Title: Dose-adaptive study designs offer benefits for proof-of-concept / Phase IIa clinical trials, as well as raise issues for continued research
1Dose-adaptive study designs offer benefits for
proof-of-concept / Phase IIa clinical trials,as
well as raise issues for continued research
- OUTLINE (Section 3) Dose-Adaptive Designs
Examples - Definition Introduction (Jim)
- Frequentist Designs, including Random Walk
Designs (Jim) - 33 Design for cancer
- UpDown Design
- Biased Coin Designs
- T-Statistic Design
- Simulations of UpDown Design for Dental Pain
Clinical Trial - Bayesian-type Designs (Inna)
- Continual Reassessment Method (CRM)
- Bayesian D-optimal Design
- Other related approaches
- Bayesian 4-parameter logistic
- Case Study (Inna)
- CytelSim Software demo
- Summary Recommendations (Inna)
- References
2Acknowledgments
- Anastasia Ivanova, UNC
- Nitin Patel, Cytel, Inc.
- Jeff Palmer, Cytel, Inc.
- Vipul Suru, Cytel Inc.
- Inna Perevozskaya, MRL
- Yevgen Tymofyeyev, MRL
- Keaven Anderson, MRL
3Dose-Adaptive DesignDefinition
- 1st patients (or groups) dose chosen based on
prior data - Each subsequent patients (or groups) dose
assigned based on pre-defined rules / algorithm - i.e., dose is adaptively assigned based on
previous responses - Rules for dose assignment developed to target
specific study objectives - Examples throughout this talk
4Dose-Adaptive DesignKey Features
- Prior response data used for sequential
allocation of dose / treatment to subsequent
(groups of) subjects - Random Walk designs only last response
- Spread observations in general region of interest
- T-statistic Design all prior responses from
current study - Concentrates observations at target dose
- Bayesian designs all prior responses from
current study - Current methods better for modeling dose-response
curve targeting particular level(s) of response
(e.g., ED50, ED90, etc.) - Limit allocation to (extreme) doses of little
interest - Maximize information gathered
- Minimize sample size
- Consistent with FDA Critical Path Initiative
5Dose-Adaptive DesignKey Requirements
- Response needs to be observable reasonably
quickly - Requires more up-front statistical work (design
simulation) - 1-2 months stat AND stat programming support
prior to protocol review - Logistical issues
- No allocation schedule requires unblinded team
to administrate - Requires ongoing communication with site
regarding allocation - Requires rapid data transfer analysis at each
stage - Duration of patient enrollment needs to be
(much?) longer than time to observed each
patients response - Results are mainly estimation-based
- Testing of hypotheses conditional on doses
assigned
6Early Phase Dose-Ranging DesignsBackground
- Typically small, sequential studies aimed at
determining the maximum tolerated dose (MTD) - Design considerations are particularly important
in cancer studies (severe side effects of
cytotoxic drugs) - Many references for dose-adaptive designs aimed
at MTD or dose that yields specific degree of
toxicity (e.g. TD30) - Current interest centers more on estimation of
efficacy - Estimate dose for a specified degree of efficacy
(e.g. ED50) - UpDown Designs inform on targeted parts of
dose-response curve - Bayesian-type Designs estimate target dose(s)
from modeling dose-response curve - Balance between individual and collective ethics
- Maximum information from the minimal number of
patients. - Maximize number of patients dosed in the range of
interest
7Single Dose-Adaptive Design can replace Typical
PoC trial and Ph.IIa Dose-Ranging Trial
Traditional Phase II Program
.
2N Patients
4N Patients
5N Patients
Phase III
PoC (Ib/IIa) (High Dose vs. Placebo)
Dose- Finding
Definitive Dose-Response (if needed)
Phase II with Dose-Adaptive PoC Trial
3N Patients
4N Patients
Phase III
Definitive Dose-Response (if needed)
PoC (High Dose vs. Placebo)
2N if futility realized
Replace 2 trials with 1?4N fewer subjects less
time N subjects / trmt group for desired
precision in PoC trial
8Single Dose-Adaptive Design can replace Typical
PoC and Ph.IIa and Ph.IIb Trials !!!
