Transcript Statistical Items

CBMTG 0801
MINIMIZATION /
STATISTICS
Tony Panzarella
Princess Margaret Hospital /
University of Toronto
Outline
 Minimization
 Sample size
 Analysis
CBMTG presentation April 7, 2010
A Randomized Trial of Thymoglobulin
to Prevent Chronic Graft versus Host
Disease in Patients Undergoing
Hematopoietic Progenitor Cell
Transplantation (HPCT) from
Unrelated Donors
CBMTG presentation April 7, 2010
Overview - Study Design
CBMTG presentation April 7, 2010
“We will seek to achieve a balanced
allocation of treatments over prognostic
factors for cGVHD.”
CBMTG presentation April 7, 2010
Prognostic factors are
• Recipient age (<10; 10-30; 31-50; > 50);
• Female donor for male recipient;
• Donor age (< 30; ≥30)
• Blood source of progenitor cells rather than
bone marrow
•
Degree of tissue type matching (full match;
one antigen/allele mismatch).
CBMTG presentation April 7, 2010
“In addition to balancing for the above
variables that are risk factors for cGVHD it
will also be necessary to balance those
variables that will influence relapse and
mortality.”
CBMTG presentation April 7, 2010
• Disease
• stage (Early vs. Late )
• comorbidity index
•
type of preparative regimen (myeloablative
vs. non-myeloablative)
• centre (to allow for differences in clinical
practice)
CBMTG presentation April 7, 2010
“Given the moderate size of the
trial and the numerous strata
that would result from our
stated aim of achieving balance
upfront for the abovementioned factors it would be
impractical to use stratified
randomization; furthermore,
potentially large overall
imbalances in treatment
allocation could occur,
defeating the purpose of
stratified randomization. ”
CBMTG presentation April 7, 2010
“Instead patients will be allocated to the
treatment groups based on a method of
dynamic allocation referred to as
minimization. As the name implies the method
attempts to minimize the differences between
treatment groups in terms of these factors .”
CBMTG presentation April 7, 2010
“ Unlike stratified randomization, where
each strata represents a combination of
each of the factors identified, minimization
tries to achieve overall balance by trying
to achieve balance within each individual
factor, not every combination of factors.
This alternative approach to balancing
factors between treatment groups allows
the possibility of balancing over more
factors. ”
CBMTG presentation April 7, 2010
“Using minimization in our
study the first patient will
have their treatment
randomly allocated (akin to
flipping a fair coin). For each
subsequent patient we will
determine which treatment
would lead to better balance
between the groups with
respect to the baseline
prognostic variables
identified a priori. ”
CBMTG presentation April 7, 2010
“Each patient is then randomized using a
weighting in favour of the treatment that
would minimize the imbalance. A weighting
of 4 to 1 will be used in this study. That is,
there will be a probability of 0.8 of receiving
the treatment that minimizes the imbalance. ”
CBMTG presentation April 7, 2010
“Thus, the study statistician will
prepare two randomization lists
using a computer random number
generator before the study begins: 1)
a simple randomization list where
both treatments occur equally often;
this list will only be used when the
two treatments have equal sums for
the levels of the baseline prognostic
factors; and 2) a list in which the
treatment with the smaller total of
patient levels occurs with probability
0.8 while the other treatment occurs
with probability 0.2.”
CBMTG presentation April 7, 2010
“Allocation will occur centrally by the Project
Manager through the project management
office. This approach ensures that the process
of treatment allocation will be concealed
from staff at the recipients centre. ”
CBMTG presentation April 7, 2010
Minimization - example
 Suppose 16
patients have been
randomized into a
trial, and their
characteristics are
distributed as in
the table
CBMTG presentation April 7, 2010
Minimization - example
 Suppose the next patient is from
hospital X, aged 38 and has stage II
disease
A : 4 + 3 + 3 = 10
B : 4 + 5 + 2 = 11
Assign A
CBMTG presentation April 7, 2010
Minimization cards
for manual allocation
Stage I/II
Treatment 1
Stage III/IV
Treatment 1
Treatment 2
1. 1002
2. 1005
3. 1006
4.
5.
6.
7.
8.
9.
10.
1. 1001
2. 1004
3.
4.
5.
6.
7.
8.
9.
10.
Treatment 2
Hospital = Z
Treatment 1
Hospital = Y
Treatment 1
Hospital = X
Age ≥ 50
Treatment 1
Age > 50
Treatment 1
Treatment 2
1. 1002
2. 1006
3.
4.
5.
6.
7.
8.
9.
10.
1. 1004
2.
3.
4.
5.
6.
7.
8.
9.
10.
Treatment 2
1. 1002
Treatment 1 2. 1005
3. 1006
1. 1005
4.
2. 1006
5.
3.
6.
4.
7.
5.
8.
6.
9.
7.
10.
8.
9.
10.
CBMTG presentation April 7, 2010
Treatment 2
1. 1001
Treatment 2 2. 1004
1. 1001
2. 3.
4.
5.
6.
7.
8.
9.
10.
3.
4.
5.
6.
7.
8.
9.
10.
Treatment 2
CIHR reviewer comments
Three of 5 reviewers commented explicitly on
the minimization approach adopted in 0801.
