Beyond ICH Q1E Opening Remarks

31
Company Confidential © 2012 Eli Lilly and Company Beyond ICH Q1E Opening Remarks Rebecca Elliott Senior Research Scientist Eli Lilly and Company MBSW 2013

description

Beyond ICH Q1E Opening Remarks. Rebecca Elliott Senior Research Scientist Eli Lilly and Company MBSW 2013. ICH Q1E. Required analysis for setting specifications Statistical details BUT what does the analysis tell us? - PowerPoint PPT Presentation

Transcript of Beyond ICH Q1E Opening Remarks

Page 1: Beyond ICH Q1E Opening Remarks

Company Confidential

© 2012 Eli Lilly and Company

Beyond ICH Q1EOpening RemarksRebecca ElliottSenior Research ScientistEli Lilly and Company

MBSW 2013

Page 2: Beyond ICH Q1E Opening Remarks

2

ICH Q1E

• Required analysis for setting specifications• Statistical details• BUT what does the analysis tell us?

• The more data, the narrower the interval on the regression line, the longer the dating.

• Assuming common slopes, the analysis provides an average change for a PRODUCT.

Page 3: Beyond ICH Q1E Opening Remarks

3

Estimate of Dating

90

92

94

96

98

100

102

104

Res

ult

0 3 6 9 12 15 18 21 24Age (months)

Linear Fit

Decreasing Property

Dating is 22 months.

Page 4: Beyond ICH Q1E Opening Remarks

4

Estimate of Dating with More Data

• Dating is now 24 months.• (Assuming common slopes.)• Slope represents average

change across batches.• Batches are a random

sample from product.• Slope represents

average change for the PRODUCT.

90

92

94

96

98

100

102

104

Res

ult

0 6 12 18 24Age (months)

Linear Fit

Decreasing Property

Page 5: Beyond ICH Q1E Opening Remarks

5

Estimate of Dating with More Data

90

92

94

96

98

100

102

104

Res

ult

0 6 12 18 24Age (months)

Linear Fit

Decreasing Property

• AND, we already know some batch results are likely to be outside of spec.• Observed• Projected

Page 6: Beyond ICH Q1E Opening Remarks

6

Why Do ICH Q1E?• Batches are released and evaluated individually.

• Individual results must meet specs. • Dating/specifications need to apply to actual test results.• ICH Q1E does not provide analysis for individual results.• ICH Q1E does not consider additional circumstances

that can cause molecule to degrade.• Shipping• Patient/customer use

Page 7: Beyond ICH Q1E Opening Remarks

7

Session Agenda• Jim Schwenke

• Consulting Statistician, Applied Research Consultants, PQRI• On the Shelf-Life of Pharmaceutical Products

• Jeff Gardner• President and Principal Consultant, DataPharm Statistical &

Data Management Services• Statistical Considerations for Mitigating the Risk of

Individual OOS Results on Stability• Becky Elliott

• Senior Research Scientist, Eli Lilly and Company• Change During Patient Use—Questions and Challenges

• Question and Answer Period

Page 8: Beyond ICH Q1E Opening Remarks

Company Confidential

© 2012 Eli Lilly and Company

Change During Patient Use—Questions and ChallengesRebecca ElliottSenior Research ScientistEli Lilly and Company

MBSW 2013

Page 9: Beyond ICH Q1E Opening Remarks

9

Stability Model

90

95

100

105

Res

ult

0 6 12 18 24 30Age (months)

Linear Fit

Decreasing Property

Release buffer is for assay variability

Release buffer is for change, change variability, assay variability

Is this picture complete?

Page 10: Beyond ICH Q1E Opening Remarks

10

More Complex Stability Model

90

95

100

105

Res

ult

0 6 12 18 24 30Age (months)

Fit Each Value Batch=="AA"

Decreasing Property

Patient use

Controlled stability chamber

Release buffer: normal change & variability, assay variability, and in-use change

Multi-use products

Page 11: Beyond ICH Q1E Opening Remarks

11

In-Use Change

• Can be large• Potentially fewer batches for analysis• Can have a different change model than routine

stability

Page 12: Beyond ICH Q1E Opening Remarks

12

Statistics are “easy”

• Determine routine and in-use models• Linear• Quadratic• Nonlinear

• Determine estimates of variability• Model• Assay

• Adjust release buffer(s)

Non-statistical questions are “hard.”

Page 13: Beyond ICH Q1E Opening Remarks

13

Today’s Topics

• Modeling in-use change1. Complexity of complete statistical model and

impact on business2. Significance of in-use change3. Correlation of results4. Groups5. Proxy data

• Other uncontrolled conditions

Page 14: Beyond ICH Q1E Opening Remarks

14

#1 Analysis Impact to Business

• One-time or yearly studies?• Requirement is often upon registration

• “Fresh” batch• “Aged” batch

• There may be no regulatory requirement to generate data yearly

• One-time estimate or yearly update?• Implications are to WHO does stat analysis

WHEN and HOW.• One complicated model• Two easier models

Page 15: Beyond ICH Q1E Opening Remarks

15

#2 Significance of Change

• Change estimates• Errors can be high depending on assay

• Is change significant?• Include estimate of change variability?

Page 16: Beyond ICH Q1E Opening Remarks

16

#2 Significance of Change

• Assay variability is included in buffer for long term stability change. Is it “double counting” to include it for in-use change?

• Is there “room” within the specifications?

