Content Uniformity (CU) Testing For The 21st Century CDER Perspective

Content Uniformity (CU) testing for
the 21st Century: CDER Perspective

Richard (Rik) Lostritto, Ph.D.
Acting Deputy Office Director for Science and Policy and
Biopharmaceutics Lead
Office of New Drug Quality Assessment (ONDQA)
Office of Pharmaceutical Science (OPS)
Center for Drug Evaluation and Research (CDER)
Outline
Introduction
USP <905>
Summary
Features
Case study 1: Levothyroxin

Case study 2: Split Tablets
Alternatives to USP <905>
Parametric Tolerance Interval Testing (PTIT)
Case study 3: PTIT
Summary and conclusions
2
Introduction: Setting the Stage

Content Uniformity (CU) testing is an important
assessment of unit dosage form performance.
Because pharmacological responses are
dynamic, variable, and their manifestations easily
confounded, clinical response alone cannot
serve as an arbiter of adequate CU performance.
CU testing is usually destructive and consumes
resources (wet chemical analyses)
Introduction: Setting the Stage

In some cases CU testing may not be
destructive
e.g., Near Infrared (NIR) and weight variation (WV)
Process Analytical Technologies (PAT) and Real

Time Release Testing (RTRT) allow for more
frequent and convenient sampling
Greater product knowledge
More efficient use of resources / less off line analysis
High assurance of product quality
Introduction: Launch point
Is USP <905> perfect? No.

Is USP <905> adequate? Not all the time.
What are the weaknesses in USP <905> ?
Are there viable alternatives?
Parametric Tolerance Interval (PTIT)
Large N
Cases of interest
Narrow therapeutic index drugs
Drug device combination products
Specialized dosage forms (e.g., inhalation)
Summary of USP<905>
Applies to many dosage forms
Weight variation (WV) versus content uniformity
(CU) for tablets (as an example)
If < 25 mg or < 25% then CU
If > 25 mg and > 25% then WV
Sample sizes for tablets are

10 units (L1) and
30 units (L2 as L1 + 20 more)
Summary of USP<905> (as CU for tablets)

Acceptance value, AV = |M-X| + ks
X = sample mean as percent of label claim (%LC)

K = 2.4 for L1 and K = 2.0 for L2
S = standard deviation of the SAMPLE
M depends on the sample mean
If X is between 98.5 %LC and 101.5 %LC them M = X

If X is < 98.5 %LC then M = 98.5
If X is > 101.5 %LC then M = 101.5
This scenario creates a zone of indifference to the mean
AV cannot exceed 15.0 at L1 or 25.0 at L2
There are also limits on individual values

7
Summary of USP<905> (as CU for tablets)

Limits on individual values in L2
No unit < 0.75 M and
No unit > 1.25 M
That means no unit outside 75 -125 %LC for on target
mean (if the mean is within 1.5% of LC)
For off target means (> +1.5%) the limits on
individuals vary
Features of USP <905> as CU for tablets

Fixed sample sizes as L1 and L2 (10 / 30)
L1 (especially) and L2 are relatively small and
may not provide for a confident estimation of the
batch compared to larger sample sizes.
There are limits on individual values.
Is that valid in a statistically relevant model?
There is a 1.5% zone of indifference around

100%LC.
Rewards off-target sample mean
Also changes M which changes individual limits
Case Study 1: Levothyroxine

Potency
Drug product specifications allow up to 10% loss of
potency over expiry
Intermediate strengths (112-150 mcg) are separated
by less than 10% of dose.
One strength has degraded to contain less active
than a lower strength tablet.
Tighter controls necessary for potency to assure

accuracy of individualized dosing which is
titrated per patient.
10
Case Study 1: Levothyroxine

CU
Variability is also observed within batch

Loss of potency and inter batch variability are linked
Both affect content uniformity performance
Using USP <905> at L1 (10 tablets) for low dose case
With 10% loss of potency an RSD < 2.7% passes

With 5% loss of potency an RSD < 4.8% passes
For reference, at 100% LC, an RSD < 6.25% passes
USP <905> has a zone of indifference of 1.5% + LC
Meeting USP <905> content uniformity with < 5%

potency loss may provide reasonable assurance that
strengths do not lose order of potency on stability.
11
Case Study 2: Split Tablets

