Use of Prognostic & Use of Prognostic & Predictive Genomic Predictive Genomic

Biomarkers in Clinical Biomarkers in Clinical Trial DesignTrial Design

Richard Simon, D.Sc.Richard Simon, D.Sc.Chief, Biometric Research BranchChief, Biometric Research Branch

National Cancer InstituteNational Cancer Institutehttp://brb.nci.nih.govhttp://brb.nci.nih.gov

BRB WebsiteBRB Websitebrb.nci.nih.govbrb.nci.nih.gov

Powerpoint presentationsPowerpoint presentations ReprintsReprints BRB-ArrayTools softwareBRB-ArrayTools software

Data archiveData archive Q/A message boardQ/A message board

Web based Sample Size Planning Web based Sample Size Planning Clinical TrialsClinical Trials

Optimal 2-stage phase II designsOptimal 2-stage phase II designs Phase III designs using predictive biomarkersPhase III designs using predictive biomarkers Phase II/III designsPhase II/III designs

Development of gene expression based predictive Development of gene expression based predictive classifiersclassifiers

Prognostic & Predictive Prognostic & Predictive BiomarkersBiomarkers

Most cancer treatments benefit only a minority Most cancer treatments benefit only a minority of patients to whom they are administeredof patients to whom they are administered

Being able to predict which patients are likely Being able to predict which patients are likely to benefit would to benefit would Save patients from unnecessary toxicity, and Save patients from unnecessary toxicity, and

enhance their chance of receiving a drug that helps enhance their chance of receiving a drug that helps themthem

Control medical costs Control medical costs Improve the success rate of clinical drug Improve the success rate of clinical drug

developmentdevelopment

Different Kinds of Different Kinds of BiomarkersBiomarkers

EndpointEndpoint Measured before, during and after treatment Measured before, during and after treatment

to monitor treatment effectto monitor treatment effect PharmacodynamicPharmacodynamic IntermediateIntermediate

Phase IIPhase II Futility analysis in phase IIIFutility analysis in phase III Patient managementPatient management

Surrogate for clinical outcomeSurrogate for clinical outcome

Surrogate EndpointsSurrogate Endpoints

It is extremely difficult to properly It is extremely difficult to properly validate a biomarker as a surrogate validate a biomarker as a surrogate for clinical outcome. It requires a for clinical outcome. It requires a series of randomized trials with both series of randomized trials with both the candidate biomarker and clinical the candidate biomarker and clinical outcome measuredoutcome measured

Intermediate Endpoints in Intermediate Endpoints in Phase I and II TrialsPhase I and II Trials

Biomarkers used as endpoints in phase I Biomarkers used as endpoints in phase I or phase II studies need not be validated or phase II studies need not be validated surrogates of clinical outcomesurrogates of clinical outcome

The purposes of phase I and phase II The purposes of phase I and phase II trials are to determine whether to trials are to determine whether to perform a phase III trial, and if so, with perform a phase III trial, and if so, with what dose, schedule, regimen and on what dose, schedule, regimen and on what population of patientswhat population of patients Claims of treatment effectiveness should be Claims of treatment effectiveness should be

based on phase III resultsbased on phase III results

Different Kinds of Different Kinds of BiomarkersBiomarkers

Predictive biomarkersPredictive biomarkers Measured before treatment to identify Measured before treatment to identify

who will or will not benefit from a who will or will not benefit from a particular treatmentparticular treatment

Prognostic biomarkersPrognostic biomarkers Measured before treatment to indicate Measured before treatment to indicate

long-term outcome for patients untreated long-term outcome for patients untreated or receiving standard treatmentor receiving standard treatment

Prognostic and Predictive Prognostic and Predictive Biomarkers in OncologyBiomarkers in Oncology

Single gene or protein measurementSingle gene or protein measurement Expression of drug targetExpression of drug target Activation of pathwayActivation of pathway

Scalar index or classifier that Scalar index or classifier that summarizes expression levels of summarizes expression levels of multiple genesmultiple genes

Disease classificationDisease classification

Types of Validation for Types of Validation for Prognostic and Predictive Prognostic and Predictive

BiomarkersBiomarkers Analytical validationAnalytical validation

Accuracy, reproducibility, robustnessAccuracy, reproducibility, robustness Clinical validationClinical validation

