1

STATISTICAL CONSIDERATIONS WHEN EVALUATING CONTENT UNIFORMITY

NJPhAST Meeting: Jim Bergum (BMS) & Kim Vukovinsky (Pfizer) September 22 2011

Outline
2

Content Uniformity Test - History Large N Tests

PhRMA

Statistics Expert Team Modified Large N European Proposals

Content Uniformity and Dissolution Acceptance Limits (CuDAL)

Batch Release - Key Attributes

3

Potency (Average Drug Substance/Dosage Unit) Dissolution (% Drug Substance Released at Specified Time) Content Uniformity (CU)

USP <905> The degree of uniformity in the amount of the drug substance among dosage units.

Typical Criteria for CU

4

Proportion of individual results within a specified range (ex: 85-115) or ranges (ex: 75-115) Relative Standard Deviation (RSD)/Coefficient of Variation (CV) Distance From Target

History: Old USP Content Uniformity Test

5

All measurements of dosage units and criteria values are in percentage label claim (%LC). At each stage, calculate the sample average, X , and the sample standard deviation s.

Stage

Number Tested CV < 6.0%

Pass stage if:

S1

10

Tablets: All Results between 85% - 115% Label Claim Capsules: No more than 1 result outside 85%- 115%LC No result outside 75% - 125% LC

CV < 7.8%
S2 20 Tablets: No more than one result outside 85% - 115%LC No result outside 75% - 125%LC Capsules: No more than two results outside 85% - 115% LC No result outside 75% - 125% LC

International Conference on Harmonization (ICH)

6

United States(US), Europe (EU), and Japan (JP) Harmonize CU

PhRMA

Statistics Expert Team Base on JP test Adjust JP test to perform similar to USP Tablet test.

History: Old Japanese Content Uniformity Test

7

All measurements of dosage units and criteria values are in percentage label claim (%LC). At each stage calculate the sample average, X , and the sample standard deviation s. Stage Number tested Pass stage if:

S1

10

Acceptance Value (AV) = | X - 100| + 2.2s

15.0

i)
S2 20 ii)

| X - 100| + 1.9s

15.0 using all 30 results (S1 + S2)

No dosage unit is outside the maximum allowed range of 75% to 125% Label Claim.

Max s

S1 :15/2.2 = 6.8 S2 :15/1.9 = 7.9

Harmonized Uniformity of Dosage Unit (UDU) Test

8

All measurements of dosage units and criteria values are in percentage label claim (%LC). At each stage calculate the sample average X and the sample standard deviation s.

Stage

Number tested

Pass stage if:

S1

10

AV = |M - X | + 2.4s

15.0, where M is defined below.

S2

20

i) |M - X | + 2.0s 15.0 using all 30 results (S1 + S2) ii) No dosage unit is outside the maximum allowed range of 0.75*M to 1.25*M.

M is defined as follows:

(i) If X is less than 98.5%LC, then M = 98.5%LC.

Indifference Zone
(ii) If X is between 98.5 and 101.5%LC, then M = X . (iii) If X is greater than 101.5%LC, then M = 101.5%LC.

Operating Characteristic (OC) Curves: UDU vs Japan vs Old USP

9

UDU Old USP Tab Japan

UDU Old USP Tablet

Japan

Batch Mean=100%LC

Batch Mean = 96%LC

Example: 10 Tablets Stage 1

10

S1 CU Data 1 98.4 2 84.7 3 101.3 4 100.2 5 97.2 6 94.2 7 91.9 8 99.0 9 102.3 10 96.2 Mean S RSD(%) Min Max = 96.5 = 5.2 = 5.4 = 84.7 = 102.3

22 4 0 2 3 3

84 86 88 90 92 94 96 98 00 02 Old USP: 1out (85-115) (Fail)

Old JP: UDU: AV = 3.5 + 2.2*5.2 = 15.0 (Pass) AV = 2.0 + 2.4*5.2 = 14.6 (Pass)

Hot Topics/Issues in Past 15 years

11

Gaining Greater Process Understanding

Quality by Design Design Space Facilitates fast and precise measures Significantly increases real time information (Real Time Release) Improves manufacturing process understanding, control and capability Provides content uniformity results for a large number of dosage units Stated in USP General Notices [Section 3.10. Applicability of Standards], the UDU procedure is not intended for inspecting uniformity of finished product for lot/batch release. Statements about whether the UDU test is met apply only to the units tested. Applying the USP UDU test for lot/batch release does not demonstrate compliance with the Current Good Manufacturing Practices (cGMPs) 21CFR Section 211.

