Trajet d'une expatri

1
Trajet d'une expatriée de la phylogénie du VIH
au traitement de la grippe, et de Paris à San
Francisco Colombe ChappeyDEA 1986, PhD 1992
2
Statistiques Cliniques
Personalized Health Care (Soins personnalisés)
Bioinformatique
DEA 1986
Reconnaissance de Formes (These 92)
Essais Cliniques
Analyse dimages
Modélisation
Epidémiologie
Analyse Exploratoire Bio-marqueurs predictifs
Epidémiologie Moleculaire
Transmission de la grippe
Programmation (Computer Science)
3
Au cours de mon trajet
Statistiques Cliniques
Personalized Health Care (Soins personnalisés)
Bioinformatique
DEA 1986
Reconnaissance de Formes (These 92)
Essais Cliniques
Analyse dimages
Modélisation
Epidémiologie
Analyse Exploratoire Bio-marqueurs predictifs
Epidémiologie Moleculaire VIH
Transmission de la grippe
Programmation (Computer Science)
4
Partager mon experience

• Transitions
• de la recherche publique en France aux Etat-Unis
• De lAcademic au privé
• de la petite Biotech a la grosse Pharma
• Données Explosion des données genetiques
  disponibles
• Nouvelles technologies de sequencages
• Limportance du to think outside the box (en
  dehors de sa bulle)
• Position unique du bioinformaticien/biostatisticie
  n entre données et idées
• Opportunities is often missed because it is
  dressed in overalls and looks like work (Thomas
  Edison

5
Reconnaissance de motifsappliquée a la
comparaison de sequences biologiques
Comparaison de séquences nucleiques/proteines -gt
Alignement des éléments/motifs en commun -gt
pondérer les différences/mutations et les
insertions/deletions
A G G T T G C A G G T C
6
Comparaison de séquences biologiques de Virus
dimmunodéficience

• Comparaison de
• 9 séquences de VIH type 1
• 1 séquences VIH type 2
• 5 séquences de VIS
• Le nombre de sequences de VIH a tres vite
  augmente.
• Certaines séquences sont plus similaires que
  dautres

1988
7
MASH Algorithme dalignement de plusieurs
séquences
Chappey C, Danckaert A, Dessen P, Hazout S. MASH
an interactive multiple alignment and consensus
sequence construction. Comp. Applic. Biosci.
1991 7195-202.
8
Applications
Distance entre séquences Classification
time
Homogénéité et hétérogénéité par region
Chappey C, Danckaert A, Dessen P, Hazout S. MASH
an interactive multiple alignment and consensus
sequence construction. Comp. Applic. Biosci.
1991 7195-202.
9
Cas du Dentiste - 1990
10
Prediction de Structure/function de la Proteine
dEnveloppe du VIF
Profile of structural constraints based on
quantification of amino acid replacements
Selection for change Profile of the ratio of
nonsynonymous to synonymous change proportions
(nsi/si, si)
Pancino G, Chappey C, Saurin W, Sonigo P. B
epitopes and selection pressures in feline
immunodeficiency virus envelope glycoproteins.
J. Virol. 1993 67664-672. Pancino G, Fossati I,
Chappey C, Castelot S, Hurtrel B, Moraillon A,
Klatzmann D, Sonigo P. Structure and variations
of feline immunodeficiency virus envelope
glycoproteins. Virology 1993 192659-62.
11
Bilan des années de These

• () Tremplin pour les collaborations
• Institut Pasteur, France
• Agence Nationale Recherche Sida (ANRS)
• Institut Cochin de Genetique Moleculaire (ICGM)
• HIV database de Los Alamos National Laboratory,
  NM
• () Publications
• Méthodes 2
• Application du logiciel dalignement
• Human immunodeficiency virus type 1 4
• Transmission HIV mother-infant 5
• Simian / human T-cell lymphotropic virus type 1
  3
• Simian immunodeficiency virus 1
• Feline immunodeficiency virus FIV 2
• (-) Occasions manquées
• Commercialisation du logiciel dalignment (alors
  que CLUSTAL)
• Analyses non-publiées