Traditional Phase II Program
.
2N Patients
4N Patients
5N Patients
Phase III
PoC (Ib/IIa) (High Dose vs. Placebo)
Dose- Finding
Definitive Dose-Response (if needed)
Phase II with Dose-Adaptive PoC Trial
3N Patients
Phase III 1 trial at Target Dose 1 Higher
dose 1 trial at Target Dose 1
Lower dose
PoC (High Dose vs. Placebo)
2N if futility realized
Replace 3 trials with 1?8N fewer subjects MUCH
less time N subjects / trmt group for
desired precision in PoC trial
9Dose-adaptive study designs offer benefits for
proof-of-concept / Phase IIa clinical trials,as
well as raise issues for continued research
- OUTLINE (Section 3) Dose-Adaptive Designs
Examples - Definition Introduction (Jim)
- Frequentist Designs, including Random Walk
Designs (Jim) - 33 Design for cancer
- UpDown Design
- Biased Coin Designs
- T-Statistic Design
- Simulations of UpDown Design for Dental Pain
Clinical Trial - Bayesian-type Designs (Inna)
- Continual Reassessment Method (CRM)
- Bayesian D-optimal Design
- Other related approaches
- Bayesian 4-parameter logistic
- Case Study (Inna)
- CytelSim Software demo
- Summary Recommendations (Inna)
- References
1033 design for cancer (Geller, 1984)Definition
- Designed under philosophy that MTD is
identifiable from the data - Patients treated in groups of 3
- Designed to screen doses quickly no estimation
involved - Probability of stopping at incorrect dose level
is higher than generally believed (Reiner,
Paoletti, OQuigley 1999)
1133 design for cancer Pros () Cons (-)
- substantially documented extensive experience
- estimates MTD at 20 toxicity level
- Weili He, et al., show how to re-estimate MTD
at intended 30 - level
- - derived estimates conditional on doses yielded
by design - - not well suited to yield efficacy info yields
little info above MTD (20-30 response level)
12UpDown DesignDefinition
- Yields distribution of doses clustered around
dose with 50 responders (ED50) - 1st subject / cohort receives dose chosen based
on prior information - Subsequent subjects / cohorts receive next lower
dose if previous subject / cohort responded, next
higher dose if no response - Data Summaries
- proportion of responders at each dose
- continuous data via summary statistics by dose
- Inference based on conditional distribution of
response given the doses yielded by the dosing
scheme - 5 MRL examples from 1980s (see back-up slides
for 1 of them)
13Biased Coin Design
- Nonparametric generalization of the UpDown
Design - UpDown Design (targets ED50)
- After Response, Probability of next lower
dose 1 - Probability of
next higher dose 0 - After non-Response, Probability of next lower
dose 0 - Probability of
next higher dose 1 - Can alter these probabilities to target a
specific level of response (EDxx) by tossing the
biased coin - Yields cluster of observed doses around EDxx
- Some patients will be assigned above and below
EDxx - informs on dose-response
- Stylianou and Flournoy, 2002
14Pros () Cons (-) of UpDown and Biased Coin
Designs
- easy to implement understand
- can be adapted so assigned doses migrate
cluster around EDxx - minimizes observations at doses too small or
too large, in comparison to randomized design - - less information on dose response away from
EDxx - can compensate by simultaneously conducting 2
or 3 clusters e.g., EDx1, EDx2, EDx3 - - derived estimates conditional on doses yielded
by design - yields information useful to design definitive
studies
15T-statistic Design - Definition 1
- Cumulative cohort design
- Goal find the dose with response level R.
- Goal of dose assignment rule assign as many
subjects as possible to a dose with mean response
R. - One dose assignment rule
- Step 1. Compute the T-Statistic comparing the
mean response at the current dose to R - T (mean-R)/SE
- Step 2.
- If T lt -1, increase the dose
- If T between -1 and 1, repeat the dose
- If T higher than 1, decrease the dose
16T-statistic Design - Definition 2
- Cumulative cohort design
- Goal find the dose with response level R.
