pre-stratified randomization is not necessary for large clinical trials

Report
PRE-STRATIFIED
RANDOMIZATION IS NOT
NECESSARY FOR LARGE
CLINICAL TRIALS
Brent Leininger, Patrick Kurkiewicz,
Lifeng Lin, Xiang Li, Bryan Trottier Jr,
Yuanyuan Wang
Pre-stratification is Insurance
Risk of Chance Imbalance
As Sample Size Increases…..
Administrative Burden
•
•
1 Allocation Schedule per Strata
1 stratification variable with 2 levels
•
Increased potential for errors
Misclassification of Strata
•
•
Multi-center variability
Timing
% Reduction in Standard Error
60%
50%
40%
30%
20%
10%
0%
0.1
0.2
0.3
0.4
0.5
Correlation
0.6
0.7
0.8
0.9
Pocock SJ, Assmann SE, Enos LE, Kasten LE. Subgroup analysis, covariate adjustment
and baseline comparisons in clinical trial reporting: current practice and problems. Stat
Med. 2002 Oct 15;21(19):2917-30.
•
•
•
111 of 258 trials balanced on prognostic factors (other
than center)
36% of trials accounted for stratification in the analysis
Accounting for stratification variables in the analysis is
recommended by both ICH and CONSORT guidelines
% Reduction in Standard Error
60%
50%
40%
30%
20%
10%
0%
0.1
0.2
0.3
0.4
0.5
Correlation
0.6
0.7
0.8
0.9
Pocock SJ, Assmann SE, Enos LE, Kasten LE. Subgroup analysis, covariate adjustment
and baseline comparisons in clinical trial reporting: current practice and problems. Stat
Med. 2002 Oct 15;21(19):2917-30.
Alternatives
• Post-stratification (i.e. adjusted analysis)
•
Efficiency Loss vs. Pre-Stratification?
•4 strata, 80 patients per strata
•Efficiency loss is <4%
McHugh R, Matts J. Post-stratification in the randomized clinical trial. Biometrics. 1983
Mar;39(1):217-25.
Post-Stratification Criticism
•Define covariates ‘a priori’ OR…..
•Define plan for covariate selection ‘a priori’
Pocock SJ, Assmann SE, Enos LE, Kasten LE. Subgroup analysis, covariate adjustment
and baseline comparisons in clinical trial reporting: current practice and problems. Stat
Med. 2002 Oct 15;21(19):2917-30.
Conclusions
CLAIM: PRE-STRATIFIED RANDOMIZATION IS OFTEN
NECESSARY FOR LARGE (N > 100 PER GROUP)
CLINICAL TRIALS
• Helps ensure that compared groups are similar with respect to
known important prognostic factors
• Potential benefits, even in large clinical trials1
• Protection against Type 1 Error
• Reduction of sample size in equivalence trials
• Facilitation of interim analyses
• Identification of subgroups prior to analysis
• Protection against treatment assignment imbalance with recruitment-center
dropout in multicenter trial
1
Kernan et al. J Clin Epidemiol (1999)
UGANDA STUDY:
IMMEDIATE VS. DELAYED IRON IN SEVERE
MALARIA
• Children with severe malaria randomized to immediate or
delayed iron
• Primary outcome = frequency of hospital admissions in next 12
months
• Severe malaria includes cerebral malaria and severe malarial
anemia
• Children with severe malarial anemia have greater risk of
readmission to hospital
UGANDA STUDY:
IMMEDIATE VS. DELAYED IRON IN SEVERE
MALARIA
THIS
NOT THIS
Pre-stratified randomization
Un-stratified randomization
79 CM
156 SEVERE MALARIA
R
R
40 Immediate
39 Delayed
77 SMA
R
39 Immediate
38 Delayed
78 Immediate
?? SMA
?? CM
78 Delayed
?? SMA
?? CM
BENEFITS OF PRE-STRATIFICATION
• Protection against Type 1 error1
1
Feinstein and Landis in Kernan et al. J Clin Epidemiol (1999)
BENEFITS OF PRE-STRATIFICATION
• Smaller sample size in equivalence trials
• Reduction in sample size by 12-42% over a range of
assumptions (Nam, Stat Med 1995)
• Significant savings in both cost and time
• Facilitation of interim analyses
• Interim DSMB meeting for iron/malaria study March 2013 (!)
• 79 CM (40 I, 39 D) ; 78 SMA (39 I, 38 D)
• Frequency of serious adverse events in I vs. D.
• If not balanced, potential safety concern undetected
BENEFITS OF PRE-STRATIFICATION
• Subgroup analysis
• Strata identified before start of study
• Protects against multiple comparisons
• Strengthens a finding of treatment effectiveness within
subgroups
• Multicenter study—protection against drop-out
• Because patients in each center are balanced for
treatment assignment, withdrawal of a center will not result
in imbalance among remainder of patients.
CLAIM: PRE-STRATIFIED RANDOMIZATION IS OFTEN
NECESSARY FOR LARGE (N > 100 PER GROUP) CLINICAL
TRIALS
For
1. Protection against
Type 1 error
2. Reduced sample size
in equivalence trials
3. Balanced interim
analyses
4. Pre-specified subgroup
analyses
5. Protection against
drop-out in multicenter
studies
Against
1. Too many strata
2. Challenge to
implement
3. Not necessary

similar documents