12
National Center Biotechnology Information(GenBank
)
National Institutes of Health
13
(No Transcript)
14
Histoire de GenBank et NCBI
Human EST
BLAST (Basic Local Alignment Search Tool)
Human Genome
GenBank demenage a NIH
international computer database of nucleic acid
sequence data Los Alamos Natl Lab, NM (NSF)
Wilbur and Lipman Algorithme de recherche de
similarites entre sequences
1979
15
Programmation dun outil dannotation et de
Soumission de Séquences Biologiques a GenBank

• La publication de nouvelles séquences biologiques
  nécessite de les rendre publiques
• Avant, elles etaient publier dans les journaux
  scientifiques
• Avec GenBank, elles sont envoyées par email au
  service qui faisait les annotations et leur
  associait un numéro dAcces (Accession Number)
• Besoin doutil informatique permettant aux
  biologistes dannotater leur séquences avant de
  les envoyer
• Types de séquences
• Gene codant (CD) -gt simple soumission
• EST (Expressed Segment T) -gt soumission en batch
• Population de Séquences -gt soumission des
  séquences alignées

16
Sequin Soumission de Sequence aux DB genetiques
1995
http//www.ebi.ac.uk/Sequin/QuickGuide/sequin.htm
17
Editeur dAnnotation de Sequences
18
Editeur dAnnotation de Sequences Alignees

• Wheeler DL, Chappey C, Lash AE, Leipe DD, Madden
  TL, Schuler GD, Tatusova TA, Rapp BA. Database
  resources of the National Center for
  Biotechnology Information.Nucleic Acids Res. 2000
  Jan 128(1)10-4.

19
PopSet de GenBank
20
CN3D Viewer de Structure de Protéines
Wang Y, Geer LY, Chappey C, Kans JA, Bryant SH.
Cn3D sequence and structure views for entrez.
Trends Biochem Sci. 2000 Jun25(6)300-2. Marchler
-Bauer A, Addess KJ, Chappey C, Geer L, Madej T,
Matsuo Y, Wang Y, Bryant SH. MMDB Entrez's 3D
structure database. Nucleic Acids Res.
199927(1)240-3.
21
(No Transcript)
22
Bilan des années NIH

• () Acquisition de connaissances dans un institut
  de renommée internationale
• Data format ASN-1 (Abstract Syntax Notation One)
• Format de répresentation de données ISO
  permettant linteroperabilité entre plateformes
  et représentation de données hétérogenes.
• Convertie en XML
• Programmer en C/C, Web server,
• Travailler dans le milieu academic américain
• Données et programmes sont disponibles au public
  (QC)
• ftp.ncbi.nih.gov
• (-) Occasion manquée (ou non)
• lopportunité de travailler sur le Génome Humain

23
1998 NCBI - Whats Next?

• Phénotype caractères observables d'un organisme
• Gene expression profiling (par Microarray
  Affymetrix, Stanford) sur RNA, comparaison de
  lexpression de génes, dans différents types
  cellulaires (traités non-traités)
• SNPs / DeCode
• HIV Drug Resistance Database in Stanford
• Données cliniques occurrence et évolution de
  maladies
• dbGaP SNPs et maladies genetiques
• Allele mutants et (partial) resistance a
  linfection par le VIH
• Reponse clinique aux antiviraux et la presence de
  virus resistance

24
(No Transcript)
25
ViroLogic Inc 2000-2009

• Mission "The right therapy to the right patient
  at the right time.
• 10 antiviraux anti-VIH
• Business Model simple

Laboratoire dAnalyses
Hopital
DB
Patient Resistance Report
Algorithm

• 100 employes, 80 dans la laboratoire danalyse,
  20 dans la recherche, ladministration

