70 likes | 274 Views
Example sample size calculations. Thomas Keegan School of Health and Medicine Lancaster University. Sample size calculators. To calculate sample sizes for your study you can Do it by hand Use a stats program (e.g. R, Stata) Use a web-based calculator
E N D
Example sample size calculations Thomas Keegan School of Health and Medicine Lancaster University
Sample size calculators • To calculate sample sizes for your study you can • Do it by hand • Use a stats program (e.g. R, Stata) • Use a web-based calculator • We are going to use a web-based calculator. It is found on: • Russ Lenth’s power and sample size page http://www.stat.uiowa.edu/~rlenth/Power/ Other calculators are available...
Two sample t test • Comparing two groups of people, one on standard treatment, one on new treatment, using continuous measure as outcome variable • Example: intervention is new type of statin. • Group 1 – old statin • Group 2 – new statin • To be useful, the new statin should decrease blood pressure by 5 mmHg more than old one (clinically relevant outcome) • We need to know: • Mean and SD of blood pressure in untreated group (160, 20mm Hg) • Assumptions: 5% significance (two tailed), required power is 80%
Test of two proportions • Comparing two groups using binary outcome (yes/no, mild/severe) • Example: trial to test the efficacy of an epilepsy drug • Outcome: no episodes or 1 or more, within 3 months • Group 1 – normal treatment, Group2 – intervention • Clinically relevant effect: 60% in no episode category rather than 50% • We need to know: • Proportion that are in in each treatment group • Significance level (0.05) two-sided, required power (0.8), assume equal group size
One sample t test • Comparing the effect of a treatment in one group of 12 y/o children • Example: test which asthma drug is better at controlling airway restriction: • In first week measure mean peak expiratory flow on standard treatment. Give new drug for a week, then measure mean again. • Clinically sig difference required: increase in peak flow of 40 l/min • We need to know: • SD of difference (50), power (0.8) and sig level (0.05, two sided) • Hypothesis is that true difference will be zero
Matched pairs • Test in one group of people whether intervention changed the outcome when outcome categorical • Example: in asthma study, only available outcome is whether participants achieved PEF more or less than 300 l/min • In first week – determine yes/no, then intervention for one week, determine yes/no, in same group of people • Or, when you have two groups with matched pairs Under these circs, ask a statistician Russ Lenth cannot help you!