imise . AREVIR Kln 12.4.2013 My Background MD & physics - - PowerPoint PPT Presentation

imise.

AREVIR – Köln – 12.4.2013

Personalized Molecular Medicine Clinical trial designs to establish biomarker based treatment decisions

• Biometrical Considerations -

Markus Loeffler Institute for Medical Informatics, Statistics and Epidemiology Center for Clinical Trials (ZKSL) Center for Bioinformatics (IZBI) University of Leipzig

Markus.Loeffler@imise.uni-leipzig.de

imise.

AREVIR – Köln – 12.4.2013

My Background

• MD & physics
• Computational models of regenerative tissues and

cancers

– eg CML treatment optimisation (Chemo/IFN TKI only)

• Biometrical support in the design and conduct of

large scale clinical trials (cancer, heart, infections)

– Interventional – Diagnostic and prognostic

• Bioinformatics of high throughput analyses

imise.

AREVIR – Köln – 12.4.2013

All men are equal (French Revolution) All men are different (Genetic Revolution) Both is true, depending on the point of view

imise.

AREVIR – Köln – 12.4.2013

All men are equal (French Revolution) All men are different (Genetic Revolution) Both is true, depending on the point of view

Biometrical point of view: Consider heterogeneity in the light of a model and resolve it only as far as needed and not more (Occam´s razor)

imise.

AREVIR – Köln – 12.4.2013

What is PMM for me as biometrician ?

A hypothesis and data driven formal model splitting a disease population into subpopulations using molecular diagnostic classifiers to design and allocate differential targeted treatments to improve efficacy and/or side effects compared with present standard treatments

Note:

• group based approach
• cinical decision making
• evidence based

imise.

AREVIR – Köln – 12.4.2013

Clinical decision tree (model)

disease +

• Classifier 1

Classifier 2 Classifier 3 Th A Th C Th B Th D +

• +

Note: Combinatorial problem  many trials needed to establish new standards

imise.

AREVIR – Köln – 12.4.2013

Binary biomarker classifier

• simple case -

Ideal: Biomarker test is positive or negative with high reproducibility and reliability and only few cases remain unclear Binary by definition:

• gene mutations: HER2, KRAS, IDH1
• surface markers (eg cd20 B-cell epitope )

imise.

AREVIR – Köln – 12.4.2013

Glioblastoma and IDH1-story

German Glioma Network cohort

imise.

AREVIR – Köln – 12.4.2013

Glioblastoma and IDH1 German Glioma Network cohort

IDH1+ GBM: 5% AIII: 57% The view changing: Genetics more important than histology in diagnostics

imise.

AREVIR – Köln – 12.4.2013

Biomarker classifier

• complex cases-

Continuous classifiers: eg gene expression signatures eg indices eg imaging signals (eg PET-SUV)  Make them binary by cut points

imise.

AREVIR – Köln – 12.4.2013

Diffuse Large B-cell Lymphoma molecular Burkitt Lymphoma - classifier

non_mBL mBL Different disease but same therapy 53 genes not selected for function Core group

imise.

AREVIR – Köln – 12.4.2013

mBL classifier external validation

• n data from Dave et al. 2006

99 aggressive B-cell lymphomas were HGU113plus2.0 data were available mBL index Gene expression mBL index Dave et al. classifier

imise.

AREVIR – Köln – 12.4.2013

Classifier construction

Multistep process important: (1) exploratory (2) define the classifier on a training sample (3) test this classifier on internal and external validation samples and demonstrate that it offers added value to classifiers already available eg

• comparing ROC-curves if diagnostic
• adjusting COX-analyses if prognostic

Mind: Statistical significance does not imply clinical relevance, It is the biology which matters !!

imise.

AREVIR – Köln – 12.4.2013

Biomarker analyses in cohort studies

Prospective clinical cohort studies play an important role for biomarker discovery with impact on diagnostis and prognosis

Requires:

• Carefully phenotyped cases
• Careful follow up (if looking for prognosis)
• High quality biomaterial
• High quality laboratories

Caveat: Cohorts outside RCTs can be biased and confounded not suited for determining which treatment is best !! RCT

imise.

AREVIR – Köln – 12.4.2013

Clinical trials to search best biomarker related treatments

• one target case, dichotomous -

imise.

AREVIR – Köln – 12.4.2013

Phase 3 biomarker trial designs

• with binary test classifiers-

Assumption:

• Valid dichotomous biomarker
• High sensitivity and specificity
• Reproducibility shown
• Practicability shown

see Freidlin et al JNCI 2010, R Simon´s group

imise.

AREVIR – Köln – 12.4.2013

(I) Biomarker stratified trial design

Note: may need different sample sizes in each stratum

imise.

AREVIR – Köln – 12.4.2013

Marker Validation for Erlotinib in Lung Cancer

imise.

AREVIR – Köln – 12.4.2013

NSCLC and metagene classifier

imise.

AREVIR – Köln – 12.4.2013

Biomarker stratified design

Advantages:

• Can provide full information on treatment benefits in each

subgroup and also regarding interactions

• Can provide information whether biomarkers are overall

prognostic and predictive for the special treatments

• Can often be implemented post hoc in RCTs that have been

conducted provided biobanking is available But:

• requires more cases than a simple RCT
• up to 4-fold if interaction is estimated
• Works only if few biomarkers are investigated
• Can only be done if same treatment in both strata can be justified

imise.

