Meta-analysis

A meta-analysis is the statistical combination of the numerical results of several studies examining a similar question. According to the Cochrane Handbook, an important step in a systematic review is to decide carefully whether it is appropriate to combine the numerical results of all or some of the studies. ^[1] In a systematic review, studies are searched for and selected with predefined criteria; the meta-analysis combines the numerical results of these studies.

How are the results combined?

An effect estimate and its standard error are taken from each study. If the studies are assumed to measure the same true effect, a fixed-effect model is used; if a real difference in effect between studies is assumed possible, a random-effects model is used. From the standard error of the pooled estimate, a confidence interval showing the precision of the estimate and a p value showing the strength of evidence against the null hypothesis are derived. ^[1]

Limits

• Study quality. If the pooled studies have a high risk of bias, the result is also unreliable. In the GRADE approach, certainty is assessed through risk of bias, inconsistency, indirectness, imprecision and publication bias. ^[2]
• Differences. If the studies differ in the group of people, dose and duration, the pooled number may not reflect any single study; inconsistency between studies (heterogeneity) and subgroups are therefore read separately. ^[1]
• Corrections. A meta-analysis can later be corrected or updated; in the example of Vitamin K2 on the site, the current analysis is taken as the basis because the integrity of some included studies was questioned.

On the site, grade A requires, apart from established physiological knowledge, multiple consistent randomized trials or a meta-analysis of them (Levels of evidence). For how the results of a meta-analysis are read: How to read a study?.