26
Test de Résistance du VIH aux antiviraux 2
approches Phénotype-Génotype
Test de Phenotype teste la capacite de chaque
antiviraux de diminuer la FONCTION de la protein
virale cible de lantivirale.
Translation
Polyprotein
Test de Genotype determine la sequence de la
proteine cible de lantiviral Un algorithme
reconnait les mutations cles qui diminue la
function de la proteine
Clivage Processing Folding
27
Database de ViroLogic
Génotype
Small studies (n 100s)
PT-GT database (n gt 100,000)
Response Clinique Reduction de la charge virale
Phénotype IC50 fold change
Small studies (n 100s) clinical cut-off pour
le phenotype
Identification de mutation associees a la
resistance du VIH aux antiviraux
28
Calling Bases and Mixtures from Raw Sequence (ABI
Chomatogram) Data
codon 184 R(A/G)TG -gt M/V
29
Fréquences des Mutations par Réponse virologic
apres 2 semaines
Zolopa, A. R. et. al. Ann Intern Med
1999131813-821
30
Régles dinterprétation du Genotype
Resistance Collaborative Group (DeGruttola et
al., 2000) Initially used in GeneSeq assay, with
some modifications Expert Consensus, derived for
meta-analysis (not intended for clinical use) UK
Drug Resistance Database (2006)
http//www.hivrdb.org.uk/ Stanford (R. Shafer),
HIVResistance.com Comprehensive, updated
frequently, good notes International AIDS Society
IAS (Hirsch et al., JAMA 2000 2008 updates)
http//iasusa.org Expert consensus updated
frequently
31
Interprétation du Génotype viral
. . . .
. . . . .
Wild-type PQITLWQRPLVTIKIGGQLKEALLDTGADDTVLEEMN
LPGRWKPKMIGGIGGFIKVRQYDQILIEICGHKAIGTVLVGPTPVNIIGR
NLLTQIGCTLNF Patient PQIALWQRPLVTIKIGGQLKEALLDT
GADNTILEEMNLPGRWKPKMVGGIGGFVKVRQYDQILIEICGHKAIGTVL
VGPTPVNIIGRNLLTQIGCTLNF
Patient virus genotype
D30N
T4A
I54V
V32I
I47V
D30N Resistance to NPV I47V, I54V Intermediate
resistance to fAMP, TPV
Regles dinterprétation du Génotype
Drug Resistance associated Mutations (RAMs)
32
How are Drug Resistance Mutations Identified?

• In vitro selection, clinical studies,
  site-directed mutagenesis BUT
• Drug resistance mutations identified during drug
  development (esp. in vitro) may not be the most
  relevant mutations in clinical settings
• Mutations that are sufficient to cause drug
  resistance may not be necessary to effect drug
  resistance
• Cross-resistance due to mutations selected by
  related drugs

33
Mesure de Résistance Phenotypique
IC50 Concentration of drug required to inhibit
viral replication by 50. Fold Change _IC50
patient_ IC50
reference Reference wild-type
reference strain NL4-3
inhibition
Log concentration of drug

• Chappey 02/23/09

34
Analysis Univariée des mutations
To determine which mutations are associated with
High or Low TPV IC50 Fold Change

• Fishers Exact test
• with the Benjamini correction for multiple tests
  (for each mutation)
• -WilcoxonMannWhitney test
• For comparison of median FC

