
4/28/2016 40
Information

 If the 80% / 80% tolerance Interval is within the
acceptance criteria for the CQA i.e. marginally
pass then adequate confidence of quality in the
CQA has been demonstrated between batches.
 Heightened monitoring and /or testing of the
CQA is needed until a readily pass level of
quality can be demonstrated for the CQA
4/28/2016 41
Information

4/28/2016 42
Information

 There are many C Q A’s with specification limits for
each C Q A.
 Examples include: A P I
◦ Color
◦ Assay
◦ Related Substances
◦ Moisture
◦ Residual Solvents
◦ Heavy Metals
◦ Particle Size distribution
4/28/2016 43
Information

 There are many C Q A’s with specification limits for
each C Q A.
 Examples include: Drug Products
◦ Appearance
◦ Assay
◦ Content Uniformity
◦ Drug related Impurities
◦ Tablet Hardness
◦ Dissolution
4/28/2016 44
Information

 This approach uses the probability that a batch will
meet all of its specifications to determine the number of
PPQ batches.
 Intuitive and relatively simple approach
 Imposes no new process performance criteria
◦ e.g Minimum values for process capability index
 Incorporates any number of specifications
 Useful for high volume processes with large number of
CQA’s all of which are likely to meet specifications
4/28/2016 45
Information

 Categorization of batches into successes &
failures results in loss of information
 This can increase sample size
 Sample size may further be increased by
batch failures
4/28/2016 46
Information

 pi = Probability of meeting the ith Specification
 i = 1 ,2 ,3….....r specifications
 p = Overall probability of a batch meeting all
its specifications
 p is a complex function of p1, p2 …. pr where
each pi can involve the true status of CQA and
measurement error.
4/28/2016 47
Information

 There are dependencies among the CQA
 Tablet hardness may be highly correlated
with dissolution
 Impurities may be correlated with one
another
 This approach does not need individual
values of p1, p2 …. pr
4/28/2016 48
Information

 Binomial distribution is used to determine
the probability of observing k successful
batches in n independent trials.
4/28/2016 49
Information

 Suppose we have observed k successful
batches in n independent trails
 A statement about the overall probability of
success p can be made.
 Statistical inference for p is a well studied
problem.
 A common estimator for of p is the maximum
likelihood estimator
4/28/2016 50
Information

 There are numerous methods for
constructing confidence intervals for p
◦ Agresti-Couli interval
◦ Clopper- Pearson interval
◦ Jeffreys Interval
◦ Wilson Score Interval
◦ Wald Interval
4/28/2016 51
Information

 Number of PPQ batches can be based on
statistical inference for p the overall
probability of a batch meeting all its
specifications.
4/28/2016 52
Information

 Consider following situation
 Successful batches = n
 Independent Trails = n
 It is possible that there could be batch failures
 These should be investigated with effective
CAPA
 Previous batch failures can be ignored if they
are considered no longer probable.
4/28/2016 53
Information

 Figure 1 in next slide shows one-sided 80%
lower confidence limits ( LCLs) for p based
on the Wilson score and Jeffrey’s intervals
 Figure 2 shows one-sided 95 % LCLs
4/28/2016 54
Information

4/28/2016 55
Information

4/28/2016 56
Information

 Fairly large sample sizes are needed to obtain 95 %
confidence in a specific value of p, relative to 80 %
confidence.
 Example 1
 With Wilson Score we are 80 % confident that p is at
least 0.93 if we observe 10 successful batches in 10
independent trials.
 While we are 95 % confident that p is at least 0.92 if we
observe 30 successful batches in 30 independent trials.
4/28/2016 57
Information

 Example 2
 10 successful clinical batches
 Regulatory filling : same process & formulation
as clinical batches
 No Batch failures
 Risk Assessment : Moderate Residual Risk
 Requirement: 90 % confidence that probability
of batch success is 90 % or greater.
4/28/2016 58
Information

 Example 2
 This requires 15 batches in total
 Therefore 5 additional PPQ batches
 If all 5 batches are successful, process will
be deemed validated
 Further information on process
performance will be obtained in stage 3
4/28/2016 59
Information

