statistical approaches to meeting emerging usp guidelines ... · 7/6/2010  · d. lansky...

29
D. Lansky Introduction Equivalence Scaling Summary Statistical Approaches to Meeting Emerging USP Guidelines for Bioassay Development, Analysis, and Validation David Lansky, Ph.D. Burlington, Vermont, USA May 12, 2010 1 / 29

Upload: others

Post on 01-Nov-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Statistical Approaches to MeetingEmerging USP Guidelines for BioassayDevelopment, Analysis, and Validation

David Lansky, Ph.D.

Burlington, Vermont, USA

May 12, 2010

1 / 29

Page 2: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Bioassay Background

I Parallel Line: Shapes similar assume:- same cmpd⇒ common slope & asymptotes- interpret horizontal shift=log potency

I Slope Ratio: Intercepts similar assume:- same cmpd ⇒ common y intercept- interpret slope ratio=potency

Parallel Line  Dose on log scale

Re

sp

on

se

-1.5

-1.0

-0.5

0.0

0.5

5 10 20 40 80 160 320 640

R R RR

R

RR

R

TT T

T

T

T

TT

Slope Ratio  Dose on arithmentic scale

Re

sp

on

se

6

8

10

12

14

16

1 2 3 4 5 6

R

RR

R

RR

T

T

T

T

T

T

2 / 29

Page 3: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Similarity in assay

Distance between curves

1 2 3 4 5 6 7 8 9 10 11 12

3 / 29

Page 4: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Similarity in assay

Distance between curves

1 2 3 4 5 6 7 8 9 10 11 12

Do the two curves have the same shape?

4 / 29

Page 5: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Similarity in assay

Distance between curves

1 2 3 4 5 6 7 8 9 10 11 12

Do the two curves have the same shape?

5 / 29

Page 6: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Potency SD and Dose Range

I SDlog potency � SDlog ED50

I PGSD (= 100(eSD − 1)) 5%, 6x-15xED50 range 68%-216%

I product potency spec. often 0.71-1.41I Generally need 3 doses on steep partI doses often 1:2 dilutionsI SDlog ED50 & potency range: need 5

dilutions

6 / 29

Page 7: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Transform not Weight

I Weight on dilution-specific variances?I SDlog ED50 inflates SD at middle dilutionsI Weighting confounds nonlinear mixed

model

0.5

1.0

1.5

2 4 6 8 10 12 7 / 29

Page 8: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Outliers

I After transform (or weight)I Fit model (w/Design Structure) to all dataI Avoid shape assumptionsI Separate outlier detection from model