- Goal of dose assignment rule assign as many
subjects as possible to a dose with mean response
R. - Another dose assignment rule
- Step 1. Compute the T-Statistic comparing the
mean response at ALL doses observed so far, to R - Ti (meani-R)/SE, i1,2,,D
doses - Step 2.
- Allocate next cohort to dose with Ti closest to 0
17Clinical Trial ExampleDental Pain Model
- Surgical removal of 2 3rd molars
- As anesthesia wears off, pain increases
- When pain reaches moderate or severe level (on
ordinal discrete scale 0none, 1mild,
2moderate, 3severe), patients randomized to
placebo, test drug, active control (e.g., 400 mg
ibuprofen) - Pain Relief measured serially over 8 hours post
treatment - 0none, 1some, 2a little, 3a lot, 4complete
- Primary endpoint is area under each patients
pain relief curve over 0-8 hours - 0-hour pain relief assumed to be 0
18Up Down DesignSimulated from Past Trial Results
- Single-dose, parallel group dental pain study
(total 399 patients) - 51 placebo patients
- 75 Dose 1 patients
- 76 Dose 2 patients
- 74 Dose 3 patients
- 76 Dose 4 patients
- 47 ibuprofen patients
- Primary endpoint is Total Pain Relief (AUC)
during 0-8 hours post dose (TOPAR8) - Dose-Adaptive Designs in sequential groups of 12
patients (3 placebo, 6 Test Drug Dose, 3
ibuprofen) - UpDown-type Group Design
- T-statistic Design (Definition 1)
19Simulated UpDown Designfrom completed Dental
Pain Study
- Sequential groups of 12 patients (3 placebo, 6
test drug, 3 ibuprofen) - First group receives Dose 2
- Subsequent group receives next higher dose if
previous group is non-response, next lower dose
if response - Response (both conditions satisfied)
- Mean test drug mean placebo 15 units TOPAR8
- Mean test drug mean ibuprofen gt 0
- Algorithm continues until all ibuprofen data
exhausted - originally planned precision for ibuprofen vs
placebo - (16 groups 191 total patients)
20Simulated T-Statistic Designfrom completed
Dental Pain Study
- Cumulative cohort design
- Goal find the dose with response level
- (mean test drug mean placebo) 15
units TOPAR8 - Goal of dose assignment rule assign as many
subjects as possible to target dose. - Dose assignment rule
- Step 1. Compute the T-Statistic comparing the
mean response at the current dose to R - T (mean test drug mean
pbo - 15)/SE - Step 2.
- If T lt -1, increase one dose increment
- If T between -1 and 1, repeat the dose
- If T higher than 1, decrease one dose increment
21Dental Pain Study Complete Results
22Simulated UpDown Resultsfrom Dental Pain Study
data (1st 8 Groups in sequence)
Test drug Dose mean of 3 placebo mean of 6 Test drug mean of 3 ibuprofen Resp/ Non-Resp
Dose 2 14.75 17.92 18.50 NR
Dose 3 7.33 25.42 23.42 R
Dose 2 3.58 21.75 21.42 R
Dose 1 2.50 16.42 21.75 NR
Dose 2 3.17 24.13 19.58 R
Dose 1 8.75 21.88 14.67 NR
Dose 2 6.42 19.71 23.83 NR
Dose 3 7.00 23.50 18.75 R
23Simulated UpDown Resultsfrom Dental Pain Study
data (last 8 Groups in sequence)
Test drug dose mean of 3 placebo mean of 6 Test drug mean of 3 ibuprofen Resp/ Non-Resp
Dose 2 10.75 25.21 21.75 NR
Dose 3 14.33 21.54 19.08 NR
Dose 4 0.00 24.29 11.33 R
Dose 3 0.50 21.17 18.42 R
Dose 2 4.50 26.79 18.00 R
Dose 1 2.50 5.88 16.67 NR
Dose 2 0.00 24.63 13.75 R
Dose 1 0.00 15.25 25.13 NR
24Example 1 Dental Pain Trial Data re-sampled
with Dose-Adaptive Designs
25Example 1 Dental Pain Trial Data re-sampled
with Dose-Adaptive Designs
26Remarks (1)
- T-Statistic Design more efficient than UpDown
