RCTs Flashcards
What are the 3 types of intervention study?
- RCTs (individual & cluster)
- controlled before & after (same as RCT, non-randomised, 2 arms)
- uncontrolled before & after (measurements taken from pop with exposure before & after study)
What sources of bias/confounding/error may arise in before & after studies?
Regression to norm
- abnormal data in previous studies (that indicated the new study) could be due to chance
Seasonal changes
- conditions change based on seasons
Change measurements/policy/practice
- difficult to compare data & could give the illusion of amplified difference
Selection bias - in controlled trials, can skew results (must randomise & blind trialist, ptpts, clinicians)
What is the importance of randomisation in RCTs?
Only difference between groups should be the intervention (for an accurate representation of effect)
- allows equivalence of prognostic RFs, characteristics that may be source of confounding btw groups
AVOIDS - selection bias
What is the importance of allocation concealment in RCTs?
(Before randomisation) concealed allocation to different arms/groups avoids exploitation of the knowledge
AVOIDS: allocation bias (based on success/characteristics. Allows for TRUE randomisation)
What is the importance of blinding in RCTs?
(After randomisation) ptpts & investigators (+/- clinicians) are not made aware of the allocation status of the ptpt (allows equal treatment/placebo effect)
AVOIDS - performance bias, interviewer bias, confirmation bias, recording bias, Hawthorne effect
What is the importance of intention to treat analysis in RCTs?
Analysing patients in the group in which they were allocated (to maintain randomisation)
AVOIDS: LTF/attrition bias, Hawthorne effect, social-acceptability bias
What are the advantages of RCTs?
- removes selection bias & confounding (if adequate randomisation, concealment & blinding)
- incidence & risk data
- can be analysed using known statistical tools
- can use for causality analysis
What are the disadvantages of an RCT?
- expensive
- time-consuming
- volunteer bias (may not be representative)
- non-compliance threatens validity (must conduct ITTA)
- not useful for rare ADEs/conditions
- blinding can have ethical limitations
What qualities must be met by the outcome measure of an RCT?
Reliable - produces consistent, reproducible estimates of true effect
Valid - measures what it claims to (reduced bias/confounding/interference)
Responsive - can detect changes that can be measured over time
What ethical challenges may arise during an RCT?
Randomisation to an intervention can only be ethical if the effect of the intervention (benefit/harm) is genuinely unknown
What is the concept of equipoise?
RCT ethics - A participant may only enter a study if there is substantial uncertainty.
What is the effect of a greater sample size?
Greater precision of results
What is the benefit of blinding patients in an RCT?
- avoid disappointment in usual care
- avoid changing behaviours (performance bias)
What is the benefit of blinding HCPs in RCTs?
- avoid compensatory action from HCP (performance bias)
- avoid complementary treatments for intervention group (performance bias)
What is the benefit of blinding researchers in RCTs?
Avoid biases that may occur from researcher measuring & interpreting outcomes
What is per-protocol analysis?
Low adherence studies - estimates effect of actually adhering to intervention as instructed
(only includes those who adhered to randomisation, can only be an exploratory aspect of study NOT main analysis)
What is a cluster RCT and when is it appropriate?
(CRT)
Randomisation of clusters/systems rather than individuals (when high risk of contamination)
Indications
- new standard of care/guideline recommendation (practice/system/hospital-wide changes)
- significant contamination potential
- HCP required to change behaviour
- impossible to separate btw access to control & intervention care
When is a power calculation performed?
Before the study - to determine the sample size needed for the population to be sufficiently represented to produce a clinically significant result
Why do you not use statistical tests to compare baseline characteristics in RCTs?
As arms of the trial are already (correctly) randomised, we know this variation is due to chance, so any statistical test is superfluous and may be misleading.
Comparisons at baseline should be based on the consideration of prognostic strength of the variables measured & size of any chance imbalances.
Differential LTF bias
(why must the LTF be due to allocation and outcome for it to be considered ‘bias’?)
Non-differential losses: The risk ratio (of outcome) is unchanged if the losses to follow up are related to allocation/exposure and not outcome - as the LTF occurred in a random proportion = random effect on RR, thus not considered bias.
Random loss by chance (ie- not related to exposure or outcome) is not bias
i.e. - Allocation/exposure - placebo
Outcome - cure
if the sickest ptpts in the placebo left the study because they were sick & wanted to find alternative treatment, systematic error has been caused in a non-random way that affects ONLY the placebo and will result in a different outcome (cure) than if they had remained in the study. This overestimates the placebo effect and hence underestimates the effect of the intervention.
Determining if there is LTF bias
- Was the LTF high (>30%)? Or low (<5%)
low LTF rates are unlikely to suggest bias - Was the LTF equivalent between all the study arms?
if random/not affected by arm/O/E then unlikely to suggest bias - Were reasons given for LTF? Could they be reasons that would occur ‘at random’? - moving from area, changing jobs/other factors unrelated to O/E
unrelated reasons unlikely to cause bias
What are the advantages of RCTs?
- gold standard for measuring effectiveness
- Randomisation accounts for confounding
- Double blinding - interviewer bias & placebo effect
What are the disadvantages of RCTs?
- Expensive if long or large/time consuming
- Ethics - unethical if exposure causes harm
- Responder, LTF & performance bias (responder = participation/volunteer bias, more likely to volunteer if middle class, white, female, middle aged etc)
