exposure-response in the presence of confounding in the presence of confounding ... ss dx c 2016...

Exposure-response in the presence of confounding

Jonathan L. French, ScD

Metrum Research Group LLC

May 3, 2016

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 1 / 24

Outline

1 Introduction

2 Example of exposure-response with multiple doses

3 Exposure-response with one dose?

4 Concluding thoughts

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 2 / 24

Introduction

Exposure-response is in the mind of the beholder

Ranging across . . .PK-PD modelK-PD modelPD model driven by a summary measure derived from PK modelPD model based on observed concentrationsDose-response

Dose   Exposure   Response  

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 3 / 24

Introduction

What summary measures?

●

●

●

●

●

●

0

10

20

30

0 50 100 150 200Time (hours)

Con

cent

ratio

n

Some commonly used include:

AUCss (Dose/(CL/F))Caveragess (AUCss/τ )CmaxCtrough

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 4 / 24

Introduction

Why do we need exposure-response analyses?

What do they give us that we can’t learn from dose-response?

May be closer to mechanism-based models

Concentration of drug→ biomarker (possibly with a delayed effect)

May allow more precise estimation when dose-response modeling isn’tinformed by the design (e.g., only 2-3 dose levels studied).

May allow more seamless interpolation (or extrapolation) to doses orpopulations that weren’t studied

May allow us to understand if a higher dose is necessary for a subset ofpatients

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 5 / 24

Example of exposure-response with multiple doses

(Simulated) Example of E-R with multiple doses

Phase 2 dose-ranging study3 treatment groups: placebo, 3, or 5 mg QDn=25 per group

Key decision-making endpoints is a continuous landmark endpoint

Objectives:Estimate E-R relationship in Phase 2 populationPredict E-R in Phase 3 population and impact on dose selection

Dose   Exposure   Response  

Covariates  

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 6 / 24

Example of exposure-response with multiple doses

Observed data show an apparant E-R relationship . . .

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

5

10

15

20

0 100 200 300Average steady−state concentration

Effi

cacy

Res

pons

e

Dose   Exposure   Response  

Covariates  

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 7 / 24

Example of exposure-response with multiple doses

Observed data show an apparant E-R relationship . . .

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

5

10

15

20

0 100 200 300Average steady−state concentration

Effi

cacy

Res

pons

e

X1●●

●●

Low X1

High X1

There also appears to be anassociation with a covariate,x1.

Dose   Exposure   Response  

Covariates  

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 7 / 24

Example of exposure-response with multiple doses

. . . and two covariates are associated with exposure

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

100

200

300

−2 −1 0 1 2X1

Ave

rage

ste

ady−

stat

e co

ncen

trat

ion

Dose (mg)●

●

3

5

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

100

200

300

−2 −1 0 1 2X2

Ave

rage

ste

ady−

stat

e co

ncen

trat

ion

Dose (mg)●

●

3

5

Dose   Exposure   Response  

Covariates  

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 8 / 24

Example of exposure-response with multiple doses

Exposure-response models

4-parameter Emax model for efficacy:

Yi = θ0 +Emaxi · Cavgγ

ss

EC50γ + Cavgγss

+ εi

εi ∼ N(0, σ2)

We will model the maximum effect as afunction of X1:

Emaxi = θ1 + θ2 x1i

EC50 = exp(θ3)

γ = exp(θ4)

EmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmaxEmax

EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50EC50

Exposure

Res

pons

e

Gammagamma = 1

gamma = 3

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 9 / 24

Example of exposure-response with multiple doses

Fitted model

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

Low X1 High X1

5

10

15

20

0 100 200 300 0 100 200 300Average steady−state concentration

Effi

cacy

Res

pons

e

X1●

●

Low X1

High X1

Parameter Estimate 95% CIθ0 9.56 (8.79 , 10.3)θ1 9.91 (8.18 , 18.1)θ2 1.43 (0.60 , 2.99)EC50 29.7 (20.0 , 44.0)γ 1.63 (0.74 , 3.60)σ 1.92

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 10 / 24

Example of exposure-response with multiple doses

Extrapolation to phase 3 population

Based on the planned inclusion criteria the Phase 3 population isexpected to have different distributions for X1 and X2.

