Randomisation and Treatment Allocation Flashcards
Double blind vs single blind
Double - no one knows, single - one or other knows
Why blind?
get rid of placebo effect and observer bias
Who is prone to bias?5
those recruiting, treating, patients, assessing outcomes, statisticians
When is blinding less important
When hard outcomes like death
Why randomise (3)
Eliminate bias, treatment groups don’t differ systematically. balances both known and unknown prognostic factors
Why simple randomisation isn’t good?
Groups can be unbalanced
What to use instead of simple randomisation?
Random permuted blocks
What do you need to do to stop prediction when using blocks? 2
Don’t reveal block length and vary it
If block size is too big
imbalance
Why use stratification?
In trials we want treatment groups to be balanced with respect to patient characteristics
When is it important to use stratification?
When there is factors of particular importance and groups need to be balanced
common stratification factors
age, disease stage, sex, country
What do you use to stratify? explain
Stratification lists - create sep random lists within each stratum- so london m and f - next patient assigned to treatment for sex/center
Other way of stratification?
minimisation
when to use minimisation?
a lot of stratification factors
What is Zelen’s design?
randomised to treatment or control before consent, if refuse treatment, move to standard of care group
playing the winner’s rule
weigh probability of allocation in favor of treatment with best results - bias
When to use unequal randomisation?
New drug vs standard, already know alot about standard
Problem with unequal randomisation
Stat efficiency
What is an example of cluster randomisation?
Randomise a whole school
What is allocation concealment?
Ensure person randomising doesn’t know what treatment
What happens if there is too few patients? 2
Important treatment effects may be missed, or may show X works when it doesn’t
What are the problems when there is too many patients 4
Unethical risk, extra time and money, delay imp results, delay more trials
What is null hypothesis?
Statement we want to reject to prove effect of our treatment
What is alternative hypothesis?
statement that we will accept if enough evidence to reject null hypothesis
what is a type II error?
False -ve fail to reject H0 but treatment actually better
What is type I error?
false positive - rejected H0 but new treatment no dif or worse
What is significance?
probability that we reject H0 given that it is correct so probability of type I error
What is significance linked to?
Prevalence
What is often the value of significance?
5%
What is power?
Probability that we reject H0 given that H1 is true- so getting the sig difference correct
Significance’s sign
alpha
Power’s formula?
1-prob of type II (1-beta)
What is usually the value of the power?
80-90%
What happens if we decrease significance?
Sacrifice power…
What are sample size calculations based on? how?
primary endpoint - formula depends on outcome measure
What does sample size depend on? 4 + signs
- significance level alpha 2. Power 1-B 3. Effect of size delta 4. variability - sigma squared
As significance increases, what happens to sample size?
decreases
As power increases, what happens to sample size?
increases
As effect size increases, what happens to sample size?
decreases
As variability increases, what happens to sample size?
increases
Formula for sample size for continuous outcome measures:
(2(variance)/(diff)^2 )* f(alpha, beta)
Variance =
SD^2
What is the treatment dif in the formula?
difference between two means
for binary outcome the formula is:
check notes
What do we do when expecting loss to follow up?
Adjust estimate using: group/ 1-rate of expected loss
