Randomisation - North Carolina State University

Download Report

Transcript Randomisation - North Carolina State University

Biometrics India, Pfizer Global R & D
The Concept of Randomization and Blinding in
Clinical Trials
Suraj P Anand
Randomization
Randomization is the process of assigning clinical trial
participants to treatment groups. Randomization gives
each participant a known (usually equal) chance of being
assigned to any of the groups. Successful randomization
requires that group assignment cannot be predicted in
advance.
Why Randomize?
If, at the end of a clinical trial, a difference in
outcomes occurs between two treatment groups (say,
intervention and control) possible explanations for
this difference would include:
the intervention exhibits a real effect;
the outcome difference is solely due to chance
there is a systematic difference (or bias) between the
groups due to factors other than the intervention.
Randomization aims to obviate the third possibility.
Forms of Randomization
Simple Randomization
Permuted Block Randomization
Stratified Block Randomization
Dynamic (adaptive) random allocation
Simple Randomization
Coin Tossing for each trial participant
Sequence of Random Numbers from statistical
textbooks
Computer generated sequence
Illustrations
The computer generated sequence:
4,8,3,2,7,2,6,6,3,4,2,1,6,2,0,…….
Two Groups (criterion:even-odd):
AABABAAABAABAAA……
Three Groups:
(criterion:{1,2,3}~A, {4,5,6}~B, {7,8,9}~C; ignore 0’s)
BCAACABBABAABA……
Two Groups: different randomisation ratios(eg.,2:3):
(criterion:{0,1,2,3}~A, {4,5,6,7,8,9}~B)
BBAABABBABAABAA……..
Permuted Block Randomization
Used for small studies to maintain reasonably
good balance among groups
In a two group design, Blocks having equal
numbers of As and Bs (A = intervention and
B = control, for example) are used, with the
order of treatments within the block being
randomly permuted
Illustration
With a block size of 4 for two groups(A,B), there are 6 possible
permutations and they can be coded as:
1=AABB, 2=ABAB, 3=ABBA, 4=BAAB, 5=BABA, 6=BBAA
Each number in the random number sequence in turn selects the
next block, determining the next four participant allocations
(ignoring numbers 0,7,8 and 9).
e.g., The sequence 67126814…. will produce BBAA AABB ABAB
BBAA AABB BAAB.
In practice, a block size of four is too small since researchers may
crack the code and risk selection bias. Mixing block sizes of
between 6 and 12 is better with the size kept unknown to the
investigator. This precaution maintains concealment. Simple
randomization should determine which block size to use next.
Stratified Block Randomization
Stratified block randomization can further restrict chance
imbalances to ensure the treatment groups are as alike as
possible for selected prognostic variables or other patient
factors. A set of permuted blocks is generated for each
combination of prognostic factors
Typical examples of such factors are age group, severity of
condition, and treatment centre. Stratification simply means
having separate block randomisation schemes for each
combination of characteristics (‘stratum’)
For example, in a study where you expect treatment effect to
differ with age and sex you may have four strata: male over 65,
male under 65, female over 65 and female under 65
Dynamic (adaptive) random allocation
Simple and block randomization methods are defined, and allocation
sequences set up, before the start of the trial.
In contrast, dynamic randomization methods allocate patients to treatment
group by checking the allocation of similar patients already randomized, and
allocating the next treatment group "live" to best balance the treatment
groups across all stratification variables. Biased coin randomization and
minimisation are two such methods.
Efron(1971) first introduced the idea of biased coin randomization as a
method for adjustment of assigning probabilities. The assigning probability
for the first patient is1/2. After k patients are enrolled, with k(A) and k(B)
patients randomized in groups A and B respectively, the idea involves
randomizing the next patient to group B with probability greater than ½ if
more patients have been randomized to group A at this stage, and viceversa. If balance is achieved, the next patient is randomized to either of the
groups with probability ½.
