You are on page 1of 108

542-03-#1

STATISTICS 542
Introduction to Clinical Trials


CLINICAL TRIAL DESIGN
542-03-#2
Types of Clinical Research
1. Case Reports
Anecdotal Problem

2. Observational
a. Case Control/Retrospective (lung cancer)
b. Cross Sectional (WESDR) Beaver Dam
c. Prospective (Framington) WESDR-II
Risk Factor Associations

3. Drug Development
(Phase 0, Phase I, & Phase II)
Dose and activity

4. Experimental (Clinical Trial) Phase III
Effect
542-03-#3
Phases of Clinical Trials (Cancer) [1]
Phase 0 - Preclinical
Preclinical animal studies
Looking for dose-response

Phase I
Seeking maximum tolerated dose (MTD)
Patients usually failed other alternatives

Phase II
Estimate of drug activity
Decide if drug warrants further testing (Phase III)
Estimate of serious toxicities
542-03-#4
Phase III
Provide effectiveness of drug or therapy
Various designs
No control
Historical control
Concurrent
Randomized
Testing for treatment effect

Phase IV
Long term post Phase III follow-up
Concern for safety
Phases of Clinical Trials (Cancer) [2]
542-03-#5
542-03-#6
Design
The choice of design depends on the
goal of the trial
Choice also depends on the
population, knowledge of the
intervention
Proper design is critical, analysis
cannot rescue improper design
542-03-#7
Phase I Design
Typical/Standard Design

Based on tradition, not so much on statistical
theory

Dose escalation to reach maximum tolerated
dose (MTD)

Dose escalation often based on Fibonacci
Series

1 2 3 5 8 13 . . . .
542-03-#8
Typical Scheme
1. Enter 3(5) patients at a given dose

2. If no toxicity, go to next dosage and repeat step 1

3. a. If 1 patient has serious toxicity, add 3 more
patients at that does (go to 4)
b. If 2/3 have serious toxicity, consider MTD

4. a. If 2 or more of 6 patient shave toxicity, MTD
reached (perhaps)
b. If 1 of 6 has toxicity, increase dose and go
back to step 1
542-03-#9
Standard Phase I Design
Designed to find dose where 1/3 of
patients experience dose limiting
toxicity (DLT)

Standard escalation design tends to
underestimate target dose

Ref: Storer, Biometrics, 1989
542-03-#10
Dose-response curves
used in simulations (1)
542-03-#11
Dose-response curves
used in simulations (2)
542-03-#12
Summary of Designs Considered (1)
(Storer, Biometrics 45:925-37, 1989)
A. Standard
Observe group of 3 patients
No toxicity increase dose
Any toxicity observe 3 or more
One toxicity out of 6 increase dose
Two or more toxicity stop

B. 1 Up, 1 Down
Observe single patients
No toxicity increase dose
Toxicity decrease dose
542-03-#13
Summary of Designs Considered (2)
(Storer, Biometrics 45:925-37, 1989)
C. 2 Up, 1 Down
Observe single patients
No toxicity in two consecutive increase
dose
Toxicity decrease dose

D. Extended Standard
Observe groups of 3 patients
No toxicity increase dose
One toxicity dose unchanged
Two or three toxicity decrease dose
542-03-#14
Summary of Designs Considered (3)
(Storer, Biometrics 45:925-37, 1989)
E. 2 Up, 2 Down
Observe groups of 2 patients
No toxicity increase dose
One toxicity dose unchanged
Both toxicity decrease dose

B, C, D, E - fixed sample sizes ranging
from 12 to 32 patients
Can speed up process to get to target
dose range
542-03-#15
Phase II Design (1)
References:
Gehan (1961) J ournal of Chronic Disorders
Fleming (1982) Biometrics
Storer (1989) Statistics in Medicine

Goal
Screen for therapeutic activity
Further evaluate toxicity
Test using MTD from Phase I
If drug passes screen, test further

542-03-#16
Phase II Design (2)
Design of Gehan
No control (?)
Two stage (double sampling)
Goal is to reject ineffective drugs ASAP

Decision I: Drug is unlikely to be effective
in > x% of patients

Decision II: Drug could be effective
in > x% of patients
542-03-#17
Phase II Design (3)
Typical Gehan Design

Let x% = 20%
That is, want to check if drug likely to
work in at least 20% of patients

