*Geek Box: Cumulative Meta-Analysis

*Geek Box: Cumulative Meta-Analysis

You’ll be familiar with the concept of a meta-analysis, in which multiple studies are included and the results combined to obtain a single summary estimate of the effects of an intervention or exposure of interest. So, what is a “cumulative” meta-analysis?

The term appears to have first been used in a 1992 paper led by Fred Mosteller and Tom Chalmers analysing the effects of various interventions for myocardial infarction (6). In effect, the cumulative meta-analysis approach retrospectively calculates a summary estimate every time a new trial published in a specific time frame became available.

For example, let’s say in that in 2010 there were 15 studies available on the effects of a drug on blood glucose levels, and a meta-analysis of all 15 indicated an overall 20% reduction in blood glucose levels. Then in 2011, 3 more studies on the same question are published, followed by 2 in 2012, and 5 in 2013. A cumulative meta-analysis would examine the effects of these subsequent published studies on the original overall effect size, to see if whether the summary available to base conclusions upon at a given time would have changed.

For example, adding the 3 studies in 2011 may have increased the effect size to 22%, while the 2 added in 2012 altered the effect size again to 19%. A cumulative meta-analysis therefore allows researchers to demonstrate how the available evidence evolved over time, and what the strength of evidence was at any given time. It provides a means of assessing accumulating evidence over time: hence the term ‘cumulative’.

An example of how cumulative meta-analysis can be important for public health policy may be seen in the example of Sudden Infant Death Syndrome (SIDS). In New Zealand in the 1980’s, a case-control study suggested that lying in the prone position was a significant risk factor for SIDS. This prompted a public health campaign which led to profound reductions in the incidence of SIDS. However, a 2005 cumulative analysis of observational evidence from 1940 demonstrated that if the evidence had been analysed as it was accumulating, the risk of lying prone could have been identified at least 10yrs earlier, preventing thousands of deaths (7).

Thus, by updating the pool of evidence each time new studies emerge, it is possible to retrospectively quantify the strength of evidence at a certain time, and to continually update the strength of evidence as new studies are published.