4

I need to draw a funnel plot for my meta-analysis, which is observational and aggregates prevalence rates of a certain medical condition. As far as I know, a funnel plot is a scatter plot of effect size measures against sample sizes. Therefore I have some difficulties in my study which I think does not have an effect size measure:

  1. I don't know if there is any effect size measures for prevalence of a single study? Each study reports a single prevalence, so I have a bunch of ratios (0-100%) only. Each study has only one single ratio, so there can be no "prevalence ratios" or "prevalence odds ratios".

  2. Can I plot the ratios (the prevalences) against sample sizes, instead of non-existent effect sizes against sample sizes?

  3. Can I calculate 95% CI for each ratio (each prevalence), and use it as the effect size? If so, does it help when the Y axis is already "Sample Size" and CI is directly affected by sample size?

  4. Do you have any other hints, in addition to the nice ones given here?

Many thanks.

Community
  • 1
Vic
  • 1,126
  • 5
  • 17
  • 28
  • 2
    The prevalence ratio **is** the effect measure in a prevalence study. Another possibile effect measure would be the *prevalence odds ratio*. [This paper](http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1247374/pdf/ehp0112-001047.pdf) discusses the two measures. I've found a funnel plots with prevalences [here](http://www.who.int/bulletin/volumes/90/2/11-091850/en/) (Fig. 3) and [here](http://www.sciencedirect.com/science/article/pii/S1438463913000448#). As the prevalence is a proportion, it might be possible to calculate the standard error if the sample size is given. – COOLSerdash Jun 25 '13 at 14:33
  • Many thanks dear Serdash. It was such a relief to see those figures and hearing that prevalence itself is the effect size measure. About the POR, I had read that article (very hastily though) but thought POR is calculable only when we have two prevalence rations at hand, right? (this is why I said each study reports a single prevalence). – Vic Jun 25 '13 at 14:52
  • 1
    Jup, for the POR, you need the prevalence rate of the cases and controls. – COOLSerdash Jun 25 '13 at 14:54
  • Ah hold on a sec please, do you mean the ratio of prevalence 1 / prevalence 2, by the term "prevalence ratio"? Or did you mean "prevalence" by "prevalence ratio"? In my case, each study has only a single prevalence (which is itself a ratio of affected people divided by the sample size). It does not have any P1/P2 ratios. Do you think can I still use such prevalence rates as effect sizes? – Vic Jun 25 '13 at 14:58
  • 1
    Yes, I think so. – COOLSerdash Jun 25 '13 at 15:01
  • Of course those figures show I can apparently. I was curious because I didn't see "prevalence" itself among effect size measures here: http://en.wikipedia.org/wiki/Effect_size – Vic Jun 25 '13 at 15:06
  • 1
    Maybe [this post](http://stats.stackexchange.com/a/59173/21054) can shed further light unto the issue. – COOLSerdash Jun 25 '13 at 15:09
  • 1
    Update: I think I can calculate SE for prevalences too, thus SE is not non-existent. :) – Vic Jun 26 '13 at 07:45

0 Answers0