Prior Elicitation in Bayesian Clinical Trial Design

Download Report

Transcript Prior Elicitation in Bayesian Clinical Trial Design

Prior Elicitation in Bayesian Clinical Trial Design
Peter F. Thall
Biostatistics Department
M.D. Anderson Cancer Center
SAMSI intensive summer research program on
Semiparametric Bayesian Inference: Applications in
Pharmacokinetics and Pharmacodynamics
Research Triangle Park, North Carolina
July 13, 2010
Disclaimer
To my knowledge, this talk has nothing to do
with semiparametric Bayesian inference,
pharmacokinetics, or pharmacodynamics.
I am presenting this at Peter Mueller’s behest.
Blame Him!
Outline ( As time permits )
1. Clinical trials: Everything you need to know
2. Eliciting Dirichlet parameters for a leukemia trial
3. Prior effective sample size
4. Eliciting logistic regression model parameters for
Pr(Toxicity | dose)
5. Eliciting values for a 6-parameter model of
Pr(Toxicity | dose1, dose2)
6. Penalized least squares for {Pr(Efficacy),Pr(Toxicity)}
7. Eliciting a hyperprior for a sarcoma trial
8. Eliciting two priors for a brain tumor trial
9. Partially informative priors for patient-specific dose finding
Clinical Trials
Definition: A clinical trial is a scientific experiment with
human subjects.
1. Its first purpose is to treat the patients in the trial.
2. Its second purpose is to collect information that may be
useful to evaluate existing treatments or develop new,
better treatments to benefit future patients.
Other, related purposes of clinical trials:
3. Generate data for research papers
4. Obtain $$ financial support $$ from pharmaceutical
companies or governmental agencies
5. Provide an empirical basis for drug or device approval
from regulatory agencies such as the US FDA
Medical Treatments
Most medical treatments, especially drugs or drug
combinations, have multiple effects.
Desirable effects are called efficacy
► Shrinkage of a solid tumor by > 50%
► Complete remission of leukemia
► Dissolving a cerebral blood clot that caused an ischemic stroke
► Engraftment of an allogeneic (matched donor) stem cell transplant
Undesirable effects are called toxicity
► Permanent damage to internal organs (liver, kidneys, heart, brain)
► Immunosuppression (low white blood cell count or platelet count)
► Cerebral bleeding or edema (accumulation of fluid)
► Graft-versus-host disease (the engrafted donor cells attack the
patient’s organs)
► Regimen-related death due to any of the above
Scientific Method
Advice from Ronald Fisher
Don’t waste information
Advice From Peter Thall
Don’t waste prior information when
designing a clinical trial
Standard Statistical Practice
Ignore Fisher’s advice and just run your favorite
statistical software package. And be sure to
record lots and lots of p-values.
A Chemotherapy Trial in Acute Leukemia
Complete Remission (CR)
T
O
X
I
C
I
T
Y
Yes
No
Yes
q1
q2
No
q3
q4
q4 = 1 – q1 – q2 – q3
Model: q = (q1, q2 , q3 , q4 ) ~ Dirichlet(a1, a2, a3, a4) ≡ Dir(a)
p(q | a) ∝ q1a1-1 q2a2-1 q3a3-1 q4 a4-1, a+ = a1+a2+a3+a4 = ESS
pTOX = q1 + q2 ~Be(a1+a2, a3+a4)
pCR = q1 + q3 ~ Be(a1+a3, a2+a4) 
E(pTOX) = (a1+a2 )/a+ E(pCR) = (a1+a3 )/a+
If possible, use Historical Data to establish a prior:
CR and Toxicity counts from 264 AML Patients Treated
With an Anthracycline + ara-C
Toxicity
No
Toxicity
CR
No CR
73
(27.7%)
101
(38.3%)
63 (23.9%)
174
(65.9%)
90
(34.1%)
P(CR | Tox) = 73/136
P(CR | No Tox) = 101/128
27
(10.2%)
= .54
= .79
136
(51.5%)
128
(48.5%)
264
CR and Tox are
Not Independent
Dirichlet Priors and Stopping Rules
S = “Standard” treatment
E = “Experimental” treatment
qS ~ Dir (73,63,101,27)  aS,+ = ESS = 264 (“Informative”)
Set mE = mS with aE,+ = 4

qE ~ Dirichlet (1.11, .955, 1.53, .409)
(“Non-Informative”)
Stop the trial if
1) Pr(qS,CR + .15 < qE,CR | data) < .025
(“futility”), or
2) Pr(qS,TOX + .05 < qE,TOX | data) > .95
(“safety”)
But what if you don’t have historical data?!!
