Title: On the Road to Genomic Predictive Medicine An Interim Analysis
1On the Road to Genomic Predictive Medicine An
Interim Analysis
- Richard Simon
- Chief, Biometric Research Branch
- National Cancer Institute
2How I got involved in genomics
- In the late 1990s genomic data was for me the
most exciting scientific data of our generation - Analysis of that data shouldnt be left to
amateurs -
- We had a great cadre of statisticians involved in
clinical trials and we know how to do reliable
clinical trials, but the drugs are often
disappointing - Statisticians should be involved in basic
research, pre-clinical target discovery and
policy
3- Biomedical leaders were looking to computer
scientists and physicists for help, not to
statisticians - Statisticians were viewed as useful for testing
hypotheses and computing p values, not for
discovery
4- Many statisticians tend to see themselves as
methods developers not as scientists focused on
subject matter area
5(No Transcript)
6(No Transcript)
7Imatinib chronology
- 1960 - Philadelphia chromosome described (P
Nowell) - 1973 Ph characterized as translocation of BCR
on chromosome 9 with ABL on chromosome 22 (J
Rowley) - 1986 BCR-ABL fusion gene characterized as
constituatively activated kinase (D Baltimore)
8Imatinib chronology
- 1988 -1995 CIBA-GEIGY develops kinase inhibitors
(A Matter, N Lydon, J Zimmermann, E Buchdunger) - 1996 B Drucker (Dana Farber -gt Oregon) screens in
ex-vivo tumors and normal lymphocytes against
compounds provided by Novartis and convinces
company to sponsor clinical trials in CML in
spite of only 5000 cases/yr in US
9- Success depended on collaboration between
industry and academia - Delayed development resulted from reluctance of
field to accept hypothesis that kinases can be
selectively inhibited or that inhibiting a single
gene could be very effective - Industry involvement dependent on vision of a
small leadership group in one company - Clinical translation dependent on vision of one
oncologist
10- Success depends on serendiptiy
- Academic medicine (NIH) is a bottom-up system not
optimized for risk taking or exploiting
scientific leads for translating basic research
to clinical products or for mounting large
cooperative programs for overcoming bottlenecks
in translation - Academic medicine is very dependent on industry
but industry has its own constraints
11(No Transcript)
12(No Transcript)
13(No Transcript)
14(No Transcript)
15(No Transcript)
16(No Transcript)
17(No Transcript)
18(No Transcript)
19(No Transcript)
20(No Transcript)
21Predictive Medicine
- Germline genetics
- GWAS
- 23andMe
- Tumor genomics
- Tumor Cell Genome Atlas
22Ioannidis et al.JNCI 102846(2010)
- 56 GWAS
- 92 statistically significant associations between
cancer phenotype and genetic variant - Median OR 1.22
- IQR OR 1.15 1.36
23(No Transcript)
24Relative Risk of Disease PrDisease PrDisease test
1.3 0.01 0.013
1.3 0.10 0.13
25- Cancers of a given histologic diagnosis are
genomically heterogeneous - Cancers are mostly caused by somatic mutations
not genetic polymorphisms - Most of the information about the disease is in
the tumor genome, not the germ-line genome
26(No Transcript)
27(No Transcript)
28Biomarkers for Early Detection
- Because of the long time between first mutation
and clinical diagnosis of human solid tumors,
there would seem to be great opportunity for
early detection
29- Phase II trials of early detection have used
samples from patients at diagnosis - Effective detection must have long lead time and
high specificity for tumors which will evolve to
be life threatening
30(No Transcript)
31Biomarkers for Informing Treatment Selection
- Prognostic biomarkers
- Measured before treatment to indicate long-term
outcome for patients untreated or receiving
standard treatment - To identify which patients have excellent
prognosis on conservative treatment - Predictive biomarkers
- Measured before treatment to identify who is
likely or unlikely to benefit from a particular
treatment
32Prognostic Markers
- Vast literature on prognostic markers
- Very few used in practice
- Most studies motivated by desire to learn about
disease biology - Broad selection of cases
- Little focus on intended use
- Little focus on analytical validation of assay
33Validation of Biomarkers
- Analytical validity
- Measures what it supposed to
- Reproducible
- Clinical validity
- Correlates with something clinically
- Clinical utility
- Is actionable
- Measuring marker leads to action that benefits
patient - Requires clarity on intended use
34If you dont know where you are going, you might
not get thereYogi Berra
35Prognostic Markers
- OncotypeDx Which patients with node negative ER
positive breast cancer who are receiving
tamoxifin will have such good prognosis that they
do not need cytotoxic chemotherapy? - Analysis focused on whether marker identifies