35
Variabilité de la résistance au SQV des virus
avec L90M
1639 samples, excludes those with gt1 mutation at
30, 48, 50, 82, 84 or mixtures at these
positions 35 are lt2.5, 69 are lt10
36
TPV Susceptibility in Groups of Samples
categorized by the TPV Mutation Score
lower clinical cutoff
(Total1411)
37
PT-GT Discordances
PT-R GT-R
PT-R GT-S
lower clinical cutoff
PT-S GT-R
PT-S GT-S
(Total1411)
At a mutation score cutoff of 4 total discordance
was 18.1
38
Performance of the New Tipranavir Mutation Score
Validation Dataset (Total1845)
N 36 76 104 166 147 145 158 126 124 112
94 99 74 68 57 70 47 34 30
16 23 14 25
At a mutation score cutoff of 4 total discordance
was 16
39
Trade off between Model Complexity, Predictive
accuracy and Biological Descriptive Meaning
Genotype Rules and Mutation Score
Genotype Rules
ML Regression
Model complexity
SVM
MLR Multiple Linear Regression
Neural Network
Model Predictive Accuracy
Increasing
SVM Non-linear Support Vector Machine
Biological Descriptive Meaning
39
40
De la bulle des Dot-Com aux Subprimes
March 10, 2000
Embauche
licenciement 2
licenciement 1
Grant 2m
NIH Grant 400K
NIH Grant 400K
Introduction en bourse
2009
Chart of NASDAQ closing values from 1994 to 2008
41
Small Business Innovation Research Grants
NIH Grants Title Dates Resume
SBIR Phase I HIV Phenotype/Genotype Database Resources Aug. 2003 July 2005 This grant supported the development of a relational database populated with phenotypic and genotypic drug resistance data collected from a large number (gt80,000) of HIV-1 patient isolates. Statistical and analytical query tools were developed to derive highly accurate genotypic-phenotypic correlations. 400.000
SBIR Phase I HIV-1 Envelope phenotype/genotype database resources . May 2004 Apr. 2006 The goal of the project was to create, populate and exploit an HIV-1 envelope database comprised of high quality data derived from genotypic and phenotypic assays recently developed at Monogram Biosciences to characterize and evaluate entry inhibitors and vaccines 400.000
SBIR Phase II The Development of a Web-based Data Retrieval System for HIV Therapy Guidance June 2007 May 2010 The goal of the project was to implement a web-based database retrieval system to search the Monogram HIV drug resistance database to support clinical management of HIV/AIDS patients and development of novel therapeutics. 2.000.000
42
Bilan

• () Organisation du travail dans un societe
  privee
• Respect des délais
• Coaching des collaborateurs
• Concrétisation de projets i.e. rédiger des
  projets aboutissant a un financement, et donc a
  une réalité
• (!) Application des connaissances acquises
• Utilisation de R, Perl
• (-) Occasions manquées
• Insuffisante priorité accordée a ma carriere au
  sein de la société (a la rue vs. promue)

43
Genentech Roche Senior Biostatistician

• Genentech 11 000 employes
• Produits les anticorps therapeutiques

44
Histoire de la collaboration entre Genentech et
Roche
Roche exercises its option to cause Genentech to
redeem its outstanding special common shares not
owned by Roche. Roche announces its intent to
publicly sell up to 19 percent of Genentech
shares and continue Genentech as a publicly
traded company on the NYSE (symbol DNA) with
independent directors. Roche signs license
agreement to sell Genentechs products in ex-U.S.
markets.
At the Roche Institute of Molecular Biology a
pure interferon alfa is isolated. Roche Nutley
and Genentech start work on a joint project to
produce a genetically engineered version of the
substance.
Pour maladies virales -HIV Saquinavir SQV -HCV
Inhibiteurs de polymerase et de protease en Phase
2 -Grippe Tamiflu (post-marketing)
1980
1990
1999
2009
Genentech and Roche complete a 2.1 billion
merger, and Genentech continues to trade on the
NYSE.
Roche and Genentech announce that they have
signed a merger agreement, and Genentech becomes
a wholly owned member of the Roche Group.
45
Personalized Health Care- Are We there Yet?
46
What is our role as Statisticians? How/when do
we get involved?The Drug/Diagnostic
Co-development
Mark Lackner
46
Phase I/II/III

Early stage research
Late stage research
Drug

• Dx Biomarker validation
• Develop validated Dx assay with partner
• Phase III strategy and implementation
• Risk mitigation plans

• Develop clinical Dx Strategy (DxST)
• Develop in house assays in Ph I

• Establish Dx hypothesis
• Identify Dx marker candidates
• Preclinical validation

• Assess need for Dx
• Initiate selected programs

Companion Dx Test
47
Ce qui me reste a faire

• Epouser un milliardaire americain
• George Soros
• Warren Buffet
• Donald Trump
• Monter une start-up Biotech
• Et la revendre a Pfizer pour 18 mds dEuros
• Ensuite racheter lUPMC
• Chirurgie esthetique
• GIS

