Shelf Life Estimation

ICH estimation methods

Current Practices in Shelf Life Estimation Typical stability study

James Schwenke Shelf Life paradigm

g Ingelheim
Boehringer g Pharmaceuticals,, Inc.
Regression methods for estimating shelf life
PQRI Stability Shelf Life Working Group
Actual stability study results

Summarize empirical distribution of estimated shelf life

Extend to random batch analyses

Pharmaceutical Stability Shelf Life August 1, 2010 1

Definitions of Shelf Life ICH Q1E

ICH Guideline Q1A defines Shelf Life (also referred to as
expiration dating period) as
The shelf life for a pharmaceutical product is the maximum
time at which a stability limiting characteristic stays within
acceptance
p criteria.
Also in Q1A, Specification Shelf Life is defined as
The combination of physical, chemical biological, and
microbiological tests and acceptance criteria that determine
the suitability of a drug substance throughout its re-test
period, or that a drug product should meet throughout its
shelf life.
Pharmaceutical Stability Shelf Life August 1, 2010
ICH Guideline Q1E defines shelf life as
The shelf life of a pharmaceutical product is the maximum
time at which the true mean response of a stability limiting
characteristic crosses the acceptance criterion.
basis for the current ICH/FDA shelf life estimation procedure
limited assurance that individual test results will comply with
the specification up to m months
focus on the mean response implies the risk to fail specification
at shelf life will be 50%
2 Pharmaceutical Stability Shelf Life August 1, 2010 3

Typical Stability Study The Shelf Life Paradigm

Minimum of 3 stability batches

Acceptance Criterion
can be 6 or more batches included in study
can be several studies combined together
Response

Study duration can be 6-12-24-36-48 months

longer
l studies
t di can h have iinterim
t i reportst
Stability Limiting R

length of study can depend on shelf life desired

Various environmental conditions
more severe environmental condition can act as an accelerated
testing for milder conditions Product
Shelf Life

Storage Time (Months)

Pharmaceutical Stability Shelf Life August 1, 2010 4 Pharmaceutical Stability Shelf Life August 1, 2010 5
The Shelf Life Paradigm ICH Shelf Life Estimation Methods

General concerns ICH methodology

Exactly what should be modeled and how does it relate to the
product shelf life?
individual tablet
composite
p sample
p of several tablets
packaged unit (bottle or blister pack)
stability batches or all future batches
How does results of content uniformity studies affect the
decision process?
If focused on the mean response, when the mean crosses the
acceptance criteria, 50% of product out of specification.
Pharmaceutical Stability Shelf Life
minimum of three batches
batches are considered fixed effects for the analysis
batches can be pooled if no significant differences
0.25 level of significance for tests involving batches
0.05 level of significance for tests involving other factors
package type, storage orientation, coating, etc.
construct 95% confidence intervals on individual (or pooled)
batch means
find minimal storage time where confidence interval crosses
acceptance criteria
August 1, 2010 6 Pharmaceutical Stability Shelf Life August 1, 2010 7

ICH Shelf Life Estimation No Pooling ICH Shelf Life Estimation Pooled Batches

Acceptance Criteria Acceptance Criteria

Confidence Band
Response

Response

Confidence Band on Pooled Batch Mean

on Worst Batch
Stability Limiting R

Stability Limiting R

Batch Pooled Batch

Mean Mean
Response Response

Labeled Product Labeled Product

Shelf Life Shelf Life Shelf Life Shelf Life

Storage Time Storage Time

Pharmaceutical Stability Shelf Life August 1, 2010 8 Pharmaceutical Stability Shelf Life August 1, 2010 9

ICH Estimation Methodology ICH Estimation Methodology

usually a simple linear (straight line) regression model is
assumed to characterize the response-time continuum
first-order nonlinear models are more appropriate for some
stability limiting characteristics
we will focus on simple linear regression models
simple linear model
yij = b0i + b1i (monthj) + ij
yij = observation at jth month for ith batch
b0i = batch intercept and b1i = batch regression slope
ij = residual error with Normal assumptions
Pharmaceutical Stability Shelf Life
ICH methodology suggest to test for equal regression slopes
among stability batches first
if batch slopes are nonsignificant ( = 0.25)
common regression slope is assumed among batches
p are tested
batch intercepts
if batch intercepts are nonsignificant ( = 0.25)
- common intercept is assumed among batches
- batches are pooled
if batch slopes are significant, no further testing is considered
ICH does not allow for a model with a common batch intercept
and unequal slopes (which can be an important model)
August 1, 2010 10 Pharmaceutical Stability Shelf Life August 1, 2010 11
ICH Estimation Issues ICH Estimation Issues