Economic driving forces encourage tablet
splitting
Typically, the 2X strength product is not doubly

expensive
Patients, prescribers, and insurers recognize this
Current practices for immediate release tablets
Splitting of scored and non scored tablets

Hand splitting
Kitchen knives and other household implements
Tablet splitters (not regulated)
12

What are some potential problems?
If scored, is the score functional?
Ease to split (splitability)

Even splitting by weight
Loss of mass (crumbs) typically 1-2%
Performance of the major pieces (i.e., halves)

regarding:
Friability
Dissolution rate
Physical and chemical stability
Content uniformity (esp low dose cases)
13

Some Points to Consider for CU of the pieces
Which tablets are appropriate for splitting (IR vs. ER)?

Is a functional score necessary?
Valid assessment of splitability?
Stability assessment (90 days post dispensing?)
New active surface exposed
New sharp edges and corners exposed
Dissolution assessment (meet same specifications?)

Loss of mass (whats a maxiumum allowable loss?)
Content Uniformity
Should each piece Meet USP <905> ?
14
Potential Alternatives to USP <905>

Large N
For RTRT approaches, FDA has approved Large N

approaches using WV and NIR
May be appropriate when large numbers of tablets can
be adequately measured; usually non destructively
The Agency is working with a goal to develop
harmonized criteria with other regions for Large N
approaches
Similar to some Large N approaches, but typically

smaller sample sizes
Discussed in FDA Draft Guidance (add reference)
15

What is PTIT?
A statistically relevant way of estimating CU by

measuring it in a sample
The test has two sample tiers Tier-1 and Tier-2
A different concept than L1 and L2 of USP<905>

Applicant chooses and specifies the tier sizes based
on confidence in their process and product
performance.
Lets explore further through an example

Case 3
16

What are the advantages of PTIT over USP<905>
Sample size choices are more flexible
Sample sizes (2 tiers); typically larger than USP but
smaller than Large N
Passing at Tier-1 rewards high quality
No zone of indifference
No restriction on individual values
No zero tolerance
17
PTIT: BATCH Distributions, Goalposts, and

the Concept of COVERAGE.
0.12
=100; sigma=4
0.10
=100; sigma=8
=100; sigma=12
Probability
0.08
Lower
Goalpost"
Upper
Goalpost"
0.06
0.04
0.02
0.00
70
80
90
100
110
120
130
Variable
18
MSD = [120 - SM] / K

If the mean is < 100 %LC MSD = [SM - 80] / K
If the mean is >100 %LC
Where
SM = sample mean
MSD = the maximum sample standard deviation
allowable to PASS at this Tier (1 or 2)
K is a PTI model derived constant and depends upon
Sample size
Tier-I or Tier-II
Relationship to Tier-I to Tier-II
The goalposts
The desired coverage between the goalposts
19

Note that:
The smaller K is, the larger the allowable MSD is
The closer Sample Mean (SM) is to 100% of target,
the larger is the allowable MSD
The smaller the sample SD is, the more off-target SM
may be and still pass the test
SM must be within the goalposts (obvious)
There is no model constraint on individual values
Here are some values of K
20
PTIT K values
N = sample size
N=10
N=30
N=20
N=60
N=30
N=90
Tier-I
Tier-II
Tier-I
Tier-II
Tier-I
Tier-II
K10
K30
K20
K60
K30
K90
82.5
2.82
1.94
2.20
1.74
2.00
1.66
85.0
2.96
2.04
2.32
1.83
2.11
1.75
87.5
3.12
2.16
2.45
1.94
2.23
1.86
90.0
3.31
2.30
2.60
2.07
2.37
1.98
%
coverage
21
Case 3: Simulated Batch for PTIT

Design a sim batch
Let batch mean = 100 %LC
Let batch Sd = 8.0 %
Let Goalposts be 80% to 120% of LC