48
ArcGIS Epidemie de grippe
49
(No Transcript)
50
(No Transcript)
51
Back-up Slides
52
The Drug/Diagnostic Co-development
Mark Lackner
52
Phase I/II/III

Early stage research
Late stage research
Drug

• Dx Biomarker validation
• Develop validated Dx assay with partner
• Phase III strategy and implementation
• Risk mitigation plans

• Develop clinical Dx Strategy (DxST)
• Develop in house assays in Ph I

• Establish Dx hypothesis
• Identify Dx marker candidates
• Preclinical validation

• Assess need for Dx
• Initiate selected programs

Companion Dx Test
Research/Research Dx
Development Dx/PDB
Companion Dx
53
Virus susceptibility to antiretroviral drugs
allows for the control of the infection
Antiviral drug susceptibility correlates with
virologic outcome
HIV resistance occurs when HIV changes or
mutates so it can escape the effect of an
antiretroviral drug -gt choosing an ART regimen in
light of resistant HIV -gt resistance testing
Deeks S. JID, 1999179137581
54
Agenda

• Phenotype (PT) and genotype (GT) assays require
  bioinformatics-based interpretation algorithms to
  interpret a patient virus as resistant (R) or
  susceptible (S) to a drug
• Phenotype assaymeasure of the ability of a virus
  to replicate in presence of a drug
• Cut-offs are used to categorize the PT measure as
  drug Resistant or Susceptible
• Genotype assay
• provides the list of mutations present in a virus
  pool and differing from the wild-type
  drug-sensitive virus
• An algorithm is used to recognize the key
  mutations associated with resistance from
  patient-specfic polymorphism

55
Application using RESIST trial for tipranavir TPV

• Boehringer Ingelheim Protease Inhibitor Aptivus
  (tipranavir)
• The RESIST trial evaluated Aptivus (tipranavir)
  in treatment-experienced HIV-1 infected patients
• Baseline samples selected were
• The study regimen did not include enfuvirtide
• Where the study PI/r was not a continuation of
  the prestudy PI/r
• Endpoint Viral Load reduction at week 4

56
Phenotype Assay Technical Process

• Isolating the viral RNA for Protease and Reverse
  Transcriptase
• 2. Constructing the test vector
• 3. Producing and testing the virus

Petropoulos CJ, ANTIMICROBIAL AGENTS AND
CHEMOTHERAPY, Apr. 2000, p. 920928
57
Phenotype Resistance Interpretation
Clinical cut-off -drug level at which a patients
probability of treatment failure increases.
-Based on outcome data from clinical
trials. Biological cut-off -based on natural
variability of wild-type viruses from
treatment-naïve HIV-1 infected patients - 99th
percentile of the IC50 FC distribution -Requires
a large number of wild-type samples. Assay/techn
ical cut-off -Based on assay variability with
repeated testing of patient samples
58
(No Transcript)
59
Conclusion 1

• 2 week process that may fail in case of viruses
  with low replication capacity
• PT may not capture the resistance in case of
  minor populations of resistant variants that are
  selected by the drug pressure
• Phenotypic Cutoffs caveats
• Biological cutoffs are assay specific
• Clinical cutoffs are method dependent

60
Genotype assay and Rule-based interpretation

• PROTEASE (1-99) and REVERSE TRANSCRIPTASE (1-305)
• Validated for samples with viral loads ? 500
  copies/mL
• Use of multiple primers Redundancy of 2 to 5
  sequence fragments
• Detects all mutations and mixtures from
  co-existing populations of virus (as minor as
  10-30)

Patient virus population (quasispecies)
61
(No Transcript)
62
(No Transcript)
63
(No Transcript)
64
(No Transcript)
65
(No Transcript)
66
(No Transcript)
67
Conclusions 2

