Eugm 2012 mehta - future plans for east - 2012 eugm

Future Plans for EastFuture�Plans�for�East

Cyrus�Mehta

President, Cytel IncPresident,�Cytel�Inc

Where�are�we�going�with�East?

Four New InitiativesFour�New�Initiatives• Enhance�the�simulations�capabilities�by�permitting�external�calls�to�R�and�SAS

• Conditional�simulation�of�the�remainder�of�the�trial�given�the�interim�data

• MultiͲarm�group�sequential�designsg p q g

• Population�enrichment�designs

External�calls�to�R�for�Adaptive�Decisions

• East already simulates patient arrivals, patientEast�already�simulates�patient�arrivals,�patient�responses�and�and�dropͲouts

• Adaptive decision rules are currently based onAdaptive�decision�rules�are�currently�based�on�conditional�power

• Permit user specified decision rules:Permit�user�specified�decision�rules:– Bayesian�criteria�for�SSR–More�general�functions�for�SSRg

Implement�through�calls�to�R�or�SAS�at�each�interim�analysis�for�each�simulated�trialy

Different�Functions�for�SSR�(J�&�T,�2012)

Conditional�Simulation�of�Remainder�of�Trial

• At an interim analysis both DMC and SponsorAt�an�interim�analysis�both�DMC�and�Sponsor�are�very�interested�in�the�question:“What is the chance that this trial will succeed”What�is�the�chance�that�this�trial�will�succeed

• Conditional�power�has�some�limitations

i l di l f h ld h– no�visual�display�of�what�could�happen– does�not�give�a�sense�of�variability– does�not�estimate�final�treatment�effect

Predictive�Interval�Plots�I.�(Li,�Evans,�Hajime and Wei, 2009)Hajime�and�Wei,�2009)

Predictive�Interval�Plots�II.�(Li,�Evans,�Hajime�and Wei, 2009)and�Wei,�2009)

MultiͲArm�Group�Sequential�Designs

• Compare�D�dose�groups�to�a�common�control�Co pa e dose g oups to a co o co t owithout�doseͲresponse�assumptions

• Standard�Approachs�include:pp

– SingleͲstage�design�with�closed�testing (MCP�in�East)– TwoͲstage�adaptive�design�with�early�stopping�or�

l i 1 bi i i htreatment�selection�at�stage�1,�combining�stages�with�preͲspecified�weights,�and�closed�testing�(Posch�et�al)

• Limitations:Limitations:�– correlation�between�test�statistics�not�exploited– Not�easy�to�generalize�to�multiple�stagesy g p g

Generalization�of�TwoͲArm�GSD

• Monitor the D test statistics over K looks:Monitor�the�D�test�statistics�over�K�looks:– C1,�C2,�…�CK are�the�K�efficacy�boundariesStop and claim efficacy if one of the test statistics– Stop�and�claim�efficacy�if�one�of�the�test�statistics�crosses�an�efficacy�boundary

– Incorporate nonͲbinding futility boundaries– Incorporate�nonͲbinding�futility�boundaries• Find�C1,�C2,�…�CK such�that�the�probability�that�the max of D multivariate normal statisticsthe�max.�of�D�multivariate�normal�statistics�exceeds�one�of�the�Cj’s�is�D

Example:�Four�Arms�and�Three�Stages

Look Cumulative Stopping Boundaries

Stopping Boundaries of J(-4) Spending Function at D=0.05

Look CumulativeSample�Size/Arm

Stopping�Boundaries

4ͲArm�Design 2ͲArm�Design

1 25 3.39 3.011

2 50 2.89 2.547

3 75 2.77 1.999

Alternative�Hypothesis Power

0.4,�0.4,�0.4 67.6%

0,�0.4,�0.4 58.7%, ,

0,�0.2,�0.4 44.6%

0,�0,�0.4 42.3%

Power of corresponding two-armdesign with 75 patients/arm is 68%

Population�Enrichment�Designs�for�Oncology

• Failure rate for late stage oncology trials is almostFailure�rate�for�late�stage�oncology�trials�is�almost�60%�(Kola�and�Landis,�2004)

