Transcript H - 44e Journées de Statistique

The Role of Statistical Methodology in
Clinical Research –
Shaping and Influencing Decision Making
Frank Bretz
Global Head – Statistical Methodology, Novartis
Adjunct Professor – Hannover Medical School, Germany
Joint work with Holger Dette & Björn Bornkamp; Willi Maurer & Martin Posch
44e Journées de Statistique – 21 au 25 mai 2012, Bruxelles
Drug development ...
 ... is the entire process of bringing a new drug to the market
 ... costs between USD 500 million to 2 billion to bring a new
drug to market, depending on the therapy
 ... is performed at various stages taking 12-15 years, where
out of 10’000 compounds only 1 makes it to the market
• drug discovery [10’000 compounds]
• pre-clinical research on animals [250]
• clinical trials on humans [10]
• market authorization [1]
2
| JDS | Frank Bretz | May 25, 2011
Drug development process
3
| JDS | Frank Bretz | May 25, 2011
Four clinical development phases
Phase
Number of Length
Study
subjects
per study population
per study
Aim
I
6 – 20
Weeks –
months
Healthy
Volunteers
Pharmacokinetics & -dynamics;
single & multiple ascending
dose studies; bioavailability
50 – 200
Months
Patients
(narrow
population)
Proof-of-concept; dose and
regimen finding; exploratory
studies
200 –
10’000
Years
Patients
(broad
population)
Confirmatory, pivotal studies
1’000 –
1’000’000
Decades
Market
New label claims & extensions;
publication studies; health
economics; pharmacovigilance
First in
human
II
First in
patients
III
Submission
IV
Post
marketing
4
| JDS | Frank Bretz | May 25, 2011
Why do we need statisticians in the
pharmaceutical industry?
Remember, one way of defining Statistics is ...
The science of quantifying uncertainty,
Dealing with uncertainty,
And making decisions in the face of uncertainty.
... and drug development is
a series of decisions under huge uncertainty !
5
| JDS | Frank Bretz | May 25, 2011
Strategic Role of Statisticians
 Decision making in drug development
• Integrated synthesized thinking, bringing together key information,
internal and external to the drug, to influence program and study design
 Optimal clinical study design
• Specify probabilistic decision rules and provide operating characteristics
to illustrate performance as parameters change
 Exploratory Data Analysis
• Take a strong supporting role in exploring and interpreting the data
 Submission planning and preparation
• Be integrally involved in the submission strategy, building the plans,
interpreting and exploring accumulating data to provide input to a robust
and well-thought through dossier
6
| JDS | Frank Bretz | May 25, 2011
Examples
7
| JDS | Frank Bretz | May 25, 2011
Four clinical development phases
Phase
Number of Length
Study
subjects
per study population
per study
Aim
I
6 – 20
Weeks –
months
Healthy
Volunteers
Pharmacokinetics & -dynamics;
single & multiple ascending
dose studies; bioavailability
50 – 200
Months
Patients
(narrow
population)
Proof-of-concept; dose and
regimen finding; exploratory
studies
200 –
10’000
Years
Patients
(broad
population)
Confirmatory, pivotal studies
1’000 –
1’000’000
Years
Market
New label claims & extensions;
publication studies; health
economics; pharmacovigilance
First in
human
II
First in
patients
III
Submission
IV
Post
marketing
8
| JDS | Frank Bretz | May 25, 2011
Example 1
Adaptive Dose Finding
9
| JDS | Frank Bretz | May 25, 2011
Notation and framework
10 | JDS | Frank Bretz | May 25, 2011
Notation and framework
11 | JDS | Frank Bretz | May 25, 2011
Optimal design for MED estimation
12 | JDS | Frank Bretz | May 25, 2011
Optimal design for MED estimation
13 | JDS | Frank Bretz | May 25, 2011
Adaptive Design for MED estimation
14 | JDS | Frank Bretz | May 25, 2011
Priors for parameters
15 | JDS | Frank Bretz | May 25, 2011
Procedure: 1) Before Trial Start
16 | JDS | Frank Bretz | May 25, 2011
Procedure: 2a) At Interim
17 | JDS | Frank Bretz | May 25, 2011
Procedure: 2b) At Interim
18 | JDS | Frank Bretz | May 25, 2011
Procedure: 3) At Trial End
19 | JDS | Frank Bretz | May 25, 2011
Example 2
Multiple testing problems
20 | JDS | Frank Bretz | May 25, 2011
Scope of multiplicity in clincial trials
 Wealth of information assessed per patient
• Background / medical history (including prognostic factors)
• Outcome measures assessed repeatedly in time: efficacy, safety, QoL, ...
• Concomitant factors: Concomitant medication and diseases, compliance, ...
 Additional information and objectives, which further complicate
the multiplicity problem
• Multiple doses or modes of administration of a new treatment
• Subgroup analyses looking for differential effects in various populations
• Combined non-inferiority and superiority testing
• Interim analyses and adaptive designs
• ...