such a subset, not on statistical significance
36B-14 ResultsRelapse-Free Survival
Paik et al, SABCS 2003
37(No Transcript)
38Major problems with prognostic studies of gene
expression signatures
- Inadequate focus on intended use
- Cases selected based on availability of specimens
rather than for relevance to intended use - Heterogeneous sample of patients with mixed
stages and treatments. Attempt to disentangle
effects using regression modeling - Overemphasis on statistical significance and
hazard ratios. - Over-fitting data
39For pgtn problems
- Fit of a model to the same data used to develop
it is no evidence of prediction accuracy for
independent data
40(No Transcript)
41Validation of Prognostic Model
- Completely independent validation dataset
- Splitting dataset into training and testing sets
- Cross-validation
42(No Transcript)
43- Partition data set D into K equal parts
D1,D2,...,DK - First training set T1D-D1
- Develop completely specified prognostic model M1
using only data T1 - Compute prognostic score for cases in D1
- Develop model M2 using only T2 and then score
cases in D2
44- Repeat for ... TK -gt MK -gt DK
- Group patients into risk groups (e.g. 2 or more)
based on their cross-validated scores - Calculate Kaplan-Meier survival curve for each
risk-group
45(No Transcript)
46Complete cross Validation
- Cross-validation simulates the process of
separately developing a model on one set of data
and predicting for a test set of data not used in
developing the model - All aspects of the model development process must
be repeated for each loop of the cross-validation - Feature selection
- Tuning parameter optimization
47Prediction on Simulated Null DataSimon et al. J
Nat Cancer Inst 9514, 2003
- Generation of Gene Expression Profiles
- 20 specimens (Pi is the expression profile for
specimen i) - Log-ratio measurements on 6000 genes
- Pi MVN(0, I6000)
- Can we distinguish between the first 10
specimens (Class 1) and the last 10 (Class 2)? - Prediction Method
- Compound covariate predictor built from the
log-ratios of the 10 most differentially
expressed genes.
48(No Transcript)
49Cross Validation
- The cross-validated estimate of misclassification
error is an estimate of the prediction error for
the model fit applying the specified algorithm to
full dataset
50- Statistical significance of the difference in
survival among risk groups is usually not the
point - But to evaluate significance, the log-rank test
cannot be used for cross-validated Kaplan-Meier
curves because the survival times are not
independent
51- Statistical significance can be properly
evaluated by approximating the null distribution
of the cross-validated log-rank statistic - Permute the survival times and repeat the entire
cross-validation procedure to generate new
cross-validated K-M curves for low risk and high
risk groups - Compute log-rank statistic for the curves
- Repeat for many sets of permutations
52(No Transcript)
53Does measuring new biomarkers have medical
utility over using a standard staging system?
- It is usually more useful to compute K-M curves
for the high and low risk groups of the new model
within each level of risk for the model based on
standard staging variables. - That representation will enable clinicians to
determine whether the refinement in risk within
groups of patients that are more homogeneous with
regard to therapeutic decision-making is of
practical significance
54Predictive Biomarkers
- Single gene or protein measurement
- ER protein expression
- HER2 amplification
- EGFR mutation
- KRAS mutation
- V600E mutation
- ALK translocation
- Index or classifier that summarizes expression
levels of multiple genes
55The standard approach to designing phase III
clinical trials is based on two assumptions
- Qualitative treatment by subset interactions are
unlikely - Costs of over-treatment are less than costs
of under-treatment
56- Cancers of a primary site often represent a
heterogeneous group of diverse molecular diseases
which vary fundamentally with regard to - the oncogenic mutations that cause them
- their responsiveness to specific drugs
- Most new cancer drugs are very expensive
- the aspirin paradigm on which current clinical
trial dogma is based can be a roadblock to
progress
57Standard Clinical Trial Approaches
- Have led to widespread over-treatment of patients
with drugs to which few benefit - Are not scientifically well founded nor
economically sustainable for future cancer
therapeutics
58- Neither current practices of subset analysis nor
current practices of ignoring subset analysis are
effective for evaluating treatments and informing
physicians in heterogeneous diseases
59- How can we develop new drugs in a manner more
consistent with modern tumor biology and obtain
reliable information about what regimens work for
what kinds of patients?