 Example 2
 Sample sizes for additional levels of
residual risk are given in Table 3.
4/28/2016 60
Information

4/28/2016 61
Information

 Example 2
 Table 4 provides a Bayesian version based
on the Jeffreys prior.
4/28/2016 62
Information

4/28/2016 63
Information

 Example 2
 These tables show that a large number of batches
are needed to obtain a high levels of confidence in
p.
 Consequently manufacturers should give
considerable thought to the level of confidence that
is needed for a specified residual risk, and develop
their own version of Table 3 and Table 4.
4/28/2016 64
Information

 Example 2
 If one or more CQA’s poses a significant risk of
not meeting specification
 Examine those sub processes in detail
 Use an alternate approach to PPQ sample size
 Model the probability of meeting those specific
specifications as in LeBlond and Mockus.
4/28/2016 65
Information

4/28/2016 66
Information

 For number of batches required
◦ Observed variation from stage 1 batches is partitioned
into the within and between batch components.
◦ Data from commercial engineering /demo batches
would be ideal to use
◦ However this may not be possible to scale differences
◦ Thus scientific and Technical knowledge should be
used to determine whether smaller scale batch data
can be used.
4/28/2016 67
Information

◦ The between batch component is then used with NIST based
calculation to determine the number of batches required.
◦ However just the use of observed variation could overestimate
number of batches required
◦ This could lengthen the time & cost for validation
◦ FDA PV guidance states that the within and between batch
variation needs to be understood.
◦ This could be what causes the variation and the ability to
qualify them
4/28/2016 68
Information

◦ It is important to understand what shift in the
mean response of the parameter is of practical
significance.
◦ The baseline for the shift consideration is the
mean of the response of the CQA as seen in the
demo or engineering runs.
4/28/2016 69
Information

◦ As specification is used in phase I and phase II
batches ,
 this shift can be defined as % of the specification window
 Leverage some form of risk analysis that looks at the
criticality of the quality attribute in question
 As well as residual risks have been identified after a
control strategy has been identified and implemented
 Figure 3 and 4 summarize this approach
4/28/2016 70
Information

4/28/2016 71
Information

4/28/2016 72
Information

◦ The values of 10 %, 15%and 20% are selected based on the
risk of significantly changing the Cpk and hence out of
specification product being produced.
◦ Table 5 shows how out of specification (parts per million)
failure and Cpk change as the shifts get bigger.
◦ Table 5 was calculated using statistical software JMP using
Avg. = 100, Std. Dev. = 2 , USL = 110 and LSL = 90.
4/28/2016 73
Information

4/28/2016 74
Information

◦ Given this understanding of mean shift that is
practically significant , a flow as shown in figure
5 be constructed to determine the number of
batches for validation as well as samples to be
taken per batch.
4/28/2016 75
Information

4/28/2016 76
Information

4/28/2016 77
Information

◦ Once the mean shift to look for has been
identified, one can use NIST method to calculate
the minimum sample size ( 7.2.2.2 of the NIST
eBook)
4/28/2016 78
Information

 Example
 Data for CQA 1 for product “W” gave a graph as
seen in Figure 6 , for 5 Engineering batches.
 The Specification for CQA is 90-110
 These batches were prepared by including as
much variation as possible
◦ Raw materials, Operators, Environmental Conditions,
Non –Key Process Parameters
4/28/2016 79
Information

4/28/2016 80
Information

4/28/2016 81
Information

4/28/2016 82
Information

4/28/2016 83
Information

4/28/2016 84
Information

4/28/2016 85
Information

4/28/2016 86
Information

◦ For this particular product , running a minimum of six batches
will allow seeing the required mean shifts, should they exist.
◦ The validation protocol can contain an acceptance criteria to
specify
 The mean of << Quality Attribute A>> should be between X ± S
 Where X = mean value of CQA derived form the demo/engineering
batches and S is the shift in the mean that is of practical
significance.
4/28/2016 87
Information

◦ If the value of the mean of the combined batches
is outside of this range
 Then an investigation should be conducted
 Appropriate corrective action should be taken
 This should be one of the several acceptance criteria
to judge the success of the validation
4/28/2016 88
Information

