biomarker-based bayesian adaptive designs for targeted ... · adaptive designs for targeted agent...

Biomarker-Based Bayesian

Adaptive Designs for Targeted

Agent Development –

Implementation and Lessons

Learned from the BATTLE TrialJ. Jack Lee, Suyu Liu, Nan Chen

Department of Biostatistics

University of Texas

M. D. Anderson Cancer Center

PremiseMany new targeted agents and many more potential

combination therapies.

Unfortunately, success rate for oncology drug

development is very low.

Targeted agents do not work for all patients.

Are there markers to guide the choice of treatments?

How to treat patients best in the trial?

How to gauge the treatment effect?

– response rate, disease stabilization, improved survival

Limited patient population enrolled in clinical trials

Time is of the essence.

How to best identify markers, gauge treatment

efficacy, and treat patients in a clinical trial?

Phase I Phase II Phase III

Traditional Drug Development

I

I

II

I

II

II

II

III

III

One dose, Schedule

A few doses/schedules

Multiple doses/schedules

Traditional Drug Development

Traditional Drug Development

Biomarkers - Evolution of Knowledge in NSCLC

Pao and Girard, Lancet Oncology 2010

How to Test Treatment Effect, Marker Effect, and

their Combined Effect (Interaction)?

crizotinib

erlotinib

trastuzumab

FTI ?

PLX4032?

?

?

?

?

Tier 1 evidence: Assign pts to treatments – single arm confirmation trials?

Tier 2 evidence: Randomized trials?

Tier 3 evidence: Single arm screening trials? Randomized screening trials?

Randomized Phase II Trial for Testing 5-Gene

Signature of Paclitaxel Sensitivity in Breast Cancer

Ito et al., Cancer Science 2011

Est. Path RR: 30% 80%

Actual Path RR: 21% 36%

(N=11)(N=56) (N=19)

Lesson:

We think we know but often we really don’t –

Need clinical trials to confirm or refute the hypothesis.

Goals for Biomarker-Based Adaptive Designs

Identify prognostic and predictive markers for targeted agents– Prognostic markers

markers that associate with the disease outcome regardless of the treatment: e.g., stage, performance status

– Predictive markers

markers which predict differential treatment efficacy in different marker groups: e.g., In Marker (-), tx does not work

but in Marker (+), tx works

Test treatment efficacy– Control type I and II error rates

– Maximize study power for testing the effectiveness between treatments

– Group ethics

Provide better treatment to patients enrolled in the trial– Assign patients into the better treatment arms with higher probabilities

– Maximize total number of successes in the trial

– Individual ethics

BATTLE (Biomarker-based

Approaches of Targeted Therapy

for Lung Cancer Elimination)Patient Population: Stage IV recurrent non-small cell lung cancer (NSCLC)

Primary Endpoint: 8-week disease control rate (DCR)

4 Targeted treatments, 11 Biomarkers

200 evaluable patients

Goal:– Test treatment efficacy

– Test biomarker effect and their predictive roles to treatment

– Treat patients better in the trial based on their biomarkers 1. Zhou X, Liu S, Kim ES, Lee JJ. Bayesian adaptive design for targeted therapy development in lung cancer - A step toward personalized medicine (Clin Trials, 2008).

2. Kim ES, Herbst RS, Wistuba II, Lee JJ, et al, Hong WK. The BATTLE Trial: Personalizing Therapy for Lung Cancer. (Cancer Discovery, 2011)

Biomarker Groups by Molecular Pathway

• EGFR marker group• Mutation

• Gene copy number, high polysomy/ amplification

5 Marker Groups

Based on 2005 data

Hierarchial-based

1

• KRAS/BRAF marker group

• Mutations (KRAS and BRAF)2

• VEGF marker group

• VEGF expression

• VEGFR-2

3

• RXR/Cyclin D1 marker group

• RXR , ,

• Cyclin D1 exp/amp

4

Inadequate tissue/ no markers

present

5

EGFR

Mutation - Sequencing

Copy Number - FISH

High Polysomy Amplification

Deletion 746E-750A

CTG858CGG (L858R) CTG858 Wild-Type

Exon 21

Exon 19

KRAS/BRAF

KRAS Mutation - Sequencing

BRAF Mutation - Sequencing

TCG446TTG (S446L)TCG446 Wild-Type

GGT12TGT(G12C) GGT12CGT(G12R)