60Develop Predictor of Response to New Drug
Using phase II data, develop predictor of
response to new drug
Patient Predicted Responsive
Patient Predicted Non-Responsive
Off Study
New Drug
Control
Targeted (Enrichment) Design
61Evaluating the Efficiency of Targeted Design
- Simon R and Maitnourim A. Evaluating the
efficiency of targeted designs for randomized
clinical trials. Clinical Cancer Research
106759-63, 2004 Correction and supplement
123229, 2006 - Maitnourim A and Simon R. On the efficiency of
targeted clinical trials. Statistics in Medicine
24329-339, 2005. - http//brb.nci.nih.gov
62- Relative efficiency of targeted design depends on
- proportion of patients test positive
- effectiveness of new drug (compared to control)
for test negative patients - When less than half of patients are test positive
and the drug has little or no benefit for test
negative patients, the targeted design requires
dramatically fewer randomized patients than the
standard design in which the marker is not used
63- Companion diagnostic test with intended use of
identifying patients who have disease subtype for
which the drug is proven effective
64Stratification Design for New Drug Development
with Companion Diagnostic
65Fallback Analysis Plan
- Compare the new drug to the control overall for
all patients ignoring the classifier. - If poverall 0.01 claim effectiveness for the
eligible population as a whole - Otherwise perform a single subset analysis
evaluating the new drug in the classifier
patients - If psubset 0.04 claim effectiveness for the
classifier patients. - The test in the subset is not dependent on
finding an overall significant finding or a
significant interaction - The trial is sized for powering both tests
- The validity of the analysis does not depend on
stratifying the randomization by the test
66(No Transcript)
67(No Transcript)
68Strong confidence in test Small r2 and large
r1 Weak confidence in test Small r2 and small
r1 p00 selected to control type I error rates
69(No Transcript)
70(No Transcript)
71(No Transcript)
72The Objectives of a Phase III Clinical Trial
- Test the strong null hypothesis that the new
treatment E is uniformly ineffective relative to
a control C while preserving the type I error of
the study - If the null hypothesis is rejected, develop an
internally validated labeling indication for
informing physicians in their decisions about
which patients they treat with the drug. - Not a hypothesis testing problem
73(No Transcript)
74The keys to developing effective drugs
- The target of the drug must be central to the
progression of the disease - Drug should be selective for the target so that
it can be administered at a concentration that
totally shuts down the de-regulated pathway - Need a test that identifies the patients who have
disease driven by de-regulation of the target
75Tumors can contain large numbers of genetic
alterations
- Passenger mutations
- Occur at rate of non-synonymous mutations
- Driver mutations
- Occur more frequently than non-synonymous
mutations and presumably have a functional role
in oncogenesis and pathogenesis of the tumor - Determined from sequencing of many tumors of a
histological type - Founder Mutations
- Originating mutation
76(No Transcript)
77Extend Previous Methods to Allow
- Background mutation rate to vary among tumors
- Background mutation rate to depend on sequence
context of mutation
78(No Transcript)
79(No Transcript)
80Founder mutations
- Mathematical modeling studies indicate that 2-4
rate-limiting events occurring at normal
mammalian mutation rates can account for
age-incidence statistics for many types of human
solid tumors - These rate limiting events may correspond to the
founder mutations of a tumor - They may be rate-limiting because they occur when
the tumor is restricted to small anatomic
compartments prior to the occurrence of genome
destabilizing mutations - They may permit the tumor to grow to a size in
which acquisition of subsequent mutations is not
rate-limiting - Additional driver mutations occur that aid tumor
invasion and metastatic dissemination
81Founder mutations may be of special importance
- They exist in all sub-clones of the tumor and so
all tumor cells may be susceptible to founder
mutation targeted treatment - Subsequent mutations develop in the context of
the founder mutations and be viable only in that
context rendering the tumors addicted to the
early mutations
82(No Transcript)
83(No Transcript)
84(No Transcript)
85(No Transcript)
86(No Transcript)
87(No Transcript)
88(No Transcript)
89Closing
- Germ-line genomics has so far had a limited
impact on individual risk prediction in oncology
and in understanding the nature of oncogenesis -
- Tumor genomics is revolutionizing our
understanding of cancer and is providing
important opportunities to identify key molecular
targets and for improving therapeutic decision
making
90Closing
- Treatment of broad populations with regimens that
do not benefit most patients is increasingly less
necessary nor economically sustainable -
- The established molecular heterogeneity of cancer
requires the use new approaches to clinical trial
design and analysis - Developments in high dimensional assays and NGS
have stimulated many areas of biostatistics and
placed greater emphasis on discovery and
prediction
91To Meet the Challenges and Opportunities
Available to Impact on Human Disease
Biostatistics Should
- Continue to broaden identity beyond probabilistic
inference and methods development - Reward statistical scientists, not just
statistical mathematicians - Embrace information technology
- Include the informatics end of bioinformatics
- Transcend hypothesis testing to include
discovery, prediction and decision making
92Joi Ito Director, MIT Media Lab
- In the old days, the world didnt change very
much, so once you became a plumber, you didnt
really need to learn that much more about
plumbing. Today you have to keep learning and
learning is somewhat of a childlike behavior. We
want the Media Lab to me more like kindergarten
and less like a lumber mill.
93- Prediction is difficult particularly the
future. - Neils Bohr
94Acknowledgements
- Boris Freidlin
- Wenyu Jiang
- Stella Karuri
- Aboubakar Maitournam
- Michael Radmacher
- Jyothi Subramanian
- Ahrim Youn
- Yingdong Zhao