adequacy

8 / 29

Page 9: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Design Structure

I Old recommendation: keep designsimple, drive bias into variance

I New: recognize design structureI Grouped (multi-channnel) and serial

dilution commonI pseudo-replicates (multiple aliquots from a

preparation)I strip-plot designs appearI (incomplete) block designs efficient

9 / 29

Page 10: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Strip-Plot Design

ABCDEFGH

1 2 3 4 5 6 7 8 9101112

1.9 1.81.81.6 1.41.2 1.2 1.11.1

1.9 1.91.81.61.4 1.31.2 1.1 1.11.1

1.9 1.91.81.71.6 1.31.2 1.1 1.11

1.9 1.91.81.81.6 1.41.3 1.2 1.11.1

1.8 1.91.81.81.7 1.61.3 1.2 1.11.1

1.9 1.91.81.81.8 1.61.4 1.3 1.11.1

1.6

1

1 2 3 4 5 6 7 8 9101112

1.9 1.91.8 1.81.6 1.4 1.31.21.11.1

1.9 1.91.8 1.71.4 1.3 1.11.11.11.1

1.9 1.81.8 1.71.5 1.3 1.21.11 1

1.9 1.91.8 1.81.7 1.5 1.21.21.11.1

1.9 1.81.9 1.81.8 1.5 1.31.21.11.1

1.9 1.91.9 1.91.8 1.7 1.41.21.11.1

2

1 2 3 4 5 6 7 8 9101112

ABCDEFGH

1.91.91.9 1.81.61.5 1.3 1.21.21.1

1.91.91.8 1.71.51.3 1.2 1.11.11.1

1.91.91.8 1.71.61.4 1.3 1.21.11.1

1.91.91.9 1.81.61.5 1.3 1.21.21.1

1.921.9 1.91.71.6 1.4 1.21.21.1

1.91.91.9 1.91.81.6 1.5 1.31.11.1

3

1 2 3 4 5 6 7 8 9101112

ABCDEFGH

1.91.9 1.91.8 1.6 1.51.31.21.11.1

1.81.9 1.81.7 1.4 1.31.21.11.11.2

1.91.9 1.81.6 1.5 1.31.31.21.11

1.91.9 1.81.8 1.6 1.41.21.21.21.1

1.91.9 1.81.8 1.6 1.51.31.31.21.1

1.91.9 1.91.8 1.8 1.61.41.31.11.2

4

1 2 3 4 5 6 7 8 9101112

1.91.91.91.71.61.4 1.31.2 1.11

1.91.91.71.61.41.3 1.21.1 11.1

21.91.81.71.61.4 1.21.1 1.11.1

1.91.91.81.81.71.4 1.31.2 1.11.1

1.91.91.91.81.71.5 1.41.2 1.11

21.921.91.81.7 1.3 1.21.11.7

5

1 2 3 4 5 6 7 8 9101112

ABCDEFGH

1.91.9 1.81.81.6 1.41.21.1 1.1 1.1

1.91.9 1.71.61.4 1.21.11.1 1 1.1

1.91.9 1.81.71.5 1.31.21.1 1.1 1.1

1.91.9 1.71.81.6 1.41.21.1 1.1 1.1

1.91.9 1.91.81.7 1.51.31.2 1.1 1.1

1.91.9 1.91.81.8 1.61.41.2 1.1 1.1

6

10 / 29

Page 11: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Strip-Plot Response

x

yhat

1.2

1.4

1.6

1.8

-10 -8 -6 -4 -2 0

R

R

RR

R

R

RR

RR

HH

H

H

H

HH

HHH

hh

hh

h

h

hh

hh

111

1

1

1

1111

ss

sss

s

s

ss

s

SSSS

S

S

S

S

SS

1

RRR

R

R

R

RR

RR

HH

H

H

H

H

HHHH

hh

hh

h

h

hhhh

1111

1

1

11

11

sssss

s

s

sss

SS

SS

SS

S

S

SS

2

-10 -8 -6 -4 -2 0

RRRR

R

R

R

RRR

HHH

H

H

H

HHHH

hhh

h

h

h

hh

hh

1111

1

1

111

1

ss

ss

s

s

s

sss

SSS

SS

S

S

S

SS

3

RRRR

R

R

R

RRR

HH

HH

H

HH

HHH

hhh

h

h

hhh

hh

1111

1

1

1111

ss

ss

s

s

ss

ss

SSS

SS

S

S

S

SS

4

-10 -8 -6 -4 -2 0

RRR

RR

R

R

RR

R

HH

HH

H

HH

HH

H

hh

hh

h

h

hhh

h

11

11

1

1

11

11

ss

ss

s

s

s

s

ss

SSS

SS

SS

S

SS

5

1.2

1.4

1.6

1.8

RRR

R

R

R

RR

RR

HH

H

H

H

H

HH

HH

hhh

h

h

h

hhhh

1111

1

1

1111

ssss

s

s

s

sss

SSS

SS

S

S

S

SS

6

11 / 29

Page 12: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Strip-Plot Model

yijk =A + ak + aik − D

1 + e−B(xijk−(Ci+ck+cjk))+ D + εijk

independent : εijk ∼ iid N(0, σ2

)ak ∼ iid N

(0, σ2

ablock

), aik ∼ iid N

(0, σ2

arow

)ck ∼ iid N

(0, σ2

ablock

), cjk ∼ iid N

(0, σ2

acol

)with i for sample (row in block), j for dilution(column in block), and k block.

12 / 29

Page 13: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Combine log Potencies

I ”weighted” and ”semi-weighted” assumebetween-assay σpotency = 0 (Finney, 1978)

I sampling SD of log potency safeI Link properties of reported value to

clinical need

13 / 29

Page 14: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Validation Params & Methods

I Relative AccuracyI ”Linearity” of log potencyI Bias limit at each targetI Bias trend limit

I PrecisionI Components (repeatability, intermediate

precision, reproducibility)I Predict for various ”formats”

I SpecificityI (Robustness)I Equivalence used broadly

USP <1032>, <1033>, and <1034> will appear in PF 36(4)(early July). These and <111> athttp://www.usp.org/meetings/workshops/2010Bioassay.html

14 / 29

Page 15: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Why Equivalence?

I Statistical similarity testedI Biological similarity assumed (stat.

similarity necessary, not sufficient)

I Assume critical differences known:I SlopeI Upper asymptoteI Lower asymptote

I Equivalence tests: ”Are reference andtest sufficiently similar”

I USP -> equivalenceI Curve parameters have meaning

I Critical quality attributesI Separate equivalence intervals

15 / 29

Page 16: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Similarity: What is Needed

I Slope ONLY: lot releaseI Asymptote of max activity:

I compare standards (i.e.; new lot)I change productionI stability

I Asymptote of min activity: checks onlymatrix effects

16 / 29

Page 17: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Difference testing for Similarity

I Practical ProblemsI Assays w/low variance fail parallelismI Assays w/high variance pass parallelism

I Theoretical Problem (one parameter)I Difference test

I H0 : βReference = βTest

I α (Type I) controls P(Falsely rejecting H0)I β∗ (Type II) controls P(Accepting H0|δ∗)

I Equivalence testI H0 :| βReference − βTest |> δ∗

I α controls Type I error of H0 : for δ∗

I In Practice: What is δ∗?