Better Non-destructive measurement techniques (NDT), such as NIR

Realization that USP tests are not batch release tests

How Can Statistics Help?

12

Develop Statistical Procedures to

Evaluate

Results from Large Sample Sizes (Large N) Assure that batches will meet USP tests ( )

What is the goal?

13

Determine Large N release criteria that provide similar performance to the UDU test.

Original: PhRMA CMC Statistics Expert Team (Sandell, D.; Vukovinsky, K;

Diener, M.; Hofer, J.; Pazdan, J.; Timmermans, J. Development of a Content Uniformity Test Suitable for Large Sample Sizes. Drug Information Journal 2006, 40, 337-344). Modified: Bergum, Vukovinsky "A Proposed Content-Uniformity Test for Large Sample Sizes", Pharm. Tech., November 2010, p 72-79

Ph. Eur. PAT working Group (European Pharmacopoeia): Evaluation of Uniformity of Dosage Units using Large Sample Sizes, 2011

Determine limits that provide assurance that a future sample taken from a batch will pass the UDU test.

Content Uniformity & Dissolution Acceptance Limits: Bergum, J.S. and

Hua Li, "Acceptance Limits for the New ICH USP 29 Content Uniformity Test," Pharmaceutical Technology, October 2007, pp. 90-100.

Large N: PhRMA SET

14

One tiered counting test Count number of results (C) outside 85% to 115% LC
Criteria: C 0.048*N

Collect N 100 Dosage Units

Express Result as % LC

Cs for Selected Ns N 100 250 500 1000 5000

# Tablets outside (85,115) %LC C? Yes No Reject Batch Pass Batch

11

23

47

239

Large N Test: Modified PhRMA SET

15

One tiered counting test Count number of results (C) outside 85% to 115% LC

Collect N Dosage Units

Express Result as % LC

Criteria: C 0.03*N

No result outside 75-125% LC?

# Tablets outside (85,115) %LC C? Yes No Reject Batch Pass Batch

Cs for Selected Ns

N C

100 3

250 7

500 15

1000 5000 30 150

OC Curves PhRMA SET vs Modified

16

Batch Mean = 100%

UDU Modified (N= 100)

PhRMA SET (N=100)

Batch Mean = 96%

UDU Modified (N= 100) PhRMA SET (N=100)

Large N: Ph. Eur. PAT WG Proposal

17

Parametic

Assuming Normal Distribution

K=2.4 stage 1 (n=10) and K=2.0 stage 2 (n=30) => Confidence Level = 84% and Coverage = 91% For given N, compute k so that CI=84% and Cov=91% Pass if |M - X | + k*s 15.0 and no more than X units are outside 0.75*M to 1.25*M where X = 0 for N < 500 and increases for larger N.

Nonparametric

Same as PhRMA SET criteria with an allowance of tablets outside Target +/- 25 if N > 500.

European Large N Proposal

Batch Mean = 100%LC
18

UDU EU Parametric (N= 100) EU Parametric (N=300)

EU Nonparametric (N=100)

CuDAL Test: Background

19

Methodology Developed in Mid 80s Application: Process Validation (Show process does what it purports to do)

Show Specific Quality Attributes will meet associated Testing Standards (eg: CU) Content Uniformity (units have similar amount of drug) Dissolution (units dissolve at required rate)

Request/Mission - Develop limits based on the process validation sample results that provide confidence that the testing standard samples will pass the testing standard.

Content Uniformity and Dissolution Acceptance Limits Example: UDU Acceptance Limit Table
20

90% Confidence Interval, 95% Coverage, n=30

Mean 97.0
Sample Results: Mean = 99.0 RSD (%) = 3.42

RSD(%) 3.79

98.0
99.0 100.0

4.03
4.26 4.47
Acceptance Limit

101.0
102.0 103.0

4.17
3.87 3.57

Meeting Relative Standard Deviation (RSD) Limit assures, with 90% confidence, that a future testing standard sample take from the batch has greater than a 95% chance of passing the UDU test.