Regression model selection Batch poolability

there are four possible linear regression models if batches cannot be pooled, shelf life is estimated on results of
1) full model: unequal intercepts and slopes among batches the worst batch
2) common intercept with unequal slopes among batches if batches can be pooled, between and within batch variation is
3) unequal intercepts with common slope among batches combined
4) common intercept and slope (pooled batches) Random batch effects
Model #2 is not allowed following ICH guidelines the 0.25 level of significance used to test hypotheses involving
still considered an important model to consider by colleagues fixed batch effects is intended to accommodate batch-to-batch
for stability limiting characteristics that should be at 0% or variation
100% at 0-months storage time available software allows for random batch analysis
would avoid batch poolability issue
Pharmaceutical Stability Shelf Life August 1, 2010 12 Pharmaceutical Stability Shelf Life August 1, 2010 13

Real-Life Example Data Set Product Shelf Life

Industry Example: 26 Stability Batches

Real-life example contributed by one of our PQRI members

Considered as Total Population to Define Product Shelf Life of 37.8

115

26 stability batches
all on same product 110

most kept on study for 24 months

Stability Limiting Response (Assay)

105

assay was measured

100

will use to study empirical distributional properties of estimated

shelf life using 3- and 6-batch studies 95

90

85

0 6 12 18 24 30 36 42 48
Storage Time (Months) Product Shelf Life
Pharmaceutical Stability Shelf Life August 1, 2010 14 Pharmaceutical Stability Shelf Life August 1, 2010 15

Empirical Study of Distributional Properties Empirical Study of Distributional Properties

use real-life stability batch data set 3-batch analysis

consider entire 26-batch data set defines the product
population of batches
use all batches to product shelf life
estimate regression
g line assumingg batches are random
product shelf life is storage time where regression line crosses
acceptance criteria
consider all possible sets of 3 and 6 batches
conduct ICH estimation methods for shelf life
summarize results
Pharmaceutical Stability Shelf Life August 1, 2010
consider all possible combinations of 3 batches from the 26
there are 2,600 combinations
conduct regression analysis allowing for all four models
does not follow ICH
allows for common intercept / unequal slope model
model is included by my analytical scientists
estimate shelf life from best fitted model
6-batch analysis
there are 230,230 possible combinations (7.5 days to run)
randomly chose 20,000 (15.5 hours to run)
16 Pharmaceutical Stability Shelf Life August 1, 2010 17
3-Batch Estimate of Shelf Life 6-Batch Estimate of Shelf Life

Comparison of ICH Shelf Life Estimation Methodology Using Industry Data Comparison of ICH Shelf Life Estimation Methodology Using Industry Data
Using 3 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time Using 6 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time
Considering All Possible Response Models Considering All Possible Response Models
Product Shelf Life = 37.8 Product Shelf Life = 37.8

450 3500
16.31%
15.27% 15.54%15.34%
400 15.03%
14.42% 3000

350 13.17%
2500 11.94%
300
Frequency

Frequency
250 2000
8.92% 8.58%
7.81%
200 7.08% 1500 6.83%
6.35% 6.13%
150 5.50%
4.96%
1000
3.85%
100 3.22%
2.46% 2.31%
1.88% 500 2.00%
50 1.27% 1.43%
0.92%
0.08% 0.31% 0.46% 0.35% 0.20% 0.08% 0.04%
0.00%
0 0
12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0 30.0 32.0 34.0 36.0 38.0 40.0 42.0 44.0 12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0 30.0 32.0 34.0 36.0 38.0 40.0 42.0 44.0
ICH Estimated Shelf Life (Total = 2600) ICH Estimated Shelf Life (Total = 230230)

Pharmaceutical Stability Shelf Life August 1, 2010 18 Pharmaceutical Stability Shelf Life August 1, 2010 19

3-Batch Estimate of Shelf Life

ICH Estimation Methodology
n = 466 (18%) mean = 22.9 months SD = 5.86
Comparison of ICH Shelf Life Estimation Methodology Using Industry Data

Comparing the two empirical distributions

Using 3 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time
Considering All Possible Response Models, Model_11: Full Model
Product Shelf Life = 37.8

there is a shifting toward shorter estimated shelf lives with an 250

increase in the number of batches included in the analysis

counterintuitive 200

increase in the number of batches should reflect an increase

150
in the amount of information about the product
Frequency

141

increase in the amount of information about the product 100

should reflect a better estimate of shelf life 88
77

should see a shift in distribution toward longer shelf lives 50 58

better estimates of product shelf life (37.8 months) 35

23
16
disincentive for industry to include more stability batches 0
12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0 30.0 32.0 34.0 36.0 38.0 40.0 42.0 44.0
ICH Estimated Shelf Life (Total = 2600)

Pharmaceutical Stability Shelf Life August 1, 2010 20 Pharmaceutical Stability Shelf Life August 1, 2010 21

