Randomization

Dr. Bhaswat S. Chakraborty

Randomization
Why Randomize in Controlled Tials?
1. To guard against any use of judgment or
systematic arrangements leading to one
treatment getting “better condition” to
succeed

2. To provide a basis for the standard methods

of statistical analysis such as significance
tests
Some Problems With Uncontrolled Trials
 Uncontrolled trials have to potential to provide
a very distorted view of therapy comparison
 Especially in the hands on over enthusiastic or
unscrupulous investigator
 E.g. cancer trials (Laetrile or Interferon therapy)
 It would be almost impossible to reproduce the
results of an uncontrolled trial to any certainty
 Recommendation from such trials are often
enthusiastic but may prove totally unrealistic
Problems With Historical Controls
 No way of ensuring that the comparison is fair
 Treatment and control groups can differ in many features
other than the treatment itself
 No guarantee that any improvement in patients response is
actually due to new treatment
 Historical control group less like to have clearly define I/E
criteria
 Type of the patients in historical control may be different
 The quality of recorded for historical control data always
inferior (since the patients were not intended to be in the trial)
 Criteria of response may be different between the two groups
Problems With Historical Controls
 Historical data are often of poorer quality so that reporting
of prognostic factors may not be consistent
 One may have only a sketchy idea of which patient factors
are important and some essential factors may go undetected
 Prognostic factors can only adjust for patient selection,
whereas bias due to changes in experimental environment
will remain
 The analysis techniques are quite complex and involve
certain assumptions, which may not be fulfilled. The
methods may be clear to only a skilled data analyst not to
may clinicians
 To propose that poor design can be corrected for by subtle
analysis techniques is not scientific
Problems with Concurrent Non- Randomized
Controls
 Systematic Assignment
 E.g. Date of Birth, alternate assignment
 The main problem with this arrangement would
investigators would know advance a patient would
receive
 Judgment Assignment
 E.g. Investigator is allow to exercise his judgment
to assign a treatment
 He may favour one particular treatment to more
serious cases to make it look inferior
Criteria for Randomization
1. Unpredictability
 Each participant has the same chance of receiving any of the
interventions.
 Allocation is carried out using a chance mechanism so that
neither the participant nor the investigator will know in
advance which will be assigned
2. Balance
 Treatment groups are of a similar size & constitution, groups
are alike in all
 important aspects and only differ in the intervention each
group receives
3. Simplicity
 • Easy for investigator/staff to implement
Simple Randomization
 This method is equivalent to tossing a coin for each subject that
enters a trial, such as
Heads = Active, Tails = Placebo.
 The random number generator is generally used. It is simple and easy
to implement and treatment assignment is completely unpredictable.
 However, imbalanced randomization can happen in smaller trials,
reducing statistical power.
 E.g., In trial of 10 participants, treatment effect variance for 5-5 split
relative to 7-3 split is (1/5+1/5)/(1/7+1/3)=.84, so 7-3 split is only 84%
as efficient as 5-5 split.
 Even if treatment is balanced at the end of a trial, it may not be balanced
at some time during the trial. For example, the trial may be balanced at
end with 100 participants, but the first 10 might be AAAATATATA.
Block Randomization
 Block randomization is balanced within each block
 The basic idea of block randomization
 divide potential patients into m blocks of size 2n
 randomize each block such that n patients are allocated to A and n
to B
 then choose the blocks randomly

