Data Loading...
Statistical Issues Multiplicity Missing data Flipbook PDF
Subgroup analyses • Assume there the results of a clinical trial are suggestive of a difference, but the effect is not s
107 Views
91 Downloads
FLIP PDF 578.99KB
Statistical Issues Multiplicity Missing data Peter Fayers
What is multiplicity? 1. Subgroups 2. Multiple outcomes 3. Repeated assessments (follow-up)
Subgroup analyses • Assume there the results of a clinical trial are suggestive of a difference, but the effect is not statistically significant. • Divide the patients into two subgroups – e.g. those mildly ill, and those severely ill. • One subgroup will show a smaller effect, the other a larger effect. • To obtain a significant result, keep forming different subgroups until you are lucky!
Subgroup analyses A subgroup analysis claiming a qualitative interaction—in which the treatment is beneficial in one subgroup but harmful in another—is unlikely to be true in a clinical trial … The overall ‘average’ result of a randomised clinical trial is usually a more reliable estimate of treatment effect in the various subgroups examined than are the observed effects in individual subgroups.
Subgroup analyses Within a complex table reporting subgroup analyses of the odds of vascular death after streptokinase, aspirin, both, or neither for acute myocardial infarction, the first “presentation feature” given is astrological birth sign. For people labouring under the star signs Gemini and Libra, aspirin was no better than placebo. For others, aspirin had a strongly beneficial effect.
(Not PROs, but the point is universally true anyway) ISIS-2 infarction. Lancet 1988; ii: 349–60.
Subgroup analyses “Investigators should be cautious when undertaking subgroup analyses. Subgroup findings should be exploratory, and only exceptionally should they affect the trial’s conclusions. Editors and referees need to correct any inappropriate, overenthusiastic uses of subgroup analyses.” Subgroup analysis and other (mis)uses of baseline data in clinical trials Susan F Assmann, Stuart J Pocock, Laura E Enos, Linda E Kasten Lancet • Vol 355 • March 25, 2000 pp1064-1069 Editorial pp 1033-1034
Multiplicity If statistical tests performed on several independent outcomes, each at 5%, the chance of at least one false positive is: – 1 5% – 2 10% – 5 23% – 10 40%
Type I error, p-value, significance
Multiple Outcomes ICH E9 Guidelines: Generally, clinical trials should have ONE primary outcome variable
Bonferroni Adjustments • Ultra conservative – assumes all tests are on independent outcomes • If the total number of significance tests is N use
α*= α/N
– E.g. For 10 tests with an overall type I error of 0.05 (i.e. 5% p-value), test each outcome and reject the null hypothesis if the p-value < 0.05 / 10 (i.e. use p