RCTs Flashcards
What needs to be considered if RCT findings are valid and applicable?
Eligibility criteria, randomisation method, allocation concealment, blinding, follow-up
What is scientific method?
Ensuring research is not shaped by personal belief but grounded in measurable (empirical) evidence
Which types of bias can be present in experimental studies?
Selection, performance, detection, attrition
What is internal validity?
Extent to which a causal conclusion is warranted, determined by the extent to which the study minimises systematic error (bias, confounding, chance)
What is external validity?
Extent to which study can be generalised to different situations and people regardless of internal validity
What are the 3 types of intervention studies?
- Uncontrolled before and after
- Controlled before and after
- RCT
What is the benefit of a controlled compared to an uncontrolled intervention study?
- Regression to the mean could give an appearance of improvement in an uncontrolled study which is minimised with control groups
- Some confounding factors may be accounted for with comparing to a control group (eg seasonal)
How are RCTs beneficial to controlled clinical trials?
Participants in a controlled clinical trial are assigned to new treatment or control by themselves, dr or researcher - this could introduce selection bias
How could participants choosing their assignment to groups impact the study?
Selection bias - participants choosing active treatment might differ to those who do not
How could doctors choosing assignment to groups impact the study?
Might put those frail or unlikely to withstand new treatment in controls, or younger or sicker people in active treatment to give them a better chance
How could researchers choosing assignment to groups impact the study?
Might only select those who will do well to be in active groups
What are the key features of an RCT?
Randomisation of allocation to groups, allocation concealment, blinding and intention to treat analysis
What are the 3 sections of the CASP tool for RCTs?
Are the results valid?
What are the results?
Will the results help locally?
How do you look at choice of outcome measure of a study?
- Is it relevant to patients (clinical effectiveness or patient experience)
- If proxy, does it measure what it should?
- Type
- Needs to be reliable, valid, responsive
What makes an outcome measure reliable?
Able to produce consistent, reproducible estimates of true effect
What makes an outcome measure valid?
Measures the construct it claims to
What makes an outcome measure responsive?
Detects changes in construct to be measured over time
What ethical issues are related to RCTs?
Equipoise (reasonable belief that interventions are equal) is required to ethically randomise participants
What are the different choices of controls in RCTs?
- Current best practice
- No treatment: only if no current best practice and do not feel new treatment is effective!
- Placebo
What is allocation concealment?
The person recruiting to a trial does not know which group the participant will be assigned to
What can a lack of allocation concealment lead to?
Confounding (imbalance of characteristics between study arms) and selection bias
Which problems with RCTs can be avoided with blinding?
- Non-adherence
- Performance bias:
A) comparator group gets extra, contamination if gets treatment anyway, seek other treatments etc
B) intervention gets extra as more regular contact with health service
What are the 4 degrees of blinding?
- Unblinded/open label
- Single blinded
- Double blinded
- Triple blinded
What does it mean if research will single blinded?
Participant is blinded to treatment or control
What does double blinded mean?
Participants and HCPs/researchers are blinded to treatment or control
What does triple blinded mean?
Participants, HCPs/researchers and statisticians blinded to treatment or control
What is the benefit of blinding statisticians?
Avoid bias in measuring or interpreting outcomes
Which study features make cross-over and contamination a particular issue?
- Information/education as intervention
2. Cannot treat people differently in same clinic
What is cluster randomisation?
Groups/clusters are randomised instead of individuals
What are benefits and disadvantages of cluster randomisation?
- Reduces contamination
2. Needs bigger study and statistical significance is more complex
How should you analyse RCTs irrespective of if people receive differently from their group allocation? What does this prevent?
Intention to treat analysis preserves randomisation and therefore reduces allocation bias.
This could underestimate an effect
What is per protocol analysis?
Estimating effect of adhering to interventions, only include those to adhered to randomisation assignment. This should be exploratory not main analysis!!
What is detection bias and how can it be prevented?
A systematic difference between groups in how outcomes are determined. This can be reduced by triple blinding.
When do losses to follow-up affect results?
If losses are related to treatment allocation and outcome
Those who drop out are systematically different
What will high losses to follow up lead to?
Incomplete outcome data affecting generalisability, and whether or not believe results
Which type of bias relates to losses to follow-up?
Attrition bias: systematic difference between groups in withdrawals from study
How can attrition bias be prevented?
Big number of participants, analyse attrition rates and compare characteristics of those lost to follow-up
How can attrition bias be related to outcome?
Sicker people more likely to drop outn
How can attrition bias be related to allocation?
Not liking intervention, disappointment in allocation to control group
What are ways to interpret outcomes of RCTs?
Mean difference
Relative risk
Odds ratio
Survival analysis (time until event)
Do larger or smaller sample sizes have greater precision?
Larger
What are sample size calculations based on?
- Expected effect size (smaller need larger samples)
- Variation in measured outcome - imprecise measurements need larger
- Significance level
- Power - how certain do we want to be to find a difference if their is one - usually 80% or 90%
How would you analyse RCT results?
- Work out absolute risk in treatment and controls
- Relative risk in treatment vs controls (<1 is likelihood of outcome less in treatment)
- Calculate absolute risk difference (difference of …% risk of dying in treatment vs controls)
- NNT
Which risk statistic do you use to calculate NNT?
Absolute risk difference
What determines whether RCT results can be applied locally?
Population of trial compared to usual practice, usual care compared to control, cost
Randomise studies across countries
What must be included in analysis for full ITT and how is this done?
Participants who did not receive assigned intervention according to protocol & those lost to follow-up.
Use imputed values
What is the strength of ITT analysis?
Preserves effect of random allocation, and thus equal distribution of confounders is maintained. It also mirrors real-world situations.
What control can be used for non-drug interventions?
Sham interventions
If a baseline characteristics table shows no significant differences between groups can you assume no selection bias?
No:
- table must include all pts recruited to trial
- consort statement says do not depend on statistical significance
- if smaller sample would have to have huge difference for it to be significant
Who’s point of view should be considered for outcomes of an RCT?
- Individual
- Policy makers and professionals
- Family/carers
- Wider community
What 3 questions should you ask to appraise loss to follow up?
- Was it high (>30%), low (<5%) or somewhere in between?
- Was it even across groups?
- Were reasons given? Are they reasons that occur ‘at random’?