1. Trials
Adrian Boyle

2. Objectives Design Measures of quality How to analyse data from an RCT
How to appraise an RCT

3. New terms Explanatory vs pragmatic Surrogate end point Equipoise
Factorial
Cross-over
Intention to treat
Number needed to treat

4. Equipoise
There should be substantial uncertainty in the clinician’s mind about which treatment is better for the patient before the patient is enrolled in a trial

5. Why randomise? To avoid / reduce selection bias
To balance confounders across groups

6. Limitations of trials
Internal validity high at expense of low external validity
Efficacy rather than effectiveness
Irrelevant narrow questions
Often a ‘Shot in the dark’
Drug companies often want to compare against placebo, not standard treatment
Expensive and time consuming

7. More jargon Phase 1 Clinical pharmacology Drug safety in volunteers
Phase 2 Initial investigation of effect Effectiveness
Phase 3 Full scale evaluation Compared to placebo or standard practice
Phase 4 Post marketing surveillance

8. Trial design
Efficacy Frontier of effect under ideal circumstances works
Effectiveness How this intervention works in ‘the real world’
Explanatory
Provide clues as to how the intervention
Pragmatic Shows how well the intervention works

9. Basic trial design Population Randomisation Exposure1 Exposure 2
Outcome
Outcome

10. Analysis of basic study design
Relative risk
incidence of outcome in group 1 DIVIDED BY
incidence of outcome in group 2
Sounds dramatic and sexy

11. Analysis of basic study design
Absolute risk reduction:
incidence of outcome in group 1
MINUS
incidence of outcome in group 2
Less sexy and more useful.
Ask the next drug rep.
Enjoy

12. Analysis of basic study design
Number needed to treat
Inverse of the absolute risk reduction

13. Example
A trial of drug A compared to drug B found that 20 out the 50 people who received A were alive at one year compared to 15 /50 who received drug B
What is the relative risk?
What is the absolute risk?
What is the NNT

14. Example
Relative risk
20/50 = 0.4
15/50 = 0.3
0.4/0.3 = 1.33
Or this could be expressed as a 33% increase in survival at one year

15. NNT Absolute risk reduction 0.4-0.3 = 0.1
This could be expressed as a risk reduction of 10% NNT 1/0.1 = 10
That is, you need to treat 10 people to save one life in one year

16. Multi-variate analysis
Adjust for potential confounders and bias
Usually with logistic regression expressed as a odds ratio

17. Factorial trial design
Population
Randomisation
Exposure2
Exposure1
Exposure1
Placebo
Placebo
Exposure2
Placebo
Outcome
Outcome
Outcome
Outcome

18. Cross over design Population Randomisation Exposure1 Exposure2 Outcome

19. Cluster
The unit of randomisation is a group of individuals e.g. GP practices or hospitals
Easier implementation of a complex design
Large studies
Seriously complicated statistics

20. Randomisation Many methods Simple
Block
Stratification
Weighted
Simple
The proof of the pudding is in the table 1

21. Sources of bias in a trial
Selection
Performance
Losses to follow up
Detection

22. What bias is there here? Population 200 Randomisation Exposure 1
100
100 15
Outcome
Outcome 12 10

23. Is this bias? Population 2000 Randomisation Exposure1 Exposure 2 100
Outcome
Outcome 50 10

24. What bias is there here? Population 200 Randomisation Exposure1
100
100
Outcome No
Outcome 60 No
Outcome 80
Outcome 20 10

25. Outcome measure Does the outcome mean anything to you?
Beware surrogate outcomes
Beware composite outcome especially if industry funded

26. Appraising a trial Identify aims Identify study design
Identify population, exposure and outcome
Consider the randomisation
Consider the blinding
Consider the measurement
What biases and confounding factors are there?
What is the result and what does it mean

27. Assessing strength / quality
JADAD scoring