1. Randomized Trials: A Brief Overview
Michael Porter
July 20, 2004
Epi 590

2. Randomized Trial Comparative study between two or more interventions
Exposure to the intervention is determined by random allocation
AKA: RCT, experimental design

3. “Hierarchy” of Evidence
Randomized Trial
(experimental)
Observational Analytic
(e.g. case-control, cohort)
Observational Descriptive
(case series, descriptive analyses)

4. Agree to participate, Informed consent
Eligible
Ineligible
Decline to
participate
Randomization
Treatment A
Treatment B
Good outcome
Poor
outcome
Population
Eligibility
criteria
Recruitment
The Basic
Model

5. Strengths Highest level of evidence
Everything can be specified up front
Ideal scenario for inferential statistics
Eliminates confounding

6. Weaknesses $$$ Time consuming
Patients and providers must relinquish treatment decision to random chance
Generalizability
Subject to dropouts, crossovers, non-compliance
Still vulnerable to bias

7. Two Broad Categories Pragmatic Explanatory
attempt to simulate clinical realities more accurately during patient recruitment and during formulation of the randomly allocated treatment groups (effectiveness)
Explanatory
attempt to answer a more specific and narrow question- in order to maximize their ability to do this, eligibility criteria may seek a more homogenous set of patients (efficacy)

8. Bias
Systematic error within the study that results in a mistaken estimate of the effect of therapy on disease
Bias can be introduced into any step of the process, including enrollment, randomization, and assessment of outcomes

9. Internal validity
The ability of a trial to come to the correct conclusion regarding the question being studied
Determined by the protocol and execution of the trial

10. External validity
The ability of a trial to produce results that are generalizable to the larger population of patients with the disease
Determined by the eligibility criteria, the protocol, and the primary outcomes

11. Randomization The key step
The biggest strength, and often the biggest challenge
Randomization breaks the link between any unmeasured confounding variables and treatment status

12. Randomization 2 important elements
concealment from the investigator
unpredictability
“Randomized trials appear to annoy human nature- if properly conducted, indeed they should”*
Still vulnerable to selection bias
*Schulz, K.F., Subverting randomization in controlled trials. JAMA, (18): p

13. Subverting Randomization
When researchers failed to adequately conceal randomization from the investigators, an average 41% percent increase in treatment effect occurred compared to trials where the randomization process was concealed appropriately
Schulz, K.F., et al., Empirical evidence of bias. Dimensions of methodological quality associated with estimates of treatment effects in controlled trials. JAMA, (5): p

14. Blinding Single blinded Double blinded
Subject unaware of treatment assignment
Double blinded
Subject + evaluator unaware of treatment assignment
Blinding provides protection against
Placebo effect
Observation bias

15. Sample Size (a.k.a. “power calculation”)
Significance level α
Type 1 error rate
p-value
Power
1-β
1-Type 2 error rate
Variability of the outcome
Degree of difference in outcomes

16. The “truth” Study conclusions Treatments do not differ
Treatments differ
Correct conclusion
Type II Error
(β)
Type I Error
(α)
(1- β)
Treatments do not differ
Study conclusions
0.95
0.10
Treatments differ
0.90
0.05

17. Outcome
Should only have one or two primary outcomes to which the study is powered
Be careful of secondary analyses
Multiple looks require adjustment of significance level

18. Time
Treatment A
(n=500)
Outcome
(n=500)
Randomize
(n=1000)
Treatment B
(n=500)
Outcome
(n=500)

19. ? Time Treatment A (n=500) Outcome (n=600) Randomize (n=1000)
Treatment B
(n=500)
Outcome
(n=400)

20. Crossovers Analyze based on treatment received?
Exclude crossovers from the analysis?
Analyze based on initial random treatment assignment?

21. Intention to treat analysis
analyze the outcomes based on the original randomized assignments, regardless of treatment actually received

22. Randomized study ≠ RCT Survey Community interventions
Clinic interventions
Be novel!

23. Ethical Issues Equipoise Data monitoring and safety committees
Eligibility criteria

24. Summary RCT- the highest level of evidence
Avoids confounding, but not necessarily bias
Expensive, time consuming
Randomization- preserve and protect it
Blinded when possible
Single important outcome
Intention to treat
Equipoise