experimental study: clinical trial - interfetpthailand€¦ · assignment of subjects to study...
TRANSCRIPT
Kam
ol U
dol
Experimental Study:Clinical Trial
Kamol Udol, MD, MSc
Faculty of Medicine Siriraj Hospital
Mahidol University
1
Kam
ol U
dol
Classification of Study Designs in Clinical Medicine and Epidemiology
Descriptive Study(No specific hypothesis)
● Case report● Case series● Descriptive study based on
rates (e.g. Prevalence or Incidence study)
Analytic study(Specific hypothesis)
● Experiment
● Non-experiment Quasi-experiment
Cohort
Retrospective
Prospective
Case-Control
Cross-sectional
Longitudinal
Ecological
Other
Unit of study can be either individual people or groups of people2
Kam
ol U
dol
Concepts
3
Kam
ol U
dol
Experiment
● A set of observations, conducted under controlled circumstances, in which the scientist manipulates the conditions to ascertain what effect, if any, such manipulation has on the observations
Rothman KJ, Greenland S, Lash TL. Modern Epidemiology 3rd ed, 2008.4
Kam
ol U
dol
Experiment
A study of
CAUSE-EFFECT association
5
Kam
ol U
dol
Validity for Causal Inference
Validity Ranking Types of Study Design
Highest Experimental study
Quasi-experimental study
Prospective cohort study
Retrospective cohort study
Nested case-control study
Time-series analysis
Cross-sectional study
Ecologic study
Case study
Lowest Anecdote
Adapted from Environmental Health Perspectives 1997;15:1079.6
Kam
ol U
dol
Health Sciences
Human subjects
Health outcomes
Exposure
Causal association?
Manipulable? Ethical?
7
Kam
ol U
dol
Experimental Study
● Exposure conditions must be amenable to manipulation
● To be ethical, exposure assignments must be expected to cause more good than harm
Therapeutic or preventive interventions
8
Kam
ol U
dol
● Possible factors other than drug A that may be related to cholesterol reduction (Extraneous or confounding factors)
Other co-interventions: E.g., Lifestyle modifications
Regression to the mean phenomenon
Subjects with hypercholesterolemia
Reduction of cholesterol
Drug A
9
Kam
ol U
dol
Subjects Outcomes
Intervention/Exposure
● Extraneous Factors Other co-interventions
Regression to the mean phenomenon
Natural course
Learning effect if measurement is done more than once
Change in properties of the measurement tools over time
Placebo effect
Various unknown factors
10
Kam
ol U
dol
Experimental Study
Main features
● There must be a control group
Not a unique feature of experimental study
11
Kam
ol U
dol
Physicians’ Health Study
Side effects Aspirin (%)
GI symptoms 34.8
Upper GI ulcers 1.5
Bleeding problems 27.0
N Engl J Med 1989;321:129-35.12
Without a control group, one might conclude that aspirin causes GI symptoms, upper GI ulcers and bleeding problems.
Kam
ol U
dol
Physicians’ Health Study
Side effects Aspirin (%)
GI symptoms 34.8
Upper GI ulcers 1.5
Bleeding problems 27.0
Control (%) p value
34.2 0.48
1.3 0.08
20.4 <0.001
N Engl J Med 1989;321:129-35.13
With a control group, it’s only bleeding problems that are related to aspirin.