Set coverage to 90% within goalposts
Try the smallest possible sampling plan
Tier-I = 10
MSD = [SM-80]/K10 or MSD = [120-SM]/K10
Tier-II = 30
MSD = [SM-80]/K30 or MSD = [120-SM]/K30
Select appropriate K values from the Table

22

N = sample size
%
coverage
N=10 N=30
N=20
N=60
N=30
N=90
Tier-I Tier-II
Tier-I
Tier-II
Tier-I
Tier-II
K10
K30
K20
K60
K30
K90
82.5
2.82
1.94
2.20
1.74
2.00
1.66
85.0
2.96
2.04
2.32
1.83
2.11
1.75
87.5
3.12
2.16
2.45
1.94
2.23
1.86
90.0
3.31
2.30
2.60
2.07
2.37
1.98
23

Tier-I = 10
MSD = [SM-80]/3.31 or MSD = [120-SM]/3.31
Tier-II = 30
MSD = [SM-80]/2.30 or MSD = [120-SM]/2.30
Pass if sample standard deviation < MSD

Do that 30 times
Heres the results of the first 10 runs
24
Case 3. SM=100% and SD=8.0%

90% coverage using smallest sampling plan: 10 / 30
Run #
N=10
N=30
SM
SD
Tier-I
Tier-II
1=pass
1=pass
SM
SD
100.5
10.7
99.5
8.6
98.9
7.1
98.0
7.8
98.3
9.5
100.4
8.8
100.6
5.9
101.9
6.3
101.4
5.7
99.4
8.7
98.9
8.8
100.2
8.6
99.0
5.8
100.3
6.7
98.6
8.8
100.5
7.6
98.6
8.5
99.5
6.1
10
101.7
8.6
101.0
8.5
0
25

Undesirable outcome
The batch passed only 21/30 times
There were no Tier-I passes
Explaining this outcome

Goalposts too far apart?
Coverage too high?
Increase coverage -> larger K
Sample sizes too small ?

Decrease sample size -> larger K
Lets try larger sample tier sizes

Larger samples are more likely to represent the batch
The requirements for larger samples are less stringent
26

N = sample size
%
coverage
N=10 N=30
N=20
N=60
N=30 N=90
Tier-I Tier-II
Tier-I
Tier-II
Tier-I Tier-II
K10
K30
K20
K60
K30
K90
82.5
2.82
1.94
2.20
1.74
2.00
1.66
85.0
2.96
2.04
2.32
1.83
2.11
1.75
87.5
3.12
2.16
2.45
1.94
2.23
1.86
90.0
3.31
2.30
2.60
2.07
2.37
1.98
27
Case 3. Sm=100% and SD=8.0%

90% coverage using LARGEST sampling plan: 30 / 90
K30 = 2.37
K90 = 1.98
Run sampling and testing 30 times
Results
Pass Tier-I about half the time
Pass Tier-II 29/30 times
28
Case 3. Sm=100% and SD=8.0%

90% coverage using LARGEST sampling plan: 30 / 90
This may be an acceptable situation
Reasonable opportunity to pass at Tier-I

(good)
High probability to pass at Tier-II
(very good)
No limit on individuals
(excellent)
The applicant has the option to select and specify the
sampling plan to suit their risk and resources
May be resource intensive at Tier-II
Typically, coverage is set at 90% between goalposts
29

What are the issues with PTIT?
Technically assumes a normal distribution of the

measured property in the batch. However, provided the
distribution is symmetrical, it should be ok.
If CU variability is not random, look for, evaluate, and control
for operative non-random causes
There are multiple PTIT approaches
Tier sizes and ratios

Various ways to weight the power of tiers
Various goalposts and coverage within goalposts
There is no universal agreement on best approach

However
30

The Agency has identified some PTIT
approaches that may be acceptable for CU
testing. For inhalation products these include:
Goalposts 80% to 100% of label claim.