Group Design for allocating more patients to
Target Dose - UpDown Group Design spreads patient
allocation across more doses than T-Statistic
Design - Still clusters dose allocations around Target
Dose - Choice of adaptive design depends on particular
objectives of drug research program
27Conclusions from Simulated UpDown Design in
Dental Pain
- UpDown T-Statistic designs are viable for
dose-ranging in Dental Pain - Yield similar dose-response information as
parallel group design - Can use substantially fewer patients than
parallel group design - T-Statistic Design yielded tighter concentration
of patients near target dose than UpDown - Logistics of implementation more complicated than
usual parallel group design - Can be accomplished in single center or small
number of centers
28Example 2 - Dose-Adaptive Design Simulated from
12-week Obesity Trial Results
- 12-week body weight loss (total 547 patients)
- 547 pts (11111) to placebo, Doses 1, 2, 3, 4
- Primary endpoint mean weight change _at_ 12wks
- Adaptive design applied to select sequential
subsets of data - Sequential groups of patients formed by each 3
days of patient entry - MonTuesWeds, ThursFriSat (2 groups / week)
- FPI to LPI 7.5 weeks yielded 15 sequential groups
- Group size ranged from 8 to 70
- Dose selection for each sequential group based on
2-week body weight change results of a previous
group (details next).
29Example 2 - Dose-Adaptive Design Simulated from
12-week Obesity Trial Results
- Sequential groups randomized (111) placebo
and 2 doses - selected all patients in each group assigned to
placebo and the 2 doses - Patients assigned to the 2 unselected doses
ignored - Total sample size 60 of total 531 patients
- Doses for first 6 groups pre-selected according
to balanced design - 7th group assigned 2 doses based on 2-week
response of corresponding group 1 sequence, 8th
based on response from group 2, etc. - Decrease one dose increment if prior group
responded - Increase one dose increment if prior group did
not respond - Response (UpDown) 50 of Drug-treated
patients with Week 2 BW loss 0.4 kg from
placebo mean
30Example 2 - Dose-Adaptive Design Simulated from
12-week Obesity Trial Results
- Two T-statistic Design Sets of Response Rules
- One for each active dose sequence group
- Dose with minimum
- T1(active-placebo0.4kg)/SE
- T2(active-placebo0.8kg)/SE
- Applied Definition 2 (Ti closest to 0)
31Example 2 - Dose-Adaptive Design Simulated from
12-week Obesity Trial Results
Group Mo-We Th-Sa Mo-We Th-Sa Mo-We Th-Sa Mo-We Th-Sa
1 Enter P, D2,D3 2-wk BW? Analyze
2 Enter P, D3,D4 2-wk BW? Analyze
3 Enter P, D1,D2 2-wk BW? Analyze
4 Enter P, D1,D4 2-wk BW?
5 Enter P, D2,D4
6 Enter P, D1,D3
7 UD
8 UD
Etc.
32Rationale for Adaptive Design in Obesity
- Patients plentiful and enter rapidly, thus, use
group sequential approach - To increase information about test drug yet
provide info on placebo, use 111 allocation
ratio (pboT1T2) - Need sufficient number of sequence groups to
allow design to span space of response (at least
8-20) - Simulation studies demonstrate that adequate
power achieved with 50 larger total sample size
than needed for MTD vs pbo proof-of-concept
trial - Thus, eliminating ½ the active drug patients
about right - Response definition (0.4 kg) chosen (based on
prior data) - reasonable predictor of at least a 1 kg weight
loss above placebo at Week 12 based on prior
analysis of several 12-week weight loss studies.