6-Batch Estimate of Shelf Life 3-Batch Estimate of Shelf Life

n = 5035 (25%) mean = 19.9 months SD = 4.60 n = 788 (30%) mean = 23.3 months SD = 5.71
Comparison of ICH Shelf Life Estimation Methodology Using Industry Data Comparison of ICH Shelf Life Estimation Methodology Using Industry Data
Using 6 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time Using 3 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time
Considering All Possible Response Models, Model_11: Full Model Considering All Possible Response Models, Model_01: Common Intercept
Product Shelf Life = 37.8 Product Shelf Life = 37.8

2500 250

2258

2000 200

1500 150
Frequency

Frequency

113
1000 100 109
97
870 87
80 82
704
500 566 50 54
45 43
35
216 243 25
129
0 0
12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0 30.0 32.0 34.0 36.0 38.0 40.0 42.0 44.0 12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0 30.0 32.0 34.0 36.0 38.0 40.0 42.0 44.0
ICH Estimated Shelf Life (Total = 230230) ICH Estimated Shelf Life (Total = 2600)

Pharmaceutical Stability Shelf Life August 1, 2010 22 Pharmaceutical Stability Shelf Life August 1, 2010 23
6-Batch Estimate of Shelf Life 3-Batch Estimate of Shelf Life
n = 6015 (30%) mean = 20.1 months SD = 4.13 n = 983 (38%) mean = 27.8 months SD = 4.61
Comparison of ICH Shelf Life Estimation Methodology Using Industry Data Comparison of ICH Shelf Life Estimation Methodology Using Industry Data
Using 6 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time Using 3 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time
Considering All Possible Response Models, Model_01: Common Intercept Considering All Possible Response Models, Model_10: Common Slopes
Product Shelf Life = 37.8 Product Shelf Life = 37.8

2500 250
252

2000 2100 200

186

164
1500 150

Frequency
Frequency

1000 100
1009
94

739 765 72
683 68
500 50
48
324 34
218
17
109 11
0 0
12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0 30.0 32.0 34.0 36.0 38.0 40.0 42.0 44.0 12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0 30.0 32.0 34.0 36.0 38.0 40.0 42.0 44.0
ICH Estimated Shelf Life (Total = 230230) ICH Estimated Shelf Life (Total = 2600)

Pharmaceutical Stability Shelf Life August 1, 2010 24 Pharmaceutical Stability Shelf Life August 1, 2010 25

6-Batch Estimate of Shelf Life 3-Batch Estimate of Shelf Life

n = 8451 (42%) mean = 27.3 months SD = 3.425 n = 363 (14%) mean = 31.2 months SD = 5.70
Comparison of ICH Shelf Life Estimation Methodology Using Industry Data Comparison of ICH Shelf Life Estimation Methodology Using Industry Data
Using 6 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time Using 3 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time
Considering All Possible Response Models, Model_10: Common Slopes Considering All Possible Response Models, Model_00: Common Intercepts and Slopes
Product Shelf Life = 37.8 Product Shelf Life = 37.8

2500 250
2413
2274

2000 200

1500 150
Frequency

Frequency

1177
1000 1080 100

63
500 554 50
49
44 43 43

235 269 28 26
20 18
159 120 13 11
0 0
12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0 30.0 32.0 34.0 36.0 38.0 40.0 42.0 44.0 12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0 30.0 32.0 34.0 36.0 38.0 40.0 42.0 44.0
ICH Estimated Shelf Life (Total = 230230) ICH Estimated Shelf Life (Total = 2600)

Pharmaceutical Stability Shelf Life August 1, 2010 26 Pharmaceutical Stability Shelf Life August 1, 2010 27

6-Batch Estimate of Shelf Life

ICH Estimation Methodology
n = 499 (3%) mean = 32.5 months SD = 4.23
Comparison of ICH Shelf Life Estimation Methodology Using Industry Data

Poolability
Using 6 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time
Considering All Possible Response Models, Model_00: Common Intercepts and Slopes
Product Shelf Life = 37.8

2500 concept is to estimate shelf life on the best fitted regression

accommodate random variation among batches by allowing
2000
pooling of batch data through regression parameter estimates
allow for common slope p and intercept
p models to characterize
1500
batch response
Frequency

1000
use = 0.25 level of significance
estimate of shelf life too heavily dependent on best model
500 assuming unequal slopes forces the shelf life estimate to be
based on worst batch
125
0
12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0 30.0 32.0 34.0 36.0 38.0 40.0 42.0 44.0 tends to minimize shelf life estimate
ICH Estimated Shelf Life (Total = 230230)

Pharmaceutical Stability Shelf Life August 1, 2010 28 Pharmaceutical Stability Shelf Life August 1, 2010 29
Random Batch Mixed Model Analysis Random Batch Mixed Models - Reflection

