Mohammad Hassan Murad professor of medicine, Zhen Wang associate professor of health services research, Haitao Chu professor, Lifeng Lin assistant professor
Murad M H, Wang Z, Chu H, Lin L.
When continuous outcomes are measured using different scales: guide for meta-analysis and interpretation
BMJ 2019; 364 :k4817
doi:10.1136/bmj.k4817
The problem of meta-analysis
This paper dives into the dilemma of meta-analysis.
The theory is that meta-analysis leads to a larger sample size, narrower confidence intervals, and thus more apparent precision.
The reality is that separate trials are conducted with different rating scales - how many diverse rating scales obtain for depression! - and the mathematical framework may be everything from an ordinary mean to Number Needed to Treat to Odds Ratio.
The paper is really about the impossibility, mathematical or otherwise, of precise scientific meta-analysis.
The resulting forest plot may look very convincing, but it may be a mere mathematical artifice.
With medical statistics, what appears mathematically convincing may not be medically worthwhile. For example, a large sample size can conjure up a statistically significant result that has no medical worth.
To an amateur statistician such as me, the paper came across as mathematical legerdemain - the complicated rabbits being pulled out of the meta-analysis hat were, in the end, an admission that this process has incurable flaws.
Competing interests: No competing interests