Justification
21

Provides high assurance that batch meets regulatory standard Assurance increases with increased sample size. Will always need a standard to define Acceptable. Can be used for more than validation or product release (ex: evaluation of NIR methods for CU used in real time release). Tied directly to regulatory requirements Ensures compliance with 21 CFR 211.165(d) - Testing and Release for distribution Can be used as a tool to meet the expectations set forth by FDA's Process Validation Guidance

Strategy Part 1
22

1. 2.

Select Testing Standard (Ex: UDU) Assume probability distribution for individual observations (ex: Normal with parameters (Mu) & (Standard
Deviation))

3.

4.

Assuming known distribution parameters, mathematically derive* the Lower Bound for each stage (Note: Each stage may have multiple criteria!) This is the hard part! Lower bound for overall test is the maximum of the individual stage lower bounds

*Bergum, J.S. and Hua Li, "Acceptance Limits for the New ICH USP 29 Content Uniformity Test," Pharmaceutical Technology, October 2007, pp. 90-100.

General Calculation
23

1.) Probability of Passing each stage m = P(Ci1 and Ci2 and Cim) 1- j=1 (1-P(Cij))
where: P(Si) is the probability of passing stage i, P(Cij) is the probability of passing the j-th criterion of the m criteria within the i-th stage.

2.) Probability of Passing a k-stage test

max{P(S1), P(S2), , P(Sk)}

UDU Test 95% Lower Bound Contour

24

Batch Standard Deviation

95% Lower Bound

Mean 90 or 110 95 or 105 100

Std Dev 2.59 4.57 6.11

Prob(Passing UDU) Simulate USP (N=1,000,000) 95.960 96.022 95.996

Batch Mean

Sampling: Population (Whole Batch)

25

Sampling: N=10 (1 Result/Location)

26

Sampling: N=20 (2 Results/Location)

27

Two Sample/Criteria Types

28

Acceptance Limit Sample (Results Compared to

Acceptance Limit Table Criteria)
Sampling

Plan 1: One unit per location Sampling Plan 2: n units per location (allows estimation of
between/within location variability)

UDU Sample (Results Compared to Testing UDU Criteria)

Sampling

Plan: Defined by the Testing Standard (Usually a random sample from the batch Sampling Plan 1)

Strategy Part 2
29

5. 6.

7.

8.

9.

Select Sampling Plan (1 or 2). Construct confidence interval for the distribution parameters based on user defined confidence level. Determine lower bound probabilities for each point in the confidence interval. Determine maximum probability across all points in confidence interval. Compare maximum probability to user defined coverage (Lower Bound).

UDU Test Lower Bound with Simultaneous Confidence Interval

30

Batch Standard Deviation

(X

Z*ULS/ n, ULS)

95% Lower Bound

Confidence Interval
( X , S)

Batch Mean ULS = Upper CI for Batch SD

Construct Acceptance Limit Table

31

To Generate Table, user selects

Confidence

Level (Usually 90 or 95%) Coverage Lower Bound - Desired Probability of future Testing Standard Samples passing Testing Standard (usually 95%). Sampling Plan/Sample Size Target (Usually 100)

Content Uniformity and Dissolution Acceptance Limits Example: UDU Acceptance Limit Table
32

90% Confidence Interval, 95% Coverage, n=30

Mean 97.0

RSD(%) 3.79

98.0
99.0 100.0

4.03
4.26 4.47

101.0
102.0 103.0

4.17
3.87 3.57

Meeting Relative Standard Deviation (RSD) Limit assures, with 90% confidence, that a future testing standard sample take from the batch has greater than a 95% chance of passing the UDU test.

Sample Size Effect on RSD Limit Sampling Plan 1

33

Sample Mean(%LC) 95.0 98.0

RSD(%) Limit 10 2.35 2.88 30 3.30 4.03 60 3.71 4.53

100.0
102.0 105.0

3.21
2.77 2.13

4.47
3.87 2.99

5.00
4.36 3.36

Evaluating Acceptance Limit Table - Determining Adequate Sample Size

34

Pick sample size such that the probability of passing the

acceptance limit table is: High for a Good Batch (Acceptable Batch Mean and Std Dev) Low for a Bad Batch (Unacceptable Batch Mean and Std Dev)

For given sample size, find probability of passing

acceptance limit table by integrating over tabled values.