Kam
ol U
dol
Experimental group
Control group
Intervention
Outcome
● Difference in outcome can be attributed to the intervention only if both groups have similar probability of developing the outcome at the beginning
● Similar prognosis
14
Kam
ol U
dol
Experimental group
Control group
Intervention
Outcome
● Difference in outcome cannot be attributed solely to the intervention as both groups have different prognosis at the beginning
15
Kam
ol U
dol
Experimental Study
Main features● There must be a control group
Not a unique feature of experimental study
● Experimental and control groups must have similar pre-intervention risk of developing the outcomes (i.e. similar prognosis)
● Investigators manipulate the assignment of exposure to study participants to ensure similar prognosis between groups The unique feature of experimental study
16
Kam
ol U
dol
Experimental Study: Basic Structure
Population
Sample
Assignment
Experiment Control
Outcome Experimental Group
OutcomeControl Group
Definition of population
Selection of subjects to be studied
Assignment of subjects to study group
Administration of intervention
Measurement of outcome
Controlled by investigator
17
Kam
ol U
dol
Experimental Study
● Clinical trials
Patients as subjects
● Field trials
Healthy individuals as subjects
● Community intervention trials
Groups of people in communities as subjects
18
Kam
ol U
dol
Phases of Clinical Trials
● Phase I Dose ranging Small group of healthy subjects
● Phase II Efficacy and side effects Small group of patients with specific diseases
● Phase III Confirmation of efficacy, effectiveness and safety Large group of patients with specific diseases
● Phase IV Postmarketing surveillance
Kam
ol U
dol
Methodology
20
Kam
ol U
dol
Validity of Research
● The extent to which the research result (answer) represents the truth
● Research result = Truth + Error
● Error
Random error (Chance)
Remedy: Appropriate sample size
Systematic error (Bias)
Remedy: Appropriate methodology
21
Kam
ol U
dol
Validity of Research
● Internal validity
Accuracy
● External validity
Generalizability, Practical utility
22
Kam
ol U
dol
Design Considerations
● Research question
● Selection of participants
● Selection of intervention and control
● Selection of outcomes (endpoints)
● Data analysis and sample size
● Measures to reduce bias
● Ethical considerations
23
Kam
ol U
dol
Research Question
● Effectiveness of therapeutic or preventive interventions Pharmacotherapy
Surgical procedures
Physical therapy
Lifestyle interventions
Educational programs
Vaccine
Etc.
24
Kam
ol U
dol
Research Question
● Primary research question The most important question the study wants to
answer The answer can be conclusive There must be at least one primary research
question (there is usually one)
● Secondary research question Less important questions The answers can only be hypothesis generation There can be any number of secondary research
question (there are usually many)
Kam
ol U
dol
Good Research Question: FINER
● Feasible Adequate number of subjects Adequate technical expertise Affordable in time and money Manageable in scope
● Interesting● Novel
Provides new findings Confirms, refutes or extends previous findings
● Ethical● Relevant
To scientific knowledge To clinical and health policy To future research
26
Kam
ol U
dol
Research Question: Components
PICO
● Patients / Population
● Intervention
● Comparison intervention (Control)
● Outcome
● Each component must be defined as specifically and clearly as possible
27
Kam
ol U
dol
Research Question
● Should be as clear and specific as possible
Not “Is drug I better than drug C?”
But “In population P, does drug I at daily dose X reduce outcome O over a period of time T more than drug C at daily dose Y by the magnitude M?”
28
Kam
ol U
dol
Design Considerations
● Research question
● Selection of participants
● Selection of intervention and control
● Selection of outcomes (endpoints)
● Data analysis and sample size
● Measures to reduce bias
● Ethical considerations
29
Kam
ol U
dol
Selection of Participants
● Eligibility criteria
Identify a population in which it is feasible, ethical and relevant to study the impact of the interventions on outcomes
30
Kam
ol U
dol
Selection of Participants
● Eligibility criteria
Inclusion criteria
Define the main characteristics of the target population that pertain to the research question
Exclusion criteria
Specific characteristics rendering individuals not suitable to enter the study despite fulfilling the inclusion criteria
Should be parsimonious (enhance generalizability)
31
Kam
ol U
dol
Exclusion criteria should fit either one of these 4 reasons
1.Conditions in which the study treatment is clearly indicated ACEI is indicated in congestive heart failure
(CHF); a study of ACEI in other population must exclude patients with CHF
2.Conditions in which the study treatment is contraindicated or would be harmful Unacceptable risk of adverse reaction, e.g. history
of severe allergic reaction to the study treatment Pregnant or lactating women Severe renal or hepatic impairment
32
Kam
ol U
dol
Exclusion criteria should fit either one of these 4 reasons
3. Subjects unlikely to benefit form study treatment Short life expectancy
High likelihood of non-adherence or being lost to follow-up
Not likely to respond to treatment due to some characteristics
4. Practical problems with participating in the protocol Impaired mental status
Language barrier
Concurrent participation in another study
33
Kam
ol U
dol
Design Considerations
● Research question
● Selection of participants
● Selection of intervention and control
● Selection of outcomes (endpoints)
● Data analysis and sample size
● Measures to reduce bias
● Ethical considerations
34
Kam
ol U
dol
Selection of Intervention
● Intensity, duration and frequency
Balance effectiveness and safety
Serious conditions: Effectiveness may be the primary concern
Highest tolerable dose
Mild conditions or prevention: Safety may be the primary concern
Lowest effective dose
Generalizability
Simple interventions are better
35
Kam
ol U
dol
Selection of Control
● Concurrent standard of care
No treatment
Placebo
Active treatment control
● Allowance or restriction of co-interventions (for both intervention and control groups)
+ Supportive care
36
Kam
ol U
dol
Design Considerations
● Research question
● Selection of participants
● Selection of intervention and control
● Selection of outcomes (endpoints)
● Data analysis and sample size
● Measures to reduce bias
● Ethical considerations
37
Kam
ol U
dol
Outcome (Endpoint)
● Primary outcome
To address primary research question
● Secondary outcome
To address secondary research question(s)
● Adverse outcomes
38
Kam
ol U
dol
Outcome
● Single event
Total mortality, Cause-specific mortality, hospitalization, occurrence of a condition e.g. MI, stroke, recurrence of cancer, etc.