Coverage within goalposts, 90% of batch population
Two sample size tiers in a 1:3 ratio
For example, 10:30, 20:60, 30:90, etc.
Power (as 1-alpha) equally distributed between tiers
The applicant chooses and specifies the tier sizes
Cases vary; recommend discussion with the Agency

before implementing
31
Case Study 4: Drug Eluting Cardiac Stents
Very low drug levels in or on the stent

Potent drugs used (e.g., paclitaxel)
Extended release design, in place a long time
Local treatment of vulnerable target tissue
Local tissue levels likely in therapeutic range.
Dose response relationships exist
The vast majority of these products comply with

USP <905> for tablets at L1 (n=10)
32
Summary and Conclusions

In some cases USP <905> works fine for CU
testing, but it has flaws:
Small fixed sample sizes for L1 and L2
Zone of indifference of 1.5%
Limits on individual values and they may vary
according to the sample mean
PTIT testing offers greater flexibility and appears

to have greater statistical relevance provided the
batch distribution is near-normal or symmetrical
33
Summary and Conclusions

Loss of potency on stability for a narrow
therapeutic drug may put more demands on
what is acceptable CU performance.
Split tablet performance is a current and
pervasive issue that merits careful consideration
in terms of CU
Pharmacokinetic and pharmacological input can
further help to determine the criticality of CU
performance to safety and efficacy
34
Acknowledgements
Yi Tsong, Office of Biostatistics, CDER
Meiyu Shen, Office of Biostatistics, CDER
Christine Moore, Office of New Drug Quality
Assessment, CDER
35
Thank you
Address Questions or Comment to:
[email protected]
36

Content Uniformity (CU) Testing For The 21st Century CDER Perspective

Uploaded by

Copyright:

Available Formats

Content Uniformity (CU) Testing For The 21st Century CDER Perspective

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Content Uniformity (CU) Testing For The 21st Century CDER Perspective

Uploaded by

Copyright:

Available Formats

What are some of the issues with USP <905>?

What are some of the issues with USP <905>?

What are some alternatives to USP <905> for content uniformity testing?

What are some alternatives to USP <905> for content uniformity testing?

Content Uniformity (CU) testing for

the 21st Century: CDER Perspective

Case study 1: Levothyroxin

Introduction: Setting the Stage

Introduction: Setting the Stage

Process Analytical Technologies (PAT) and Real

Introduction: Launch point

Is USP <905> perfect? No.

Sample sizes for tablets are

Summary of USP<905> (as CU for tablets)

X = sample mean as percent of label claim (%LC)

If X is between 98.5 %LC and 101.5 %LC them M = X

AV cannot exceed 15.0 at L1 or 25.0 at L2

There are also limits on individual values

Summary of USP<905> (as CU for tablets)

Features of USP <905> as CU for tablets

There is a 1.5% zone of indifference around

Case Study 1: Levothyroxine

Tighter controls necessary for potency to assure

Case Study 1: Levothyroxine

Variability is also observed within batch

Using USP <905> at L1 (10 tablets) for low dose case

With 10% loss of potency an RSD < 2.7% passes

Meeting USP <905> content uniformity with < 5%

Case Study 2: Split Tablets

Typically, the 2X strength product is not doubly

Current practices for immediate release tablets

Splitting of scored and non scored tablets

Case Study 2: Split Tablets

Ease to split (splitability)

Performance of the major pieces (i.e., halves)

Case Study 2: Split Tablets

Which tablets are appropriate for splitting (IR vs. ER)?

Dissolution assessment (meet same specifications?)

Potential Alternatives to USP <905>

For RTRT approaches, FDA has approved Large N

Parametric Tolerance Interval Testing (PTIT)

Similar to some Large N approaches, but typically

Parametric Tolerance Interval Testing (PTIT)

A statistically relevant way of estimating CU by

The test has two sample tiers Tier-1 and Tier-2

A different concept than L1 and L2 of USP<905>

Lets explore further through an example

Parametric Tolerance Interval Testing (PTIT)

PTIT: BATCH Distributions, Goalposts, and

Parametric Tolerance Interval Testing (PTIT)

MSD = [120 - SM] / K

Parametric Tolerance Interval Testing (PTIT)

Here are some values of K

Case 3: Simulated Batch for PTIT

Let Goalposts be 80% to 120% of LC

Select appropriate K values from the Table

Case 3: Simulated Batch for PTIT

Case 3: Simulated Batch for PTIT