Title: Modification of Sample Size in Group Sequential Clinical Trials
1Modification of Sample Size in Group Sequential
Clinical Trials
- Madan Gopal Kundu
- PhD (Biostatistics) student, IUPUI
2Sample Size
- Sample size (N) No. of patients
- Estimation of Sample size depends on - Type I
error, Power and Expected effect size - Reasonable power at reasonable cost (Best deal!!)
- Sample size is determined at the beginning of
trial.
3Group Sequential Design
Determine N
Final Analysis
Conduct of Clinical Trial
- Scope of Early termination of trial
- Overwhelming efficacy
- futility of the drug
4Outline
- Introduction
- A case study
- Group sequential Z test
- A sequential test procedure with sample size
modification (based on Z test) - Generalization Brownian motion
5Introduction
Less than
Planned sample size is NOT sufficient
There is scope to modify sample size when trial
is ongoing Q Does it increase overall type I
error? Q Is there any testing strategy to modify
sample size without increasing overall type I
error?
6Motivation A case study
Planning
- Phase III, comparative, placebo-controlled trial
for prevention of myocardial infarction. - Assumption
- Incidence rate in Placebo 22
- Incidence rate in New Drug 11
- Planned sample size 600 (powergt95)
Expected Effect size 0.30
Interim Analysis
- After evaluation of 300 patients
- Incidence rate in Placebo 22
- Incidence rate in New Drug 16.5
- Need to increase the sample size
Observed Effect size 0.14
7Motivation A case study
Concern
- Does it inflate overall type I error?
- No valid testing procedure was available to
account for such an outcome dependent adjustment
of sample size
Finally
- It was decided not to increase the sample size
- Trial eventually failed to show a statistically
significant effect
8Solution to this Dilemma
- To have accurate estimate of treatment effect
size at the beginning of trial - - Less likely!!
- Implementation of valid inferential procedure
that allows adjustment of sample size in the
mid-course of trial - - Cui, Hung and Wang method
9Theoretical set-up
Population I N (µ1, s21)
Population II N (µ2, s21)
x1, x2, . , xN
y1, y2, . , yN
Effect size, ? µ1 - µ2
Our interest is to test (using two sample
Z-test) Ho ? 0 vs Ha ? gt 0
Assuming ? d, total sample size (N) per
population
10Group-sequential structure
Additional Subjects
n1
n2
nL-1
nL
nK-1
nK
Cumulative Subjects
N1
N2
NL-1
NL
NK-1
NKN
Information Time
Observed Effect size
? 1
? 2
? L-1
? L
? K-1
? K
2-sample Z Test Statistic
T1
T2
TL-1
TL
TK-1
TK
C1
C2
CL-1
CL
CK-1
CK
Critical values
Reject Ho Stop trial if
T1gtC1
T2gtC2
TL-1gtCL-1
TLgtCL
TK-1gtCK-1
TKgtCK
11Conditional Power
- The conditional power evaluated at the Lth
interim analysis
- Sample size may be modified based on conditional
power. - is the Rejection Region.
12Sample Size Modification
- Calculate and
- Decide two positive constants 1
- If or
, N should be
modified to
- This adjustment of sample size preserves the
unconditional power at 1-ß when - If is smaller than d then it gives large M.
13Does Sample Size adjustment inflates Type I
error??
- Simulation studies
- Increase in sample size
- Substantial inflation in Type I error
rate - Decrease in sample size
- Mild effect on Type I error rate and
power
14Sample size modification
- Modify sample size in Lth interim analysis N?M
Cumulative sample size (with adjust)
N1
N2
NL
M
15Effect on Test Statistic because of sample size
modification
- When sample size is NOT allowed to increase
Test statistic at (Lj)th interim analysis,
(Eq. 1)
Where,
16Effect on Test Statistic because of sample size
modification
- When sample size is allowed to increase
(Eq. 2)
Where,
(Eq. 1) versus (Eq. 2)
Note Here is replaced by
Note Weights are also changed and become random
as is a function of
17A different group sequential test procedure (CHW)
- Here weights are kept fixed but are
replaced by
(Eq. 3)
Note ULj reduces to TLj, when MLj NLj
Test procedure
(K-1) Interim Analyses
Test Statistic
T1
T2
TL
ULj
UK-1
UK
UL1
Critical values
C1
C2
CL
CLj
CK-1
CK
CL1
Reject Ho Stop trial if
T1gtC1
T2gtC2
TLgtCL
UL1gtCL1
ULjgtCLj
UK-1gtCK-1
UKgtCK
18Distribution of ULj
Under H0 µ1 - µ2 0
19Distribution of ULj
20Impact on Type-I error
D
So,
D
Overall Type I error of the new test Procedure
Overall Type I error of the Original test
Procedure a Monte Carlo Simulation
New test has its type I error rate
maintained at a. Conclusion New test procedure
allows to modify sample size without increase in
overall type I error.
21Generalization
Scope
Repeated significance test with Brownian motion
process and Independent increment
Steps
- B(t) be such repeated significance test at
information time t. - T(t) B(t)/t1/2
- Let at ttL sample size increased to M
- w N/M
- b (w tL)/(1-tL)
Test Statistic
U(t) T(t) , if ttL
, if tgttL
22Brownian Motion
- B(tt?T) is known as Brownian motion process if
- Multivariate Normal Distribution
- Mean 0
- Var B(t) t
- Var B(t2) B(t1) t2 t1
- Cov B(t2), B(t1) min(t2, t1)
23Brownian Motion Z test statistic
24Brownian Motion Other Test statistic
T-test
Approx. Brownian motion (see Pocock 1977)
Log-rank test
- Approx. Brownian motion (see Tsiatis 1982, Sellke
Siegmund 1983, Slud 1984)
Wilcoxon test
Approx. Brownian motion (see Slud Wei 1982)
25Reference
- Cui L, Hung H J and Wang S J (1999). Modification
of sample size in Group Sequential Clinical
Trials. Biometrics 55 853-857. - Pocock S J (1977). Group sequential methods in
the design and analysis of clinical trials.
Biometrika 64 191-199. - Lan K K G and Wittes J (1988). The B-value A
tool for monitoring data. Biometrics 44 579-585. - Lan K K G and Zucker D M (1993). Sequential
monitoring of clinical trials the role of
information and brownian motion. Statistics in
Medicine 12 753-765. - Reboussin D M, DeMets D L, Kim K M and Lan K K G
(2000). Computation for group sequential
boundaries using the Lan-DeMets spending function
method. Controlled Clinical Trials 21 190-207. - Lan K K G and DeMets D L (1983). Discrete
sequential boundaries for clinical trials.
Biometrika 70 659-663.
26Reference
- Shih W J (2003). Group Sequential Methods.
Encyclopedia of Biopharm. Statistics 11 423-432. - Tsiatis A A (1982). Repeated significance testing
for a general class of statistics used in
censored survival analysis. JASA 77 855-861. - Sellke T and Siegmund D (1983). Sequential
analysis of the proportional hazard model.
Biometrika 70 315-326 - Slud E V (1984). Sequential linear rank tests for
two sample censored survival data. Annals of
Statistics 12 551-571. - Slud E and Wei L J (1982). Two-sample repeated
significance tests based on the modified Wilcoxon
test statistic . JASA 77(380) 862-867.
27Thank You!
28BACK-UP Slides
29(No Transcript)
30Now,
Therefore,