An Easy Solution: To obtain the prior on qS
1) Elicit the prior marginal outcome probability means
E(pTOX) = (a1+a2 )/a+ and E(pCR) = (a1+a3 )/a+
2) Assume independence and solve algebraically for
(m1, m2, m3, m4) = (a1, a2, a3, a4)/ a+
3) Elicit the effective sample size ESS = a+ that the elicited
values E(pTOX) and E(pCR) were based on
4) Solve for (a1, a2, a3, a4)
Sensitivity Analysis of Association in the desirable case where
Pr(CR) ↑ 0.15 from .659 to .809 and Pr(TOX) = .516
i.e. there is no increase in toxicity.
p11p00
p10p01
True qE
.007
(.027,.489,.782,.102)
.138
Probability of Stopping Sample Size
the Trial Early
(25%,50%,75%)
>.99
4 7 14
(.227,.289, .582,.102)
Oops!!
.56
14 44 56
1.28
(.427,.089,.382,.102)
.16
56 56 56
52.6
(.510,.006,.299,.185)
.16
56 56 56
If you don’t have historical data . . .
A slightly smarter way to obtain prior(qS) :
1) Elicit the prior means
E(pTOX) = (a1+a2 )/a+ and
E(pCR) = (a1+a3 )/a+
2) Elicit the prior mean of a conditional probability, like
Pr(CR | Tox) = q1/(q1 + q2), which has mean a1/(a1 + a2), and
solve for (m1, m2, m3, m4) = (a1, a2, a3, a4)/ a+ . That is, do not
assume independence.
3) Elicit the effective sample size ESS = a+ that the values
E(pTOX) and E(pCR) were based on
4) Solve for (a1, a2, a3, a4)
Rocket
Science!!
Example
Elicited prior values
E(pTOX) = (a1+a2 )/a+ = .30
E(pCR) = (a1+a3 )/a+
= .50
E{ Pr(CR | Tox) } = E{ q1/(q1 + q2)} = a1/(a1 + a2) = .40
ESS = a+ = 120 
(a1, a2, a3, a4) = (14.4, 21.6, 45.6, 38.4)
(m1, m2, m3, m4) = (a1, a2, a3, a4)/ a+ = (.12, .18, .38, .32)
Determining the Effective Sample Size of a
Parametric Prior (Morita, Thall and Mueller, 2008)
A Fundamental question in Bayesian analysis:
How much information is contained in the
prior?
Prior
p(θ)
The answer is straightforward for many
commonly used models
E.g. for beta distributions
Be (16,19)
ESS = 16+19
= 35
Be (3,8)
Be (1.5,2.5)
ESS = 3+8 = 11
ESS = 1.5+12.5 = 5
But for many commonly used parametric
Bayesian models it is not obvious how to
determine the ESS of the prior.
