Design and Analysis of Clinical Study 12. Randomized Clinical Trials

1
Design and Analysis of Clinical Study 12.
Randomized Clinical Trials

• Dr. Tuan V. Nguyen
• Garvan Institute of Medical Research
• Sydney, Australia

2
Basic Design of Clinical Trials
Cured
Same
Treatment
Sample
Randomise
Placebo
Cured
Same
Blinding
Subjects
Blocking
3
Variations in Basic Design - 1
Run-in Design
Admission
On placebo
Compliers randomised
A S S E S S
A S S E S S
Cross-over Design
Treatment 1
Recruitment
Randomised
Treatment 2
4
Variations in Basic Design - 2
Time - Series Design
No Treatment
Recruitment
Treatment
Treatment
Assess
Assess
Assess
Low Dose
Factorial Design
Treatment 1
High Dose
Recruitment
Randomise
Treatment 2
Low Dose
High Dose
5
Issues of Methodology - I

• Entry criteria
• Strict for explanatory trials
• Less strict for pragmatic trials
• Diagnosis
• How accurate?
• Intervention
• Compliance
• Drop-out
• Competing intervention

6
Issues of Methodology - II

• Subject allocation
• Different rates of drop-out between groups cause
  under or over estimate of outcome
• Treatment allocation
• Randomisation and blinding help to remove bias.
  Blocking needed if outcomes vary because of age,
  sex or other attributes.

7
Challenges in Designing Clinical Trials

• Control of bias
• In allocation of subjects to treatment
• In assessment of outcome
• Sample size
• How small a difference is clinically important?
• What tests of significance will be used?
• What outcome is expected in the control group?
• Drop-outs and withdrawals
• How to handle them during analysis?

8
Generalizability of Results
Treated
Outcome
Sample
Population of Patients
Control
Outcome
Difference in Outcome by Chance?
Rigour of study
Other Populations
Generalisability
9
Randomization works!

• Volunteerism
• Eligibility
• Placebo effect
• Hawthorne effect
• Regression towards the mean

10
Volunteerism, Eligibility, Placebo Effect

• Volunteerism
• People who agree to participate in clinical
  trials are an elite group of patients with
  extremely good prognosis.
• Eligibility
• Patients have to meet stringent eligibility
  criteria before randomization, or they would be
  excluded
• Placebo can do just about anything (prolong life,
  cure cancer).
• Placebo can also cause side effects.
• Placebo effect is very useful in medicine but in
  epidemiology it causes problems, so we try to
  equalize it between the 2 groups.

11
Regression Towards the Mean

• Weather game
• Individuals with initially abnormal results tend
  on average to have more normal (closer to the
  mean) results later.
• Lab tests, BP etc.
• Recheck before randomization. Run-in period.
• Sophomore slump, medical school, Airforce landing
  feedback

12
Objectives of Subgroup Analysis

• Support the main finding
• Check the consistency of main finding
• Address specific concerns re efficacy or safety
  in specific subgroup
• Generate hypotheses for future studies

13
Inappropriate Uses of Subgroup Analysis

• Rescue a negative trial
• Rescue a harmful trial
• Data dredging find interesting results without a
  prespecified plan or hypothesis

14
To Avoid Inappropriate Uses of Subgroup Analysis

• Prespecify analysis plan
• Prespecify hypotheses to be tested based on prior
  evidence
• Plan adequate power in the subgroups
• Avoid the previous pitfalls

15
Problems with Subgroup Analysis

1. Low power
2. Multiplicity
3. Test for interaction
4. Comparability of the treatment groups maybe
   compromized
5. Over interpretation

16
ITT

• Intention to treat analysis
• Once randomized always analyzed
• Why ?
• 1. Change in therapy may be related to outcome or
  eligibility
• 2. To get the full benefit of randomization
• 3. Effectiveness versus efficacy

17
Five-Year Mortality in Coronary Drug Project
18
Screening Mammography
19
Descriptions of Trials

• 34 relative decrease in the incidence of MI. The
  decrease is statistically significant. The 95 CI
  ranges from 55 relative decrease to a 9
  relative decrease.
• 1.4 decrease in . (2.5 versus 3.9). The
  decrease is statistically significant. The 95
  confidence interval ranges from a 2.5 decrease
  to a ..
• 77 persons must be treated for an average of just
  over 5 years to prevent 1 MI.

20
Ethical Issues

• When is it unethical to randomize ?
• When Do you stop a trial?
• Data Safety Monitoring Board
• Early Termination rules
• OBrien Fleming
• Early vs. late
• Benefit vs. harm (blinding?)
• Multiplicity
• Rules. Scenarios.