How to run a proper meta-analysis when the selected studies only report the median and 95% Confidence Interval of a parameter of survival?
Should I treat the data in the same way as a mean, and apply the same meta-analysis techniques?
Some proposed methods such as Hozo and Wan address this question, but either IQR, Quartiles or full data distribution are required for estimating mean and SD, which trials do NOT provide. Please assume I DO NOT have the range, so estimating the mean and SD would not be possible also.
I have already emailed the authors for the full data, though.
Can anyone with experience in the topic help?
Thanks in advance!
It is nearly imposible to convert from only median to mean without knowing the interquatiles range and min-max value. In your case, I would suggest the calculation and plotting of median of median differences. Here is an excellent articles discussing your problem: Article Meta-analysis of the difference of medians
Thank you so much for your reply!
Some trials share the min-max value. How would this change the calculation?
And I'm proposing a single arm meta-analysis of those medians. Is this a possibility?
Lucas Mitre Knowing min, max and median allows you to calculate interquatile range. Knowing interquatile range, median, min, max allows you to approximate mean and SD by transformation based approaches that discussed in the article
Lucas Mitre Did you mean the meta-analysis of single arm trials? If yes, that depends on your data. Keep in mind that sometimes median or mode is more helpful than mean (e.g. presenting result of a likert scale). However, as what I experienced, most of meta-analysis for clinical trials use binary data analysis (Odds ratio, RRs, etc) or mean differences. If you can tell me what type of data you are going to collect, I may suggest an effect size estimation for your data
Here I would suggest you another excellent article with more than 7000 citation! read this and I believe you can understand what I have explained as the method is really simple.
https://bmcmedresmethodol.biomedcentral.com/articles/10.1186/1471-2288-5-13
Sure thing! Thank you so much for your elaborate response. I am developing a single-arm meta-analysis of clinical trials that do not have a control/comparison group. Most of the report Median, min-max (not all), and the 95% CI of the Median - for *survival outcomes* (PFS, OS, ORR...) of this only one group!
If the survival outcomes were only presented as median (95% CI), and min-max, I would suggest to plot them in form of forrest plot with median as effect size and 95% CI of median as 95% CI of effect size (you may have to manually calculate and plot these as the forest plot commands on STATA or R are developed for conventional effect size but not median). However, I think the best way is to tranform them to mean and SD, then apply the normal protocol for meta-analysis to analyse them. Those without reporting min-max should be excluded from meta-analysis and you can systematically review them by traditional way. When you meta-analyse the median, I assume that your data do not point out the differences between groups, so your meta-analysis should be more like meta-analysis of observational study.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS