Book: Introduction to Clinical trials: RCTs
What is RCT?
Randomly allocated intervention
Benefits
Prospective
Planned, as are all clinical trials - not obs studies
Consistent
All follow-up can be timed from the same place for each pt
Controlled
control group for comparison
Contemporary
Control group and intervention - same period of time
Effect of outcome due to change over time removed
Comparable groups
before and at time of randomisation
Analytic benefit
randomisation provides basis for direct comparison between groups to quantify differences that intervention provides
Why RCTs preferred?
Can't compare one pt with outcome that might have happened without treatment
Outcome by chance or certain to happen?
Would effect be the same in other people?
Comparison with control group gives the treatment effect relativity
To attribute difference in outcomes between groups - groups must be the same in all aspects except the intervention
Differences between groups are all random, even those not measured, except intervention
Difficulties in interpreting treatment effect
Can compare OUTCOME difference between groups, but can't say definitively that this was due to treatment by just measuring outcome
Reasons other than treatment that might have caused difference in outcomes between groups:
Personal characteristics of pts
Regression to the mean phenomenon: poor health at outset, imperfect measurement methods - both return to more typical, healthy state regardless of treatment received.
Hawthorne effect: Being involved in study affects outcomes alone
Placebo effect: Perceived benefits by pts that aren't really caused by treatment
Quantifying attributable effects of treatment difficult but RCT makes comparison of treatment effect between groups simple and quantifies effect attributable to choice of treatment
Unbiased experiment
As well as randomisation for unbiased results we also need no:
Selection bias: groups formed so that one treatment appears better than other
Performance bias: ways treatment implemented favours one group
Randomisation addresses this
e.g. oral antibiotics in one group
e.g. specialist hospital setting for one group
Attrition bias: some pts with AEs may drop out and not be able to be used in analysis if no FU
Must encourage FU even if withdrawn from treatment
Must make treatment received by each group as similar as possible
Detection bias: way outcome assessed differs between groups
e.g. Which outcome is assessed - one treatment may not be able to be repeated so will automatically have better outcome
e.g. when outcome assessed might mean outcome captured at a time when pt is particularly poorly
e.g. how outcome assessed - person assessing subjective outcome might judge symptoms more severely in one grp than other, or e.g. radiologist may judge subjective scans differently knowing about the treatment
Solution isn't to just have objective outcomes - subjective are often most helpful to pts/drs
Blinding allocation is desirable
Not always possible - e.g. surgical trials
Can often blind or use independent assessor, e.g. radiologist, to avoid detection bias though
Reporting bias: findings are presented in a misleading/selective way
What does randomisation achieve? :
similar groups
Provides groups that can be analysed statistically to evaluate treatment diff
any diff between groups is random chance
could be adjusted for (if baseline value known)
don't need to know about disease area to know what to control for when allocating groups
Can ensure trial is carried out exactly the same between the 2 groups - can't with obs studies
Blinding also possible
How to randomise
1) choose appropriate rand method
2) Generate rand sequence according to rand method
3) Deliver rand allocation so that benefits are honoured
Simple rand
Only constraint is rand ratio - all pts have 1 in 2 chance of receiving treatment (if 1:1 ratio). Doesn't depend on other pts
restricted rand: to achieve balance in group sizes or balance in key characteristics
Outcome adaptive rand: where it depends on outcome and allocations of prev pts
random permuted blocks - equal distribution to each group (if a block is finished)
stratification with random permuted blocks
groups balanced on key characteristics
minimisation - minimises imbalance of specified characteristics between groups
Debate between types of rand
Against simple rand: may be unlucky and have v unbalanced groups
But loss of statistical precision likely to be modest
More of an issue: small trials may be unable to adjust for baseline vars or perform subgroup analyses reliably
Key prognostic factor imblance may cast doubt on validity of findings but can still be adjusted for during analysis
Choose based on statistical efficiency and predictability
Restricted randomisation more necessary in smaller trials as imbalance can't necessarily be adjusted for
Only restrict factor if it is prognostic, except for site
minimum number of factors
to minimise errors in allocation or in giving correct information
Needs to be delivered and info collected at convenient time, e.g. not in middle of surgery (so final assessment of disease status cannot be controlled for)
Generating allocation sequence
Simple randomisation: use random number sequences, odd numbers for Intervention, even for control. Prepared in advance or next number generated at time.
If advance, could reject sequences with big imbalance BUT this is form of restricted randomisation. Best if done formally.
Similar but more formal = randomise order of fixed number of Intervention/control allocations
MAXIMALLY TOLERATED IMBALANCE METHODS: minimises imbalance between groups at any time by minimally restricting. Complex
RANDOM PERMUTED BLOCK ALLOCATION: maintains balance in group sizes throughout allocation. Equal number of allocations to each group within each block
Check of sequence just to check that block randomisation carried out correctly
Stratification (with random permuted blocks): 2 separate randomisation blocks with next allocation dependent on e.g. sex
Balanced number of men vs women in treatment groups if full blocks used
often big difference in spread between treatment groups: depends on no. of pts randomised, stratification factors used, characteristics of pts, size of blocks within strata combinations - higher imbalance with more strat factors - but still modest compared to size of study
Minimisation: similar to stratification but combination of characteristics of next pt used to determine which allocation would provide least imbalance
various forms - Taves' method most commonly used: simple randomisation where there is a tieand first allocation
Delivering allocations
Instant allocation by recruiter - e.g. preprepared envelopes
allocation done at separate location, e.g. trial office and communicated to centre.
use of individual or group at centre to provide allocation (e.g. pharmacist)
But e.g. minimisation can't be used if using this as depends on allocations to previous pts. So method must work for allocation type chosen.
Methods that can track allocations (i.e. watch for misuse) are preferred
Need to ensure future allocations are concealed from recruiter
Randomisation must be available whenever needed
If used, must be opaque envelopes and must be sequentially numbered so break in sequence noticed
Why not have all trials RCTs?
If treatment/intervention undergoing refinement still, e.g. surgery
May be unethical - i.e. if benefits are not uncertain (equipoise) equally between control and intervention
Judgement may change over time with more knowledge
Not necessary if objective, e.g. dose escalation (to only pts with intervention) can be achieved without it
Comparative assessment only needed in later studies
Not feasible, due to e.g. cost, or otherwise impractical (e.g. too much additional work for clinical staff) or not enough pts will be able to be recruited