● Combination of events: Composite outcome
Should be capable of meaningful interpretation such as being related through a common underlying conditions
Example: Cardiovascular death, non-fatal MI, recurrent severe angina or unplanned coronary revascularization
39
Kam
ol U
dol
Composite Outcome
● Advantages
Reduce sample size
Reduce bias
● Disadvantages
Assume equal importance of each component
Death=MI=Angina
Assume similar effect of the intervention on each component
Complicate interpretation of the result
40
Kam
ol U
dol
Choice of Outcome (1)
● Responsiveness to intervention
● Relevance Clinical or “hard” outcome preferable to
“surrogate” or “physiologic” outcome (relevant to patients & physicians)
Quality of life (relevant to patients & physicians)
Economic outcome (relevant to government, policy makers, payers of health care cost)
● Credibility Assessment can be objective and unambiguous
41
Kam
ol U
dol
Choice of Outcome (2)
● Capability of being assessed in all participants, in an unbiased fashion, and as completely as possible
Similar outcome for all subjects
Similar technique of measurement
Outcome with objective criteria
Requiring no or minimal co-operation of subjects
42
Kam
ol U
dol
Adverse Outcomes
● Hardly any intervention is 100% safe
● Though rarely specified as primary outcome in most clinical trials, adverse outcome is very important
● Most clinical trials are not designed for the purpose of addressing adverse outcomes Inadequate sample size, short duration of follow-
up, restricted subject selection
Statistical requirements for comparison of adverse outcome should not be too strict
43
Kam
ol U
dol
Adverse Outcomes
● Definitions Difficult to define clearly as there are many
possibilities and some are unexpected However, important adverse outcomes should be
well-defined
● Ascertainment Systematic elicitation using checklist and lab tests
Standardization
Spontaneous report by subjects Clinically important events Unexpected events
44
Kam
ol U
dol
Outcome Measurements
● Clear, appropriate and acceptable definitions
● Standardized measurement tools
Reliability
Validity
● Consistent measurement procedures
Kam
ol U
dol
Design Considerations
● Research question
● Selection of participants
● Selection of intervention and control
● Selection of outcomes (endpoints)
● Data analysis and sample size
● Measures to reduce bias
● Ethical considerations
46
Kam
ol U
dol
Is the intervention effective?(Does the intervention make any difference?)
Study population
Intervention group
Control group
Outcome Intervention
group
OutcomeControl group
Comparison
Kam
ol U
dol
Comparison of Outcome Between Groups
● Absolutely no difference
The intervention is not effective
● Some degree of difference
What is the probability that the observed difference is only a “chance” finding?
This probability is called “p value” (obtained by the approach of “Hypothesis testing”)
If this probability (p) is very small (< 5%), the observed difference is unlikely to be a chance finding There is a true difference
The intervention is effective
Kam
ol U
dol
Comparison of Outcome Between Groups
Type of Outcome Statistical Test
CategoricalFisher’s ExactChi-Square (Approximate)
Time-to-Event Log-rank (Survival analysis)
Continuous Student’s t
Kam
ol U
dol
What is the magnitude of effect?