21 | JDS | Frank Bretz | May 25, 2011
Impact of multiplicity on Type I error rate
Probability to commit at least one Type I error when performing m
independent hypotheses tests (= FWER, familywise error rate)
22 | JDS | Frank Bretz | May 25, 2011
Impact of multiplicity on treatment effect estimation
Distribution of the maximum of mean estimates from m independent
treatment groups with mean 0 (normal distribution)
23 | JDS | Frank Bretz | May 25, 2011
Phase III development of a new diabetes drug
 Structured family of hypotheses with two levels of multiplicity
1. Clinical study with three treatment groups
• placebo, low and high dose
• compare each of the two active doses with placebo
2. Two hierarchically ordered endpoints
• HbA1c (primary objective) and body weight (secondary objective)
 Total of four structured hypotheses Hi
H1: comparison of low dose vs. placebo for HbA1c
H2: comparison of high dose vs. placebo for HbA1c
H3: comparison of low dose vs. placebo for body weight
H4: comparison of high dose vs. placebo for body weight
 In clinical practice often even more levels of multiplicity
24 | JDS | Frank Bretz | May 25, 2011
How to construct decision strategies that
reflect complex clinical constraints?
25 | JDS | Frank Bretz | May 25, 2011
Basic idea
 Hypotheses H1, ..., Hk
 Initial allocation of the significance level α = α1 + ... + αk
 P-values p1, ..., pk
α-propagation
If a hypothesis Hi can be rejected at level αi, i.e. pi ≤ αi,
reallocate its level αi to other hypotheses (according to a
prefixed rule) and repeat the testing with the updated
significance levels.
26 | JDS | Frank Bretz | May 25, 2011
Bonferroni-Holm test (k = 2)
27 | JDS | Frank Bretz | May 25, 2011
Bonferroni-Holm test (k = 2)
Example with α = 0.05
28 | JDS | Frank Bretz | May 25, 2011
Bonferroni-Holm test (k = 2)
Example with α = 0.05
29 | JDS | Frank Bretz | May 25, 2011
Bonferroni-Holm test (k = 2)
Example with α = 0.05
30 | JDS | Frank Bretz | May 25, 2011
Bonferroni-Holm test (k = 2)
Example with α = 0.05
31 | JDS | Frank Bretz | May 25, 2011
Bonferroni-Holm test (k = 2)
Example with α = 0.05
32 | JDS | Frank Bretz | May 25, 2011
General definition
33 | JDS | Frank Bretz | May 25, 2011
Graphical test procedure
34 | JDS | Frank Bretz | May 25, 2011
Main result
35 | JDS | Frank Bretz | May 25, 2011
Example re-visited
 Two primary hypotheses H1 and H2
• Low and high dose compared with placebo for primary endpoint (HbA1c)
 Two secondary hypotheses H3 and H4
• Low and high dose for secondary endpoint (body weight)
 Proposed graph on next slide
• reflects trial objectives, controls Type I error rate, and displays possible
decision paths
• can be finetuned to reflect additional clinical considerations or treatment
effect assumptions
36 | JDS | Frank Bretz | May 25, 2011
Resulting test procedure
37 | JDS | Frank Bretz | May 25, 2011
Resulting test procedure
38 | JDS | Frank Bretz | May 25, 2011
Resulting test procedure
39 | JDS | Frank Bretz | May 25, 2011
Resulting test procedure
40 | JDS | Frank Bretz | May 25, 2011
Resulting test procedure
41 | JDS | Frank Bretz | May 25, 2011
Resulting test procedure
42 | JDS | Frank Bretz | May 25, 2011
Resulting test procedure
43 | JDS | Frank Bretz | May 25, 2011
Resulting test procedure
44 | JDS | Frank Bretz | May 25, 2011
Now and future
 In addition to building and driving innovation internally,
important to leverage strengths externally at the scientific
interface between industry, academia, and regulatory agencies
 At its best, cross-collaboration is greater than the sum of the
individual contributions
• Synergy on different perspectives and strengths
 Provides opportunity to more deeply embed change throughout
industry and to have greater acceptance by stakeholders
An exciting time to be a statistician !
45 | JDS | Frank Bretz | May 25, 2011
Selected References
 Bornkamp, B., Bretz, F., and Dette, H. (2011) Response-adaptive dose-finding under model uncertainty.
Annals of Applied Statistics (in press)
 Bretz, F., Maurer, W., and Hommel, G. (2011) Test and power considerations for multiple endpoint analyses
using sequentially rejective graphical procedures. Statistics in Medicine (in press)
 Maurer, W., Glimm, E., and Bretz, F. (2011) Multiple and repeated testing of primary, co-primary and
secondary hypotheses. Statistics in Biopharmaceutical Research (in press)
 Dette, H., Kiss, C., Bevanda, M., and Bretz, F. (2010) Optimal designs for the Emax, log-linear and
exponential models. Biometrika 97, 513-518.
 Bretz, F., Dette, H., and Pinheiro, J. (2010) Practical considerations for optimal designs in clinical dose
finding studies. Statistics in Medicine 29, 731-742.
 Dragalin, V., Bornkamp, B., Bretz, F., Miller, F., Padmanabhan, S.K., Patel, N., Perevozskaya, I.,
Pinheiro, J., and Smith, J.R. (2010) A simulation study to compare new adaptive dose-ranging designs.
Statistics in Biopharmaceutical Research 2(4), 487-512.
 Bretz, F., Maurer, W., Brannath, W., and Posch, M. (2009) A graphical approach to sequentially rejective
multiple test procedures. Statistics in Medicine 28(4), 586-604.
 Dette, H., Bretz, F., Pepelyshev, A., and Pinheiro, J.C. (2008) Optimal designs for dose finding studies.
Journal of the American Statistical Association 103(483), 1225-1237.
 Bretz, F., Pinheiro, J.C., and Branson, M. (2005) Combining multiple comparisons and modeling
techniques in dose-response studies. Biometrics, 61(3), 738-748.
46 | JDS | Frank Bretz | May 25, 2011