-Two (including statistical reviewer) said it was
appropriate
- One would have preferred a simpler approach;
namely, stratified randomization using the most
important prognostic factor, and adjusting for
other imbalances in the analysis
CBMTG presentation April 7, 2010
Primary Endpoint
“The primary endpoint is the freedom from
chronic graft versus host disease, as
indicated by the withdrawal of all systemic
immunosuppressive agents and without
resumption up to 12 months after
transplantation. ”
CBMTG presentation April 7, 2010
“The primary endpoint could be construed
as a ‘focused stringent-positive’ variation of
‘chronic GVHD-free survival.’"
CBMTG presentation April 7, 2010
Withdrawal of immunosuppressive Tx
Death
S
0 +1
+2
+3
+4
+5
+6
+7
+8
+9 +10 +11 +12
Months from Transplant
CBMTG presentation April 7, 2010
Death (On immunosuppressive Tx)
F
0 +1
+2
+3
+4
+5
+6
+7
+8
+9 +10 +11 +12
Months from Transplant
CBMTG presentation April 7, 2010
Withdrawal of immunosuppressive Tx prompted by
imminent death / persistent malignancy
F
0 +1
+2
+3
+4
+5
+6
+7
+8
+9 +10 +11 +12
Months from Transplant
CBMTG presentation April 7, 2010
Effect Size
“If we assume that the no treatment
group (i.e. the control group) has a
probability of response of 0.4 (higher
than previously described and erring on
the side of caution) it would be clinically
worthwhile to know if patients
administered Thymoglobulin® could
increase this response proportion by at
least an additional absolute difference of
0.2, to 0.6.”
CBMTG presentation April 7, 2010
Sample Size
“To be able to detect
this difference or more
as statistically
significant at the type I
error (2-sided) level of
0.05 with a power of
0.8, a total sample size
of 194 patients would
be required (East
version 5.1). Assuming,
conservatively, that 2%
of patients are lost to
follow-up in each group
a total of 198 patients
would be recruited.”
CBMTG presentation April 7, 2010
CBMTG presentation April 7, 2010
CIHR reviewer comments
Definition of primary endpoint
most contentious issue
Sample size calculation appropriate
but sample size thought to be small
CBMTG presentation April 7, 2010
Analysis
“The primary endpoint will
be compared between
treatment groups using
logistic regression adjusted
for covariates employed in
the design. This will yield
significance tests of proper
size. ”
CBMTG presentation April 7, 2010
Secondary endpoints
• the incidence of cGVHD (regardless of
need for treatment)
• the incidence of “extensive” cGVHD,
• time to non-relapse mortality
• time to all-cause mortality
• time to relapse of leukemia
CBMTG presentation April 7, 2010
Secondary endpoints continued…
• graft rejection or failure (Yes vs. No)
• serious infection (Yes vs. No)
• CMV activation (Yes vs. No)
• quality of life
• organ specific grading of chronic graft versus host
disease
• resumption of immunosuppressive agents after 12
months (Yes vs. No)
• doses of immunosuppressive drugs still required at
12 months.
CBMTG presentation April 7, 2010
“Comparisons of time to
failure endpoints will
incorporate Kaplan-Meier
probability estimates, log
rank testing and, when
adjustment of covariates
is made, use of the Cox
proportional hazards
model. If the analysis of a
time to failure endpoint
involves competing risks
then the probability of
failure will be estimated
using the cumulative
incidence function. ”
CBMTG presentation April 7, 2010
Binary/categorical secondary endpoints will be
compared between treatment groups using
logistic regression.
Doses of immunosuppressive drugs required at
12 months will be compared between treatment
groups using multiple regression.
CBMTG presentation April 7, 2010
Quality of life will be measured
before transplant and at 6, 12,
18, and 24 months post
transplant using a number of
instruments.
CBMTG presentation April 7, 2010
The results for each of the primary and
secondary endpoints will be summarized
by a significance test and 95 per cent
confidence interval.
CBMTG presentation April 7, 2010
“Analysis of secondary endpoints and sub-group
analyses will be considered exploratory and
hypothesis generating. Given that multiple
comparisons increase the probability of a Type
I error, adjustment of individual statistical tests
using a more strict cut point (p<0.01) will be
used to facilitate interpretation.”
CBMTG presentation April 7, 2010
All statistical tests quoted will be 2-tailed.
Analysis will follow the intention-to-treat principle.
Most analyses will be conducted using SAS version
9.2.
However, competing risk failure time data will be
analyzed using the library cmprsk in R
(http://www.r-project.org).
CBMTG presentation April 7, 2010
“Sub-group analysis will be
conducted as follows: The two
treatment groups will be
compared among subgroups
based on the covariates
diagnosis, gender, type of
preparative regimen
(myeloablative vs. nonmyeloablative), donor age and
recipient age. Differences
between treatments by
subgroup will be tested using
interaction effects. Results will
be considered hypothesisgenerating. ”
CBMTG presentation April 7, 2010
“One interim analysis will be conducted after half of the
evaluable patients have been recruited and followed for one
year. A group sequential design utilizing the Lan-DeMets
spending function with an O’Brien-Fleming stopping
boundary will be incorporated. As a result of the interim
look the trial could be: 1) stopped early by rejecting the
null hypothesis of no treatment difference; 2) stopped
early by rejecting the alternative hypothesis that the
difference in proportions is at least 0.2; or 3) the study is
continued. ”
CBMTG presentation April 7, 2010
CBMTG presentation April 7, 2010
CIHR reviewer comments
The only analysis issue identified
had to do with stopping early for
safety rather than efficacy
CBMTG presentation April 7, 2010
The foundation of success in clinical trials is
teamwork…and I think the study chair has
assembled a fine team!
CBMTG presentation April 7, 2010