Randomised control trials Flashcards
What type of study is an RCT
Analytic intervention study
RCT design
A group of people are randomly assigned to a control or intervention arm, then followed up and compared
Key features of an RCT
Randomisation- confounders should be equally shared between groups. Method of randomisation should explicitly stated
Allocation concealment- minimises selection bias (conceal allocation sequence from everyone- should state how it was)
Blinding- single (particpants)/ double (participants and researcher)/ triple (participants, researcher and statistician)
Intention to treat analysis- analysing according to allocation (regardless of adherence/ drop-out/ contamination)
Importance of outcome measure
Need to relevant to the participants
If it is proxy, need confidence it is suitable
Reliable (produces consistent, reproducible estimates of true effect)
Valid (measures what it says it does)
Responsive (can detect changes)
Other aspects of RCTs
Equipoise (genuinely don’t know if intervention or control is better)
Ethics- cannot cause harm in either intervention or control
Choice of control (sham intervention/usual care/placebo)
Sample size (need sample size calculations- larger the sample, the greater the precision (CI))
Problems with RCTs
Non-adherence
Performance bias (differences in how groups treated)
Contamination
Detection bias (if outcomes detected differently between groups)
Potentially attrition bias if certain groups dropped out/ lost to follow-up
What is per-protocol analysis
A secondary, exploratory analysis. It estimates the effect of adhering to a treatment
May be done if there was high contamination
Interpretation of outcomes in RCTs
Mean difference
Relative risk
Odds ratio
Survival analysis
Need to consider if the results are applicable to your local population
What are cluster RCTs
An RCT where groups of people are randomly assigned to an arm, rather than individuals
Why are cluster RCTs done
Done when there is a high risk of high contamination
Random methods of sampling
Simple random (random number generator)
Cluster sampling
Stratified sampling (when you want to compare relatively small sub-groups (selectively recruit higher proportion of small groups to increase statistical power and sample size) or when measurement is likely to vary between sub-groups)
Non-random methods of sampling
Systematic sampling (pick every nth person)
Convenience sampling (those most available- easiest but volunteer bias)
Snowball sampling (ask participants to recommend others who meet the pre-specified criteria)
Street survey (pick people who appear to meet the criteria on the street, often they ask a specified number of men/women/age groups- quota sampling)