33Example 2 Dose-Adaptive Design Simulated from
Week 2 Mean Body Weight Change from Baseline
34Example 2 Dose-Adaptive Design Simulated from
Obesity Trial - Week 2 Number of Patients Sampled
35Example 2 Dose-Adaptive Design Simulated from
Week 12 Mean Body Weight Change from Baseline
36Example 2 Dose-Adaptive Design Simulated from
Obesity Trial - Week 12 Number of Patients Sampled
37Example 3 - Dose-Adaptive Design Simulated from
6-week Osteoarthritis Trial Results
- 6-week osteoarthritis parallel group trial
- 609 pts (122222) to placebo and 5 doses
- Primary endpoint average mean Patient Global
Assessment of Response to Therapy (PGART 0-4
Likert scale) - Adaptive design applied to select subset of data
in sequence - Sequential groups of patients formed by each week
of patient entry - FPI to LPI 18 weeks yielded 18 sequential groups
- Group size ranged from 10 to 74
- Dose selection for each sequential group based on
2-week PGART results of a previous group (lag
similar to obesity trial).
38Example 3 - Dose-Adaptive Design Simulated from
6-week Osteoarthritis Trial Results
- Sequential groups randomized (111) placebo
and 2 doses - selected all patients in each group assigned to
placebo and the 2 doses - Patients assigned to the 3 unselected doses
ignored - Total sample size lt50 of total 609 patients
- Doses for first 4 groups pre-selected according
to balanced design - 5th group assigned 2 doses based on 2-week
response of corresponding group 1 sequence, 6th
based on response from group 2, etc. - Decrease one dose increment if prior group
responded - Increase one dose increment if prior group did
not respond - Response (UpDown) 50 of active pts with
Week 2 PGART good or excellent
39Example 3 - Dose-Adaptive Design Simulated from
6-week Osteoarthritis Trial Results
- Two T-statistic Design Sets of Response Rules
- One for each active dose sequence group
- Dose with minimum
- T(active-placebo1)/SE
- T(active-placebo2)/SE
- Applied Definition 2 (Ti closest to 0)
40Example 3 - Dose-Adaptive Design Simulated from
OA Trial Week 2 Mean PGART Results
41Example 3 - Dose-Adaptive Design Simulated from
OA Trial Week 2 No. of Patients Sampled
42Example 3 - Dose-Adaptive Design Simulated from
OA Trial Mean Time-Weighted Average PGART Wks
2-6
43Example 3 - Dose-Adaptive Design Simulated from
OA Trial No. Patients Sampled Weeks 2-6
44Remarks (2)
- Logistics of implementation more complicated than
usual parallel group design - Frequent data calls / brief simple analyses
- Close contact with sites re dose assignments
- Special packaging (IVRS??)
- Drug Supply needed sufficiently for many
possibilities - Tolerability rule(s) can be added for downward
dose-assignment if pre-specified AE criteria are
encountered - This has been studied in context of Bayesian
dose-adaptive designs, but not in context of
updown designs - Number of placebo patients maintained as designed
for intended precision vs. that group could be
down-sized, though
45Simulations of UpDown Design in Dental Pain
Model
- Sequential groups of 8-16 patients enter each
study day - Each group randomized to placebo, dose X of study
drug, ibuprofen (varying from 121 to 131) - (242), (252), (262), (362), (372),
(382), (392), (393), (3103) - Initial group receives dose 3 from doses 1-7
- Subsequent groups receive dose based on of
