Randomized controlled trials

Question 1

How doe we know a treatment is effective?

Answer 1

Someone tells us - authority figures (but how do they know?)
underlying theory - pathophysiology
observational evidence - scientific method

Question 2

What factors need to be taken into account when deciding whether to treat?

Answer 2

benefits worth harm
resource costs
patient preference
ethics

Question 3

What are the 4 factors you need to consider and how can you correct for them to explain your observations ?

Answer 3

1) bias - study design can reduce this
2) confounding - randomization can reduce this (study design)
3) chance - statistics can check this
4) hypothesis correct= effective

Question 4

What comes under external validity (generalizability)?

Answer 4

benefits worth harms, resource costs, patient preferences, ethics and is this applicable to my patient/population

Question 5

What comes under internal validity?

Answer 5

explanation of observations

- bias, confounding, chance and hypothesis correct

Question 6

What are the 2 groups of randomization’s in RCTs?

Answer 6

• individual randomization

- cluster randomisation

Question 7

What does regression to the mean mean?

Answer 7

most things vary to some extent by chance
if you observe an extreme value (unusually high or unusually low), part of the reason for it being extreme is chance
Therefore the next observation is likely to be closer to the mean - which can give the appearance of improvement

Question 8

What does bias mean?

Answer 8

patient’s or researchers beliefs affect outcome measurements

Question 9

What does confounding mean?

Answer 9

Differences between groups at the start

Some other difference in the way in which groups are treated

Question 10

What is the purpose of randomization?

Answer 10

ensures known and unknown characteristics that might affect the outcome (confounders) are distributed by chance
- any differences between groups t the start are due to chance
Minimizes confounding

Question 11

What are the issues with RCTs?

Answer 11

choice of outcome measure
ethics
choice of control
contamination (crossover)
bias in assessment of outcome 
losses to follow up
sample size

Question 12

What are the different choices of outcome that can be measured?

Answer 12

• ideally measure outcome relevant to the patients - if a proxy or process measure: are we confident it is linked to the outcome? e.g. immunization rate proxy for % of population immune
• clinical effectiveness – measure clinical outcome e.g. cure or a proxy
• patient experience - generic quality of like, disease specific measure e.g. pain score

Question 13

What are some of the ethical issues associated with RCTs?

Answer 13

are individuals disadvantaged by being randomized to intervention or control?
if we believe one is better we cannot ethically randomize - equipoise
if we don’t know which is better is it ethical NOT to evaluate effectiveness

Question 14

What are the different options for choice of control?

Answer 14

usual care
no treatment -if there is no usual care and we don’t believe new treatment is effective
Placebo - something that appears similar to intervention - distinguishes intervention effects from placebo effects - helps to blind the patient and researcher

Question 15

What does it mean by contamination or crossover?

Answer 15

participants randomized to the control group may unintentionally receive the intervention
participants randomized to the intervention group may NOT receive the intervention

Question 16

What happens if crossover or contamination occurs?

Answer 16

if you analyst according to the treatments received you have allocation bias

Question 17

What options do you have if your study undergoes crossover or contamination?

Answer 17

1) analyst according to treatment received - but no longer randomized
2) analyst by intention to treat - preferred option
- this reduces risk of allocation bias, however it may underestimate the effect

Question 18

What are the final numbers in this analysis?

Answer 18

It includes everyone, even those that were lost to follow up

Question 19

When does contamination or crossover tend to be a problem?

Answer 19

if the intervention is information/education

and/or if we can’t treat patients differently in the same clinic

Question 20

How can you reduce crossover/contamination?

Answer 20

cluster randomisation - randomize groups or clusters rather than individuals, reduces crossover but the statistical analysis is more complex

Question 21

How can bias occur when assessing outcomes?

Answer 21

influenced by patient’s expectations, researcher’s expectations and analysts expectations

Question 22

What was found when assessors were non-blinded?

Answer 22

substantially exaggerate treatment effects - misclassification of assessed patients

Question 23

Other than blinding how else can bias arise?

Answer 23

by stopping an intervention too early - treatment may be effective initially but over time it might not be

Question 24

What are the degrees of blinding?

Answer 24

unblinded/open
single blind - participants don’t know if treatment or control
double blind - researcher measuring outcomes does not know
triple blind - analyst reviewing results doesn’t known

Question 25

What is allocation concealment and why is it important ?

Answer 25

when randomization is undertaken the researcher should not know whether a participant will be allocated to intervention or control

Allocation concealment prevents researchers from influencing which participants are assigned to the intervention or control group

Question 26

When does losses to FU affect results and when doesn’t it?

Answer 26

little effect on result if:

• losses related to treatment allocation or
• losses related to outcome

Affects results if losses related to treatment allocation and to outcome (bias)
Helpful to describe the characteristics of those lost compared to those followed up - are there systematic differences

Question 27

Why is sample size important?

Answer 27

The correct sample size is required for results to be interpreted

• expected effect = smaller effects will need larger sample sizes
• variation in measured outcome = outcomes measured imprecisely need a larger sample size
• significance - at what significance level will we accept there is a difference
• power = how certain do we want to be to find a difference if there is one

Question 28

What does selective losses in fu mean?

Answer 28

bias