Statistical Evaluation of Dissolution for Specification Setting and Stability Studies
Fasheng LiAssociate Director, Pharmaceutical Statistics
Worldwide R&DPfizer, Inc
37th Annual MBSW Muncie, IN
May 20, 2014
Dissolution routinely tested to provide in vitro drug release information
Drug development: prediction of in vivo drug release profiles Quality control: assessment of batch-to-batch consistency
Decision making during dissolution method and drug development
Data based specification setting for USP <711> dissolution test Dissolution monitoring on stability
Statistical assessment integral to decision making process.
Motivation
2 2
Outline
Setting Extended Release Dissolution Specifications
Number of time points needed Case 1: Two-point spec Case 2: Three-point spec
Evaluation of possible specifications
Dissolution on Stability
No significant linear trend observed Non-linear trend observed
3 3
Dissolution Specification Setting
How many time points are necessary for setting dissolution specifications?
Based on “Guidance for Industry: Extended Release Oral Dosage Forms: Development, Evaluation, and Application of In Vitro/In Vivo Correlations”, at least three points (early, middle, late) on the dissolution profile should be used to have specifications
Are fewer than three time points sufficient?Are three time points enough?
4 4
Dissolution Spec Setting – Case 1 Mean disso
profiles of three typical batches of a sustained release drug product
Proposed to have specs at two time points (30 and 180 minutes)
Team discussed to add a spec at either 15 or 60 minutes
Specs at 30 and 180 minutes
Add either 15 or 60 minutes?
5 5
Dissolution Spec Setting – Case 1 An empirical
first-order two-parameter non-linear regression model fit to the release profiles
Goodness-of-fit of the model evaluated by a coefficient of determination R2-type measure
Model appropriateness evaluated by the lack-of-fit test
R2 = SSTotalSSE
1
,
6 6
Release = A(1-e-bt) is a two-parameter Weibull model
Dissolution Spec Setting – Case 1 Dissolution profiles
defined well by a two-parameter release model
Mathematically, any two points on the profile would be able to sufficiently define the release profile
No need to add a third time point for specification
Team agreed to set disso specifications without a third point
,
Two-point spec
7 7
Dissolution Spec Setting – Case 2 Mean disso
profiles of three typical batches of a extended release drug product
Originally specs at 5 time points proposed; should 1 more be added to improve control?
Question: How many time points are needed for setting dissolution specifications? 8 8
Dissolution Spec Setting – Case 2 An empirical
three-parameter non-linear regression model (Weibull ) fit to the release profiles
Goodness-of-fit of the model evaluated by a coefficient of determination R2-type measure
Model appropriateness evaluated by the lack-of-fit test
R2 = SSTotalSSE
1
,
The three-parameter Weibull model is sufficient and adequate to define dissolution profiles in this case
Recommend three-time points for dissolution specifications
Three-point spec
9 9
Brief Summary ,
Three-point specifications are apparently more advantageous than six-point specifications:
Cost Savings
Save 50% reducing from 6 to 3 time points
Quicker Analytical Results
Conformity Risk Reduction: Assuming the probability of passing USP <711> dissolution test at each time point is the same (e.g. 98%), the overall probability to pass:
0.983 = 94% with three time points
vs. 0.986 = 89% with six time points
10 10
Evaluation of Dissolution Specifications,
After Determining Number of Time Points
Evaluate proposed dissolution specifications against USP <711> at each time point
Recommend new dissolution specifications if necessary
Statistical Approach
Simulations performed on individual dissolution data at each of the specification time points to check the probabilities of passing different stages (L1, L2, and L3) of USP <711> dissolution test
11 11
USP <711> Dissolution Test
12 12
Acceptance Criteria for Extended Release Drug Products
Controlled release product specs:
@1 hour: <= 30% @4 hours: 40-60% @24 hours: >= 80%
Re-evaluate specs due to method change
Data: 46 unique dissolution conditions Each have 6 to 96
individual disso profiles A total of 1578 disso
profiles collected
Evaluation of Dissolution Specifications
13
50000 simulations performed on 46 dissolution data sets to check probabilities of passing USP <711> stages (L1, L2, and L3)
Current three-point specifications
13
Evaluation of Dissolution Specifications
14
Specifications Stage
% of Disso TestsAcceptance Probability
by Stage Lx
Need Stage Lx Pass 100% by Stage Lx Mean StDevQ1: <= 25% Q4: 35-55% Q24: >= 80%
L1 100.0 13.0 95.3 9.0L2 87.0 52.2 98.5 2.9L3 47.8 80.4 99.5 1.4
Q1: <= 30% Q4: 35-55% Q24: >= 80%
L1 100.0 15.2 96.2 7.6L2 84.8 58.7 99.0 2.1L3 41.3 87.0 99.8 0.7
Q1: <= 30% Q4: 40-60% Q24: >= 85%
L1 100.0 4.3 59.4 38.2L2 95.7 17.4 87.1 19.1L3 82.6 47.8 91.0 19.9
Proposed Specs
ComparableSpecs
14
Controlled release product - recommended specifications for new dissolution method
@1 hour: <= 30% @4 hours: 35-55% @24 hours: >= 80%
Evaluation of Dissolution Specifications
15
Revised three-point specifications
Revised three-point specifications
Individual Dissolution Profiles
Mean Dissolution Profiles
15
Brief Summary – Disso Spec Setting,
The number of time points on dissolution profiles used for specification setting
Can be justified by fitting a non-linear release model Based on the number of parameters of the non-linear release
model
Specifications at each time points
Can be evaluated by performing simulations on dissolution data against USP <711> criteria
Calculate the probability to pass USP criteria
16 16
Dissolution on Stability
Dissolution usually monitored on stability as a numerical quality attributes with numeric specifications
e.g. %Release at 6 hours should be between 40-60%
Dissolution data may not have a significant linear trend along stability time
Linear trends not significantNon-linear trends observed
How to evaluate dissolution data on stability? Typical Q1E shelf life analysis not appropriate.