• - Genotype algorithms evolve over time with
  increased clinical experience and more clinical
  data on cross-resistance and reverse
  susceptibility
• Use of large database combining phenotype and
  genotype results to generate more accurate
  genotype interpretive algorithms
• Minimizing PT-GT Discordance tradeoff between
  false negatives (PT-S GT-R) and the false
  positives (PT-R GT-S)
• PT-R GT-S
• New mutations
• Cross-resistance
• PT-S GT-R
• Suppression of resistance or re-sensitization
• Presence of mixtures
• Use of more complex prediction models yield to
  more accurate algorithms but with less biological
  descriptive meaning

68
Monogram Technologies for Resistance Testing
Patient virus
RT-PCR
PR-RT DNA
PhenoSense
GeneSeq
Vector Assembly
Transfection
Sequencing
Recombinant Virus
Infection
Resistance Mutations
Measure of Drug Susceptibility
Rules forgenotype Interpretation
Categorize R if FC gt cut-off S if FC lt cut-off
Pheno-Geno Database
Prediction of DrugSusceptibility
Categorization of DrugSusceptibility
69
Discussion

• Interpretation of phenotypic (cutoffs) and
  genotypic (algorithms) resistance assays is an
  evolving science
• Large databases of phenotypic and genotypic
  information are essential tools to understand and
  improve discordance rates
• The use of both types of assay in many cases
  provides the most complete picture of an
  individual patients virus resistance profile

70
Acknowledgements
Increasing Genetic Complexity
Genotypic testing
Phenotypic testing
Utility
Treatment rounds

• All my colleagues at Monogram Biosciences
  (Clinical Reference Laboratory and Research and
  Development)
• And my collaborators (Steve Deeks, UCSF, Andy
  Zolopa, Stanford, Sebastian Bonhoeffer,
  Swizerland, R. Shafer ,Stanford..)

71
Biological Cut-off Definition

• Biological cut-off based on natural variability
  of wild-type viruses from treatment-naïve HIV-1
  infected patients (infected by patient who is
  also drug naïve)
• When the treatment history is not known,
  wild-type virus WT is defined by the absence of
  any drug-selected mutation in PR or RT
• PR 23, 24, 30, 32, 33F, 46, 47, 48, 50, 54, 82
  (not 82I), 84, 90
• RT 41, 65, 67, 69 (incl. ins.), 70, 74, 75, 100,
  101E or P, 103N or S, 106A or M, 151, 181, 184,
  190, 210, 215F or Y, 219, 225, 227, 230, 236

72
Biological Cut-off for TPV
Natural Variation of TPV FC Among Wild-type
Samples
99th percentile 2.1
0.16 0.25 0.40 0.63 1.0
1.6 2.5 4.0
TPV fold change
N2848 , no PI or RTI recognized resistance
mutations
73
Genotype Interpretation for Tipranavir (TPV)

• TPV susceptibility based on genotype uses an
  algorithm that counts mutations associated with
  reduced in vitro susceptibility or in vivo
  virological response.
• The TPV mutation score was derived from
  analysis of a limited number of patient samples
  collected during phase 2 and 3 clinical trials
  and considers the following mutations L10V,
  I13V, K20M, R, or V, L33F, E35G, M36I, K43T,
  M46L, I47V, I54A, M, or V, Q58E, H69K, T74P, V82L
  or T, N83D, I84V1.

Kohlbrenner et al., HIV DART, 2004
74
Mutations Associated with PT-R GT-S
Mutation N mut Odds ratio P-Value
I54A 16 15.1 0.00253
A71L 18 8.0 0.00497
V11L 20 4.0 0.03667
V82T 65 2.8 0.00076
I47V 122 2.8 lt0.0001
G73T 66 2.5 0.00329
L89V 105 2.3 0.00034
I84V 356 2.2 lt0.0001
V32I 169 2.0 0.00008
M36L 77 2.0 0.02024
I66 94 1.9 0.01722
D60E 217 1.6 0.00265
K55R 169 1.6 0.01546
L90M 787 1.3 lt0.0001
M46I 495 1.3 0.00424
L10I 625 1.2 0.02199
underlined mutations in existing TPV mutation
score the ratio of H samples with the
mutation to L samples with the mutation
75
Phenotype-ClinicalWeek 4 HIV-1 VL Change vs.
Baseline IC50 Fold Change to TPV
(N 176)
76
Clinical Cutoffs Definitions
Lower clinical cutoff The IC50 fold change at
which the HIV RNA response first begins to decline
Upper clinical cutoff The fold change above
which a clinically meaningful HIV RNA response
(gt0.3 log10) is unlikely
Probability of response
Zone of Intermediate Response
Fold Change
77
Clinical Cutoffs Methods
Lower clinical cut-off Comparison of HIV RNA
responses between two adjacent groups across a
moving IC50 FC cut-off (Kruskal-Wallis test)