E.g. usual normal linear regression model
E (Y ) =  0 + 1 X , Var (Y ) = 
2
( 0 , 1 ) ~ bivariate
2
q = ( 0 , 1 ,  2 )
normal,  ~ inverse 
2
Intuitive Motivation
Saying Be(a, b) has ESS = a+b implicitly refers to the wel
known fact that
θ ~ Be(a, b) and Y | θ ~ binom(n, θ) 
θ | Y,n 〜 Be(a +Y, b +n-Y) which has ESS = a + b + n
So, saying Be(a,b) has ESS = a + b implictly refers to an
earlier Be(c,d) prior with very small c+d = e, and solving
for m = a+b – (c+d) = a+b – e for a very small e > 0
General Approach
1) Construct an “e-information” prior q0(θ), with same
means and corrs. as p(θ) but inflated variances
2) For each possible ESS m = 1, 2, ... consider a
sample Ym of size m
3) Compute posterior qm(θ|Ym) starting with prior q0(θ)
4) Compute the distance between qm(θ|Ym) and p(θ)
5) The interpolated value of m minimizing the distance
is the ESS.
A Phase I Trial to Find a Safe Dose for
Advanced Renal Cell Cancer (RCC)
Patients with renal cell cancer, progressive after treatment
with Interferon
Treatment = Fixed dose of 5-FU + one of 6 doses of
Gemcitabine: {100, 200, 300, 400, 500, 600} mg/m2
Toxicity = Grade 3,4 diarrhea, mucositis, or hematologic
(blood) toxicity
Nmax = 36 patients, treated in cohorts of 3
Start the with1st cohort treated at 200 mg/m2
Adaptively pick a “best” dose for each cohort
Continual Reassessment Method (CRM, O’Quigley et al.
1990) with a Bayesian Logistic Regression Model
1) Specify a model for p(xj,q) = Pr(Toxicity| q, dose xj)
and prior on q
2) Physician specifies pTOX* = a target Pr(Toxicity)
3) Treat each successive cohort of 3 pats. at the “best”
dose for which E[p(xj,q) | data] is closest to pTOX*
4) The best dose at the end of the trial is selected
p(xj,q) =
exp( m+ xj )
1 + exp( m+ xj )

using xj = log(dj) - {S j=1,…k log(dj)}/k ,
Prior:
m ~ N(nm, m2),
 ~ N(n, 2)
q = (m, )
j=1,…,6.
CRM with Bayesian Logistic Regression Model
Elicit the mean toxicity probabilities at two doses.
In the RCC trial, the elicited prior values were
E{p(200, q)} = .25 and E{p(500,q)} = .75
1) Solve algebraically for nm = -.13 and n = 2.40
2) m=  = 2  m ~ N(-.13, 4),  ~N(2.40, 4)
which gives prior ESS = 2.3
Alternatively, one may specify the prior ESS and solve for
 = m = 
Plot of ESS as a function of 
For cohorts of size 1 to 3,
 =1 is still too small since it
gives prior ESS = 9.3
ESS{ p(m,)| }

0.1
0.2
0.3
0.4
0.5
ESS 928 232
103
58.0
37.1 18.9
These  values give a
prior with far more
information than the data
in a typical phase I trial.
0.7
1
2
3
9.3
2.3
1.0
4
5
10
0.58 0.37 0.09
These ESS values are
OK, so  = 2 to 5 is OK.
?

Prior of p = Prob(tox | d = 200, )
p(p |)
ESS=928
 =0.1
ESS=37.1
 =0.5
ESS=0.09
ESS=2.3
 =2.0
 =10.0
p
Why not just set m=  = a very large number, so ESS = a
tiny number, and have a very “non-informative” prior ?
Example: A “non-informative” prior is m ~ N(-.13,100) and
 ~ N(2.40,100), i.e.  =10.0  ESS = 0.09.
But this prior has some very undesirable properties :
Prior Probabilities
of Extreme Values
Dose of Gemcitabine (mg/m2)
100
200
300
400
500
600
Pr{p(x,q)<.01}
.45
.37
.33
.31
.31
.31
Pr{p(x,q)>.99}
.30
.30
.32
.35
.38
.40
This says you believe, a priori, that
1)
Pr{p(x,q) < .01} = Prob(toxicity is virtually impossible) =
.31 to .45
2) Pr{p(x,q) > .99} = Prob(toxicity is virtually certain) =
.30 to .40

Making  =m=  too large (a so-called “non
informative” prior) gives a pathological prior.