Does the biomarker predict a clinical Does the biomarker predict a clinical endpoint or phenotypeendpoint or phenotype

Clinical utilityClinical utility Does use of the biomarker result in Does use of the biomarker result in

patient benefitpatient benefit By informing treatment decisionsBy informing treatment decisions Is it actionableIs it actionable

Pusztai et al. The Oncologist 8:252-8, Pusztai et al. The Oncologist 8:252-8, 20032003

939 articles on “prognostic markers” 939 articles on “prognostic markers” or “prognostic factors” in breast or “prognostic factors” in breast cancer in past 20 yearscancer in past 20 years

ASCO guidelines only recommend ASCO guidelines only recommend routine testing for ER, PR and HER-routine testing for ER, PR and HER-2 in breast cancer2 in breast cancer

Most prognostic markers or prognostic models are Most prognostic markers or prognostic models are not used because although they correlate with a not used because although they correlate with a clinical endpoint, they do not facilitate therapeutic clinical endpoint, they do not facilitate therapeutic decision making; i.e. they have no demonstrated decision making; i.e. they have no demonstrated medical utilitymedical utility

Most prognostic marker studies are based on a Most prognostic marker studies are based on a “convenience sample” of heterogeneous patients, “convenience sample” of heterogeneous patients, often not limited by stage or treatment. often not limited by stage or treatment.

The studies are not planned or analyzed with clear The studies are not planned or analyzed with clear focus on an intended use of the markerfocus on an intended use of the marker

Retrospective studies of prognostic markers should Retrospective studies of prognostic markers should be planned and analyzed with specific focus on be planned and analyzed with specific focus on intended use of the markerintended use of the marker

Design of prospective studies depends on context of Design of prospective studies depends on context of use of the biomarkeruse of the biomarker Treatment options and practice guidelinesTreatment options and practice guidelines Other prognostic factorsOther prognostic factors

OncotypeDx as a Model for OncotypeDx as a Model for Development of a Development of a

Therapeutically Relevant Gene Therapeutically Relevant Gene Expression SignatureExpression Signature

<10% of node negative ER+ breast <10% of node negative ER+ breast cancer patients require or benefit from cancer patients require or benefit from the cytotoxic chemotherapy that they the cytotoxic chemotherapy that they receivereceive

Identify patients with node negative Identify patients with node negative ER+ breast cancer who have low risk ER+ breast cancer who have low risk of recurrence on tamoxifen aloneof recurrence on tamoxifen alone

B-14 Results—Relapse-Free B-14 Results—Relapse-Free SurvivalSurvival

338 pts

149 pts

181 pts

0 2 4 6 8 10 12 14 16

Time (yrs)

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Relap

se-Free S

urvival

Low R isk (R S < 18) Intermediate R isk (R S 18 - 30) H igh R isk (RS 31)

p<0.0001

Paik et al, SABCS 2003

Key Features of OncotypeDx Key Features of OncotypeDx DevelopmentDevelopment

Focus on important therapeutic Focus on important therapeutic decision contextdecision context

Staged development and validationStaged development and validation Separation of data used for test Separation of data used for test

development from data used for test development from data used for test validationvalidation

Development of robust analytically Development of robust analytically validated assayvalidated assay

Potential Uses of a Potential Uses of a Prognostic BiomarkerPrognostic Biomarker

Identify patients who have very good Identify patients who have very good prognosis on standard treatment and prognosis on standard treatment and do not require more intensive do not require more intensive regimens regimens

Identify patients who have poor Identify patients who have poor prognosis on standard chemotherapy prognosis on standard chemotherapy who are good candidates for who are good candidates for experimental regimensexperimental regimens

Predictive BiomarkersPredictive Biomarkers

Predictive BiomarkersPredictive Biomarkers

In the past often studied as In the past often studied as exploratory post-hoc subset analyses exploratory post-hoc subset analyses of RCTs.of RCTs. Numerous subsets examinedNumerous subsets examined No pre-specified hypothesesNo pre-specified hypotheses No control of type I errorNo control of type I error