● Measure of effect
Parameter estimation
Point estimate
Interval estimate (95% confidence interval)
Kam
ol U
dol
Magnitude of effect: Binary outcome
Treatment (n = NT) Control (n = NC)
Outcome+ pT = nT/NT pC = nC/NC
Measures Formula
Risk Difference (RD) orAbsolute risk reduction (ARR)
pC - pT
Relative Risk (RR) pT/pC
Relative Risk Reduction (RRR) 1 - RR
Number needed to treat (NNT) 100/ARR (%)
Odds Ratio (OR) [pT/(1-pT)]/[pC/(1-pC)]
Hazard Ratio (HR) Cox regression model
Relative Hazard Reduction(Frequently presented as RRR)
1 - HR
Kam
ol U
dol
Subgroup Analysis
● Examine if treatment effects differ (in direction or magnitude) among different subgroups Male vs. Female
Young vs. Old
Severe vs. Mild
etc.
● By statistical test for heterogeneity / interaction / subgroup effect
52
Kam
ol U
dol
● Apixaban is more effective than warfarin in subjects without heart failure (CI excludes 1), but not in subjects with heart failure (CI includes 1)
● Misconception of subgroup analysis!!!
53
Kam
ol U
dol
● Comparison between subgroups, not looking at each subgroup separately (inadequate sample size)
● Are HRs for subjects with and without heart failure too different to be a chance finding?
● Look for p value for interaction / heterogeneity / subgroup effect● If this p is < 0.05, then there is subgroup effect (HR in subjects with HF is
different from HR in subjects without HF), which has to be viewed as hypothesis generation rather than a definitive conclusion.
54
Kam
ol U
dol
Sample Size Estimation
● Too small sample size can prove nothing Probability of type II error
Wasteful, unethical, may mislead conclusions
● A clinical trial should have sufficient statistical power to detect clinically important effect
● Calculation of sample size with provision for adequate levels of significance and power is an essential part of planning
Kam
ol U
dol
Sample Size Estimation
● Parameters required for calculation
Acceptable type I () error (level of significance)
Acceptable type II () error (complement of “power”)
Smallest clinically important effect that needs to be detected
Expected variability of outcome variable
From other trials or a pilot study
Kam
ol U
dol
Sample Size Formula
● Binary outcome, 2 groups, equal sample size in each group
𝑛 𝑝𝑒𝑟 𝑔𝑟𝑜𝑢𝑝 =𝑧 ൗ𝛼 2
2ത𝜋 1 − ത𝜋 + 𝑧𝛽 𝜋1 1 − 𝜋1 + 𝜋2 1 − 𝜋22
𝜋1 − 𝜋22
• (1 - ) represents the variance of binary outcome• 1 = Proportion of outcome in group 1• 2 = Proportion of outcome in group 2• 1 - 2 = Smallest clinically important difference• ഥπ = (1 + 2)/2
Kam
ol U
dol
Example
● Drug A to reduce mortality of avian influenza
● Current mortality rate = 60%
● Gr 1 = Control (No drug A)
● Gr 2 = Drug A
● = 0.05 z/2 = 1.96
● Power = 90% = 0.1 z = 1.28
𝑛 𝑝𝑒𝑟 𝑔𝑟𝑜𝑢𝑝 =𝑧 ൗ𝛼 2
2ത𝜋 1 − ത𝜋 + 𝑧𝛽 𝜋1 1 − 𝜋1 + 𝜋2 1 − 𝜋22
𝜋1 − 𝜋22
Kam
ol U
dol
● 1 = 0.6, 2 = ?
Define 1 – 2
If 1 – 2 = 0.1, then 2 = 0.5
● ഥπ = (0.6 + 0.5)/2 = 0.55
𝑛 𝑝𝑒𝑟 𝑔𝑟𝑜𝑢𝑝 =𝑧 ൗ𝛼 2
2ത𝜋 1 − ത𝜋 + 𝑧𝛽 𝜋1 1 − 𝜋1 + 𝜋2 1 − 𝜋22
𝜋1 − 𝜋22
𝑛 𝑝𝑒𝑟 𝑔𝑟𝑜𝑢𝑝
=1.96 2 × 0.55 × 0.45 + 1.28 0.6 × 0.4 + 0.5 × 0.5
2
0.6 − 0.5 2
= 517.5 = 518Total N = 2 x 518 = 1036
Kam
ol U
dol
What if both 1 and 2 are unknown?
