Kate Leslie

Royal Melbourne Hospital, Parkville, VIC, Australia

Sample size calculation may seem daunting initially and they always seem to result in a sample size projection that is disappointingly large! However, without a proper, prospective sample calculation, your research is likely to lack power and will almost certainly be unpublishable. The aims of this lecture are to:

1. Introduce two-sample calculation for differences between two means and two proportions
2. Describe testing for equivalence and non-inferiority
3. Identify when you need special statistical help.

Sample size calculators are available on the web (e.g. www.stat.ubc.ca/~rollin/stats/ssize/). Standard statistical software packages usually are capable of calculating sample size as well. nQuery Advisor is one specialist package that allows sophisticated sample size calculations.
Information you need to start your sample size calculation
Type I error = the chance of rejecting the null hypothesis when the null hypothesis is true or in other words saying there's a difference where none exists (usually accepted as 1 in 20 [alpha = 0.05]
Type II error = the chance of accepting the null hypothesis when the null hypothesis is false or in other words saying there's no difference when there is (usually accepted as 1 in 5 or 1 in 10 [beta = 0.2 or 0.1]
Power = the chance of detecting a difference when it exists (1 - beta [usually described as 80% or 90%])
Effect size = the expected difference between the mean values (e.g. 40 mmHg difference) or proportions (i.e. 10% difference). Determine this from the literature or a pilot study, or you can choose a clinically important difference.
Variability = the expected standard deviation within each group (e.g. 20 mm Hg). Determine this from the literature or a pilot study.
Reference
Myles PS, Gin T. Statistical methods for anaesthesia and intensive care. Butterworth-Heineman, Oxford. Pages 24-28