What  is “too large” numerically is not obvious without
computing the corresponding ESS.
Dose-Finding With Two Agents
(Thall, Millikan, Mueller, Lee, 2003)
Study two agents used together in a phase I clinical trial, with
dose-finding based on p(x,q) = probability of toxicity for a
patient given the dose pair x = (x1, x2)
Find one or more dose pairs (x1, x2) of the two agents used
together for future clinical use and/or study in a randomized
phase II trial
Elicit prior information on p(x,q) with each agent used alone
Single Agent Toxicity Probabilities :
p1 (x1,q1) = p(x1,0, q) = Prob(Toxicity | x1, x2=0, q1)
p2 (x2,q2) = p(0, x2, q) = Prob(Toxicity | x1=0, x2, q2)
Hypothetical Dose-Toxicity Surface
80
70
60
50
40
30
20
900
10
600
1,400
1,000
1,200
Cyclophosphamide
800
600
400
400
200
0
0
P(tox)
0
Gemcitabine
Probability Model
x2=0  p1 (x1,q1) = a1 x11 / ( 1 + a1 x11 ) = exp(h1)/{1+exp(h1)}
x1=0  p2 (x2,q2) = a2 x22 / (1 +a2 x22 ) = exp(h2) / {1+ exp(h2)}
where hj = log(aj)+j log(xj)
for j=1,2
q= ( q1 , q2 , q3), where q1 = (a1 , 1) and q2 = (a2 , 2) have elicited
informative priors and the interaction parameters
q2 = ( a3 , 3) have non-informative priors.
Single-Agent Prior Elicitation Questions
1. What is the highest dose having negligible
(<5%) toxicity?
2. What dose has the targeted toxicity p* ?
3. What dose above the target has
unacceptably high (60%) toxicity?
4. At what dose above the target are you
nearly certain (99% sure) that toxicity is
above the target (30%) ?
Resulting Equations for the Hyperparameters
Denote g(h) = h / (1+h) so p(x,a,)} = g(ax).
Denote the doses given as answers to the questions by
{ x(1), x(2), x(3) = x*, x(4) }, and zj = x(j) / x*.
Assuming a ~ Ga(a1 , a2 ) and  ~ Ga(b1 , b2 ), solve the
following equations for (a1 , a2 , b1 , b2 ) :
1. Pr{ g(az1) < .05 } = 0.99
2. E(a(z*)) = a1 a2 E(1) = p* / (1 - p* )
3. E(az3) = a1 a2 E(z3) = 0.60 / 0.40 = 1.5
4. Pr{ g(az4) > p* } = 0.99
The answers to the 4 questions for each single agent
Randy Millikan, MD
An Interpretation of this Prior
The ESS of p(θ) = p(θ1, θ2, θ3) is 1.5
Since informative priors on θ1 and θ2 and a vague prior on θ3
were elicited, it is useful to determine the prior ESS of
each subvector :
ESS of marginal prior p(θ1) is 547.3 for p(x1,0 | a1, 1)}
ESS of marginal prior p(θ2) is 756.3 for p(0,x2 | a2, 2)}
ESS of marginal prior p(θ3) is 0.01 for the interaction
parameters θ3 = (a3, 3)
This illustrates 4 key features of prior ESS
1.
ESS is a readily interpretable index of a prior’s
informativeness.
2.
It may be very useful to compute ESS’s for both the
entire parameter vector and for particular subvectors
3.
ESS values may be used as feedback in the elicitation
process
4.
Even when standard distributions are used for priors, it
may NOT be obvious how to define a prior’s ESS.