1– (1–)
0.1 0.9 0.09
0.2 0.8 0.16
0.3 0.7 0.21
0.4 0.6 0.24
0.5 0.5 0.25
0.6 0.4 0.24
0.7 0.3 0.21
0.8 0.2 0.16
0.9 0.1 0.09
The greatest value of (1 –) = 0.25
Kam
ol U
dol
● Set all the variance term (1 - ) to be 0.25 (the highest possible value)
● Define 1 – 2 (smallest clinically important difference) = 0.1
𝑛 𝑝𝑒𝑟 𝑔𝑟𝑜𝑢𝑝 =𝑧 ൗ𝛼 2
2ത𝜋 1 − ത𝜋 + 𝑧𝛽 𝜋1 1 − 𝜋1 + 𝜋2 1 − 𝜋22
𝜋1 − 𝜋22
𝑛 𝑝𝑒𝑟 𝑔𝑟𝑜𝑢𝑝 =1.96 2 × 0.25 + 1.28 0.25 + 0.25
2
0.1 2
= 524.9 = 525Total N = 2 x 525 = 1050
Kam
ol U
dol
Sample Size Formula
● Continuous outcome, 2 groups, equal sample size in each group
𝑛 𝑝𝑒𝑟 𝑔𝑟𝑜𝑢𝑝 = 2𝑧 ൗ𝛼 2
+ 𝑧𝛽 𝜎
𝜇1 − 𝜇2
2
• = Standard deviation of outcome
• 1 - 2 = Smallest clinically important difference (of means)
Kam
ol U
dol
Example
● Drug B to reduce cholesterol in hypercholesterolemia
● Mean SD cholesterol = 250 40 mg/dL
● Group 1 = Control (No drug B)
● Group 2 = Drug B
● = 0.05 z/2 = 1.96
● Power = 90% = 0.1 z = 1.28
● = 40
𝑛 𝑝𝑒𝑟 𝑔𝑟𝑜𝑢𝑝 = 2𝑧 ൗ𝛼 2
+ 𝑧𝛽 𝜎
𝜇1 − 𝜇2
2
Kam
ol U
dol
● Define 1 – 2 = 30
𝑛 𝑝𝑒𝑟 𝑔𝑟𝑜𝑢𝑝 = 2𝑧 ൗ𝛼 2
+ 𝑧𝛽 𝜎
𝜇1 − 𝜇2
2
𝑛 𝑝𝑒𝑟 𝑔𝑟𝑜𝑢𝑝 = 21.96 + 1.28 × 40
30
2
= 37.3 = 38Total N = 2 x 38 = 76
Kam
ol U
dol
Design Considerations
● Research question
● Selection of participants
● Selection of intervention and control
● Selection of outcomes (endpoints)
● Data analysis and sample size
● Measures to reduce bias
● Ethical considerations
Kam
ol U
dol
Measures to Reduce Bias
● Treatment allocation by randomization
● Concealment of randomization
● Blinding of treatment allocation
● Maximizing adherence and follow-up
● Analysis using intention-to-treat principle
66
Kam
ol U
dol
Randomization
● A process analogous to coin tossing
Subjects are equally likely to be assigned to either the intervention or control group
The best method for obtaining comparability of groups
67
Kam
ol U
dol
Randomization
● Merits of randomization More likely than other methods to balance the
distribution of prognostic factors (both known and unknown) between the intervention and control groups
Avoid “selection bias”: allocation of treatment based on physician or patient preference, which (either consciously or unconsciously) is usually based on many prognostic factors
Guarantee the validity of statistical tests
68
Kam
ol U
dol
Experimental Study = Randomized Controlled Trial (RCT)
Population
Sample
Randomization
Experiment Control
Outcome Experimental Group
OutcomeControl Group
Definition of population
Selection of subjects to be studied
Assignment of subjects to study group
Administration of intervention
Measurement of outcome
Controlled by investigator
69
Kam
ol U
dol
Randomized Controlled Trial
● The best study design to evaluate efficacy/effectiveness of a therapeutic / preventive intervention
● Matching instead of randomization?
Possible only for known prognostic factors
Low feasibility with increasing numbers of prognostic factors to be matched
70
Kam
ol U
dol
Randomization Process
● Simple randomization
● Blocked randomization
● Stratified randomization
● Adaptive randomization
● Cluster randomization
71
Kam
ol U
dol
Simple Randomization
● Procedures equivalent to “coin tossing”
Random-number table
Computer generated randomization list
● Advantage
Easy
● Disadvantages (esp. for small sample size)
Risk of imbalance in number of subjects between groups
Risk of imbalance in prognosis between groups
72
Kam
ol U
dol
Blocked Randomization
● Avoid serious imbalance in the number of subjects assigned to each group
● Guarantee that
at no time during randomization will the imbalance be large (maximum difference = 0.5 x block size)
at certain points the number of subjects in each group will be equal
73
Kam
ol U
dol
Blocked randomization
● At the end of each block the number of subjects in each group will be equal
● Order of treatment within the block and order of consecutive blocks are at random
● Assignment to the last person in each block can be known if treatment is not blind
Solution: randomly varying block size
Block of 4
Block of 6
TCTC
TCCTCT
CTTC CCTT CTCT TTCC TCCT . . . .