 One of the limitations of this approach is the it is
not possible to make confidence statements
◦ “ X % confidence that Y % will be within specifications”
 The intent of the outlined approach was never to show that the
lots would be within specifications;
 This should have been demonstrated as part of stage 1
development & engineering studies.
 The intent of this approach is to
◦ show a consistent performance of the process being validated
◦ And that there is no difference between the commercial scale process at stage 1
and stage 2.
4/28/2016 89
Information

 Another limitation of this approach is that the
stage 1 batches are assumed to be at commercial
scale which may not be common for small-
molecule manufacturing.
 One unintended consequences of this approach is
that the shift in mean to detect could be widened.
 This could lead to fewer batches needed for stage
2 purposes.
4/28/2016 90
Information

4/28/2016 91
Information

 One criteria for successful validation is that
the process exhibits low variability.
 In this approach a predetermined estimate
of variability is used as the basis for
selecting the number of batches for
validation.
4/28/2016 92
Information

 This approach is applicable when
◦ Validating a process following a significant change to
the current process or
◦ When a process is being transferred from one site to
another.
 In such cases there is a well-established
estimate of process variability
 This is based on the historical process data.
4/28/2016 93
Information

 The Goal of the validation is to ensure that
◦ The process being validated continues to show variability
similar to , or no worse than the current process.
 There may be differences between Current & New
Process average
 Important to have process average on target
 This can vary due to differences in equipment,
setup and raw materials
4/28/2016 94
Information

 Adjustments to achieve the target process
average can be made more easily once the
process variability is deemed acceptable.
4/28/2016 95
Information

 Assumptions
◦ Current process is stable prior to the PPQ
◦ Historical estimate can be thought of as known parameter
or population value.
 If the data from validation batches can be shown to
have the same variability as the current process
◦ The estimate of variability from current process can be
used to determine the capability of the process being
validated.
4/28/2016 96
Information

4/28/2016 97
Information

◦ The confidence level ( 1- α ) and power ( 1-β ) should
be specified
◦ α = Probability of rejecting the null hypothesis when it is true
◦ β = Probability of accepting null hypothesis when it is false.
◦ Choice of R depends on
 The process knowledge available
 Impact of process change to the process being validated
 R = 2 can be used to indicate similarity of variability when risk to
the process is low
 A lower value of R can be preferred when the risk is high
4/28/2016 98
Information

 Choice of α & β depends on Residual Risk
 CQA associated with moderate risk = Smaller α & β
values may be chosen. ( 0.05 and 0.2)
 CQA associated with Lower risk = Higher α & β
values may be chosen. ( 0.05 and 0.3)
4/28/2016 99
Information

 Higher values of β mean
◦ There is s greater probability that we would find the
validation variability acceptable when it is actually
unacceptably high.
◦ If the validation is being performed for technology transfer
or after changes to the current process the residual risk is
generally determined to be small.
◦ This is due to abundance to information available form the
current process.
4/28/2016
10
0
Information

 Number of batches for a given residual risk is
summarized for in Table 8.
 The Risk levels are as defined in Topic 1 – Stage 2 process
validation” discussion paper.
 Other choices for α & β are possible , and these should be
determined by each organization during their Risk
assessment process.
4/28/2016
10
1
Information

4/28/2016
10
2
Information

4/28/2016
10
3
Information

 There is a certain level of arbitrariness in selecting
a particular confidence level or power for a given
risk level.
 It is up to organization to decide whether
◦ the confidence level and/or power are consistent with their
understanding of the risk level identified and
◦ The impact of the process and product
4/28/2016
10
4
Information

 If product has multiple variants , the total number of
batches may be made up across different variants.
 Based on the risk assessment certain variants may be
replicated more than others
 The total number of batches may include
◦ Process Development batches or
◦ Any other batches made by the same manufacturing process
as that used for Stage 2.
4/28/2016
10
5
Information

 Perform statistical analysis on the data collected for
CQA’s
 Perform a hypothesis test on the sample Spv
 Determine whether its ratio to the historical standard
deviation (σp) does not exceed a specified value.
 If multiple variants have been used, then it may be
possible to combine data for some CQA’s across some
or all variants.
4/28/2016
10
6
Information