1. Enter 14 patients

2. If 0/14 responses, stop and
declare true drug response s20%

3. If 1+/14 responses, add 15-40
more patients

4. Estimate response rate & C.I.
542-03-#18
Phase II Design (4)
(Why 14 failures?)
Compute probability of consecutive failures









If drug > 20% effective, there would be
~95.6% chance of at least one success

If 0/14 success observed, reject drug
Patient Prob
1 0.8
2 0.64 (0.8 x 0.8)
3 0.512 (0.8 x 0.8 x 0.8)
--- ---
8 0.16
--- ---
14 0.044
542-03-#19
Phase II Design (5)
Stage I Sample Size

Table I

Rejection Effectiveness (%)
Error 5 10 15 20 25 40 50
5% 59 29 19 14 11 6 5
10% 45 22 15 11 9 5 4
542-03-#20
Stage II Sample Size (1)
Based on desired precision of effectiveness
estimate
r
1
= # of successes in Stage 1
n
1
= # of patients in Stage 1



Now precision of total sample N=(n
1

+ n
2
)


Let


1 1 1
/n r p =

1
n
) p

(1 p

) p

SE(
1 1
1

=
N
) p (1 p
) p SE(
* *
*


=
1 *
p p =
542-03-#21
Stage II Sample Size (2)
) p

1.15SE( p

1 1
*
+ =

To be conservative, Gehan suggested





upper 75% confidence limit from first sample

Thus, we can generate a table for size of
second stage (n
2
) based on desired precision

542-03-#22
Additional Patients for Stage II (n
2
)
(Rejection Rate 5% for Stage I)
Therapeutic Effectiveness (%)
Required
Precision
(SE)
Number of
Successes
Stage I
5 10 15 20 25 30
n
1
59 29 19 14 11 9
1 0 4 30 45 60 70
2 0 17 45 63 78 87
3 0 28 58 76 87 91
4 0 38 67 83 89 91
+1 SE
5%
5 0 46 75 86 89 91
1 0 0 0 1 7 11
2 0 0 0 6 12 15
3 0 0 1 9 14 16
4 0 0 3 11 14 16
+1 SE
10%
5 0 0 5 11 14 16
542-03-#23
Additional Patients for Stage II (n
2
)
(Rejection Rate 5% for Stage I)

We might require 10% precision with 20% desired
effectiveness. Assuming 4 or 5 successes in the
first stage ....

n
1
= 14

n
2
= 11


N = 25

We will use estimate p (= r/N) to design a Phase III
study where r = r
1
+ r
2
.
542-03-#24
Phase II Trials
Many most cancer Phase II trials follow
this design
Many other diseases could there seems
to be no standard non-cancer Phase II
design
Might also randomize patients into
multiple arms each with a different dose
can then get a dose response curve
542-03-#25
The foundation for the design of controlled
experiments established for agricultural
experiments

The need for control groups in clinical studies
recognized, but not widely accepted until 1950s

No comparison groups needed when results
dramatic:
Penicillin for pneumococcal pneumonia
Rabies vaccine

Use of proper control group necessary due to:
Natural history of most diseases
Variability of a patient's response to intervention

Phase III Introduction
542-03-#26
Phase III Design
Comparative Studies
Experimental Group vs. Control Group
Establishing a Control
1. Historical
2. Concurrent
3. Randomized
Randomized Control Trial (RCT) is the
gold standard
Eliminates several sources of bias
542-03-#27
Purpose of Control Group
To allow discrimination of patient
outcomes caused by experimental
intervention from those caused by
other factors
Natural progression of disease
Observer/patient expectations
Other treatment
Fair comparisons
Necessary to be informative
542-03-#28
Choice of Control Group
Goals of Controlled Clinical Trials

Types of Control Groups

Significance of Control Group

Assay Sensitivity
542-03-#29
Considerations in Choice of
Control Group
Available standard therapies

