Meta analysis
Useful/important when:
- many studies with insufficient sample sizes
- summarising results of many trials
- to help guide researches planning new trials
Most in medical literature are of clinical trials, recently epidemiological reanalysed as well.
Main purposes:
- increase statistical power
- resolve uncertainty between conflicting reports
- improve estimates of size of effect
- answer new questions not posed at start of original trials
- to balance “overflow of enthusiasm” after introduction of new clinical
- therapy
Practical steps:
- identify studies of a topic
- define criteria for inclusion
- abstract data from eligible studes
- analyse data
If original data are available then analyse like multicentre, with trials replacing centres. Trial.treatment variance larger than centre.treatment variance because of different protocols. Taking trial effects as random can increase accuracy of treatment estimates.
Meta-analysis usually done on binomial data – with normal data individual trials usually achieve desired conclusion.