Led to conventional wisdomLed to conventional wisdom Only hypothesis generationOnly hypothesis generation Only valid if overall treatment Only valid if overall treatment

difference is significant difference is significant

Prospective Co-Prospective Co-Development of Drugs and Development of Drugs and

Companion DiagnosticsCompanion Diagnostics1.1. Develop a completely specified Develop a completely specified

genomic classifier of the patients genomic classifier of the patients likely to benefit from a new druglikely to benefit from a new drug

2.2. Establish analytical validity of the Establish analytical validity of the classifierclassifier

3.3. Use the completely specified Use the completely specified classifier in the primary analysis classifier in the primary analysis plan of a phase III trial of the new plan of a phase III trial of the new drugdrug

Guiding PrincipleGuiding Principle

The data used to develop the classifier The data used to develop the classifier should be distinct from the data used should be distinct from the data used to test hypotheses about treatment to test hypotheses about treatment effect in subsets determined by the effect in subsets determined by the classifierclassifier Developmental studies can be exploratoryDevelopmental studies can be exploratory Studies on which treatment effectiveness Studies on which treatment effectiveness

claims are to be based should not be claims are to be based should not be exploratoryexploratory

Using phase II data, develop predictor of response to new drugDevelop Predictor of Response to New Drug

Patient Predicted Responsive

New Drug Control

Patient Predicted Non-Responsive

Off Study

BRB-ArrayToolsBRB-ArrayTools Architect – R SimonArchitect – R Simon Developer – Emmes CorporationDeveloper – Emmes Corporation

Contains wide range of analysis tools that I have Contains wide range of analysis tools that I have selectedselected

Designed for use by biomedical scientistsDesigned for use by biomedical scientists Imports data from all gene expression and copy-Imports data from all gene expression and copy-

number platformsnumber platforms Automated import of data from NCBI Gene Express Automated import of data from NCBI Gene Express

OmnibusOmnibus Highly computationally efficientHighly computationally efficient Extensive annotations for identified genesExtensive annotations for identified genes Integrated analysis of expression data, copy number Integrated analysis of expression data, copy number

data, pathway data and data other biological datadata, pathway data and data other biological data

Predictive Classifiers in Predictive Classifiers in BRB-ArrayToolsBRB-ArrayTools

ClassifiersClassifiers Diagonal linear discriminantDiagonal linear discriminant Compound covariate Compound covariate Bayesian compound covariateBayesian compound covariate Support vector machine with Support vector machine with

inner product kernelinner product kernel K-nearest neighborK-nearest neighbor Nearest centroidNearest centroid Shrunken centroid (PAM)Shrunken centroid (PAM) Random forrestRandom forrest Tree of binary classifiers for Tree of binary classifiers for

k-classesk-classes Survival risk-groupSurvival risk-group

Supervised pc’sSupervised pc’s With clinical covariatesWith clinical covariates Cross-validated K-M curvesCross-validated K-M curves

Predict quantitative traitPredict quantitative trait LARS, LASSOLARS, LASSO

Feature selection optionsFeature selection options Univariate t/F statisticUnivariate t/F statistic Hierarchical random variance Hierarchical random variance

modelmodel Restricted by fold effectRestricted by fold effect Univariate classification Univariate classification

powerpower Recursive feature eliminationRecursive feature elimination Top-scoring pairsTop-scoring pairs

Validation methodsValidation methods Split-sampleSplit-sample LOOCVLOOCV Repeated k-fold CVRepeated k-fold CV .632+ bootstrap.632+ bootstrap

Permutational statistical Permutational statistical significancesignificance

BRB-ArrayToolsBRB-ArrayToolsJune 2009June 2009

10,000+ Registered users 10,000+ Registered users 68 Countries68 Countries 1000+ Citations1000+ Citations

AcknowledgementsAcknowledgements

NCI Biometric Research BranchNCI Biometric Research Branch Kevin DobbinKevin Dobbin Alain DupuyAlain Dupuy Boris FreidlinBoris Freidlin Wenyu JiangWenyu Jiang Aboubakar MaitournamAboubakar Maitournam Michael RadmacherMichael Radmacher Jyothi SubramarianJyothi Subramarian George WrightGeorge Wright Yingdong ZhaoYingdong Zhao

BRB-ArrayTools Development TeamBRB-ArrayTools Development Team Soon Paik, NSABPSoon Paik, NSABP Daniel Hayes, U. MichiganDaniel Hayes, U. Michigan