Study Designs - Lecture Twenty-One Flashcards
Randomised Controlled Trials
Only thing better than a randomised controlled trial?
Having more than one randomised controlled trial
Observational analytic studies
Cross-sectional
Cohort
Ecological
Case-control
Intervention analytic studies
Randomised controlled trials
Non-randomised controlled trials
Randomised
Participants randomly allocated to groups
Controlled
There’s a comparison (control) group
Trial
Testing effect of treatments/interventions
What is the exposure determined by in observational studies?
Luck, environment, by people themselves
Why do we randomise?
People who decide to take a treatment are often different to those who don’t e.g. age, sex, health risks, views of the health professions treating them, health beliefs and habits
Random allocation
Each participant has an equal chance of being in either group, if enough people are randomised then both known and unknown confounders should be balanced
Variants of randomisation
Cluster
Stratified or Block Randomisation
Cluster
Entire practices are randomised to treatment or control. All participants in each practice get the same intervention and GPs don’t have to do different things for different patients.
Stratified or Block Randomisation
Participants are randomised to treatment or placebo in blocks (or strata) at each hospital. Differences between hospitals will be balanced between treatment and control groups
Cross-over studies
Each person gets both treatments - confounding is effectively eliminated, however, this can only be done for long-term conditions and treatments that are not curative
How do we preserve the benefits of randomisation
Concealment of allocation
Intention-to-treat
Concealment of allocation
Make sure that people can’t cheat and pick the treatment that they prefer, otherwise bias could be introduced
Intention-to-treat analysis
Once participants have been randomised, you don’t change the groups
Intention-to-treat analysis information
Analyse participants as randomised
Reflects the ‘real-world’: people often don’t take treatments
Difficult if data are missing – you can’t analyse data you don’t have
Per-protocol analysis information
Analyse as treated (not necessarily as randomised)
Lose the benefit of randomisation
Can be appropriate for efficacy trials
Potential source of bias
Lack of blinding
Loss of follow-up
Non-adherence
Lack of blinding
If participants (or researchers) know which treatment they are on, they may act differently
Loss of follow-up
If people withdraw because of side effects, we may underestimate the harms of treatment.
Confounding and bias may occur if there is loss to follow-up
Non-adherence
If people don’t take the treatment, we will not learn about its true benefits and harms
Non-adherence
Participants often don’t do what you ask them to: Some only take some treatment or stop it altogether, some don’t turn up for appointments, or some take alternative treatments (including the intervention or control). If there is too much non-adherence, it is difficult to interpret the study
Strengths of randomised controlled trials
The best study design to test an intervention
Well conducted studies should eliminate confounding and bias
You can calculate Incidence, Relative Risks, and Risk Differences
The strongest design for testing cause-and-effect associations
Clinical equipoise
Genuine uncertainty about benefit or harm of intervention
Unethical to
Give known harmful interventions to people
Give interventions known to be less effective than current treatments
Waste resources and risk harm if we already know the answer
Practical issues about randomised controlled trials
Can be very expensive
Often funded by pharmaceutical companies
Practical issue: Can be very expensive
May need large numbers of participants
Ensure complete follow-up
Can take a long time
Practical issue: Often funded by pharmaceutical companies
Potential for big profits if the drug works
Unlikely to fund studies of cheap treatments with little chance of profit
Limitations of randomised controlled trials
Often no representative
Not efficient for rare outcomes