active dose responses in previous group - Up 2 dose increments if lt20 responders
- Up 1 dose increment if 20-39 responders
- Repeat same dose if 40-60 responders
- Down 1 dose increment if 61-80 responders
- Down 2 dose increments if gt80 responders
- Response TOPAR8 gt 15 units more than group
placebo mean - Study stops when total sample size exceeds 140
46Simulations of UpDown Design in Dental Pain Model
- 7 true underlying dose-response curves (SD9)
- Assumed true underlying mean
TOPAR8 - Distribution pbo dose1 dose2 dose3 dose4
dose5 dose6 dose7 - 0no effect 5 5 5 5
5 5 5 5 - 1lweak 5 6 8 10
10 10 10 10 - 1r 5 5 5 5
5 6 7 10 - 2lmodest 5 7 9 13
13 13 13 13 - 2r 5 5 5 5
7 9 11 13 - 3libuprofen 5 8 12 16
19 19 19 19 - 3r 5 5 5 5
8 12 15 19 - 5000 simulations of updown design for each
distribution - Computed mean TOPAR8 at each dose for each
simulation of the updown design (total sample
size 140 patients)
47Distributions of Simulation Results (observed
TOPAR8 means by dose) for Left-Shifted True
Distributions
75th percentile
median
25th percentile
True values connected circles
48Distributions of Simulation Results for
Right-Shifted True Distributions
49Distributions of Simulation Results for
No-effect True Distribution
Each pair right true distribution
Each pair left simulation results
95th percentile
75th percentile
median
25th percentile
5th percentile
33 lt1 lt1 7 1 7 2 70 ? average
sample sizes
50Distributions of Simulation Results fromTrue
Distributions 0, 1l, 2l, 3l
51Distributions of Simulation Results fromTrue
Distributions 0, 1r, 2r, 3r
52Power of UpDown Designsfrom Simulations
Power to reject null hypothesis of zero slope
across doses for average sample sizes 33 for
placebo, 87 on active drug
Distribution / Design Distn 0 Diff0 Distn 1 Diff5 Distn 2 Diff8 Distn 3 Diff14
Right-shifted / UpDown 5 84 gt99 gt99
Left-shifted / UpDown 5 81 99 gt99
/ Randomized 5 77 99 gt99
53Dose-adaptive study designs offer benefits for
proof-of-concept / Phase IIa clinical trials,as
well as raise issues for continued research
- OUTLINE (Section 3) Dose-Adaptive Designs
Examples - Definition Introduction (Jim)
- Frequentist Designs, including Random Walk
Designs (Jim) - 33 Design for cancer
- UpDown Design
- Biased Coin Designs
- T-Statistic Design
- Simulations of UpDown Design for Dental Pain
Clinical Trial - Bayesian-type Designs (Inna)
- Continual Reassessment Method (CRM)
- Bayesian D-optimal Design
- Other related approaches
- Bayesian 4-parameter logistic
- Case Study (Inna)
- CytelSim Software demo
- Summary Recommendations (Inna)
- References
54Logistics for Conduct of a Dose-Adaptive
Designed Trial
- Response observable reasonably quickly
- Increased statistical computations / simulations
to justify dose-adaptive scheme in protocol - Need on-call person to assess previous response
data and generate dose for next subject - For model-based dose-adaptive designs, need
on-call unblinded statistician for associated
analyses - OR, this could be automated via web-based
interface (increases cost) - Rapid transfer of needed data
- Need special packaging or unblinded pharmacist at
site to package selected dose for each patient
55More Remarks
- Logistics of implementation more complicated than
usual parallel group design - Frequent data calls / brief simple analyses
- Close contact with sites re dose assignments
- Special packaging (IVRS??)