90

95

100

105

Res

ult

0 5 10 15 20 25 30 35Age in Days

Linear Fit

Lot B1

90

95

100

105

Res

ult

0 5 10 15 20 25 30 35Age in Days

Linear Fit

Lot B2

Page 17: Beyond ICH Q1E Opening Remarks

17

#2 Significance of Change

• Is change meaningful?• Science vs. statistical significance

90

95

100

105

Res

ult

0 5 10 15 20 25 30 35Age in Days

Linear Fit

Lot B3

90

95

100

105

Res

ult

0 5 10 15 20 25 30 35Age in Days

Linear Fit

Lot B4

p-value = 0.02 p-value = 0.06

Page 18: Beyond ICH Q1E Opening Remarks

18

#3 Correlation in Results• Multiple batches can be manufactured close together in

time (e.g., validation batches, special studies).• Timepoints to be assayed are close together.• Lab wants to maximize resources.

• Hold samples• Test them together

• Common timepoints across batches are put on same assay run.

Testing batches together dependent slopes

Page 19: Beyond ICH Q1E Opening Remarks

19

#3 Correlation—Shared Assay Dates

96

98

100

102

Res

ult

0 10 20 30Age in days

12/15/200802/16/200902/17/2009

Test Date

Linear Fit

Lot A1

96

98

100

102

Res

ult

0 10 20 30Age in days

01/01/200902/16/200902/17/2009

Test Date

Linear Fit

Lot A2

96

98

100

102

Res

ult

0 10 20 30Age in days

12/28/200802/16/200902/17/2009

Test Date

Linear Fit

Lot A3

96

98

100

102

Res

ult

0 10 20 30Age in days

01/09/200902/16/200902/17/2009

Test Date

Linear Fit

Lot A4

Are these 4 independent estimates of the slope?

Page 20: Beyond ICH Q1E Opening Remarks

20

#3 Correlation—Solution• Backload samples• E.g.: 30 day study tested on day 0, 7, 15, 30

• Study day 1: put 30-day samples on stability• Study day 15: put 15-day samples on stability• Study day 23: put 7-day samples on stability• Study day 30: test all samples on same assay run

• Independent slope estimates without run-to-run assay variability

• More planning with lab• Protocols are more complicated

Page 21: Beyond ICH Q1E Opening Remarks

21

#4 Groups

• What about group differences?• Sites, components, raw materials?• Different testing labs

• Do we have “enough” data to tell meaningful differences?

• Should we expect group differences?

Page 22: Beyond ICH Q1E Opening Remarks

22

#4 Groups—Are they different?

2.5

3.0

3.5

4.0

4.5

Res

ult

0 5 10 15 20 25 30 35Age in Days

Fit Each Value Lot=="B1"Fit Each Value Lot=="B2"Fit Each Value Lot=="B3"Fit Each Value Lot=="B4"Fit Each Value Lot=="B5"Fit Each Value Lot=="B6"

Increasing Property: Groups

No technical or scientific reason for these groups to be different. Therefore, there is no practical difference here. Sums of squares is small due to low variability within batches.

Group x age effect p-value < 0.0001

Age in DaysGroupLot[Group]Group*Age in DaysLot*Age in Days[Group]

Source12323

Nparm12323

DF2.31153643.24465950.01256730.15561130.0253697

Seq SS465.5347326.7311

0.843715.66981.7031

F Ratio<.0001*<.0001*0.4853<.0001*0.1970

Prob > FSequential (Type 1) Tests

Page 23: Beyond ICH Q1E Opening Remarks

23

#5 Proxy Data

• Patient use involves simulating dosing regimen• Does this impact the molecule?

• Accelerated studies may be held under the same ambient conditions as patient use• Do these studies have same change?• What are timepoints? Are there enough during

the in-use period?

Page 24: Beyond ICH Q1E Opening Remarks

24

In-Use Change

• Non-statistical questions can impact• Conclusions• Analysis• Cost to the business

Page 25: Beyond ICH Q1E Opening Remarks

25

Other Uncontrolled Environments

• Manufacturing wait times• Transfer times between production steps• Transfer times to packaging• Packaging/labeling time• Transfer time shipping• Shipping excursions• When in the process are stability samples

assayed?

Page 26: Beyond ICH Q1E Opening Remarks

26

Estimating Routine Stability Change

Manufacturing Shipping Customer Use

Page 27: Beyond ICH Q1E Opening Remarks

27

Estimating Routine Stability Change

Manufacturing• Controlled temps• Uncontrolled temps• Wait times• Packaging

Shipping Customer Use

Page 28: Beyond ICH Q1E Opening Remarks

28

Estimating Routine Stability Change

Manufacturing• Controlled temps• Uncontrolled temps• Wait times• Packaging

Shipping• Controlled temps• Uncontrolled temps• Warehouse• Loading• Shipping excursions

Customer Use

Page 29: Beyond ICH Q1E Opening Remarks

29

Estimating Routine Stability Change

Manufacturing• Controlled temps• Uncontrolled temps• Wait times• Packaging

Shipping• Controlled temps• Uncontrolled temps• Warehouse• Loading• Shipping excursions

Customer Use• Controlled temps• Uncontrolled temps

Page 30: Beyond ICH Q1E Opening Remarks

30

Estimating Routine Stability Change

Manufacturing• Controlled temps• Uncontrolled temps• Wait times• Packaging

Shipping• Controlled temps• Uncontrolled temps• Warehouse• Loading• Shipping excursions

Customer Use• Controlled temps• Uncontrolled temps

Time 0

• Batch release

Stability Chamber

• End of shelf-life

Where is time 0 sample drawn?Are we missing changes?

Page 31: Beyond ICH Q1E Opening Remarks

31

Conclusions

• Estimating stability change goes beyond statistical computations.• Consider business processes

• Impact to statistical modeling• Consider data structure

• Correlated data points• Data groups

• Consider science AND statistical significance• Consider proxy data