Probability Model for Dose-Finding Based on Bivariate Binary
Efficacy (Response) and Toxicity Indicators YE and YT
(Thall and Cook, 2004)
For indices a=0,1 and b=0,1, and x = standardized dose,
pa,b (x, q) = Pr(YE = a , YT = b | x, q)
= pEa(1-pE)1-a pTb(1-pT)1-b + (-1)a+b pE(1-pE)pT(1-pT) (ey-1)/(ey+1)
with marginals
logit pT(x,q) = mT + xT
logit pE(x,q) = mE + xE,1 + x2E,2

The model parameter vector is q = (mT , T , mE , E,1 , E,2 , y)
Establishing Priors
1) Elicit mean & sd of pT(x,q) & pE(x,q) for several values of x.
2) Use least squares to solve for initial values of the
hyperparameters x in prior(q | x)
3) Each component of q is assumed normally distributed,
qr ~ N(mr, r), so x = (m1,1,…, mp,p)
4) mE,j = prior mean and sE,j = prior sd of pE(xj,q)
mT,j = prior mean and sT,j = prior sd of pT(xj,q)
5) # elicited values > dim(x)  find the vector x that minimizes
the objective function
Penalty term to keep the
’s on
the same numerical domain, c = .15
Example: Elicited Prior for the illustrative application in
Thall and Cook (2004)
A trial of allogeneic stem cell transplant patients:
Up to 12 cohorts of 3 each (Nmax = 36) were treated to determine
a best dose among {.25, .50, .75, 1.00 } mg/m2 of
Pentostatin® as prophyaxis for graft-versus-host disease.
E = drop from baseline of at least 1 grade in GVHD at week 2
T = unresolved infection or death within 2 weeks.
 ESS(q) = 8.9 (equivalent to 3 cohorts of patients!!)
ESS(qE) = 13.7,
ESS(qT) = 5.3,
ESS(y) = 9.0
A Slightly Smarter Way to Think About Priors
A Strategy for Determining Priors in the Regression Model
Fix the means
and use ESS contour plots to choose
Example:
To obtain desired overall ESS = 2.0 and
ESSE = ESST = ESSy = 2.0,
one may inspect the ESS plots to choose the variances
of the hyperprior. One combination that gives this is
Eliciting the Hyperprior for a Hierarchical Bayesian
Model in a Phase II Trial (Thall, et at. 2003)
A single arm trial of Imatinib (Gleevec, STI571) in sarcoma,
accounting for multiple disease subtypes.
pi = Pr( Tumor response in subtype i )
Prior:
logit(pi) | m, t ~ i.i.d Normal( m, t ), i=1,…,k
Hyperprior:
m ~ N( -2.8, 1), t ~ Ga( 0.99, 0.41 )
Stopping Rule: Terminate accrual within the ith subtype if
Pr( pi > 0.30 | Data ) < 0.005
“Data” refers to the data from all 10 subtypes.
But where did these numbers come from?
Eliciting the Hyperprior
Denote Xi = # responders out of mi patients in subtype i.
1) I fixed the mean of m at logit(.20) = -1.386, to correspond to
mean prior response rate midway between the target .30 and
the uninteresting value .10.
2) I elicited the following 3 prior probabilities :
Pr( p1 > 0.30 ) = 0.45
Pr( p1 > 0.30 | X1 / m1 = 2/6) = 0.525
Pr( p1 > 0.30 | X2 / m2 = 2/6) = 0.47
Prior Correlation Between
Two Sarcoma Subtype Response
Probabilities p1 and p2
Two Priors for a Phase II-III Pediatric Brain Tumor Trial
A two-stage trial of 4 chemotherapy combinations :
S = carboplatin + cyclophosphamide + etoposide + vincristine
E1 = doxorubicin + cisplatinum + actinomycin + etoposide
E2 = high dose methotrexate
E3 = temozolomide + CPT-11
Outcome (T,Y) is 2-dimensional :
T = disease-free survival time
Y = binary indicator of severe but non-fatal toxicity
Both p(T | Y,Z,q) and p(Y | Z,q) account for patient covariates:
Age, I(Metastatic disease), I(Complete resection)
I(Histology=Choriod plexus carcinoma)
Probability Model
1)
T| Z,Y, j ~ lognormal with variance T2 and
mean mT,j(Z,Y,x) = gT,j + T(Z,Y)

gT,j = effect of trt j on T, after adjusting for Z and Y
 For j=0 (standard trt), xT = (gT,0 , T)
2) logit{Pr(Y=1 | Z, j)} = gY,j + Y Z 
gY,j = effect of trt j on Y, after adjusting for Z
 For j=0 (standard trt), xY = (gY,0 , Y)
Toxicity Probability as a Function of Age
Elicited from Three Pediatric Oncologists
Probability Model for Toxicity
logit{Pr(Y=1 | Z, x, j=0)} = gY,0 + Y,1 Age1/2 + Y,2 log(Age)
was determined by fitting 72 different fractional polynomial
functions and picking the one giving the smallest BIC.
