Transcript Lecture8

Blinding or Masking of
Treatments and of Other Aspects
of the Trial
Two Important Methods for
Removing Bias in Clinical Trials
1. Randomization
2. Blinding
– Treatment assigned and application
– Endpoint assessment
Bias (def.) - A systematic error usually
introduced (conscious or unconscious)
by investigator / trial participant which
leads to incorrect estimates of the
treatment effect
Blinding of Treatments
• Feature of design to eliminate bias
associated with physician/patient being
aware of treatment that is given
Note: This is different from the type of
bias that randomization prevents and
from allocation concealment
Examples of Bias Eliminated or
Reduced with Blinding
• Differential/preferential ancillary/compensatory treatment (cointervention bias)
• Psychological impact of being treated with what might be
perceived to be superior treatment (placebo effect)
• Differential ascertainment/diagnosis of endpoints (primary
outcomes and toxicities) – particularly important for subjective
outcomes
• Differential compliance/visit attendance/record
keeping/withdrawal
ICH Guidelines (E 10)
• Potential biases blinding can prevent/minimize:
– Patients may be more likely to report benefit on
active drug.
– Observers may be more likely to report favorable
outcomes and adverse effects on active drug.
– Knowledge of treatment may affect rigor of follow-up.
– Knowledge of treatment could affect decisions to use
concomitant treatment or to stay on study drug.
– Knowledge of treatment could affect decision to
leave in the analysis.
– Knowledge of treatment could affect choice of
statistical analysis.
Overview of Trials in Pregnancy or
Childbirth*
• 250 trials; authors studied the association of
methodological rigor with treatment effect as
measured by odds ratio (i.e., interaction of
treatment effect with quality measure)
• Two significant predictors: 1) unclear allocation
concealment resulted in more extreme treatment
differences than adequate measures (p<0.001);
and 2) trials that were not double-blind resulted in
more extreme treatment differences than doubleblind studies (p=0.01).
*
Schulz et al., JAMA, pp.408-412, 1995.
Pocock identifies four areas to
consider to determine whether
blinding is feasible
1) Ethics - undue harm to patient
2) Practicality - similarity of treatments
3) Avoidance of bias - How much bias?
4) Compromise – partial blinding
MRC Trial on Tuberculosis
• MRC committee considered whether the trial
on streptomycin should be blinded and
whether the investigator should be allowed
to modify therapy (e.g., collapse therapy).
• Use of placebo would have required IM
injection 4xday for 4 months
• Hill argued “no need in the search for
precision to throw common sense out the
window”
FDA Panel Discussion on Blinding
• A randomized trial (RE-LY) was carried out for patients with
atrial fibrillation comparing dabigatran, an oral direct
thrombin inhibitor that provides stable anticoagulation at a
fixed dose without any need for laboratory control, with
warfarin,a vitamin K antagonist, that requires strict
laboratory control.
• The trial was not blinded and the FDA asked the panel to
consider if open-label warfarin was reasonable.
• The trial was published in Lancet 2010; 376:975-983.
Panel questions can be found at
ww.fda.gov/downloads/AdvisoryCommittees/CommitteesM
eetingMaterials/Drugs/CardiovascularandRenalDrugsAdvis
oryCommittee/UCM226006.pdf
Blinding
1) of study participant
2) of treatment team
3) of endpoint assessment (e.g., endpoint
review committee)
4) of data monitoring committee
5) of data analysts and management staff
Blinding of Treatments
Non-blind
- Investigator and patient know
treatment assigned (“open label”)
Single-blind
- Investigator knows treatment
assigned but patient does not (in
some cases participants knows and
investigator does not)
Double-blind - Neither investigator nor patient
knows treatment assigned
General view - Double-blind > single blind > nonblind
Blinding in a Non-Blind Study
1. Accumulating data: investigators should be
blinded to group data; an external independent
data monitoring committee (DMC) should not be.
2. Endpoint committee assessments and laboratory
measurements (can be performed blinded to
treatment group even in an open trial).
PROBE Design
• Prospective, Randomized, Open-label,
Blinded Endpoint Design
• Phrase coined by Hannsson et al (Blood
Pressure 1992)
• Motivation:
– More similar to clinical practice
– Easier to enroll trials
– Better patient compliance
– Cheaper (?)
Example: Non-Blind Study
Aspirin trial in British male doctors
Aspirin
(500 mg Daily)
Control
(Avoid Aspirin)
No.
3429
1710
Age < 60(%)
46.8
47.0
Never smoker
25.1
23.1
Hypertension
10.2
9.3
Diabetes
2.0
1.9
BMJ 1988; 296: 313-316.
British Aspirin Study
Stopped taking
aspirin(%)
Started taking
aspirin(%)
*
Aspirin
44.3*
–
Control
–
12
19.5% in first year
The potential for non-compliance with the treatment
assignment needs to be considered in the design – very
important in a non-blind study of a readily available treatment.