Adequacy of the control evidence for
the chosen design

Ethical considerations
542-03-#30
Significance of Control Group
Inference drawn from the trial
Ethical acceptability of the trial
Degree to which bias is minimized
Type of subjects
Kind of endpoints that can be studied
Credibility of the results
Acceptability of the results by regulatory
authorities
Other features of the trial, its conduct, and
interpretation
542-03-#31
Type of Controls
External
Historical
Concurrent, not randomized
Internal and concurrent
No treatment
Placebo
Dose-response
Active (Positive) control
Multiple
Both an Active and Placebo
Multiple doses of test drug and of an active
control
542-03-#32
Use of Placebo Control
The placebo effect is well documented
Could be
No treatment + placebo
Standard care + placebo
Matched placebos are necessary so patients and
investigators cannot decode the treatment
E.g. Vitamin C trial for common cold
Placebo was used, but was distinguishable
Many on placebo dropped out of study
Those who knew they were on vitamin C
reported fewer cold symptoms and duration
than those on vitamin who didn't know
542-03-#33
A new treatment used in a series of subjects

Outcome compared with previous series of
comparable subjects

Non-randomized, non-concurrent

Rapid, inexpensive, good for initial testing of new
treatments

Two sources of historical control data:
Literature Subject to publication bias
Data base
Historical Control Study (1)
542-03-#34
Vulnerable to bias

Changes in outcome over time may
come from change in:
underlying patient populations
criteria for selecting patients
patient care and management peripheral
to treatment
diagnostic or evaluating criteria
quality of data available
Historical Control Study (2)
542-03-#35
Changes in Definitions
542-03-#36
Time Trend
Age-adjusted Death Rates for Selected Causes: United States, 1950-76
542-03-#37
Stat Bite
Cancer and Heart Disease Deaths
Cancer and heart disease are the leading causes of death in the United States. For people less than
age 65, heart disease death rated declined greatly from 1973 to 1992, while cancer death rates declined
slightly. For people age 65 and older, heart disease remains the leading killer despite a reduction in
deaths from this disease. Because cancer is a disease of aging, longer life expectancies and fewer
deaths from competing causes, such as heart disease, are contributing to the increasing cancer
incidence and mortality for those age 65 and older
JNCI 87(16): 1206, 1995
542-03-#38
542-03-#39
Historical Control Study (3)
Tend to exaggerate the value of a new treatment
Literature controls particularly poor
Even historical controls from a previous trial in
the same institution or organization may still be
problematic
Pocock (1977, Brit Med J)
In 19 studies where the same treatment was used in
two consecutive trials, differences in survival ranged
from 46 to 24 , with four differences being statistically
significant
Adjustment for patient selection may be made, but
all other biases will remain
542-03-#40
PRAISE I vs. PRAISE II
Placebo arms
542-03-#41
Concurrent Controls
Not randomized
Patients compared, treated by
different strategies, same period
Advantage
Eliminate time trend
Data of comparable quality
Disadvantage
Selection Bias
Treatment groups not comparable
Covariance analysis not adequate
542-03-#42
Biases in Concurrent Control Study
Types
Magnitude of effects
False positive

Sources
Patient selection
Referral patterns
Refusals
Different eligibility criteria

Experimental environment
Diagnosis/staging
Supportive care
Evaluation methods
Data quality
542-03-#43
Randomized Control
Clinical Trial
Reference: Byar et al. (1976)
New England J ournal of Medicine

Patients assigned at random to either
treatment(s) or control

Considered to be Gold Standard

542-03-#44
Advantages of Randomized
Control Clinical Trial
1. Randomization "tends" to produce comparable
groups

Design Sources of Imbalance

Randomized Chance
Concurrent Chance & Selection Bias
(Non-randomized)
Historical Chance, Selection Bias,
(Non-randomized) & Time Bias

2. Randomization produces valid statistical tests

Reference: Byar et al (1976) NEJM
542-03-#45
Disadvantages of Randomized
Control Clinical Trial
1. Generalizable Results?
Subjects may not represent general
patient population volunteer effect

2. Recruitment
Twice as many new patients

3. Acceptability of Randomization Process
Some physicians will refuse
Some patients will refuse

4. Administrative Complexity
542-03-#46
Bias of Non-RCTs
Example - Peto (1979) Biomedicine
Trials of anticoagulant therapy

Design #Patients P<0.05 Observed
Effect
18 Historical 900 15/18 50%
8 Concurrent 3000 5/8 50%
6 Randomized 3000 1/6 20%

Biases
False positives
Magnitude of effect
542-03-#47
Ethics of Randomization (1)
Statistician/clinical trialist must sell benefits of
randomization

Ethics MD should do what he thinks is best for his
patient
Two MD's might ethically treat same patient quite differently

