I'm performing a meta-analysis, attempting to produce a summary statistic for incidence rate of events of sudden death per 100,000 person years with 95% CI, and eventual subgroup analysis.
I am having difficulty finding a statistical methodology/instruction on using any stats package that can produce this statistic, along with the usual heterogeneity, publication bias, etc.
Can anyone provide advice about which package to use (i.e. Stata, R, Meta XL, etc.) with guidance about where to find specifics on the steps towards doing this?
Hello Aaron,
There are a number of survival analysis statistics that may be used for meta-analytic studies, for instance, hazard ratio and log hazard ratio. I don't know whether for your intended research goals whether these or some other summary statistic having to do with (what I imagine would be fairly low incidence events) your outcome of interest would be best.
The program, comprehensive meta-analysis (https://www.meta-analysis.com/) is capable of handling these and a wide variety of ES types. However, while you can use it for a free trial, it's not a free program.
The Cochrane collection software, Revman5 (https://community.cochrane.org/help/tools-and-software/revman-5) is freely available for academic/scholarly use. The user guide isn't really clear on whether the software will or won't actually compute ES values for you.
This link is to free guidance on using R for executing a meta-analysis: https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/
Finally, here's a more complete list of R libraries that help with the various phases of conducting meta-analysis: https://cran.r-project.org/web/views/MetaAnalysis.html
Good luck with your work!
The following is an example of a meta-analysis concerned with rates. It is completely descriptive--there are no inferential tests.
Article Nurse exposure to physical and nonphysical violence, bullyin...
I agree that you would likely calculate risk ratios: (group deaths / group population) / (total deaths / total population). Common practice is to treat anything greater than 1 as significant. When you are using “complete” data for the population (and subgroups), you do not use inferential statistics because you are not estimating any values. If you are using samples, then your confidence interval is determined by the sample size, population size of each group, and the group’s variance.
I hope this helps!
Thanks all.
As it happens, I'm doing two things. The first is actually attempting to meta-analyze the incidence of sudden death in athletes. This is not being compared to anything, thus statistics such as Risk or rate ratios aren't appropriate, as there is no comparison group.
The second goal is to compare the outcomes of screening athletes with ECG. In this case, I will be using the Revman tool.
I have gotten some advice that summarizing the incidence rates is possible in R, using metaprop. I have lots to do to learn R in the coming weeks. Thanks for all your help and advice.
As for the R book, I managed to contact Dr. Harrer, and his colleague quickly responded with advice, and instruction for R. Very kind. God bless the Dutch.
I think you need to describe your method more completely.
You say that incidence is not being compared to anything. The what are you doing for the subgroup analysis?
If you are only reporting incidence rates, you will be using descriptive statistics (as has been stated several times). Any statistics software can do this reporting — even basic spreadsheet software like OpenOffice.
- Badges
- Science method
- Similar topics
- Methodology