Physician’s Health Study
Double-Blind Study
Aspirin
(N=11,037)
Compliance
85.71
Placebo
for Aspirin
(N=11,034)
85.74
N Engl J Med 1989; 321: 129-135.
Example: Non-Blind Study
MRFIT
Risk Factor Changes
After 6 Years
SI
UC
-10.5
-7.4
∆ DBP (mm Hg)
∆ cholesterol (mg/dl)
% quitting smoking
-12.3
-6.4
46
29
JAMA 1982; 248: 1465-1477.
MRFIT
Endpoint Ascertainment
• Mortality review blinded to treatment group
for assigning cause of death
Issues which had to be dealt with:
• Rapidity with which deaths were ascertained
• Completeness of data
• Assuring blinded review
Examples of Studies Where Blinding
of Treatment is Difficult or Impossible
• Surgical (e.g., device) vs. medical treatment (even in this
situation, one may be able blind some members of the
treatment team)
• Non-pharmacologic and behavioral interventions (e.g., diet,
rehabilitation)
• Utility of genotypic resistance testing versus not using it for
choosing salvage treatments for patients with HIV
• Interventions to improve patient adherence
• Strategic trials of how to use treatments based on disease
markers (e.g., START trial on when to start ART)
Design:
Randomized, single/double-blind 2x2
factorial, multi-center clinical trial.
Randomization
Placebo
+
Alt Points
Placebo
+
Acupuncture
Amitriptyline
+
Alt Points
Sample Size: 260 patients (65 in each arm)
JAMA 1998; 280: 1590-1596.
Amitriptyline
+
Acupuncture
Example: Single Blind Study
Pulmonary Embolism Trial
(UPET)
12 hours urokinase + heparin
vs.
heparin alone
Endpoint: 24-hour clot resolution;
symptom relief;
complications
Circulation 1973; Supplement II
Nature of Blinding
• Patient was blinded
• Two members of medical staff were aware
of treatment because of clotting studies for
patient management
• Daily evaluations of symptoms by blinded
staff
• Blinded evaluation of angiograms and lung
scans.
How well was the blind maintained?
of patient?
Very well; no objective data
of treatment team?
Not very well because of bleeding
complications
of endpoint evaluation?
Example: Double Blind Study
TOMHS
AIM: Among mild hypertensive men and
women does the addition of drug to intensive
nutrition intervention result in a reduction of
CVD morbidity/mortality?
JAMA 1993; 270:713-724
TOMHS
Weight Loss + Na Reduction +
Alcohol Reduction
and
(1)
Placebo
(2)
Acebutolol
(400 mg)
(3)
Amlodipine
(5 mg)
(4)
Chlorthalidone
(15 mg)
(5)
Doxazosin
(2 mg)
(6)
Enalapril
(5 mg)
TOMHS: Double-Dummy
Weight Loss + Na Reduction +
Alcohol Reduction
and
(1)
Placebo ( )
Placebo ( )
(2)
Acebutolol ( )
Placebo ( )
(3)
Placebo ( )
Amlodipine ( )
(4)
Placebo ( )
Chlorthalidone ( )
(5)
Placebo ( )
Doxazosin ( )
(6)
Placebo ( )
Enalapril ( )
Blinding in Influenza-IVIG Study
•INSIGHT 005: FLU-IVIG Pilot Study *
PID Number: [write 8-digit PID number here]
Anti-Influenza Hyperimmune IVIG or Placebo
Directions for use: Infuse entire contents over a
continuous period (approximately 2 hours)
Start infusion as soon as possible after the date
and time treatment was prepared
FOR INVESTIGATIONAL USE ONLY
Time for preparing placebo should mimic that for IVIG (e.g., thawing and
preparing proper dose weight based dose.
Blinding can complicate the treatment
regimen and change the nature of the
question being addressed
Design Considerations for the
Alzheimer’s Disease Anti-inflammatory
Prevention Trial (ADAPT)
• Randomization to naproxen (220 mg bid),
celecoxib (200 mg bid) or placebo (1:1:1.5)
• Four placebo design options were
considered to allow masking
Cont Clin Trials 2002; 23:93-99.
Obtain drug in powder form
and repackage.
Obtain in existing formulation
and encapsulate.
Partially blinded with existing
formulations.
Fully blinded, double dummy.
