Sample size for clinical trials

An outcome variable is one which we hope to change, predict or estimate in a trial. Examples: • Systolic blood pressure in a hypertension trial • Caesarean section rate in an obstetric trial • Survival time in a cancer trial How many outcome variables should I have? If we have many outcome variables: • all possibilities would be covered, • we would be less likely to miss something, • the risk of false positives, finding an apparent effect when there is none in reality, would be increased. If we have few outcome variables: • the trial would be easier and quicker to check and analyse, • the trial would be cheaper, • the trial would be easier for research subjects, • we would avoid multiple testing and the high risk of a false positive result. We get round the problem of multiple testing by having one outcome variable on which the main conclusion stands or falls, the primary outcome variable. If we do not find an effect for this variable, the study has a negative result. Usually we have several secondary outcome variables, to answer secondary questions. A significant difference for one of these would generate further questions to investigate rather than provide clear evidence for a treatment effect. The primary outcome variable must relate to the main aim of the study. Choose one and stick to it. A significance test for comparing two means is more likely to detect a large difference between two populations than a small one. The probability that a test will produce a significant difference at a given significance level is called the power of the test.