Two rationale to suggest an alternative random batch analysis

Acceptance Criteria
can extend inference of estimated shelf life to future batches
avoids dependence of shelf life estimate on best model fit
Confidence

Response
mixed model analysis would Band

Stability Limiting R
model between-batch variation as a random effect Mean of Random Batch
quantify both between and within-batch variation separately Mixed Model Analysis

allows broad and narrow inferences

allows estimation of shelf life through calibration techniques
defined by a one-sided (lower) interval estimate on Labeled
Shelf Life
calibration storage time point
Storage Time

Pharmaceutical Stability Shelf Life August 1, 2010 30 Pharmaceutical Stability Shelf Life August 1, 2010 31

3-Batch Estimate of Shelf Life Confidence Interval 6-Batch Estimate of Shelf Life Confidence Interval

Alternatives to ICH Shelf Life Estimation Methodology Using Industry Data Alternatives to ICH Shelf Life Estimation Methodology Using Industry Data
Using 3 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time Using 6 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time
Estimating Shelf Life Using Random Batch Mixed Model Analysis Estimating Shelf Life Using Random Batch Mixed Model Analysis
Product Shelf Life = 37.8 Product Shelf Life = 37.8

450 3500
16.47%
15.76%
400 15.00% 15.01%
3000
13.59%
350 12.65%
11.65% 2500
11.46% 11.37%
300
Frequency

Frequency

250 2000 9.48%

8.62% 8.98%
8.35% 8.50%

200 1500
6.65%

150
4.62% 1000 4.69%
4.31%
100
2.54% 2.23% 500 2.12%
50 1.31% 1.12% 1.38%
1.00% 0.81%
0.38%
0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00%
0 0
12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44
Reflection Calibration Method (Total = 2600) Reflection Calibration Method (Total = 230230)

Pharmaceutical Stability Shelf Life August 1, 2010 32 Pharmaceutical Stability Shelf Life August 1, 2010 33

Random Batch Mixed Models Distribution of -hat 3-Batch Estimate of Shelf Life Confidence Interval

Alternatives to ICH Shelf Life Estimation Methodology Using Industry Data

Using 3 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time
Estimating Shelf Life Using Random Batch Mixed Model Analysis
Product Shelf Life = 37.8
Acceptance Criteria
450

400
14.46%
13.69%
Response

350
12.46%

300
Stability Limiting R

9.88%
Frequency

Mean of Random Batch 250 8.85%

8.54%
Mixed Model Analysis 8.23%
200
6.27%
5.69%
150

100 3.50%
2.42% 2.46%
1.96%
1.54%
Labeled Calibration 50

Shelf Life Point 0.00% 0.00% 0.04%

( 0
12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44
Storage Time B_Hat Calibration Method (Total = 2600)

Pharmaceutical Stability Shelf Life August 1, 2010 34 Pharmaceutical Stability Shelf Life August 1, 2010 35
6-Batch Estimate of Shelf Life Confidence Interval Conclusions

Alternatives to ICH Shelf Life Estimation Methodology Using Industry Data

Moving from a fixed-batch to a random-batch analysis has

Using 6 Batches with 24 Months of Data, Extrapolating to 48 Months of Storage Time
Estimating Shelf Life Using Random Batch Mixed Model Analysis
Product Shelf Life = 37.8

several advantages
3500
16.32%16.27%
can be used to extend the inference of a shelf life statement to
3000
13.72% future batches
depending on how well the stability data represent the
12.54%
2500 12.16%
11.71%

manufacturing process
Frequency

2000

1500 7.19%
avoids the batch poolability issue
breaks the dependence of the estimate of shelf life on which
reduced or pooled regression model is selected
1000 4.23%
3.49%

500
1.44%
0.88%
avoids the issue of a common intercept / different slope
0
0.00% 0.00% 0.00% 0.00% 0.00% 0.09%
regression model being appropriate or not
12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44
B_Hat Calibration Method (Total = 230230)

Pharmaceutical Stability Shelf Life August 1, 2010 36 Pharmaceutical Stability Shelf Life August 1, 2010 37

Conclusions Conclusions

better quantifies between and within-batch variation by using Random-batch analysis

an appropriate statistical model as presented, is still estimating an average batch shelf life
allows additional stability batches to benefit the estimate of confidence interval approach was discussed which is
shelf life consistent with ICH
for fixed-batch analysis, additional stability batches increases still have conceptual
p p
problem that at shelf life,, half of p
product
chance that shelf life is estimated by worst-batch unit is above specification limit
for random-batch analysis, additional stability batches adds does not reflect desired quality statement
to the information on between-batch variation alternatives methods for shelf life estimation
better distributional characteristics of shelf life estimate tolerance intervals
quantile regression

Pharmaceutical Stability Shelf Life August 1, 2010 38 Pharmaceutical Stability Shelf Life August 1, 2010 39