Exon 11

Codon 12

Biomarker analysis performed in

Thoracic Molecular Pathology Research Lab

VEGF/VEGFR-2

VEGF - IHC

VEGFR-2 - IHC

RXR/Cyclin D1

Cyclin D1 - IHC

CCND1 - FISH

RXRs - IHC

Amplification

RXR

RXR

RXR

Summary of Marker Group Assignment and

Treatment Randomization

• Randomized open-label phase II

• 250 patient enrollment

• Initial “equal” followed by “adaptive” randomization

• Novel clinical design

EGFR KRAS VEGFRXR/

CycD1None Total

Erlotinib

Vandetanib

Erlotinib+

Bexarotene

Sorafenib

Total

Tre

atm

en

ts

Marker Groups

5 marker groups

Marker Group assignment was based on:

1. Results of the individual biomarkers

2. Higher ranking biomarker took precedence

4 treatments

Treatment assignment was based on:

1. Equal randomization initially

2. Adaptive randomization based on marker group and

posterior probability

BATTLE Schema

Erlotinib SorafenibVandetanib Erlotinib + Bexarotene

Randomization:

Equal Adaptive

Primary end point: 8 week Disease Control (DC)

Umbrella Protocol

EGFR KRAS/BRAF

VEGF RXR/CyclinD1

Core BiopsyBiomarker

Profile

Bayesian Hierarchical Probit Model

1 if 0

0 otherwise

Pr( 1) Pr( 0)

ijk

ijk

jk ijk ijk

zy

y z

2

2

~ ( , 1), for 1,...,

~ ( , ), for 1,...,5

~ (0, ), for 1,...,4

ijk jk jk

jk j

j

z N i n

N k

N j

• Notation

-- ith : subject, i=1 , ..., njk

-- jth : treatment arm, j=1 , …, 4

-- kth : marker group, k=1 , …, 5

-- yijk: 8-week progression-free survival status: 0(no) vs 1(yes)

-- zijk : latent variable

-- jk : location parameter

-- j : hyper-prior on jk

-- jk : disease control rate (DCR)

-- 2,2 : hyper-parameters control borrowing across MGs

within and between treatments

• Probit model with hyper prior (Albert et al, 1993)

ER is applied in the first stage for model developmentAR will be applied after enrolling at least one patient in each (Treatment x MG) subgroup.Adaptively assign the next patient into the treatment arms proportional to the marginal posterior disease control rates.

set a minimum RR to 10% to ensuring a reasonable probability of randomizing pts in each arm

Suspend randomization of a treatment in a biomarker group if – Probability(DCR > 0.5 | Data) < 10%

Declare a treatment is effective in a biomarker group if – Probability(DCR > 0.3 | Data) > 80%

ˆ ˆ/ ( )jk wk

w

Equal Randomization (ER) Followed By

Adaptive Randomization (AR)

Adaptive Randomization (AR) vs.

Equal Randomization (ER)Consider two treatments, binary outcome

First n pts equally randomized (ER) into TX1 and TX2

After ER phase, the next patient will be assigned to

TX1 with probability , where

1 1 2 2

1 1 2 2 2 1

ˆ ˆ, or

Pr( ), Pr( )

B p B p

B p p B p p

1 1 2/( )B B B

Note that the tuning parameter

– = 0, ER

– = , “play the winner”

Continue the study until reaching early stopping

criteria or maximum N

1 1 2ˆ ˆ ˆ/( )p p pAR rate to TX 1=

Demo 1

Adaptive vs. Fixed Randomization

Korn and Freidlin (JCO 2011)

– Compare AR versus FR in two-arm trials using simulations.

– AR has no benefits over FR if accrual rate is the same.

– If using AR can increase the accrual rate, use a fixed 1:2

randomization in favor of the experimental arm.

Berry (JCO 2011)

– I agree with their calculations, but not with their conclusion.

– Greatest potential for adaptive randomization is in multi-

armed Trials.

– Adaptive designs promise shorter cancer drug development

and better identification of responding patient populations.

– Adaptive trials have disadvantages that must be considered

along with the potential advantages.