Example of randomization using the minimisation method in a trial of
chemotherapy for breast cancer, with stratification factors of clinic site,
estrogen receptor status (ER+ or ER–) and menopausal status
Characteristic
Treatment A
Treatment B
Site 1
7
8
Site 2
10
9
ER+
5
6
ER–
12
11
Premenopausal
8
9
Postmenopausal
9
8
Total
17
17
The next participant (no. 35) is from Site 2, ER+, postmenopausal. Subtotals for
treatment allocation to this profile of characteristics are 10 + 5 + 9 = 24 for Treatment A
and 9 + 6 + 8 = 23 for Treatment B (note subjects are counted more than once).
Participant no. 35 would therefore be allocated to Treatment B. When the tallies on A
and B are equal within a profile, the next participant is randomly allocated. This process
is equivalent to a permuted block size of two within the profile.
Inappropriate randomisation methods
Assigning patients alternately to treatment group is
not random assignment
Assigning the first half of the population to one group
is not random assignment
Assignments by methods based on patient
characteristics such as date of birth, order of entry
into the clinic or day of clinic attendance, are not
reliably random
Allocation Concealment
It is very important that those responsible for recruiting people into a trial are
unaware of the group to which a participant will be allocated, should that
subject agree to be in the study. This avoids both conscious and unconscious
selection of patients into the study.
For multicentre clinical trials, central randomization by telephone, interactive
voice response system, fax or the Internet are ideal methods for allocation
concealment. The clinician or data manager at the participating site assesses
eligibility, gains consent, and makes the decision to enroll a patient, then calls
the randomization service to get the treatment allocation.
For single-centre clinical trials, it is usually possible to identify a staff member
not involved with the trial who can keep the randomization list or envelopes.
They should be instructed to keep the list private, and to only reveal a
treatment allocation after receiving information demonstrating that the patient
is eligible and has consented to the trial.
In situations where remote randomization may not be feasible or desirable, a
set of tamper-evident envelopes may be provided to each participating site.
The envelopes should look identical, and each should have the trial
identification and a sequential number on it. Inside is the treatment allocation
and usually a trial identifier for the patient (e.g., unique sequential number).
After assessing eligibility and consent, the next envelope in sequence is
opened. Care needs to be taken that the envelopes are opaque and well sealed,
and that the sequence of opening the envelopes is monitored regularly.
Issues leading to Blinding
Most investigators have firm views about which of a range of
alternative treatments is more effective and often, which is
more appropriate for particular groups of patients. As a result,
there is a strong temptation by investigators to channel
particular groups of patients to particular treatments
(channeling effect )
There is also a risk of the investigators subconsciously losing
their objectivity in their assessments of treatment effects simply
because of their clear preference for particular treatments
There is a risk of having other forms of bias, which can be
satisfactorily controlled by proper blinding
Bias
Bias is said to have occurred if the results observed reflect
other factors in addition to (or even instead of) the effect of the
treatment:
Some potential sources of bias:
Patient bias
Care Provider bias
Assessor bias
Laboratory bias
Analysis and Interpretation bias
Patient Bias
the patient's knowledge that the patient is receiving a
"new" treatment may substantially affect the patient's
subjective assessment
there is a subject x disease interaction in at least some
diseases (and virtually all diseases)
thus, the patient's knowledge of the treatment being
received may affect the outcome of the study
Care Provider Bias
the care provider's knowledge of which treatment a
patient is receiving may affect the way the provider
– deals with the patient
– treats the patient
these differences may give the patient information
(even if incorrect) about the treatment the patient is
receiving, which then may affect the outcome of the
study
Assessor Bias
the assessor's knowledge of which treatment the
patient is receiving may affect the way the assessor
assesses outcome
such a bias would directly affect the validity of the
conclusions of the study
if the assessment is done while the patient is still
receiving treatment, this may provide the patient with
information about the treatment being received
Laboratory Bias
the knowledge of which treatment the patient received
may affect the way in which the test is run or interpreted,
or be retested.
although this is most severe with subjectively graded
results (pathology slides, photographs, ECG, etc.), this
can also be a problem with "objective tests" such as
laboratory assays which may be run subtly differently by
the technician.