• Upper clinical cut-off
• Phenotypic susceptibility scoring to account for
  background effect
• Define the HIV RNA change attributable to the
  PI/r
• Define the fold change associated with an HIV RNA
  reduction of -0.3 log10 copies/mL

• Chappey 02/23/09

78
LCCO First difference from reference Expanding
Window method
79
LCCO First difference from reference Expanding
Window method
80
LCCO First difference from reference Expanding
Window method
81
LCCO First difference from reference Expanding
Window method
82
LCCO First difference from reference Expanding
Window method
83
LCCO First difference from reference Expanding
Window method
84
LCCO First difference from reference Fixed
Window Method
85
LCCO First difference from reference Fixed
Window Method
86
LCCO First difference from reference Fixed
Window method
87
LCCO First difference from reference Fixed
Window method
88
LCCO First difference from reference Fixed
Window method
89
LCCO First difference from reference Fixed
Window method
90
LCCO First difference from reference Fixed
Window method
91
Comparing LCCO with the Biological Cut-off
Natural Variation of TPV FC Among Wild-type
Samples
In order to minimize misclassification of
wildtype isolates as resistant a TPV/r LCO at 2.0
was chosen
LCCO 1.5
99th percentile 2.1
0.16 0.25 0.40 0.63 1.0
1.6 2.5 4.0
TPV fold change
N2848 , no PI or RTI recognized resistance
mutations
92
Clinical Cutoffs Methods
Lower clinical cut-off Comparison of HIV RNA
responses between two adjacent groups across a
moving IC50 FC cut-off (Kruskal-Wallis test)

• Upper clinical cut-off
• Phenotypic susceptibility scoring (PSS) to
  account for background effect
• Define the HIV RNA change attributable to the
  PI/r
• Define the fold change associated with an HIV RNA
  reduction of -0.3 log10 copies/mL

• Chappey 02/23/09

93
UCCO DeterminationCalculate the proportion of
HIV RNA change attributed to PI/r
HIV RNA reduction attributable to each drug
2 NRTI
2 NRTI
50


PSS0
PSS1
TPV 50
PSS1
PSS1
TPV 100
TPV/r
TPV/r
Adjust HIV RNA change attributable to TPV/r
94
Phenotypic Susceptibility Scoring (PSS)
95
Scatter plots of drug susceptibility versus week
4 HIV RNA change
Regimen phenotypic susceptibility score (PSS)
versus HIV RNA change (R²0.19, plt0.0001)
TPV FC (log10) versus unadjusted Week 4HIV-1 RNA
(log10) change, N176, (R²0.22, plt0.0001)
-0.3log10c/mL
96
TPV FC versus Adjusted Week 4 HIV RNA Change
97
Adjusted Week 4 HIV RNA outcomes by TPV
susceptibility category
98
What is our role as Statisticians? How/when do
we get involved?
99
What is Our Responsibility

• We are strategic partners
• PHC strategy is part of the Development Plan
• Embrace the PHC strategy
• Engage the DST in strategic/prioritization/timelin
  es discussions related to PHC
• Raise the right issues
• Plan for resources
• Work with DST and your manager
• Network with the Biomarker Experts/Dx sub-teams
• Be proactive/Stay informed
• Get Involved!