Number of Non-RespondersTrue

Response

Rate

FR 1:1

[N=132]

FR 1:2

[N=153]

AR

[N=140]

Cntl Exp

0.2 0.05 115.5 137.7 118.7

0.2 0.1 112.2 132.6 117.3

0.2 0.2 105.6 122.4 112.0

0.2 0.3 99.0 112.2 103.5

0.2 0.4 92.4 102.0 92.9

0.2 0.5 85.8 91.8 81.1

0.2 0.6 79.2 81.6 68.9

0.2 0.7 72.6 71.4 56.5

0.2 0.8 66.0 61.2 44.0

0.2 0.9 59.4 61.0 31.4

AR: Randomization probability bounded at 0.9.

Percent of RespondersTrue

Response

Rate

FR 1:1

[N=140]

AR

[N=140]

FR 1:1

[N=153]

FR 1:2

[N=153]

AR

[N=153]

Cntl Exp

0.2 0.05 12.9 15.2 13.5 10.0 15.6

0.2 0.1 15.3 16.2 15.7 13.3 16.5

0.2 0.2 20.0 20.0 20.0 20.0 20.0

0.2 0.3 25.0 26.0 25.0 26.6 25.9

0.2 0.4 30.5 33.7 31.1 33.3 34.1

0.2 0.5 35.8 42.1 37.0 40.0 42.7

0.2 0.6 41.1 50.8 42.7 46.6 51.6

0.2 0.7 46.4 59.6 48.4 53.3 60.5

0.2 0.8 51.7 68.6 54.1 60.0 69.6

0.2 0.9 57.0 77.2 59.8 66.6 78.3

AR: Randomization probability bounded at 0.9.

Three-Arm Trials: FR versus ARTrue Response

Rate

Fixed Ratio Randomization

(1:1:1)

Adaptive Randomization

[Bound 0.9]

CntlExp

1

Exp

2

Pr(resp)

%

[N=171]

# of Non-

Responders

[N=171]

Rejection

Rate

[N=171]

Pr(resp)

%

[N=183]

Pr(resp)

%

[N=183]

# of Non-

Responders

[N=183]

% on

Control

[N=183]

% on

Exp1

[N=183]

% on

Exp2

[N=183]

Rejection

Rate

[N=183]

0.2 0.05 0.05 10.0 153.9 0 10.7 11.6 161.8 43.7 28.1 28.2 0

0.2 0.1 0.1 13.3 148.2 0 13.8 14.0 157.4 39.6 30.2 30.2 0

0.2 0.2 0.2 20.0 136.8 0.10 20.0 20.0 146.4 33.3 33.3 33.3 0.10

0.2 0.3 0.3 26.7 125.4 0.52 26.6 27.2 133.2 27.8 36.1 36.1 0.53

0.2 0.4 0.4 33.3 114.0 0.90 33.6 35.4 118.3 23.2 38.4 38.4 0.90

0.2 0.5 0.5 40.0 102.6 0.99 40.6 44.0 102.5 19.9 40.0 40.0 0.99

0.2 0.6 0.6 46.7 91.2 1.00 47.5 52.9 86.3 18.0 41.0 41.0 0.99

0.2 0.7 0.7 53.3 79.8 1.00 54.4 61.7 70.0 16.5 41.7 41.7 1.00

0.2 0.8 0.8 60.0 68.4 1.00 61.3 70.7 53.6 15.5 42.2 42.2 1.00

0.2 0.9 0.9 66.7 57.0 1.00 68.2 79.7 37.1 14.7 42.6 42.6 1.00

0.2 0.3 0.5 33.3 114.0 0.97 34.4 36.3 116.5 23.8 32.7 43.5 0.98

0.2 0.4 0.6 40.0 102.6 0.99 41.3 44.7 101.1 20.3 35.7 44.0 0.99

0.2 0.4 0.8 46.7 91.2 1.00 48.9 56.0 80.6 18.5 32.3 49.2 1.00

0.2 0.1 0.6 30.0 119.7 1.00 32.0 38.6 112.2 28.3 20.0 51.7 1.00

0.2 0.2 0.4 26.7 125.4 0.78 27.3 28.5 130.8 28.7 28.7 42.6 0.83

0.2 0.2 0.6 33.3 114.0 0.99 35.1 39.9 110.0 25.2 25.2 49.6 1.00

0.2 0.2 0.8 40.0 102.6 1.00 42.6 53.7 84.8 21.9 21.9 56.2 1.00

Allocation % on Exp. Arm vs. # of

Patients Accrued, No Early Stopping

0 50 100 150

2040

6080

100

Number of Patients Accrued

Allo

catio

n R

ate

on E

xp.