17 / 29

Page 18: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Difference test for Similarity

Incorrectly sensitive to variance

Log(dose)

Response

-1

0

1

20 40 80 160 320

R

R

R

R

T

T

T

Tp< 0.03612

assay

R

R

R

R

T

T

T

T

p< 0.00577

assay

R

R

R

R

T

T

T

Tp< 0.53529

assay

20 40 80 160 320

-1

0

1

RR

R

R

T

T

T

Tp< 0.07369

assay

18 / 29

Page 19: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Equivalence test for Similarity

Correctly sensitive to variance90% CI inside indifference zone⇒ equivalence95% CI does not include 0 ⇒ difference

Test slope as % of Reference

1

2

3

4

50 100 200

Log(dose)

Response

-101

20 40 80 160 320

RR

RR

TT

TTp< 0.03612

assay

-101

RR

RR

TT

TT

p< 0.00577

assay

-101

RR R

R

T TT

Tp< 0.53529

assay

-101

R RR

R

TT

T Tp< 0.07369

assay

19 / 29

Page 20: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

What can we say about δ∗?

I What if we knew δ∗?I Assay dependentI Scale dependentI Parameter link to quality attribute weakI Standardize meaning of δ?

20 / 29

Page 21: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Seeking Scale Invariance

I Assays have different responsesI Asymptotes in response units

I Slope units logit(response)log(dose)

21 / 29

Page 22: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Simple Scale Invariance

y ∗ =Ai

1 + e−Bi (log(x)−Ci )+ Di

with C = log (ED50) and i = [Ref|Test](Ratkowsky & Reedy, 1986)

Range= A: %∆A = 100ATest−ARef

ARef

Lower Asy= D: %∆D = 100DTest−DRef

ARef

Slope= B : %∆B = 100BTest−BRef

BRef

Concerns:I Is meaning consistent?I Are these useful across assays?I Variances of %∆A, %∆B , and %∆D

22 / 29

Page 23: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Interpretation of %∆A:B :D

     

A

     

B

     

D

A in {2A/3,A, 3A/2}, %∆A = 10B in {B/3,B , 3B}, %∆B = 50D in {2D/3,D, 3D/2}, %∆D = 10

23 / 29

Page 24: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Interpretation of %∆A:B :D

I scaling mostly worksI Requires some explaining

24 / 29

Page 25: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

%∆A:B :D 10:50:10

D=-10

     

{ -50 }{ -10 }

{ 0 }{ -10 }

     

{ 50 }{ -10 }

{ -50 }{ 0 }

{ 0 }{ 0 }

{ 50 }{ 0 }

{ -50 }{ 10 }

     

{ 0 }{ 10 }

{ 50 }{ 10 }

D=0

     

{ -50 }{ -10 }

{ 0 }{ -10 }

     

{ 50 }{ -10 }

{ -50 }{ 0 }

{ 0 }{ 0 }

{ 50 }{ 0 }

{ -50 }{ 10 }

     

{ 0 }{ 10 }

{ 50 }{ 10 }

D=10

     

{ -50 }{ -10 }

{ 0 }{ -10 }

     

{ 50 }{ -10 }

{ -50 }{ 0 }

{ 0 }{ 0 }

{ 50 }{ 0 }

{ -50 }{ 10 }

     

{ 0 }{ 10 }

{ 50 }{ 10 }

50% slope change seems small in comparison

25 / 29

Page 26: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

%∆A:B :D 5:35:5

D=-5

     

{ -35 }{ -5 }

{ 0 }{ -5 }

     

{ 35 }{ -5 }

{ -35 }{ 0 }

{ 0 }{ 0 }

{ 35 }{ 0 }

{ -35 }{ 5 }

     

{ 0 }{ 5 }

{ 35 }{ 5 }

D=0

     

{ -35 }{ -5 }

{ 0 }{ -5 }

     

{ 35 }{ -5 }

{ -35 }{ 0 }

{ 0 }{ 0 }

{ 35 }{ 0 }

{ -35 }{ 5 }

     

{ 0 }{ 5 }

{ 35 }{ 5 }

D=5

     

{ -35 }{ -5 }

{ 0 }{ -5 }

     

{ 35 }{ -5 }

{ -35 }{ 0 }

{ 0 }{ 0 }

{ 35 }{ 0 }

{ -35 }{ 5 }

     

{ 0 }{ 5 }

{ 35 }{ 5 }

35% slope change small?5% range small vs. 5% on lower asy?

26 / 29

Page 27: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Experience with %∆A:B :D

I excellent assays can use 5:35:5I many cell fail 5:35:5, ok w/10:50:10I noisy assays struggle with 15:50:15I Equiv. in linear: longer subsets

27 / 29

Page 28: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Summary

I TransformI Detect outliers (w/smooth) modelI Use Design StructureI Assess similarity with equivalenceI Asymptote sim. needed at timesI Assess Validation with equivalence

28 / 29

Page 29: Statistical Approaches to Meeting Emerging USP Guidelines ... · 7/6/2010  · D. Lansky Introduction Equivalence Scaling Summary Validation Params & Methods I Relative Accuracy I

©

D. Lansky

Introduction

Equivalence

Scaling

Summary

Acknowledgements

I Consulting clientsI USP and USP bioassay panel membersI Carrie WagerI NSF EPSCoRI NIH SBIR 3R44RR02198-03S1

29 / 29