 Perform statistical analysis on the data collected for CQA’s
 Perform a hypothesis test on the sample Spv
 Determine whether its ratio to the historical standard
deviation (σp) does not exceed a specified value.
 If multiple variants have been used, then it may possible to
combine data for for some CQA’s across some or all variants.
 These CQA’s do not vary across all variants
 This helps ensure appropriate power across for the analysis.
4/28/2016
10
7
Information

 If the null hypothesis is not rejected and the ratio does
not exceed the specified value, the validation batches
may be considered no different from the current
process with respect to the standard deviation
 The historical estimate of the process standard
deviation can be used for any further evaluation of the
process.
◦ e.g. Calculating Process Capability
4/28/2016
10
8
Information

 Since a known estimate of variability is being used there
is less uncertainty in the estimate of process capability.
 If the null hypothesis is rejected and if the ratio is grater
than the specified value
◦ The Standard Deviation of the validation batches is
unacceptably higher than that of the current process.
 Additional work may be needed to better understand
and mitigate the sources of variability before validation
can be considered complete.
4/28/2016
10
9
Information

 Perform additional analysis to access the capability
of the process
 Use historical estimate of the S.D. and sample
mean from the validation batches
 This should be determined when planning process
validation
 This will depend on the risk assessment
performed.
4/28/2016
11
0
Information

 Various analysis and/or criteria for passing
validation are possible
◦ All validation batches are within specifications
◦ Calculate Cpk and compare to Historical Cpk
◦ Calculate Cpk and compare it against a stated acceptance
criteria.
4/28/2016
11
1
Information

 If Cpk is being calculated then the target Cpk should be
determined prior to validation
 This should be based on the product being validated
 Acceptance criteria based on Cpk were previously proposed in
“Topic 1- Stage 2 Process Validation” discussion paper.
 These may be used to judge the validations.
 These criteria use the confidence Interval for the Cpk to
determine the acceptability of the batch
 This is necessary as very few batches are available for this
calculation , leading to a greater uncertainty in the estimate.
4/28/2016
11
2
Information

 An example is provided in Table 9 using the
number of batches proposed in Table 8
4/28/2016
11
3
Information

4/28/2016
11
4
Information

 If the observed Cpk ≥ 1.6 , the true Cpk value is at least 1 in each
case in Table 9.
 Thus there is high confidence that the sample mean is also close
to target value.
 In this case we can consider to have passed easily.
 Therefore enhanced monitoring will not be necessary
 However the validation batches may not represent all sources of
process variability.
 Therefore the process variability should continue to be
monitored to ensure that it continues to meet the target set for
validation.
4/28/2016
11
5
Information

 When 1.0 ≤ Cpk ≥ 1.6 , then there is less
confidence that that the process mean is at or close
to the target.
 Further work is needed to
◦ Better understand and improve the mean
◦ Reduce the variability further
◦ Evaluate the process
◦ Identify & adjust the parameters that impact the mean
4/28/2016
11
6
Information

 Further work is needed to
◦ Depending on the residual risk this may done before
validation is considered or during enhanced monitoring
stage.
 Finally if Cpk < 1.0 ,then the process is at high risk
of producing OOS batches during manufacture and
cannot be validated.
4/28/2016
11
7
Information

4/28/2016
11
8
Information

4/28/2016
11
9
Information

 This presentation is compiled from freely
available resources like the websites of FDA and
ISPE specifically the Discussion Paper of ISPE on
the subject
 “Drug Regulations” is a non profit organization
which provides free online resource to the
Pharmaceutical Professional.
 Visit http://www.drugregulations.org for latest
4/28/2016
12
0
Drug Regulations : Online
Resource for Latest Information

US FDA - EU Process Validation : Determination of Number of PPQ Batches

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Viewers also liked

Viewers also liked (20)

Similar to US FDA - EU Process Validation : Determination of Number of PPQ Batches

Similar to US FDA - EU Process Validation : Determination of Number of PPQ Batches (20)

More from GMP EDUCATION : Not for Profit Organization

More from GMP EDUCATION : Not for Profit Organization (10)

Recently uploaded

Recently uploaded (20)

US FDA - EU Process Validation : Determination of Number of PPQ Batches