CCTCTT TTTCCC TCTCCT . . . . . .
74
Kam
ol U
dol
Stratified Randomization
● Avoid imbalance in prognosis (baseline characteristics) between groups
● Subjects are classified into stratum based on stratification factor(s)
Important prognostic factors regarding the outcome
● Blocked randomization is then performed for each stratum separately
75
Kam
ol U
dol
Stratified Randomization
Example● Suppose there are 2 important prognostic
factors that should not be imbalance between groups Age
40-49 50-59 60+
History of MI Yes No
76
Kam
ol U
dol
Stratified Randomization
Strata Age History of MI Assignment
1 40-49 Yes TCCT CTCT
2 40-49 No TCTC CCTT
3 50-59 Yes etc.
4 50-59 No
5 60+ Yes
6 60+ No
77
Kam
ol U
dol
Stratified Randomization
● Numbers of prognostic factors used for stratification should be kept minimum, otherwise there would be too many strata and some of them would contain only a few subjects if the sample size is small
● In most multi-center clinical trials, randomization is usually stratified by centers
78
Kam
ol U
dol
Adaptive Randomization
● Randomization procedures that change allocation probability as the study progresses, based on imbalance in numbers of subjects, in prognosis, or in response to treatment Baseline adaptive randomization Correct imbalance in number or prognosis
Response adaptive randomization Maximize the number of subjects on the superior
intervention
79
Kam
ol U
dol
Cluster randomization
● Randomization with naturally occurring groups or cluster of participants as units of randomization (schools, physicians’ practices, communities, etc.)
All subjects in the same cluster receive the same intervention
Suitable for interventions that cannot be delivered to individual subjects
80
Kam
ol U
dol
Measures to Reduce Bias
● Treatment allocation by randomization
● Concealment of randomization
● Blinding of treatment allocation
● Maximizing adherence and follow-up
● Analysis using intention-to-treat principle
81
Kam
ol U
dol
Concealment of randomization
● Concealed randomization
Inability to predict the next assignment
● Unconcealed randomization
Awareness of the next assignment may result in sicker or less sick patients being systematically enrolled to either intervention or control groups, resulting in groups with different prognosis
82
Kam
ol U
dol
Concealment of randomization
● Commonly used methods of concealed randomization
1. Central (or remote) randomization, usually by using a telephone call or internet
2. Preparation of blinded medication in a pharmacy
3. Sequentially numbered, opaque, sealed envelopes (SNOSE)
83
Kam
ol U
dol
Pseudo- (or Quasi-) Randomization
● Superficially similar to randomization, but it is systematic alternate treatment allocation
allocation according to birth date, chart or ID number, day of the week, attending physicians, etc.
● Lack of concealment of randomization
● The factor used for assignment of intervention may somehow relate to prognosis Patients presenting on Monday may in general be
sicker than those presenting on other days
84
Kam
ol U
dol
Measures to Reduce Bias
● Treatment allocation by randomization
● Concealment of randomization
● Blinding of treatment allocation
● Maximizing adherence and follow-up
● Analysis using intention-to-treat principle
85
Kam
ol U
dol
Blinding of treatment allocation
● Unawareness of treatment assignment after randomization
● Ensure comparability between intervention and control groups after randomization
● Should be considered whenever it is possible to blind participants Blinding is not always possible surgical trials, devices, behavioral intervention,
etc.