Analysis and Interpretation bias
knowledge of the treatment group may affect the results of
the analysis of the data by
– seeking an explanation of an "anomalous” finding when one is
found contrary to the study hypothesis
– accepting a "positive" finding without fully exploring the data
knowledge of the treatment group may affect the decisions
made by external monitors of a study by
– terminating a study for adverse events because they fit the
expectations of the monitors
– terminating a study for superiority of treatment because it fits
the expectations of the monitors
Blinding
All of these potential problems can be avoided
if everyone involved in the study is blinded to
the actual treatment the patient is receiving.
Blinding (also called masking or concealment
of treatment) is intended to avoid bias caused
by subjective judgment in reporting, evaluation,
data processing, and analysis due to knowledge
of treatment.
Hierarchy of Blinding
open label: no blinding
single blind: patient (usually;occasionally may be assessor)
blinded to treatment
double blind: patient and assessors (who often are also the
health care providers and data collectors) blinded to
treatment
complete blind: everyone involved in the study blinded to
treatment
Open Label Studies
These may be useful for
• pilot studies
• dose ranging studies
However, even these applications may be substantially biased
by knowledge of the treatment given and may result in
• toxicity over (or under) reported
• efficacy over estimated
Even a small fraction of patients assigned at random to
placebo will reduce these potential problems substantially.
Single Blind Studies
single blind studies are usually done to blind the patient to the
treatment given. Health care providers and assessors
usually know the actual treatment given
justification is usually that double-blind is "impractical" because
of need to adjust medication, medication affecting laboratory
values, potential side effects, etc.
a single blind study should be used only when it would be
unacceptable ethically to give an appropriate placebo treatment
to a patient, and in such a case, the assessor (not the patient)
should be the one blinded to the treatment
Double Blind Studies
When both the subjects and the investigators are kept from
knowing who is assigned to which treatment, the experiment is
called “double blind"
Serve as a standard by which all studies are judged, since it
minimizes both potential patient biases and potential assessor
biases
Should be used whenever possible, which is whenever it is
ethically
permissible to blind a patient
Double Blinding:Techniques
Coded treatment groups
Placebo for each possible treatment
- tablets identical in physical appearance
- tablets with similar taste and smell
- IV infusions would normally be the same carrier as used for
active
medications
Other treatments "shammed" as far as
possible:
– minimal power ultrasound therapy when testing effect of
physical therapy in back pain
– breathing exercises when assessing the effect of conditioning
exercises
Double Blinding:always feasible??
Situations when double blinding might not be possible
it might not be ethically permissible to blind a patient.
As an example, it is unlikely that sham surgery would
be considered ethical in a study
it might not be possible to blind a patient. For
example, it would be hard to blind a patient to the
therapy given in an exercise study
it might not be possible to blind a patient while
comparing utility of different invasive procedures
Double Blind Studies:stumbling blocks
Side effects:
side effects (observable by patient) are much harder to blind
in general, there are significant ethical problems using placebos to induce
side effects in patients
side effects are, in fact, one of the major ways in which blinding is broken
a way to avoid it is that the side effects of all the potential therapies be
combined into a single list, so that knowledge of side effects would not
indicate therapy (at least to patient)
Efficacy:
a truly effective treatment can be recognized by its efficacy in patients
although rare, some new treatments truly are major leaps. when this
happens, it is usually very clear which treatment a patient is receiving, at
least for the health care providers involved in the trial
Complete Blinding
probably the best approach which can be used, but
requires two groups for data processing, one group to
encode the
data/analysis and one group to perform the analysis
normally only available in major drug company studies,
and not routinely used even then
Complete Blinding:Techniques
analysis uses coded treatment groups
analysis uses coded side effects (e.g., side
effects coded using non-standard scheme, with
only numeric codes available at time of analysis)
analysis uses coded laboratory tests (e.g., name
of test coded numerically at time of analysis,
using non-standard code)