17 17
Dissolution on Stability – No Linear Trend
18
No Statistically significant
Linear Trend
18
Linear Trend
Clear linear trend for the chemical impurity data ICH Q1E Analysis
Appropriate
ICH Q1E Analysis is not meaningful
No overall statistically significant trend in
dissolution at 10 hours
19
Dissolution on Stability – No Linear Trend
Shelf-life predicted based on the major chemical impurities: Apply linear regression analyses following the ICH Q1E guidance
The risk of failing dissolution on stability will be quantified
Make sure the risk of failing dissolution spec is low Utilize dissolution profiles tested at each of the stability
time points
19
20
Dissolution on Stability – No Linear Trend
20
A three-parameter Weibull model:%Release = A(1-exp(-b*tm))
fit to all mean or individual dissolution profiles at each of the stability time points for all registration stability batches
The risk of failing dissolution at a future stability test time since time is not relevant can be quantified by
1. Constructing prediction limits with confidence level p%2. Checking the limits against the spec of (45, 65)3. If the prediction limits are within the spec limits, the risk
of failing a future average dissolution would be no more than 100-p%
Storage Condition Strength
Nonlinear Model Parameters and Fit Statistics
99 % Pred Limits for
Dissolution at 10hr
99 % Pred Limits Meet
Spec (45, 65)?
%Chance for a Future Disso Test
to FailA b m R2 P-value
25°C/60%RH
1 94.5 0.0176 1.711 0.9939 0.0000 49.6, 63.0 Yes 0.0452 94.0 0.0147 1.760 0.9952 0.0000 47.8, 59.7 Yes 0.0123 94.8 0.0159 1.739 0.9960 0.0000 49.5, 60.6 Yes 0.0034 95.7 0.0142 1.777 0.9964 0.0000 49.5, 60.1 Yes 0.0035 96.2 0.0140 1.783 0.9960 0.0000 49.5, 60.8 Yes 0.0036 94.4 0.0135 1.800 0.9955 0.0000 48.3, 60.1 Yes 0.0037 93.6 0.0107 1.867 0.9945 0.0000 44.5, 57.4 Not the lower limit 0.865
30°C/75%RH
1 95.8 0.0180 1.714 0.9932 0.0000 50.9, 65.3 Not the upper limit 0.6692 94.4 0.0150 1.764 0.9952 0.0000 48.8, 60.9 Yes 0.0033 95.7 0.0165 1.734 0.9940 0.0000 49.7, 63.4 Yes 0.0744 97.5 0.0149 1.756 0.9951 0.0000 49.6, 62.1 Yes 0.0125 96.6 0.0137 1.796 0.9950 0.0000 49.2, 62.0 Yes 0.0076 96.1 0.0132 1.811 0.9952 0.0000 48.9, 61.3 Yes 0.0037 95.1 0.0108 1.867 0.9955 0.0000 46.2, 57.9 Yes 0.112
)e-A(1 Release%m-bt
• Model fit mean dissolution profiles of stability times points very well (R2 > 0.99)
• The risk of failing dissolution test on stability at a future time is no more than 0.9%
Risk of Dissolution on Stability
21
22
Dissolution on Stability – No Linear Trend
Risk of failing disso on stability is < 0.9%
23
Dissolution on Stability – No Linear Trend
Risk of failing disso on stability is < 0.7%
Dissolution on Stability – Non-linear Trend
24 24
Extended release product: with a clear non-linear trend for dissolution data at x hours
Dissolution on Stability – Non-linear Trend
25 25
Empirical model of: %Release at x hours = A(1-e-b*(t+t0))
can be fit to dissolution at x hours from manufacturing age for all registration stability batches
Shelf life (in terms of manufacture age) can be predicted when the 95% confidence interval intersects with the spec limits
Shelf life (in terms of stability storage age) can be determined by subtracting the manufacturing age of the initial stability time point (stability time 0 month)
Dissolution on Stability – Non-linear Trend
26 26
Stability program started at 7.4 months of manufacturing age
Predicted shelf life is about 54.5 -7.4 = 47.1 months
Brief Summary – Dissolution on Stability,
Stability dissolution data often show no significant linear trends
No significant linear or non-linear trend
Dissolution profile data can be utilized to remediate the risk of meeting dissolution specifications• More information used versus evaluate at 1 time
point on the profile
Non-linear trend
Empirical non-linear model fit to stability data could help justify the prediction of shelf life
27 27
Summary
28
Dissolution for extended release drug products facing decision makings in areas such as
Setting Specifications Number of time points on the profile for spec Specification limits at the time points
Dissolution on Stability No significant linear trend Non-linear trend
• Statistics will be able to contribute greatly in the above areas to make regulatory appealing decisions
Statisticians need to work proactively with team scientists
Acknowledgment
29
Kim Vukovinsky, Senior Director, Pharmaceutical Statistics, Worldwide R&D, Pfizer Inc.