Tre

atm

ent

(%)

AR

ER

P1 = 0.2

P2 = 0.4

Allocation % on Exp. Arm vs. # of

Patients Accrued, With Early Stopping

0 50 100 150 200 250

2040

6080

100

Number of Patients Accrued

Allo

catio

n R

ate

on E

xp.

Tre

atm

ent

(%)

AR

ER

P1=0.2

P2=0.4

Simulation – Scenario 1

One effective treatment for MG 1-4, no effective treatment

for MG 5, adaptive randomization (AR), with vague

prior

MG 1 MG 2 MG 3 MG 4 MG 5

TX 1 0.8 0.3 0.3 0.3 0.3

TX 2 0.3 0.6 0.3 0.3 0.3

TX 3 0.3 0.3 0.6 0.3 0.3

TX 4 0.3 0.3 0.3 0.6 0.3

Simulation Results, Scenario 1

(with early stopping)

Conventional Design

Simon’s optimal two-stage design in each of the 4 x 5 = 20 trt x MG combinations

H0: p p0 vs. H1: p p1

p0 = 0.3, p1 = 0.5, = 0.20, 1- = 0.80

n1 = 6, r1 = 1, n = 20, r = 7,

Total N up to 20 x 20 = 400

MG 1 MG 2 MG 3 MG 4 MG 5

Erlotinib

Sorafenib

Vandetanib

Erlotinib +

Bexarotene

Registration

Biomarker

Information

(Determine

MarkerGroup)

Evaluate at

8 Week

And determine

Response

Adaptive

Randomization

Assign

Treatment

Clinical Visits

Patient

Follow Up

Visits

Off Study

InformationBiopsy

Two Weeks Eight Weeks

Schematic Diagram to run the web based

“BATTLE” application

Report

s

Medical History

Physical Exams

Lab Tests

Diagnostic

Procedures

Drug

Compliance

Concomitant

Medications

Adverse

Events

Sample

Collection

Tumor

Measurements

Adaptive

Randomization

(R Code, Web Svc)

Web

Inte

rface

SQL 2005 DatabaseCORe

Menu

Randomization ProcessPatient Consented and

Registered in Database

Information sent to

Surgical Team for biopsy

Biopsy sent to Thoracic

Molecular Path Lab

Randomization of

Patient to Trial

Biomarker results entered

into database

Research Nurse Notified

Automatically

Patient Consented to

Appropriate Trial

Randomize

BATTLE Timeline and Accrual

• Patients Enrolled 341

• Randomized 255

• Evaluable 244

Late 2004: Grant Planned

Mid 2005: Grant Submitted

April 2006: Grant Approved

Nov 2006: Trials Activated- 1st pt

Oct 2009: Trials completed accrual!

2004 2005 2006 2007 2008 20092007 2009

Accrual

0

50

100

150

200

250

300

350

1 4 7 10 13 16 19 22 25 28 31 34

Est. reg

Est. rand

Cum. reg

Cum. rand

Study Accrual and Randomization

Time (months)

Pati

en

ts E

nro

lled

341

255

BATTLE Accrual Summary

Trial activation: November 2006

Last patient enrolled: October 2009

Total patients registered 341

Total patients randomized 255

Total patients with 8-wk DC status 244

Patients equally randomized 95 (39%)

Patients adaptively randomized 149 (61%)

Patients with complete markers 203 (83%)

Patients w/o complete markers 41 (17%)

Registration and Randomization of

Patients

Total RegisteredN = 341

Total Randomized (N = 255)Equal Randomization (N = 97, 38%)

Adaptive Randomization (N = 158, 62%)

ErlotinibEqual (N = 25)

Adaptive (N = 33)

VandetanibEqual (N = 23)

Adaptive (N = 29)

Erlotinib+ BexaroteneEqual (N = 21)

Adaptive (N = 15)

SorafenibEqual (N = 26)

Adaptive (N = 72)

Reasons not randomized (N = 86)-Concurrent illness/Poor PS (N = 51)

-Not biopsied/other (N = 21)

-Alternate or No Tx (N = 14)