 Example: Two treatments of A, B and Block size of

2 x 2= 4
 Possible treatment allocations within each block are (1) AABB, (2)
BBAA, (3) ABAB, (4) BABA, (5) ABBA, (6) BAAB
 Block size depends on the number of treatments, it should be short
enough to prevent imbalance, and long enough to prevent guessing
allocation in trials
Block Randomization..
 The block size is not stated in the protocol – blind the
investigator to the block size
 In open-label trials, the sequence becomes somewhat
predictable (e.g. 2n= 4): B A B ? Must be A. A A ? ?
Must be B B.
 This could lead to selection bias. The solution to avoid
selection bias is
 Do not reveal blocking mechanism
 Use random block sizes
 If treatment is double blinded, selection bias is not likely
 Note if only one block is requested, then it produces a single
sequence of random assignment, i.e. simple randomization
The Urn Design
 The urn design is the most widely studied member of
the family of adaptive biased-coin designs
 Such designs are a compromise between designs that
yield perfect balance in treatment assignments and
complete randomization which eliminates experimental
bias
 The urn design forces a small-sized trial to be balanced
but approaches complete randomization as the size of
the trial (n) increases.
 The urn design is not as vulnerable to experimental
bias as are other restricted randomization procedures
The Urn Design..
Stratified Randomization
 An RCT may not be considered valid if it is not well balanced
across prognostic factors
 E.g., Age Group: < 40, 41-60, >60; Sex: M, F; Total number of
strata = 3 x 2 = 6
 Stratification can balance subjects on baseline covariates, tend to
produce comparable groups with regard to certain characteristics
(e.g., gender, age, race, disease severity)thus produces valid
statistical tests
 The block size should be relative small to maintain balance in small
strata.
 Increased number of stratification variables or increased number of
levels within strata leads to fewer patients per stratum.
 Large clinical trials without IAs don’t use stratification. Unlikely to
get imbalance in subject characteristics in a large randomized trial
Impact of Treatment Imbalance & Selection
Bias
 First order selection bias – when patients select their own treatments or
treatments are assigned based on patient characteristics, such as disease
severity
 eliminated by randomization, but subconsciously or otherwise, an investigator
uses advance knowledge of upcoming treatment allocations as the basis for
deciding whom to enroll
 e.g., patients more likely to respond may be preferentially enrolled when the
active treatment is due to be allocated, and patients less likely to respond may
be enrolled when the control group is due
 Second order selection bias – if upcoming allocations can be observed in
their entirety
 allocation concealment minimizes the ability to observe upcoming allocations,
yet upcoming allocations may still be predicted (imperfectly), or even
determined with certainty, if at least some of the previous allocations are
known, and if restrictions (such as randomized blocks) were placed on the
randomization
 Third order selection bias– prediction but not observation of upcoming
allocations
 controlled by perfectly successful masking
 the smaller the block sizes, the more accurately one can predict future treatment
assignments in the same block as known previous assignments
Impact of Treatment Imbalance & Selection
Bias..
 For most randomization procedures, the treatment
imbalance may affect statistical power
 However, treatment imbalance must be substantial
before power is more than trivially affected
 Expected selection bias associated with a
randomization procedure is a function of the
predictability of the treatment allocations and is readily
evaluated for any sequence of treatment assignments
 In an unmasked study, the potential for selection bias
may be substantial with highly predictable sequences
Unequal Randomization
 When two or more treatments under evaluation have a
cost difference
 substantial cost savings can be achieved by adopting a
smaller randomization ratio such as a ratio of 2:1, with only a
modest loss in statistical power
 Another scenario – when one arm saves lives and the
other such as placebo/medical care only does not
 Generally, randomization ratio of 3:1 will lose
considerable statistical power
Unequal Randomization and Power

0.8
1
0.6
0.75
0.4
0.5 m
0.2 0.33
0 0.25
0 20 40 60 80 100 120 140 160 180 200
Sample size

m is number of patients experimental over control

Double Blind (Masked) Studies

Neither the patient nor those

responsible for his care and
evaluation know which treatment he
is receiving
Conduct of Double Blind Studies
 Matched Placebos
 Oral placebo which is identical in all respects to the active
oral drug except the absence of active ingredient
 Coding of randomization
 The randomization list must be prepared by statistician
(preferably)
 A pharmacist then makes up identical packages containing
active drug or placebo for each patient
 Have a simple coding system linking the drug packages to
randomization list
Conduct of Double Blind Studies ..
 Breaking the code
 Interim analysis may not need breaking the code
 DMC decides the future course of the trial based on IA results
 Code must be broken correctly
 Objective evaluation of side effect
 In case of SAEs or clear failure of response the investigator may
have to be given the broken code for ethical reasons
 Other types of double blind studies
 E.g. Two blinded packages of active drugs
 More complex situation where two drugs have different dosing
schedules
 Package 1 contains Once-daily 200mg + Placebo
 Package 2 contains two conventional 100mg tablet
When is Blinding Possible
 Ethics : The double-blind procedure should not
result in any harm or undue risk to a patient
 Practicality : For some treatments it would be
totally impossible to arrange a double-blind trial
 Avoidance of bias: One needs to assess just how
serious the bias might be without blinding
 Compromise: Sometimes partial blinding (e.g.
independent blinded evaluators) can be sufficient
to reduce bias in treatment comparison
Blinding..
 Placebos are commonly used as an inactive treatment to achieve double
blinding. Active placebos, with which symptoms or side effects are
imitated, can also be used. Placebos are justly used when no existing
effective treatment is available.
 If the blinding of the placebo arm is not effective then the protection
against expectation effects, biased assessment, contamination, and co-
intervention are all lost.
 The observed superiority of a new treatment over placebo could merely be a
consequence of loss of this control—and an ineffective new treatment would
spuriously seem to be superior.
 It is not sufficient that trials describe themselves as double blind.
 It is also important that the efficacy of the blinding is actually assessed.
In other words, an assessment of the face validity of the double
blinding is needed.
 Success of blinding challenges the notion that placebo controlled trials
inherently possess assay sensitivity
Blinding
Case Study: RCT comparing tamoxifen and
anastrozole
 For gynecomastia and breast pain reduction
 Development of bicalutamide- related breast changes
 Optimum dose of tamoxifen 20mg/day
 for both prophylaxis and treatment and to assess any impact
on prostate cancer control
 Anastrozole 1mg/day
 does not appear to be viable management option for
bicalutamide induced gynecosmastia and breast pain

 Examine the Randomization and Blinding

References & Further Reading
 Stuart J Pocock’s Clinical Trials
 Statistical properties of randomization in clinical trials.
Control Clin Trials. 1988
 Properties of simple randomization in clinical trials.
Control Clin Trials. 1988
 Randomization in clinical trials: conclusions and
recommendations. Control Clin Trials. 1988
 Properties of permuted-block randomization in clinical
trials. Control Clin Trials. 1988
 Adaptive biased urn randomization in small strata
when blinding is impossible. Biometrics. 1995