Estimated linear term with posterior mean subscripted by the
posterior sd is
This determined the prior of xY
64 Elicited EFS Probabilities
Johannes Wolff, MD
How do you use
these 64
probabilities to
solve for 10
hyperparameters?!!
Prior for xT
xT = (gT,j , T, T) has prior
Regard each prior mean EFS prob as a func of
Use nonlinear least squares to solve for
by minimizing

E(T) = (0.44, -0.41, 0.56, -0.53) with common
variance 0.152
and log(T) ~ N(-0.08, 0.142)
A Phase I/II Dose-Finding Method Based
on E and T that Accounts for Covariates
YE = indicator of Efficacy
YT = indicator of Toxicity
d = assigned dose
Z = vector of baseline patient covariates
Model the marginals
pE(d, Z) = Prob(E if d is given to a patient with covs Z)
pT(d, Z) = Prob(T if d is given to a patient with covs Z)
Use a copula to define the joint distribution :
pa,b = Pr(YE=a, YT=b) is a function of pE(d, Z) and pT(d, Z)
Model for pE(d,Z) and pT(d,Z)
pE = link{ hE(d,Z) }
&
pT = link{ hT(d,Z) }
where hE(d,Z) & hT(d,Z) are functions of
[ dose effects ] + [ covariate effects ]
+ [ dose-covariate interactions ]
pa,b = Pr(YE=a, YT=b) = func(pE, pT ,y )
for a, b = 0 or 1
Linear Terms of the Model for pE(t,Z)
For the trial:
hE(x, Z) = f(x,aE) + EZ + x gEZ
Dose effect
Covariate effects
Dose-Covariate
Interactions
For the historical treatment j :
hE( j, Z) = mE,j + E,HZ + xE,j Z
Historical trt effect
Historical trtcovariate interactions
Linear Terms of the Model for pT(t,Z)
For the trial:
hT(x, Z) = f(x,aT) + TZ + x gTZ
Dose effect
Covariate effects
Dose-Covariate
Interactions
For the historical treatment j :
hT( j, Z) = mT,j + T,HZ + xT,j Z
Historical trt effects
Historical trtcovariate interactions
Using Historical Data
In planning the trial, historical data are used to
estimate patient covariate main effects :
Prior(T) = Posterior(T,H | Historical data)
Prior(E) = Posterior(E,H | Historical data)
The estimated covariate effects are incorporated
into the model for pE(d,Z) and pT(d,Z) used to
plan and conduct the trial
Establishing Priors
For a reference patient Z*, elicit prior means
of pT(xj, Z*) and pE(xj, Z*) at each dose xj to
establish prior means of the dose effect
parameters
Assume non-informative priors on dose
effects and dose-covariate interactions
Use prior variances to tune prior effective
sample size (ESS) in terms of pE and pT
Control the prior ESS to
make sure that the data
drives the decisions,
rather than the prior on
the dose-outcome
parameters
Application
A dose-finding trial of a new “targeted” chemoprotective agent (CPA) given with idarubicin +
cytosine arabinoside (IDA) for untreated acute
myelogenous leukemia (AML)patients age < 60
Historical data from 693 AML patients
Z = (Age, Cytogenetics)
-5 or -7
Inv-16 or t(8:21)
where Cytogenetics = (Poor, Intermediate, Good)
Application
Efficacy = Alive and in Complete
Remission at day 40 from the start of
treatment
Toxicity = Severe (Grade 3 or worse)
mucositis, diarrhea, pneumonia or