• Two�recent�scientific�developments�can�improve�this�p p

track�record– development�of�molecularly�targeted�agentsp y g g

– statistical�methodology�of�adaptive�trial�design�applied�to�timeͲtoͲevent�data

• Fact:�Some�subgroups�benefit�differentially�from�others�when�treated�with�the�targeted�agent

Oncology�Products�Approved�in�the�USA�for�Selected�Patient�Population

Compound/Target Indication (prevalence target)Compound/Target Indication�(prevalence�target)

Crizotinib (Xalkori®)/�ALKͲrearrangement

•NonͲsmall�cell�lung�cancer�with�ALKͲrearrangements�(5%)

Vem rafenib (Zelboraf®)/ BRAFVemurafenib (Zelboraf®)/ BRAF�mutation •Advanced melanoma�with�mutant�BRAF�(30Ͳ40%)

Trastuzumab (Herceptin®);�Lapatinib (Tykerb®/�Her2

•Her2�expressing�breast�cancer�(25%)•Her2 expressing metastatic gastric cancer (20 30%)p ( y / •Her2 expressing�metastatic�gastric�cancer�(20Ͳ30%)

Aromatase inhibitors (letrozole,�exemestane) •ER(+) breast�cancer�(60Ͳ70%)

Rit i b (Rit ®)/ CD20 ( ) ll l h ( )Rituximab (Rituxan®)/�CD20 •CD20(+)�BͲcell lymphomas (90%+)

Cetuximab (Erbitux®);�Panitumumab (Vectibix®)�/�EGFR

•Advanced Head/neck�cancer�(~100%)•EGFR(+)�metastatic�colorectal�cancer�(60Ͳ80%)

WT l l ( )EGFR •KRASWT metastatic�colorectal�cancer�(60%)

DIA�Adaptive�Design�Scientific�Working�Group�

Proposed�TwoͲStage�Adaptive�Design

Statistical�Method

• and are null hypotheses for the subͲH Hand�������are�null�hypotheses�for�the�subpopulation�and�the�full�population,�respectively

• Specify a two stage combination test for

SH FH

H H• Specify�a�twoͲstage�combination�test�for�����,������and�����������������D id h i i l i h h

F SH H�SH FH

• Decide�at�the�interim�analysis�whether�to�proceed�with�the�full�population�or�the�subͲ

l ipopulation

• Perform�a�test�of�both�����������������and�������if�you�F SH H� SHenrich

Software�Challenges�for�Event�Driven�Trials

• Provide simulation tools to generate patientProvide�simulation�tools�to�generate�patient�arrivals�with�different�arrival�rates,�dropout�rates and hazard rates for the two subgroupsrates�and�hazard�rates�for�the�two�subgroups

• Provide�ability�for�user�to�specify�flexible�decision rules for population enrichmentdecision�rules�for�population�enrichment

• Provide�visulation�tools�for�determining�when�f h i i l i ( li dto�perform�the�interim�analysis�(generalized�eͲ

charts)

Generalize�the�eͲchart�to�handle�subͲpopulationssub populations

Population�enrichment�designs�can�havea very large payͲoffa�very�large�pay off�

• Design Parameters:Design�Parameters:

– 4.5�months�median�PFS�on�control�armSuppose HR = 0 55 for Bio+ and HR=1 for Bio– Suppose�HR�=�0.55�for�Bio+�and�HR=1�for�BioͲ

– 40%�of�population�is�Bio+.�Thus�overall�HR�=�0.8C id t i l ith 370 ti t 300 t• Consider�a�trial�with�370�patients;�300�events:– NonͲadaptive�design�has�50%�overall�power– Adaptive�design:�has�78%�overall�power�and�94%�power�conditional�on�enriching�

Thank�you�for�participating

• Lots of good discussion• Lots�of�good�discussion•Many ideas for new software•Many�ideas�for�new�software�• Cytel is well on its way.Cytel�is�well�on�its�way.�Looking�forward�to�the�next�25�years�of�growth

Thank�you!�Questions?�

[email protected]