Demo 2

37

BATTLE Results: Disease Control in % (n)

EGFR KRAS VEGFRXR/

CycD1None Total

Erlotinib 35% (17) 14% (7) 40% (25) 0% (1) 38% (8) 34% (58)

Vandetanib 41% (27) 0% (3) 38% (16) NA (0) 0% (6) 33% (52)

Erlotinib +

Bexarotene55% (20) 33% (3) 0% (3) 100% (1) 56% (9) 50% (36)

Sorafenib 39% (23) 79% (14) 64% (39) 25% (4) 61% (18) 58% (98)

Total 43% (87) 48% (27) 49% (83) 33% (6) 46% (41) 46% (244)

Tre

atm

en

ts

Marker Groups

AACR Presentation: http://app2.capitalreach.com/esp1204/servlet/tc?cn=aacr&c=10165&s=20435&e=12587&&m=1&br=80&audio=false

Posterior Probabilities for DCRs

BATTLE Results: Disease Control %

[Probability(DCR > 30%) > 0.8]

EGFR KRAS VEGFRXR/

CycD1None Total

Erlotinib35%

[0.66]

14%

[0.14]

40%

[0.85]0%

38%

[0.67]34%

Vandetanib41%

[0.92]

0%

[0]

38%

[0.75]NA

0%

[0]33%

Erlotinib +

Bexarotene

55%

[0.99]

33%

[0.54]

0%

[0]100%

56%

[0.92]50%

Sorafenib39%

[0.76]

79%

[1.00]

64%

[1.00]25%

61%

[1.00]58%

Total 43% 48% 49% 33% 46% 46%

Tre

atm

ents

Marker Groups

EGFR and KRAS Marker Groups

EGFR Group

35%39%41%

0

10

80

20

40

30

60

70

50

Erlotinib

Vandetanib

Erlotinib + Bexarotene

Sorafenib

90

55%

Achie

ved 8

week D

C (

%)

100

Primary Endpoint: Overall DCR at 8 weeks was 46%

N =17 27 20 23

KRAS Group

14%

79%

0

33%

7 3 3 140

10

80

20

40

30

60

70

50

90

100

N =

EGFR Group- mutation

- copy number

KRAS Group- KRAS mutation

- BRAF mutation

EGFR and KRAS Mutations:

Novel Discovery Findings

KRAS Mutation

0

10

80

20

40

30

60

70

50

90

Achie

ved 8

week D

C (

%)

100

22%

37%

56%

61%

+ - + - + - + -

EGFR Mutation

+ -+ -+ -+ -

23%

64%

0

10

80

20

40

30

60

70

50

90

100

Erlotinib Vandetanib Erlotinib + Bexarotene Sorafenib

71%

29%

Erlotinib Sorafenib Erlotinib Sorafenib

N = 7 45 13 67 N = 9 43 18 62

Individual Biomarkers for Response

and Resistance to Targeted Treatment:

Exploratory Analysis

Drug Treatment Biomarker P–value DC

Erlotinib EGFR mutation 0.04 Improved

Vandetanib High VEGFR-2 expression 0.05 Improved

Erlotinib +

BexaroteneHigh Cyclin D1 expression 0.001 Improved

EGFR FISH Amp 0.006 Improved

Sorafenib EGFR mutation 0.012 Worse

EGFR high polysomy 0.048 Worse

Did AR work?

Prob(RAND)

AV

E(a

rm=

=1

)

0.0 0.2 0.4 0.6 0.8

0.0

0.2

0.4

0.6

0.8

1.0

Randomized to Arm 1

Prob(RAND)

AV

E(a

rm=

=2

)

0.0 0.2 0.4 0.6 0.8

0.0

0.2

0.4

0.6

0.8

1.0

Randomized to Arm 2

Prob(RAND)

AV

E(a

rm=

=3

)

0.0 0.2 0.4 0.6 0.8

0.0

0.2

0.4

0.6

0.8

1.0

Randomized to Arm 3

Prob(RAND)

AV

E(a

rm=

=4

)

0.0 0.2 0.4 0.6 0.8

0.0

0.2

0.4

0.6

0.8

1.0

Randomized to Arm 4

What are the DCRs in ER vs. AR?

DCR

– in ER = 52% (N=95)

– in AR = 42% (N=149)

What a bummer !?

Why?