Example: Sampling Plan 1 P(Passing 90/95 Acceptance Limit Table)

35

Sample Size
Batch Mean Batch RSD 2 3 2 3 10 99.0 59.7 96.0 46.2 30 100.0 99.9 100.0 98.8 60 100.0 100.0 100.0 100.0

100.0 98.0 96.0

2 3

84.5 26.0

100.0 84.9

100.0 99.7

All Together Now (N=200)

USP

Modified Large N
Euro Nonparametric Parametric CuDAL 90% CI/95% Cov 95% CI/95% Cov

36

Example: Sampling Plan 2 (15 X 4)

37

Result
Location
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Summary statistics
3 4
98.37 101.01 98.88 98.94 99.66 99.06 98.65 98.63 97.67 100.26 97.49 97.28 100.48 98.49 100.14 97.5 100.29 97.96 97.78 97.2 98.8 99.98 98.06 98.95 98.74 97.5 98.8 98.62 100.93 99.2

1
97.08 99.72 99.9 98.78 96.32 100.97 97.02 99.39 99.59 97.97 96.09 98.87 101.1 100.8 99.7

2
99.72 100.32 98.27 98.17 96.61 102.17 97.35 98.81 97.8 98.54 98.61 97.81 102.6 100.34 100.09

Mean
98.17 100.34 98.75 98.42 97.45 100.25 98.25 98.72 98.50 98.88 97.42 98.19 100.70 100.14 99.78

Variance
1.36 0.28 0.73 0.29 2.31 2.57 1.83 0.30 0.86 0.96 1.07 0.60 2.71 1.27 0.19

Std Dev
1.17 0.53 0.86 0.54 1.52 1.60 1.35 0.55 0.93 0.98 1.03 0.78 1.65 1.13 0.44

Example: Sampling Plan 2

38

Example: Sampling Plan 2

39

Descriptive Statistics

Mean SE (within-location Std Dev) Standard deviation of location means

98.93 1.07

1.06

Standard Deviation of Location Means 0.9 SE 0.9 1.0 1.1

LL UL LL

1.0
UL LL

1.1
UL LL

1.2
UL

88.1 88.2 88.4

111.9 111.8 111.6

88.5 88.6 88.7

111.5 111.4 111.3

88.9 89.0 89.1

111.1 111.0 110.9

89.3 89.4 89.5

110.7 110.6 110.5

1.2 1.3

88.5 88.7

111.5 111.3

88.9 89.0

111.1 111.0

89.2 110.8 89.4 110.6

89.6 110.4 89.7 110.3

Sampling Plan 2: Sample Size Evaluation

40

Prob(Passing Acceptance Limit Table) Variance Component (SD) Batch Mean Between Location* Within Location
(Example = 0.91) (Example = 1.1)

Sampling Plan (Loc x #/Loc) 15x4 100.0 100.0 100.0 99.6 15x2 100.0 100.0 100.0 98.2 10x2 100.0 96.6 99.7 81.7

1 100 2

1 2 1 2

1 97 2

1
2 1 2

100.0
98.6 100.0 90.4

100.0
97.9 99.9 81.9

100.0
82.3 94.8 53.3

* = ((SD Location Means)2 - SE2/n)

CuDAL: Current Status

41

General Methodology: ASTM E11 Standard E2709

Original
Lower

(Done)

Bound Confidence Intervals Acceptance Limit Table

Revision
Add

(In 2nd Ballot)

Sampling Plan 2

Content Uniformity: ASTM E55 Standard (3rd Ballot)

Lower

Bound Calculations Specific to UDU Test

Future Other Tests?

Overview
42

API

Design Space
NOR

Q b D

Testing/Monitoring
CuDAL

Formulation

Analytical
Large N

Process Design Process Qualification

Submission Post Approval Statistical Process Control

Control Charts Relate Response to Inputs CuDAL Large N

Continued Process Verificaton Reduced Testing and/or Criteria?

Summary/Take Home Messages

43

UDU is a compromise between Old USP and Japan CU test Modified large N

Easy to use. Equivalent or more conservative than UDU for N < 250. Plans

Sampling

1: One result per location 2: More than one result per location Allows more complete evaluation of between and within location variability

Batch size doesnt matter when batch size is much larger than sample size.
Can be used for any sample size. Easy to use. Provides assurance that batch will pass UDU test if tested.

CuDAL

Need Strategy for reducing testing as process knowledge increases (Ex: CuDAL => Modified Large N)