AREVIR – Köln – 12.4.2013

(II) Biomarker enrichment design

CD20+ DLBCL 18-60 years IPI 0,1 Stages II-IV, I with bulk 6 x CHOP-like 6 x CHOP-like + Rituximab Random. CD20 neg: T-cell Lym Off Study N = 800, 10% difference in 1.EP

DL DLBCL-Ritux uxim imab ab

median observation time: 22 months

Tim Time t to

• Tr

Treatment Fail Failure

Months

50 45 40 35 30 25 20 15 10 5

Probability

1.0 .9 .8 .7 .6 .5 .4 .3 .2 .1 0.0 79.9% R-CHEMO 60.8% CHEMO p = 0.000 000 007

Pfreundschuh et al Lancet Oncol 2006

imise.

AREVIR – Köln – 12.4.2013

Biomarker enrichment designs

Advantage:

• Can only resolve treatment question in one subgroup

Disadvantage:

• No answer for other subgroup
• Cannot provide information about overall prognostic
• r predictive relevance of biomarker

CAVE:

Severe problems if the classifier is not correctly classifiying  overtreatment, undertreatment, wrong treatment

imise.

AREVIR – Köln – 12.4.2013

(III) Biomarker Strategy Design

imise.

AREVIR – Köln – 12.4.2013

Biomarker strategy design NSCLC-example

imise.

AREVIR – Köln – 12.4.2013

Biomarker Strategy Design standard vs fully individualized treatments

Biomarker Assessment & Random Fully individualized treatment Standard treatment Process of treatment engineering eg: Dentritic cell based vaccination therapy HIV targeted drug combinations CAVE: Regulatory problems

imise.

AREVIR – Köln – 12.4.2013

Design standard vs fully individualized treatments

Assessment Individualized treatment Standard treatment Computational treatment design eg: HIV targeted drug combinations CAVE: Regulatory problems

R

imise.

AREVIR – Köln – 12.4.2013

Design two different individualized treatments

Assessment Individualized treatment (Comp) Computational treatment design eg: HIV targeted drug combinations CAVE: Regulatory problems Traditional treatment design Individualized treatment (trad)

R

imise.

AREVIR – Köln – 12.4.2013

Biomarker Strategy Design

Compares a complex treatment strategy as a whole vs a standard treatment Advantage: Provides information on the overall strategy, but not

• n subgroups

can cope with many treament options Disadvantage: Interpretation problems which parts of the strategy are successful or harmful

imise.

AREVIR – Köln – 12.4.2013

Remark

Splitting the diseases into smaller subpopulations we often see stronger benefits with targeted treatment hence we need to do many trials but sample sizes and power is not necessarily a major concern

imise.

AREVIR – Köln – 12.4.2013

Special designs

imise.

AREVIR – Köln – 12.4.2013

Which of two classifiers is superior ? example Mindac trial

BRCA Two biomarkers (1) Immun histo (2) Gene profile Consensus Low risk: moderate Th Consensus High risk: intensive Th Discordant: R Moderate Th Intensive Th

imise.

AREVIR – Köln – 12.4.2013

Multimarker-multitarget designs Vach et al 2006

2 Biomarkers: a and b which can coexist Therapy: A- Targeted B- Targeted S- Standard Disregards a & b cases a & b cases not fully exploited

imise.

AREVIR – Köln – 12.4.2013

Multimarker designs Vach et al 2006

Best design

• balanced
• many comparisons
• value of each

single marker/Th Full benefit if many independent targeted therapies are considered

imise.

AREVIR – Köln – 12.4.2013

Phase 2 and 3 trial sequence

Traditional trial sequence:

• 1. Phase 2 – one armed experimental trial
• 2. Phase 3 – RCT with control vs experimental

Problems:

• Unknown selections in phase 2
• Phase 2 information not used in phase 3
• Regulatory delays

 Integrated randomized phase 2 & 3 designs helpful

imise.

AREVIR – Köln – 12.4.2013

Phase 2 selection designs

• Principle -

Procedure to identify a promissing experimental therapy from a list of candidates for phase 3

Random ExpTh 2 ExpTh 1 ExpTh 3 ExpTh 4 ExpTh 5 1.Stage Meets specified minimum success rate 2.Stage Meets specified success rate

STOP STOP STOP

Select best Phase 3 RCT

STOP

imise.

AREVIR – Köln – 12.4.2013

Summary

There is no reason to let PMM escape from the principles of evidence based medicine PMM needs well designed prospective trials

• for defining and validating diagnostic classifiers
• to demonstrate that they unravel heterogeneity
• to select promissing novel targeted therapies
• to demonstrate their efficacy and effectiveness

imise.

AREVIR – Köln – 12.4.2013

Conceptual questions

PMM makes us rethink

• What is a disease ?
• I argue it is a (formal) reductionist cognitive concept,

abstracting from a set of observations to deduce decisions and actions,  we need to get better and more detailled models to deal with information overflow on irrelevant data

• n irrelevant heterogeneity and find what is really

important and ignore the rest

imise.

AREVIR – Köln – 12.4.2013

Thank you