86
Kam
ol U
dol
Blinding of Treatment Allocation
Avoid bias due to Placebo effect
Fake response in order to please the physician or investigators
Differential co-intervention
Differential interpretation of lab results
Differential follow-up procedure and schedule
Differential effort to detect outcomes
Differential verification of outcomes
87
Kam
ol U
dol
Participants Purposes
Patients To avoid placebo effects
CliniciansTo prevent differential administration of therapies that affect the outcome of interest (co-intervention)
Data collectors To prevent bias in data collection
Adjudicators of outcomeTo prevent bias in decisions about whether or not a patient has had an outcome of interest
Data analystsTo avoid bias in decisions regarding data analysis
Five groups that should be blind to treatment assignment, if possible
88
Kam
ol U
dol
Blinding of Treatment Allocation
● Achieved by
Placebo
Sham procedures
Blind adjudication committee
89
Kam
ol U
dol
Difference between Blinding and Concealment of Randomization
● Blinding
Unawareness of treatment allocation afterrandomization
● Concealment of randomization
Unawareness of treatment allocation beforerandomization
90
Kam
ol U
dol
Measures to Reduce Bias
● Treatment allocation by randomization
● Concealment of randomization
● Blinding of treatment allocation
● Maximizing adherence and follow-up
● Analysis using intention-to-treat principle
91
Kam
ol U
dol
Adherence and Follow-up
● Nonadherence (noncompliance) and lost to follow-up Loss of power of the trial
Biased results
● Subjects not adherent to the protocol or lost to follow-up have different prognosis from others
● Efforts to maximize adherence and follow-up should be carried out
92
Kam
ol U
dol
Coronary Drug Project
Compliance5-Y Mortality (%)
Clofibrate
Total group 20.0
Good (≥ 80%) 15.0
Poor (< 80%) 24.6
Placebo
28.3
15.1
20.9
N Engl J Med 1980;303:1038-1041.
Even in the placebo group, those with good compliance had better prognosis than those with poor compliance.
93
Kam
ol U
dol
Follow-up
● In general, if lost to follow-up is ≥20%, the study validity is usually considerably compromised
● If lost to follow-up is <20%, the study validity may or may not be considerably compromised
If want to be sure, do the sensitivity analysis assuming the worst case scenario
94
Kam
ol U
dol
When does loss to F/U seriously threaten validity?
Trial A
Treatment Control
Number of patients randomized 1000 1000
Number (%) lost to F/U 30 (3%) 30 (3%)
Number (%) of deaths 200 (20%) 400 (40%)
RR not counting patients loss to F/U 0.2/0.4 = 0.5
RR for worst case scenario 0.23/0.4 = 0.57
Trial B
Treatment Control
1000 1000
30 (3%) 30 (3%)
30 (3%) 60 (6%)
0.03/0.06 = 0.5
0.06/0.06 = 1.0
95
Kam
ol U
dol
Follow-up
● Consider the number lost to follow-up to the number of outcome events
No. lost to F/U << No. outcome events
Small effect of bias
No. lost to F/U ≈ No. outcome events
Considerable effect of bias
96
Kam
ol U
dol
Adherence and Follow-up
● Efforts to maximize adherence and follow-up should be carried out
Run-in period
Placebo run-in
Active treatment run-in
Incentives
Monetary and non-monetary
Social events
97
Kam
ol U
dol
Measures to Reduce Bias
● Treatment allocation by randomization
● Concealment of randomization
● Blinding of treatment allocation
● Maximizing adherence and follow-up
● Analysis using intention-to-treat principle
98
Kam
ol U
dol
Analysis
● Intention-to-treat analysis
Analysis of outcomes based on the treatment arm to which patients were randomized rather than which treatment they actually received
● Preserve the value of randomization
● Results reflecting true population effect when the intervention is widely used in real practice
99
Kam
ol U
dol
Analysis
● Per-Protocol analysis
Analysis restricted to subjects who comply to the study protocol
Can introduce bias (See example)
100
Kam
ol U
dol
Analysis
Example
● Drug A in acute myocardial infarction to reduce in-hospital mortality
● Drug A can cause severe hypotension in subjects with poor left ventricular function
● Truth: Drug A has no effect in acute myocardial infarction regarding in-hospital mortality.
101
Kam
ol U
dol
R
350
350
150
150
Placebo
Drug A
Death
70Mortality 20%
Mortality 40%60
130/500 = 26%
150
Mortality 20%70
60
70/350 = 20%
p = 0.04
Per-Protocol Analysis
Drug A reduces
mortality!!!