Prevention of Toxoplasmic Encephalitis
Design and Recruitment Goals
Lancet 1992; 339:333-334 and JID 1994;169:384-394
750 Patients
Unblinded
375 Clindamycin Arm
Blinded
250 Active
Treatment
375 Pyrimethamine Arm
Blinded
125
Placebo
250 Active
Treatment
125
Placebo
Combination Nucleoside (NuCombo) Study
N Engl J Med 1996;335:1099-1106
Unblinded
ddI Arm
Blinded
ddI
+
AZT
ddC Arm
Blinded
Placebo (ddI)
+
AZT
ddC
+
AZT
Placebo (ddC)
+
AZT
No. Tablets
Morning
4
4
4
4
Afternoon
2
2
4
4
Evening
4
4
4
4
10
10
12
12
NuCombo Study
Alternative Design
“Double-Dummy”
Blinded
ddI
+
ddC Placebo
+
AZT
ddI Placebo
+
ddC
+
AZT
ddI Placebo
+
ddC Placebo
+
AZT
Morning
6
6
6
Afternoon
4
4
4
Evening
6
6
6
16
16
16
No. Tablets
NuCombo Study
Alternative Design
Unblinded
ddI
+
AZT
ddC
+
AZT
Morning
4
4
2
Afternoon
2
4
2
Evening
4
4
2
10
12
6
AZT
No. Tablets
Example: Clinical Trial of Radiotherapy (R) Alone vs.
Radiotherapy Preceded by Drug Treatment (DR) for 30
Days
DR
R
Drug
Pragmatic Approach
R
R
Time
Which treatment is better when administered under usual conditions?
DR
Drug
R
No Drug
R
Explanatory Approach
R
Time
Does drug have a sensitizing effect?
Schwartz D and Lellouch J, J Chronic Dis, 1967.
Vaginal Microbicide
to Prevent HIV Infection
(1)
Microbicide
Placebo
(2)
Microbicide
“Condom only”
(no gel)
(3)
Microbicide
Placebo
“Condom only”
(no gel)
Double blind
Non- blind
Partially blind
Concern: Use of microbicides might decrease use of condoms.
Implementation of Double Blind
Design
1. Maintenance of blind
– Intermediate response variables
– Laboratory data
2. Preparation and packaging of drugs
– Similar in appearance, taste, weight
– Labeling; unique bottle numbers
– Quality control
3. Breaking the blind (this should be tracked and
reported)
4. Evaluation of blinding
Maintenance of Blind
in the CPPT Study of Cholestyramine
Cholestryamine
Placebo
Patients
56.0
54.6
Treatment team
55.2
52.9
% assignments guessed correctly
JAMA 1984; 251:351-64
Maintenance of Blind
in Mt. Sinai Hypertension Trial:
Potassium Supplementation versus
Placebo
% Correct
Participant
60
Nutritionist
49
Nurse
56
N Engl J Med 1990; 322: 569-574.
How Blind is Blind?
• Should the blind be assessed? before the
trial begins? as the trial is ongoing? at the
end?
• Is the bias that results from “unblinding”
different if it is due to substantial efficacy
versus minor side effects?
• How should the blind assessment, if done,
be used to adjust/interpret the primary
results?
Evaluation of Blinding in 191 Trials
Published in General Medicine and Psychiatric Journals
• 7 of 97 trials (7%) in general medicine journals
reported success of blinding
• 8 of 94 (9%) in psychiatric journals reported
success of blinding
Ferguson D et al., BMJ 2004;328:432-437.
Evaluation of Methods of Blinding in 819 Trials
of Pharmacologic Treatments Published in Major
Journals in 2004
• Reporting of blinding in trials is poor – 58% of trials
reported method of blinding (should always state who was
blinded and how in addition to using terms like single- and
double-blind).
• 28% blinding of patients, health care providers and
outcome assessors; 14% blinding of patients and health
care providers.
• 24% patients only
• 0.1 % health care providers only
• 21% outcome assessors but not health care providers
Boutron et al., PLoS Med 2006;3:1931-1939.
CONSORT Guidelines for Reporting
Results of Trials
• Blinding
– Who was blinded to the interventions and how
was the blind implemented.
– If relevant, description of the similarity of the
interventions.
– TOMHS: “To facilitate the double-blind design,
active drugs and placebo were prepared in
capsule form. Since all active treatments could
not be provided in the same size capsule,
participants took two different-sized capsules
daily as the initial dose”.
Trial Reports of Blinding (cont.)
• Acupuncture study: “To maintain blinding and to determine the need for
supplemental points, the acupuncturists asked all patients a series of
standard questions, irrespective of treatment arm…The placebo
capsules were identical in appearance and taste to the active
capsules.”
• Weight loss diets (N Engl J Med 2009;360:859-873): Except for the
interventionists (dieticians and behavioral psychologists) investigators
and staff were kept blind to diet assignments…The trial adhered to
established procedures to maintain separation between staff that take
outcome assignments and staff that deliver the intervention. Staff who
obtained outcome assignments were kept blind to diet group
assignment. All investigators…were kept masked to outcome
measurements and trial results.”
Summary
• Blinding is an effective way to reduce/eliminate
bias in clinical trials – do it when you can.
• Blinding does not guarantee valid results
• Willingness to compromise is essential
– Feasibility (cost, time, patient/investigator interest,
patient safety)
– Necessity and common sense (how much bias)
• In some circumstances, blinding can change the
nature of the research question.
• Consider opportunities for blinding carefully
before the trial begins.