Chalmers & Shaw (1970) Annals New York Academy of
Science
1. If MD "knows" best treatment, should not participate in trial
2. If in doubt, randomization gives each patient equal chance to
receive one of therapies (i.e. best)
3. More ethical way of practicing medicine
542-03-#48

Byar et al. (1976) NEJ M

1. RCT honest admission best is not
known!
2. RCT is best method to find out!
3. Reduces risk of being on inferior
treatment
4. Reduces risk for future patients
Ethics of Randomization (2)
542-03-#49
Ethics of Randomization (3)
Classic Example -
Reference: Silverman (1977) Scientific Amer

1. High dose oxygen to premature infants was
common practice

2. Suspicion about frequency of blindness

3. RCT showed high dose cause of blindness
542-03-#50
Comparing Treatments
Fundamental principle
Groups must be alike in all important aspects and only differ in the
treatment each group receives
In practical terms, comparable treatment groups means
alike on the average

Randomization
Each patient has the same chance of receiving any of the
treatments under study
Allocation of treatments to participants is carried out using a
chance mechanism so that neither the patient nor the physician
know in advance which therapy will be assigned

Blinding
Avoidance of psychological influence
Fair evaluation of outcomes
542-03-#51
Randomized Phase III
Experimental Designs
Assume:
Patients enrolled in trial have satisfied eligibility
criteria and have given consent
Balanced randomization: each treatment group will
be assigned an equal number of patients

Issue
Different experimental designs can be used to
answer different therapeutic questions
542-03-#52
Commonly Used Phase III Designs
Parallel
Withdrawal
Group/Cluster
Randomized Consent
Cross Over
Factorial
Large Simple
Equivalence/Non-inferiority
Sequential
542-03-#53
Parallel Design
Screen


Trt A
Randomize -
Trt B
H
0
: A vs. B
Advantage
Simple, General Use
Valid Comparison
Disadvantage
Few Questions/Study
542-03-#54
Fundamental Design
Eligible Consent
R
A
N
D
O
M
I
Z
E
No
No
Dropped Dropped
Yes Yes
B
A
Comment: Compare A with B
542-03-#55
Examples of Parallel Designs
VEST
CAST
DCCT
NOTT
IPPB
542-03-#56
Run-In Design
Problem:

Non-compliance by patient may seriously impair
efficiency and possibly distort conclusions

Possible Solution: Drug Trials

Assign all eligible patients a placebo to be taken
for a brief period of time. Patients who are
judged compliant are enrolled into the study.
This is often referred to as the Placebo Run-In
period.
Can also use active drug to test for compliance
542-03-#57
Run-In Design
Screen &
Consent
Run-In
Period
R
A
N
D
O
M
I
Z
E
Unsatisfactory
Dropped
B
A
Note: It is assumed that all patient entering the run-in
period are eligible and have given consent
Satisfactory
542-03-#58
Examples of Run-In Trials
Cardiac Arrhythmia Suppression
Trial (CAST)
Diabetes Control and Complications
Trial (DCCT)
Physicians Health Study (PHS)
542-03-#59
Withdrawal Study
I Trt A
Trt A -
II Not Trt A

H
0
: How long should TRT A continue?
Advantage
Easy Access to Subjects
Show continued Tx Beneficial
Disadvantage
Selected Population
Different Disease Stage
542-03-#60
Cluster Randomization Designs
Groups (clinics, communities) are randomized to treatment
or control
Examples:
Community trials on fluoridization of water
Breast self examination programs in different clinic setting in USSR
Smoking cessation intervention trial in different school district
in the state of Washington
Advantages
Sometimes logistically more feasible
Avoid contamination
Allow mass intervention, thus public health trial
Disadvantages
Effective sample size less than number of subjects
Many units must participate to overcome unit-to-unit variation,
thus requires larger sample size
Need cluster sampling methods
542-03-#61
Randomized Consent Design
Zelen (NEJM, 1979)

Group I: Regular Care
(TRT A)
Patient Randomize
Group II:
Experimental Consent
(TRT B)
NO
(TRT A)
YES
(TRT B)
542-03-#62
Randomized Consent
(Zelen, 1979 NEJM)
Usual Order Proposed Order
Screen Screen

Consent Randomize

Randomize Consent
(from Exp. Group only)

Advantages
Easier Recruitment
Disadvantages
Need Low Refusal Rate
Control Must Be Standard
Unblinded
Ethical?
Refusal Rate Dilution Increase Sample Size
15% 2x
542-03-#63
Cross Over Design
H
0
: A vs. B
Scheme
Period
Group I II
AB 1 TRT A TRT B
BA 2 TRT B TRT A