N = 500
N = 500Mortality 40%
Discontinue due to
severe hypotension 102
Poor LV function
Good LV function
Kam
ol U
dol
R
350
350
150
150
PlaceboDeath
70Mortality 20%
Mortality 40%60
130/500 = 26%
150
Mortality 20%70
60
130/500 = 26%
p = 1.00
N = 500
N = 500Mortality 40%
Get it right
Drug A
Discontinue due to
severe hypotension 103
Poor LV function
Good LV functionIntention-to-treat Analysis
Kam
ol U
dol
Design Considerations
● Research question
● Selection of participants
● Selection of intervention and control
● Selection of outcomes (endpoints)
● Data analysis and sample size
● Measures to reduce bias
● Ethical considerations
104
Kam
ol U
dol
Ethical Issues
● Is randomization ethical?
● Is using placebo ethical?
● IRB or EC approval
● Informed consent
105
Kam
ol U
dol
Ethical Issues
● Is randomization ethical? Withholding a drug from patients
● “Randomization is ethical only when we do not know whether drug A is better than drug B”
● Equipoise The state of uncertainty about the relative merits
of new treatment and standard treatment
Current evidence does not prove that either arm is superior
106
Kam
ol U
dol
Ethical Issues
● “Studies are only ethical if they have a reasonable likelihood of producing the correct answer to the research question, and randomized studies are more likely to lead to a conclusive and correct result than nonrandomized designs”
● Is it ethical “not to randomize”?
107
Kam
ol U
dol
Ethical Issues
● Is using placebo ethical?
“Pure placebo” is not ethical if there is a standard treatment for the condition
Standard treatment must be provided to all patients; experimental treatment is then compared to placebo on top of the standard treatment
108
Kam
ol U
dol
IRB or EC Approval
● Risk minimization
● Risks are reasonable in relation to anticipated benefits
● Informed consent
● Maintenance of confidentiality
109
Kam
ol U
dol
Informed Consent
● Consent provided after subjects fully understand the objectives, procedures, anticipated benefits and risks of the study
● “Written” consent is standard
● “Verbal” consent is acceptable in some situations
● Decision to participate must be made free of any influence
110
Kam
ol U
dol
Other Forms of Experiment
● Randomized cross-over design
● Factorial design
111
Kam
ol U
dol
Cross-Over Design
Study population
Intervention
Control
Control
Intervention
R Period 1 Period 2
● Each subject serves as his/her own control
● Reduced variability, resulting in smaller sample size
● Suitable for
Chronic, stable diseases, episodic conditions
Treatment is non-curative
● Based on the assumption that there is no “carry over”effect
112
Kam
ol U
dol
Factorial Design
● Evaluate 2 or more interventions in a single trial (save cost and time)
● Interaction between the 2 (or more) interventions
Absence: much smaller sample size than conducting 2 separate trials
Presence: the only trial design that enables assessment of interaction
113
Kam
ol U
dol
Factorial Design: ISIS-2 Trial
● Evaluate the effect of 2 drugs, namely Aspirin (ASA) and Streptokinase (SK), in patients with acute myocardial infarction
● Patients were randomized 2 times
Randomized to receive ASA or ASA Placebo, then
Randomized to receive SK or SK Placebo
Kam
ol U
dol
Factorial design: ISIS-2 Trial
ASA ASA Placebo
SK SK + ASA SK only
SK Placebo ASA only Nothing
Eventually, there were 4 groups of patients SK + ASA
SK only
ASA only
Nothing
115
Kam
ol U
dol
Factorial design: ISIS-2 Trial
ASA ASA Placebo
SK SK + ASA SK only
SK Placebo ASA only Nothing
Effect of SK was estimated by comparing the outcome between SK group (SK+ASA combined with SK only) and SK Placebo group (ASA only combined with Nothing)
116
Kam
ol U
dol
Factorial design: ISIS-2 Trial
ASA ASA Placebo
SK SK + ASA SK only
SK Placebo ASA only Nothing
Effect of ASA was estimated by comparing the outcome between ASA group (SK+ASA combined with ASA only) and ASA Placebo group (SK only combined with Nothing)
117
Kam
ol U
dol
Conclusions (1)
● Experimental study (= RCT) is most suitable for studying the effectiveness of a therapeutic or preventive intervention
● Main features Control group
Similar prognosis between experimental and control groups
Investigators manipulate or control assignment of subjects to groups using randomization technique (unique for experiment study)
118
Kam
ol U
dol
Conclusions (2)
● Measures used to reduce bias
Treatment allocation by randomization
Concealment of randomization
Blinding of treatment allocation
Maximizing adherence and follow-up
Analysis using intention-to-treat principle
119