- Drug Supply needed sufficiently for many
possibilities - Tolerability rule(s) can be added for downward
dose-assignment if pre-specified AE criteria are
encountered - This has been studied in context of Bayesian
dose-adaptive designs, but not in context of
updown designs - Number of placebo patients maintained as designed
for intended precision vs. that group could be
down-sized, though
56Dose-Adaptive DesignSummary
- Allocation of dose for next subject based on
response(s) of previous subject(s) - Random Walk designs only last subjects response
- T-statistic (frequentist) designs all previous
subjects responses - Bayesian-type designs all previous subjects
responses - High potential to limit subject allocation to
doses of little interest (too high / too low) - Maximize information gathered from fixed N
- Ethical advantage over fixed randomization
- More attractive to patients / subjects
- Inference conditional on doses assigned by
design, but not overly important in early
development - Requires more statistical up-front work
(simulation) - No pre-specified allocation schedule requires
ongoing communication with site regarding
allocation
57Dose-Adaptive Design Recommendations
- Bayesian-type designs preferable to estimate
dose-response curve can also estimate a
dose-response quantile of interest (e.g., EDxx)
or (part of?) region of increasing dose-response - Complex heavy computations
- Random Walk T-statistic Designs focus on
quantile(s) of interest - Easy to understand program
- Consider as starting point for implementing
dose-adaptive design - Let other design features guide towards other
adaptive techniques based on particular
experimental situation - Ongoing incomplete simulations have yet to
identify major advantage of Bayesian-type designs
over RW T, unless prior information is
important to consider. - Study, comparison, refinement of these
dose-adaptive designs continues
58Conclusions
- Adaptive dose-finding designs are viable for PoC
and dose-ranging in a single study - Yield similar, maybe better, dose-response
information as larger parallel group designs - Likely to use substantially fewer patients than
parallel group design - Likely to save development time in Phase II
- Logistic details need to be workable
- Response observable reasonably quickly relative
to patient entry - Dynamic allocation issues
- Drug Supply Labeling more complicated
59References
- Rosenberger WF, Haines LM. Competing designs for
phase I clinical trials a review. Statistics in
Medicine. 2002 212757-2770 - Haines LM, Perevozskaya I, Rosenberger WF.
Bayesian optimal designs for phase I clinical
trials. Biometrics. 2003 59561-600 (Bayesian
D-optimal Design) - Stylianou M and Flournoy N. Dose Finding Using
the Biased Coin Up-and-Down Design and Isotonic
Regression. Biometrics. 2002 58171-177.
(Biased Coin and UpDown Designs) - Krams M, Lees KR, Hacke W, et al. Acute Stroke
Therapy by Inhibition of Neutrophils (ASTIN), an
adaptive dose-response study of UK-279,276 in
Acute Ischemic Stroke. Stroke. 2003
342543-2548 - Dougherty TB, Porche VH, Thall PF. Maximum
tolerated dose of nalmefene in patients receiving
epidural fentanyl and dilute bupivacaine for
postoperative analgesia. Anesthesiology, 2000
92(4)1010-1016. - Bolognese JA. A monte carlo comparison of three
up-and-down designs for dose ranging. Controlled
Clinical Trials, 1983 4187-196. (UpDown
Designs) - Bolognese JA, Gomez HJ, Tobert JA, Rucinska EJ.
The up-and-down design for dose ranging and its
use in various clinical settings. ASCPT,
Atlanta, 1984. - Ivanova, A., Bolognese, J. and Perevozskaya, I.
(2008). Adaptive design based on t-statistic for
dose-response trials. Statistics in Medicine, 27,
1581-1592. (T-statistic Design) - He W, Liu J, Bincowitz B, Quan, H. A model-based
approach in the estimation of maximum tolerated
dose in phase I cancer clinical trials. MRL
Technical report 102, June 2004. (33 Design)
60References (cont.)
- Geller NL. Design of phase I and II clinical
trials in cancer a statisticians view. Cancer
Investigations 1984 2483491. (33 design in
cancer) - Durham SD, Flournoy N. Random walks for quantile
estimation. In Statistical Decision Theory and
Related Topics, Gupta SS, Berger JO (eds).
Springer New York, 1994 467476. (RWR) - OQuigley J, Pepe M, Fisher L. Continual
reassessment method a practical design for phase
I clinical trials in cancer. Biometrics 1990
463348. (CRM) - Babb J, Rogatko A, Zacks S. Cancer phase I
clinical trials efficient dose escalation with
overdose control. Statistics in Medicine 1998
171103 1120. (EWOC) - Whitehead J, Brunier H. Bayesian decision
procedures for dose determining experiments.
Statistics in Medicine1995 14885893 (
decision-theoretic approach) - Dragalin V., Fedorov V. Adaptive model-based
designs for dose-finding studies. To appear in
Journal of Statistical Planning and Inference,
2005 (bivariate models) - Reiner E, Paoletti X, OQuigley J. Operating
characteristics of the standard phase I clinical
trial design. Computational Statistics and Data
Analysis 1999 30303 315. - Atkinson AC, Donev AN. Optimum Experimental
Designs. Oxford Clarendon, 1992.
