meta analysis of median survival times

[Mashaal Dhir]

I am attempting a meta analysis where only median survival times (say variable x) are known along with sample size (n) of each study using stata. Upper and lower limits of the median survival is not mentioned by all studies. (Other methods odds ratio, IPD, hazard ratio not an option)

One of the previous studies have suggested the following methodology. Could some one suggest how to input the information with metan command (or any other command if appropriate) to get the output in stata.


The Weighted estimate of population median

where mi denotes the median survival within a study population i (with i from 1 to k) and wi refers a study specific weight function, and summation wi=1.. Weight of each study is equivalent to , number of study participants divided by the total number of evaluable patients. Confidence intervals for median estimates were not calculable and therefore the range of medians was provided instead of confidence limits.

[Sandy Kim]

Hi, what were the studies that suggested using this formula? Im trying to do the exact same thing and canno figure out how atm

[Clyde Schechter]

I'm pretty sure that -metan- does not handle this.

With your variables x and n, the code to calculate the formula in your post yourself would be:

gen inverse_x = 1/x
summ n
local n_total = r(sum)
gen w = n/`n_total'
summ inverse_x [aweight = w]

display "Weighted estimate of population median = " =1/`r(sum)'

[Tom Hsiung]

I like to learn this too.

[evank23]

Just to add that the original formula seems to have been first described here (in the appendix): https://www.ncbi.nlm.nih.gov/pubmed/3576013
Annoyingly almost impossible to find online, I had to scan at the library.

Btw I don't like this approach since it ignores heterogeneity.