Why didn’t AR work better? Time drift. ER pts tends to have more– never smokers (28% vs 18%)

– PS=0 pts (13% vs 6%)

– Erlotinib naïve (58% vs 53%)

– Female (51% vs 44%)

Percent of patients eligible for trials– Only 1 trial: 21%

– Two trials: 35%

– Three trials: 31%

– All four trials: 14%

PI override (N=3, all progressed)

Actual outcomes changed by the final endpoint review committee (N=14)

AR started too late (~40%, ER) and not aggressive enough.

46

Equal versus Adaptive Randomization –

Population Drift

Trial Enrollment

Pro

port

ion o

f D

isease C

ontr

ol (D

C)

Oct

2006

Nov

2009

Equal Randomization

(N = 95)

DC = 52%

Adaptive

Randomization

(N = 149)

DC = 42%

If equal

randomization had

continued

DC = 37%

5%

Lessons Learned from BATTLE?Biomarker-based adaptive design is doable! It is well received by clinicians and patients.

Prospective tissues collection & biomarkers analysis provide a wealth of information

Treatment effect & predictive markers are efficiently assessed.

Pre-selecting markers is not a good idea. We don’t know what are the best predictive markers at get-go.

Bundling markers into groups, although can reduce the total number of marker patterns, is not the best way to use the marker information either.

Watch for time drift and avoid it.

AR should kicks in early & closely monitored.

AR works well only when we have good drugs and good predictive markers.

Discovery Platform versus

Confirmatory PlatformEarly phase of drug developing is about discovery and learning.

Adaptive design provides an ideal platform for learning.

Due to the large number of tests, the overall false positive rate may be large.

Results found in the discovery platform need to be validated in the confirmatory platform– Validation of treatment efficacy

– Validation of predictive markers

After narrowing down the biomarkers and treatments combination(s), validation trials can be more focused.– Traditional designs or innovative designs

BATTLE-2 SchemaProtocol enrollment

Biopsy performed

Stage 1:

Adaptive RandomizationKRAS mutation

Primary endpoint: 8-week disease control

N = 400

Erlotinib Erlotinib+AKTi MEKi+AKTi

Stage 2:

Refined Adaptive Randomization“Best” discovery markers/signatures

Principles• Better specific drugs

• Better specific targets

• No biomarker grouping

• Selection, integration and validation of novel predictive biomarkers

Sorafenib

Tools for Conducting

Bayesian / Adaptive Trials at MDAClinical Trial Conduct (CTC) Website

Secured web application for conducting

Bayesian clinical trials

Can be used to

– Register patients

– Log in key information for randomization

Baseline characteristics

Outcome (toxicity, efficacy)

– Randomize patients

Connect to statistical software via web services

– Capture endpoints for interim analysis

New Trial Request Form

Trial Information and

Administration

Design Methods

Patients Information Input

Pocock-Simon Design

Outcome based Adaptive

Randomization Trial

Monitoring Efficacy and

Toxicity

Multi-Center Supported

Clinical Trial Conduct (CTC) Website

(as of August 2011)

Adaptive Randomization 44

Pocock-Simon Design 42

CRM 29

One Arm Time-To-Event Monitoring 11

Equal Randomization 6

Efftox 1

Total trials 133

Total # of patients ~4,360

Software Toolshttps://biostatistics.mdanderson.org/SoftwareDownload/

Over 70 programs freely available

Adaptive Designs

for

Personalized Medicine

Are we there?

Are Bayesian adaptive designs useful for

the development of personalized medicine?

Absolutely!

Are Bayesian adaptive designs ready for

the prime time?

Getting there. Need more work on education,

software development, and implementation.

It is an exciting time for advancement in

medicine and statistics.

Adapt!Adapt! Adapt!

Summary

Adaptive designs enable us to continue to learn about the new agents’ activities and identify the predictive markers during the trial in order to apply this knowledge to better treat patients in real time.

It can increase the study efficiency, allow flexibility in study conduct, and provide better treatment to study participants. – Speed up drug development– A step towards personalized medicine

Extra steps need to be taken to ensure the integrity of study conduct, e.g., objective and timely evaluation of endpoints, monitoring AR continuously.

Roll up your sleeves! The proof of the pudding is in the eating. We need to do more such innovative trials to learn and to improve the trial designs and conduct so we can turn promise into progress.