Topic 6: Sample Size Flashcards
What three things do we need enough participants for
To be generalisable, to estimate the treatment effect precisely, to have a good chance of reliably detecting a treatment effect, or to conclude no treatment effect if the results don’t show it.
What things aside from validity do we need to consider in terms of sample size
Ethics - exposing to risk. Economics - may be an upper limit to sample size due to cost.
What are the three main kinds of endpoint
Continuous, binary, time to event
When comparing proportions, what kind of regression is used
Logistic regression (binomial)
When comparing rates, what kind of regression is used
poisson regression
When comparing means between treatment groups, what kind of regression is used
Linear
What are the 6 key considerations in determining sample size (Think sample size formula)
Nature of primary outcome, method of analysis, what results are anticipated in control, treatment difference to be detected, variability of the response, degree of certainty needed to detect a treatment difference.
How do we know the standard deviation of endpoint and the results anticipated in the control group?
Prior data from similar trials
What is a type 1 error
False positive
What is a type 2 error
False negative
What does increasing the sample size do the type 1 and 2 error rates
Reduces them
When do you declare statistical significance
When the test statistic lies beyond the threshold (critical value) determined by the type 1 error rate (the significance level).
How does the sample size formula change when changing what is being compared in the analysis
The form of the standard error is different
Do you need a smaller or larger sample size to detect a smaller treatment difference
Larger
If we reduce the power needed, what does this do to the sample size
Reduces sample size needed
When the standard deviation of the outcome is larger, what does this do to required sample size
Larger sample size needed.
What does an imprecise estimate of the true effect say about the conifidence intervals
Wide confidence intervals around the estimate
Give three ways to increase sample size
Increase accrual rate, relax scientific requirements, run a small pilot study.
How can you increase the accrual rate
Multi-centre trials, relax eligibility
How do pilot studies help when considering sample size
Give more accurate estimates of the various parameters we need in our sample size calculation.
Give 3 things that may affect sample size, and therefore that sample size can be inflated to account for
Incomplete outcome data/missing data, patients moving between arms, unequal allocation.
Why might we get incomplete outcome data
Loss to follow up, death before outcome can be measured, withdraw consent, becomes ineligible, randomised in error, technical problems collecting outcome.
Why do we need to inflate sample size for patients moving arms
Because when this happens generally the groups in the trial will become more similar and the treatment effect will get smaller and smaller.
What is the design effect
How much we need to inflate the sample size when using a more complex design compared to an individually randomised trial
What is used to calculate the design effect in a trial with clustered data
ICC and cluster size.
As the clusters increase in size, what does this to do the required sample size
The bigger the cluster, the more we need to inflate the sample size by.
Does required sample size change for a non-inferiority trial
Yes
Does required sample size change for an equivalence trial
Yes