−4 −2 0 2 4 6

0.0

0.1

0.2

0.3

0.4

X1

Den

sity

Phase 2 populationPhase 3 population

−4 −2 0 2 4

0.0

0.1

0.2

0.3

0.4

X2

Den

sity

Phase 2 populationPhase 3 population

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 11 / 24

Example of exposure-response with multiple doses

Expected exposure and response

Higher response given exposure

12

15

18

0 100 200 300Average steady−state concentration

Res

pons

e PopulationPhase 2

Phase 3

Lower exposure given covariates

0.000

0.005

0.010

0.015

0.020

0 100 200 300 400Average steady−state concentration at 5 mg dose

dens

ity

PopulationPhase 2

Phase 3

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 12 / 24

Example of exposure-response with multiple doses

How do we translate this to dose-response?

If we focus on the expected response, then

E(Y | dose,X ) =

∫E(Y | Cavgss,X ) f (Cavgss | dose,X )dCavgss

and

E(Y | dose,Popn) =∫

E(Y | Cavgss,X ) f (Cavgss | dose,X ) f (X |Popn) dCavgss dX

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 13 / 24

Example of exposure-response with multiple doses

Comparison of expected dose-response

10.0

12.5

15.0

17.5

20.0

0 5 10 15Dose (mg)

Res

pons

e PopulationPhase 2

Phase 3

A 4 mg dose gives ∼85% of maximalresponse in Phase 2population.

Equivalent expectedresponse at 6 mg dose inPhase 3 population.

Potential for higherresponse rates in Phase3 population at higherdoses?

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 14 / 24

Example of exposure-response with multiple doses

Dose   Exposure   Response  

Covariates  

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 15 / 24

Exposure-response with one dose?

Can we use E-R modeling when we’ve only studiedone dose?

Sometimes (e.g. in oncology), a single dose is studied in Phase 2(and Phase 3).We may have some variation in exposure . . .. . . but what if exposure is related to measured (or unmeasured)confounding variables?

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 16 / 24

Exposure-response with one dose?

An apparent exposure-response relationship. . .

Yang et al. (2012) The Combination ofExposure-Response and Case-ControlAnalyses in Regulatory Decision Making

Suppose we want to . . .

Estimate exposure-responserelationship on the hazard ratiofor trastuzumab relative to control

Estimate the hazard ratio for Q1relative to control

Predict the effects of a higherdose in patients with lowerexposure

. . . in an unbiased (causal) manner.

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 17 / 24

Exposure-response with one dose?

What if there are covariates associated with exposureand outcome . . .

Yang et al. (2012) The Combination of Exposure-Response and Case-Control Analyses inRegulatory Decision Making

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 18 / 24

Exposure-response with one dose?

. . . and the distribution is different across the exposurerange?

Yang et al. (2012) The Combination of Exposure-Response and Case-Control Analyses inRegulatory Decision Making

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 19 / 24

Exposure-response with one dose?

Potential solutions to confounded E-R

Include potential confounders as covariates in the E-R modelTreat the analysis as if it were from an observational study (e.g.,using matching)

Categorize exposure (e.g., quantiles)Calculate propensity scores (for each exposure category relative toplacebo)Use for matched or weighted regression analyses

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 20 / 24

Concluding thoughts

Exposure-response modeling is a key ingredient indrug development

”Exposure-response information is at the heart of any determination of the safetyand effectiveness of drugs.” – FDA Guidance for Industry: Exposure-ResponseRelationships ? Study Design, Data Analysis, and Regulatory Applications

Uses an understanding of pharmacology of the drug (dose→ exposure)

May allow the use of models closely tied to the action of the drug (exposure→response)

Need to consider potential for confounding/effect modification (particularly if E-Rbased on one dose level)

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 21 / 24

Concluding thoughts

Many opportunities for interdisciplinary collaboration

Identifying development questions to address throughexposure-response modeling (at portfolio level as well as compoundlevel)

Design of studies to inform exposure-response modeling

Execution of M&S work

What can statisticians do to get more involved?

Be open to learning a new topic

Share what you do know in a collaborative manner (e.g., knowledge ofstudy population vs. knowledge of pharmacology)

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 22 / 24

Concluding thoughts

Back-up Slides

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 23 / 24

Concluding thoughts

E-R for what endpoints? (editorial aside)

E-R is most believable when you’re modeling endpoints that aremost closely related to the mechanism of action of the drug.

Concentration of drug→ biomarker (possibly with a delayed effect)

As we move farther away from MOA, K-PD or summary measuresof exposure may be reasonable

e.g., average concentration→ tumor growth inhibition

Even farther away, we may need to rely on somescientifically-based hand waving

e.g., Average concentration in Cycle 1→ survival

c©2016 Metrum Research Group LLC TICTS, Durham, NC, 2016 May 3, 2016 24 / 24