61 Dose-adaptive study designs offer benefits for
proof-of-concept / Phase IIa clinical trials,
as well as raise issues for continued research
- QUESTIONS
- / COMMENTS
- / DISCUSSION
62 - BACK-UP SLIDES FOLLOW THIS ONE
63UpDown DesignMRL (circa 1982) Example
- 10 subjects received a single dose from among
doses 1-6 in each of 3 periods - Response 50 inhibition of isoproterenol-induce
d tachycardia (Period 1) 75 during periods 2
and 3, because Period 1 worked so well - Subject 1, Period 1 received dose 1
- Subsequent subjects in Period 1
- Up 1 dose level after no response
- Down 1 dose level after response
- Dose range bounded by Dose 1 and Dose 6
- Periods 2 3 followed same rules within Subject
64UpDown DesignMRL (1982) Example - Results
Note minimized obs at sub-effective doses
65UpDown DesignMRL (1982) Example - Conclusions
- Some evidence of response exists beginning with
Dose 3 - Period 1 info sufficient to move to higher
level of response - Magnitude of response does not quite reach 75
inhibition - From dose response data, response appears to
plateau Dose 3 - Observations at ineffective doses minimized
- Formal inferences limited, but yielded info
sufficient to identify doses for future
definitive trials
66Rationale for UpDown Designin Dental Pain
- Patients plentiful and enter rapidly ? group
sequential approach - Need both active and placebo control treatments
- ? 3 treatments per sequence group (pbo, test,
comparator) - To increase information in test trmt group yet
provide information on placebo and comparator,
use 121 allocation ratio - Need sufficient number of sequence groups for
design to span space of response (8-20) - Example trial has 47 comparator patients
therefore, aim for 16 groups of 12 (363) - Last group will have only 2 comparator patients
to avoid re-sampling - Response definition derives from meta-analyses of
analgesia studies - 1. Mean vs pbo 13 units TOPAR8
- 2. Mean vs pbo 15 units TOPAR8
67Biased Coin Design (a Random Walk Rule)
- Nonparametric generalization of the up-and-down
design - Creates unimodal distribution of doses around the
dose with targeted level of response - Consequently, some patients will be assigned
above and below the dose for targeted response
level
68Bayesian-type Dose-Adaptive DesignContinual
Reassessment Method (CRM)
Choose initial estimate of response
distribution choose initial dose
Update Dose Response Model estimate Prob.
(Resp.) _at_ each dose
Obtain next Patients Observation
Next Pt. Dose Dose w/ Prob. (Resp.) Closest
to Target level
Stop. EDxx Dose w/ Prob. (Resp.) Closest
to Target level
Max N Reached?
no
yes
69Distributions of Simulation Results fromTrue
Distribution 1l
Avg sample sizes ?33 lt1 lt1 8 3 7 6
67
70Distributions of Simulation Results fromTrue
Distribution 2l
Avg sample sizes ? 33 lt1 1 10 5 9
9 53
71Distributions of Simulation Results fromTrue
Distribution 3l
Avg sample sizes ? 33 1 5 18 15 17
12 18
72Distributions of Simulation Results fromTrue
Distribution 1r
Avg sample sizes ? 33 lt1 lt1 7 1 7
4 67
73Distributions of Simulation Results fromTrue
Distribution 2r
Avg sample sizes ? 33 lt1 lt1 7 1 9
7 62
74Distributions of Simulation Results fromTrue
Distribution 3r
Avg sample sizes ? 33 lt1 lt1 8 3 15 17
45
75Average Sample Size Assigned by UpDown Design to
Each Dose
Distn Dose1 Dose2 Dose3 Dose4 Dose5 Dose6 Dose7
0 lt1 lt1 7 1 7 2 70
1r lt1 lt1 7 1 7 4 67
2r lt1 lt1 7 1 9 7 62
3r lt1 lt1 8 3 15 17 45
1l lt1 lt1 8 3 7 6 67
2l lt1 1 10 5 9 9 53
3l 1 5 18 15 17 12 18