Advantage
Each patient their own control
Smaller sample size

Disadvantage
Not useful for acute disease
Disease must be stable
Assumes no period carry over
If carryover, have a study half sized
(Period I A vs. Period I B)
542-03-#64
Factorial Design
Schema


Factor I
Placebo Trt B
Factor II
Placebo N/4 N/4
Trt A N/4 N/4
B vs. Placebo
A vs. Placebo
542-03-#65
Factorial Design
Advantages
Two studies for one
Discover interactions

Disadvantages
Test of main effect assumes no interaction
Often inadequate power to test for interaction
Compliance

Examples
Physicians' Health Study (PHS) NEJM 321(3):129-135, 1989.
Final report on the aspirin component
Canadian Cooperative Stroke Study (1978) NEJM p. 53
542-03-#66
Physicians Health Study
542-03-#67
Physician Health Study
542-03-#68
Physicians Health Study
542-03-#69
Physicians Health Study
542-03-#70
Superiority vs.
Non-Inferiority Trials
Superiority Design: Show that new treatment
is better than the control or standard
(maybe a placebo)

Non-inferiority: Show that the new treatment
a) Is not worse that the standard by more than
some margin
b) Would have beaten placebo if a placebo arm
had been included (regulatory)
542-03-#71
Non-inferiority Trial
Trial with active (positive) controls
The question is whether new (easier or cheaper)
treatment is as good as the current treatment
Must specify margin of equivalence or non-
inferiority
Can't statistically prove equivalency -- only show
that difference is less than something with specified
probability
Historical evidence of sensitivity to treatment
Small sample size, leading to low power and
subsequently lack of significant difference, does not
imply equivalence
542-03-#72



Possible outcomes in a non-inferiority trial
(observed difference & 95% CI - Pocock)
New Treatment Better New Treatment Worse
Superior
Non-inferior
Non-inferior
Tricky (& rare)
Inconclusive
Inconclusive
Inferior, but
Inferior
H
G
F
E
D
C
B
A

0 Delta
Treatment Difference

542-03-#73

Difference in Events
Test Drug Standard Drug
542-03-#74
Superior vs NonInferiority Designs
1.0
(
)
(
)
(
)
.8 1.25
Benefit Harm
RR
A
A A
Better Worse
RR
Active Control
Placebo
Harm
Non-significant
Benefit
(
)
(
)
(
)
1.0
Standard
Worse
Non-Inferior
Better
Modified from Fleming, 1990
X
X
X
X
X
X
542-03-#75
Non-Inferiority Challenges (1)
Requires high quality trial

Poor execution favors non-inferiority

Requires strong control; weak
control favors non-inferiority
542-03-#76
Non-Inferiority Challenges (2)
Treatment margin somewhat
arbitrary

Imputed Trt vs. Plbo effect
Uses historical control concept
Imputed estimate not very robust
542-03-#77
Steering Committee
J. Kjekshus (Chair), K. Dickstein (Coordinator),
S. G. Ball, A. J. S. Coats, R. Dietz, A. Kesniemi, E. S. P. Myhre,
M. S. Nieminen, K. Skagen, K. Swedberg, K. Thygesen, H. Wedel,
R. Willenheimer, A. Zeiher, J. C. Fox and K. Kristianson

Endpoint Committee
J. G. F. Cleland and M. Romo

Data Safety and Monitoring Board
D. Julian (Chair), A. Bays de Luna, D. L. DeMets,
C. D. Furberg, W. W. Parmley and L. Rydn
OPtimal Trial In Myocardial infarction with the
Angiotensin II Antagonist Losartan
OPTIMAAL
Lancet 2002; 360:752-60
542-03-#78
Rationale
ACE inhibitors reduce mortality in
high risk post MI patients
Selective Angiotensin II Receptor
Antagonists are an alternative
because of more complete blockade
of tissue RAAS
Better tolerability
542-03-#79
Hypothesis
Losartan (50 mg) is superior or non-inferior
to captopril (150 mg) in decreasing all-cause
mortality in high-risk patients following AMI
Double-blind, randomized, parallel,
investigator initiated, no placebo control
Event driven (all-cause death = 937)
Multicentre (Denmark, Finland, Germany,
Ireland, Norway, Sweden, UK)
Study design
542-03-#80
Captopril as Comparator
Captopril has well documented
benefits
Captopril 50 mg 3 times daily has
indication for CHF worldwide
Widely used, available as generic