100
What is our role as Statisticians? How/when do
we get involved?The Drug/Diagnostic
Co-development
Mark Lackner
100
Phase I/II/III

Early stage research
Late stage research
Drug

• Dx Biomarker validation
• Develop validated Dx assay with partner
• Phase III strategy and implementation
• Risk mitigation plans

• Develop clinical Dx Strategy (DxST)
• Develop in house assays in Ph I

• Establish Dx hypothesis
• Identify Dx marker candidates
• Preclinical validation

• Assess need for Dx
• Initiate selected programs

Companion Dx Test
101
PHC strategy Development Strategy
PHC Strategy PHC Strategy PHC Strategy
Strong Dx hypothesis No activity in Dx- Strong Dx hypothesis Some activity in Dx- No strong Dx hypothesis Exploratory Stage
Development Strategy Development Strategy Development Strategy
Patient selection through all phases of development Complex, larger phase IIs with stratification Complex phase IIIs No selection or stratification Possible data mining trap
102
Impact on components of CDP

• Target product profile
• Parallel development of companion diagnostic
• Phase I trials
• Selection for quick signal seeking
• Phase II trials
• Complex issues become more complex
• More unknowns, more questions to answer
• Phase III trials
• Clinical Validation of Dx
• Design depends on Phase II outcome
• Selection, stratification or all-comers

103
Phase II Considerations

• Objective simultaneous Rx/Dx evaluation
• Scientific rationale and pre-clinical data - main
  determinants of the scenario prior to Phase II
• Statistical considerations
• Co-primary endpoints
• Value added and feasibility of stratification
• Defining cut-offs for continuous biomarker
• Go/No Go decision algorithm
• Dedicated studies to investigate assay or
  biomarker properties
• Reproducibility, prevalence, prognostic value

104
Phase III Considerations

• Study Objective
• Assess/determine risk/benefit
• Clinical Validation of Dx
• Implementation issues
• Analytically validate Dx assay before applying it
  to specimens in pivotal trials
• Accruing / prospective stratification based on
  non-final assay can result in discordance
• Analysis method
• Test two hypothesis,
• All comers
• Dx positive subgroup
• Appropriately control for type I error
• Clearly define your decision tree there are no
  freebies

105
End of Phase III Decision Criteria
Phase III outcome Not statistically significant in all comers Statistically significant in all comers
Statistically significant in Dx group SELECTION CLAIM All comers claim if no diff. b/w Dx- Dx groups
Statistically significant in Dx group SELECTION CLAIM Greater benefit claim if clinically meaningful diff. b/w Dx- Dx
Statistically significant in Dx group SELECTION CLAIM Selection Claim if no improvement in Dx- group
106
Old Drugs New Tests

• Biomarker not known at the time of study
  initiation
• Data not analyzed with that biomarker as part of
  the hypothesis
• New scientific advancements/new technologies
• Biomarker discovery generation of new
  hypotheses
• Prospective-Retrospective Study

Exploratory Analysis
107
Prospective/Retrospective Study

• Completed or post-interim-analysis trial
• Patient samples collected prior to treatment
  initiation
• Clinical outcomes data unblinded and analyzed
  without the biomarker data
• Diagnostic hypothesis/analysis plan -
  prospectively specified
• Analysis is retrospective

108
Components of good biomarker analysis plan

• Role of randomization - fairness of comparison
• Marker availability impact of convenience
  samples
• Bias due to missing data
• Marker performance
• Marker performance and prevalence may explain
  study to study heterogeneity
• Statistical control of false positive conclusions
  
• How many hypothesis
• How many outcomes
• Model selection
• Over-fitting can lead to bias
• Validation methods
• Data to generate the hypothesis vs. data to
  confirm the hypothesis

109
Summary

• Companion diagnostics are at the heart of
  personalized health care
• Predictive claims rely on understanding the
  effect of the drug in biomarker positive and
  negative patients
• Optimal approach Adequate and well-controlled
  trials, prospectively designed to assess
  risk/benefit in biomarker subgroups
• Late emergence of critical biomarkers for
  existing drugs - revision of drugs use
• As strategic partners, we need to be involved in
  all stages of the co-development process