sepsis within 40 days from the start
of treatment
Doses and Rationale
The CPA is hypothesized to decrease the risk of
IDA-induced mucositis and diarrhea and thus
allow higher doses of IDA
Fixed CPA dose = 2.4 mg/kg and ara-C dose =
1.5 mg/m2 daily on days 1, 2, 3, 4
IDA dose = 12 (standard), 15, 18, 21 or 24 mg/m2
daily on days 1, 2, 3 (five possible IDA doses)
Models for the linear terms used to fit the historical data
Interactive
hE( j, Z) = mE,j + EZ + xE,j Z
hT( j, Z) = mT,j + TZ + xT,j Z
Additive
hE( j, Z) = mE,j + EZ
hT( j, Z) = mT,j + TZ
No treatmentcovariate interactions
Reduced
hE( j, Z) = mE + EZ
hT( j, Z) = mT + TZ
No differences
between the
historical treatment
effects
Model Selection for Historical Data
Posteriors of pE(t, Z) and pT(t, Z) based on
Historical Data from 693 Untreated AML Patients
Dose-Finding Algorithm
1) Choose each patient’s most desirable dose
based on his/her Z
2) No dose acceptable for that Z :
Do Not Treat
3) At the end of the trial, use the fitted model
to pick ( d | Z ) for future patients
The trial’s entry criteria may change
dynamically during the trial :
1) Different patients may receive different
doses at the same point in the trial
2) Patients initially eligible may be ineligible
(no acceptable dose) after some data have
been observed
3) Patients initially ineligible may become
eligible after some data have been
observed
Hypothetical Trial Results :
Recommended Idarubicin Doses by Z
AGE
Cyto Poor
Cyto Int
Cyto Good
18 – 33
18
24
24
34 – 42
18
21
24
43 – 58
15
18
21
59 – 66
12
15
18
> 66
None
12
15
Currently being used to conduct a 36-patient trial to select
among 4 dose levels of a new cytotoxic agent for
relapsed/refractory Acute Myelogenous Leukemia
Y=
(CR, Toxicity) at 6 weeks
Z = (Age, [1st CR > 1 year], Number of previous trts)
Marina Konopleva, MD, PhD
is the PI
Bibliography
[1] Morita S, Thall PF, Mueller P. Determining the effective sample size of
a parametric prior. Biometrics. 64:595-602, 2008.
[2] Morita S, Thall PF, Mueller P. Evaluating the impact of prior
assumptions in Bayesian biostatistics. Statistics in Biosciences. In
press.
[3] Thall PF, Cook JD. Dose-finding based on efficacy-toxicity trade-offs.
Biometrics, 60:684-693, 2004.
[4] Thall PF, Simon R, Estey EH. Bayesian sequential monitoring designs
for single-arm clinical trials with multiple outcomes. Statistics in
Medicine 14:357-379, 1995.
[5] Thall PF, Wathen JK, Bekele BN, Champlin RE, Baker LO, Benjamin
RS. Hierarchical Bayesian approaches to phase II trials in diseases
with multiple subtypes. Statistics in Medicine 22: 763-780, 2003.
[6] Thall PF, Wooten LH, Nguyen HQ, Wang X, Wolff JE. A geometric
select-and-test design based on treatment failure time and toxicity:
Screening chemotherapies for pediatric brain tumors. Submitted for
publication.
[7] Thall PF, Nguyen H, Estey EH. Patient-specific dose-finding based on
bivariate outcomes and covariates. Biometrics. 64:1126-1136, 2008.