542-03-#81
Statistical Methods
937 deaths required for 95% power to
detect a 20% difference between groups

Non-inferiority margin of 10% chosen
based on placebo-controlled trials of
ACE-inhibitors

Analysis by Intention-to-Treat and Cox
regression model
542-03-#82
All-cause death
losartan (n) 2744 2504 2432 2390 2344 2301 1285
captopril (n) 2733 2534 2463 2423 2374 2329 1309
Month
6 12 18 24 30 36
0
5
10
15
20
25
E
v
e
n
t

r
a
t
e

(
%
)

losartan (n=499 events)
captopril (n=447 events)
Relative Risk = 1.13 (0.99 to 1.28); p=0.069
0
542-03-#83
Subgroup Analyses
0.6

1 1.5 2
losartan better captopril better
Age <65 2170
65-74 1840
>75 1467
Gender Female 1575
Male 3902
Diabetes Non-diabetic 4537
Diabetic 940
Killip class Killip class 1 1735
Killip class 2 3131
Killip class 3-4 609
Heart failure No heart failure 1060
Heart failure 4417
Infarct location Infarct ant/lat 3821
Infarct inf/post 1152
Prior MI No prior MI 4479
Prior MI 998
Thrombolytic use No thromb use 2499
Thromb use 2978
0-blocker use No 0-blocker use 1171
0-blocker use 4306
Overall 5477

n Hazard ratio (95% CI)
542-03-#84
Effect of losartan
relative to placebo?
Rel. Risk % change
captopril vs. placebo* 0.805 - 19.5
losartan vs. captopril (OPTIMAAL) 1.126 12.6

losartan vs. putative 0.906 - 9.4
placebo (0.805 x 1.126)
* SAVE, AIRE. TRACE, SMILE, GISSI III, CONSENSUS II and ISIS IV
542-03-#85
Non-Inferiority Methodology
a) Comparison: New Treatment vs. Standard
RR
a


b) Estimate of standard vs. placebo RR
b

(based on literature)

c) Imputed effect of New Trt vs. placebo (RR
c
)
RR
c
= RR
a
x RR
b

542-03-#86
Assay Sensitivity
Ability to distinguish an effective treatment from a
less effective or ineffective treatment

Different implications of lack of assay sensitivity

Superiority trials
Failing to show that the test treatment is superior
Thus failing to lead to a conclusion of efficacy

Non-inferiority trials
Finding an ineffective treatment to be non-inferior
Thus leading to an erroneous conclusion of efficacy
542-03-#87
Large, Simple Trial
Advocated for common pathological conditions

To uncover even modest benefits of intervention

That are easily implemented in a large population

Intervention unlikely to have different effects in
different patient subpopulations

Unbiased allocation to treatments

Unbiased and easily ascertained outcome

Very limited data collection
542-03-#88
CAPRIE
Design
Ischemic stroke, MI, atherosclerotic PAD
Clopidogrel
75 mg/day PO
Aspirin
325 mg/day PO
Completed Trial
(N = 9,577)
Completed Trial
(n = 9,566)
Source: CAPRIE Steering Comm. Lancet. 1996; 348:1329
542-03-#89
CAPRIE
Risk Reduction by Major Outcomes
Ischemic stroke

MI

Vascular death

All events
Percentage Relative Risk Reduction
-40 -20 0 20 40
8.7
7.6
19.2
5.2
p = 0.419
p = 0.008
p = 0.29
p = 0.043
542-03-#90
Sequential Design
Continue to randomize subjects until H
0
is
either rejected or accepted

A large statistical literature for classical
sequential designs

Developed for industrial setting

Modified for clinical trials
(e.g. Armitage 1975, Sequential Medical Trials)
542-03-#91
Classical Sequential Design (1)
Continue to randomize subjects until H
0
is either rejected or accepted

Classic
-
Net

Trt

Effect
100 200 300
No. of Paired Observations
Trt Worse
Continue
Accept H
0

Trt Better
Continue
-20
0
20
542-03-#92
Classical Sequential Design (2)
Assumptions
Acute Response
Paired Subjects
Continuous Testing

Not widely used

Modified for group sequential designs
542-03-#93
Beta-blocker Heart Attack Trial
(BHAT)
Design Features

Mortality Outcome 3,837 patients
Randomized Men and women
Double-blind 30-69 years of age
Placebo-controlled 5-21 days post-M.I.
Extended follow-up Propranolol-180 or 240 mg/day

Preliminary Report. JAMA 246:2073-2074, 1981
Final Report. JAMA 247:1707-1714, 1982
542-03-#94
BHAT GSB
542-03-#95
Therapeutic vs. Prevention Trials
Prevention Trials
Primary - Prevent disease
Secondary - Prevent recurrence

Therapeutic Trials
Treat disease

Basic fundamentals apply equally

Some differences exist
Complexity
Recruitment Strategies
Compliance
Length of Follow-up
Size
542-03-#96
Confounding Bias
Suppose you are interested in the effects of
a treatment T upon an outcome O in the
presence of a predictor P

Randomization takes care of bias due to
factors P before treatment

Blinding takes care of bias due to factors P
after treatment

542-03-#97
Masking or Blinding (1)
Keeping the identity of treatment assignments
masked for:
1. Subject
2. Investigator, treatment team or evaluator
3. Evaluation teams

Purpose of masking: bias reduction
Each group masked eliminates a different
source of bias
Masking is most useful when there is a subjective
component to treatment or evaluation
542-03-#98
Blinding or Masking (2)
No Blind
All patients know treatment
Single Blind
Patient does not know treatment
Double Blind
Neither patient nor health care provider know
treatment
Triple Blind
Patient, physician and statistician/monitors do
not know treatment
Double blind recommended when possible

542-03-#99
Blinding or Masking (3)
Assures that subjects are similar with regard
to post-treatment variables that could affect
outcomes

Minimizes the potential biases resulting from
differences in management, treatment, or
assessment of patients, or interpretation of
results

Avoids subjective assessment and decisions
by knowing treatment assignment
542-03-#100
Feasibility of Masking
Ethics: The double-masking procedure should not
result in any harm or undue risk to a patient
Practicality: It may be impossible to mask some
treatments
Avoidance of bias: Masked studies require extra
effort (manufacturing look-alike pills, setting up coding
systems, etc.)
Compromise: Sometimes partial masking, e.g.,
independent masked evaluators, can be sufficient to
reduce bias in treatment comparison
Although masked trials require extra effort, sometimes
they are the only way to obtain an objective answer to
a clinical question
542-03-#101
Reasons for Subject Masking
Those on no-treatment or standard treatment
may be discouraged or drop out of the study

Those on the new drug may exhibit a placebo
effect, i.e., the new drug may appear better when it
is actually not

Subject reporting and cooperation may be biased
depending on how the subject feels about the
treatment
542-03-#102
Unbiased Evaluation
Subject Bias (NIH Cold Study)
(Karlowski, 1975)

Duration of Cold (Days)
Blinded Unblinded
Subjects Subjects
Placebo 6.3 8.6
Ascorbic Acid 6.5 4.8
542-03-#103
Reasons for
Treatment Team Masking
Treatment decisions can be biased by knowledge of
the treatment, especially if the treatment team has
preconceived ideas about either treatment

Dose modifications

Intensity of patient examination

Need for additional treatment

Influence on patient attitude through enthusiasm
(or not) shown regarding the treatment
542-03-#104
Unbiased Evaluation
Investigator Bias - (Taste & Smell Study)
(Henkin et al, 1972 & 1976)

Single Blind Double Blind
Zinc 8/8* 5/8
Placebo 0/8 7/8


*Number of variables with significant
improvement/Number of variables
542-03-#105
Reasons for Evaluator
(Third Party) Masking
If endpoint is subjective, evaluator bias will lead to
recording more favorable responses on the preferred
treatment

Even supposedly hard endpoints often require
clinical judgment, e.g., blood pressure, MI
542-03-#106
Reasons for Monitoring
Committee Masking
Treatments can be objectively evaluated

Recommendations to stop the trial for ethical
reasons will not be based on personal biases

Sometimes, however, triple-mask studies are hard
to justify for reasons of safety and ethics

A policy not recommended, not required by FDA
542-03-#107
Design Summary
Design used must fit goals of trial
RCT minimizes bias
Superiority vs. Non-Inferiority trial
challenges
Use blinding when feasible to
minimize bias after randomization
542-03-#108
West Campus