2

I want to perform a meta analysis of a particular subjects of cancer. My effect sizes are Overall Response Rate (ORR), Disease Control Rate (DCR), Duration of Response (DOR), Progression-Free Survival (PFS), and Overall Survival (OS). While ORR and DCR can be performed by meta analysis of proportions, PFS, OS, and DOR is in median months. I've managed to figure out how to do Cochran Q test for ORR and DCR. However, I can't manage to figure out how to perform heterogeneity test for DOR, PFS, and OS since it's in months and not in binary outcomes.

I have considered using PFS6m(%) and OS1y(%), wich units are in percentages. However, out of 7 studies, only 1 provided PFS6m and OS1y, thus I can't make a meta analysis using those

Furthermore, which statistical software package is eligible to perform I2 test? And, I performed Spearman correlation between median PFS and median OS, and it shows medium correlation with bad significance (p = 0.294). Is it possible performing meta analysis with those data?

Thanks!

amedicalenthusiast
  • 199
  • 7
  • 1
    There would seem to bee plenty published on this. Did you Google scholar the question and check your reference book? – Björn Feb 03 '19 at 07:58
  • Yes, but I can't find meta analysis of median PFS nor median OS, the only effect size I can find is Hazard Ratio (HR). Nevertheless, since the studies included in my analysis are single-armed, they don't yield HR. Furthermore, I can't manage to find heterogeneity test for median PFS/median OS, which units is in months. As far as I know, Cochran Q test and I2 statistic value are only available for dichotomous variables – amedicalenthusiast Feb 03 '19 at 09:22
  • If you search this site for question about meta-analysis and medians you may come across my answer https://stats.stackexchange.com/questions/88813/meta-analysis-of-means-and-medians-in-r/372139#372139 which gives references to two articles and a software package. – mdewey Feb 03 '19 at 14:20
  • Incidentally you are not correct that Q and $I^2$ can only be used for binary outcomes. – mdewey Feb 03 '19 at 14:21
  • @mdewey thanks for the information! I happened to have read your answer beforehand and tried to convert the data from median and range to mean and SD. But still, I can't find which heterogeneity test is appropriate for that cause. About Cochran Q test and I2 statistics, aren't they only compute dichotomous outcome (0 and 1)? – amedicalenthusiast Feb 03 '19 at 14:35
  • @mdewey #p.s. Also, thanks for the information about the statistical software. I'm aware that R is very eligible for single-armed meta analysis. But, it's very difficult and I'm still trying to learn it from the basic. While doing so, I try to overcome the problem by using another software, which is Comprehensive Meta-Analysis (CMA) ver. 3.0. Yet, I haven't figured out how to overcome the CI asymmetry problem. I have asked about this issue in another question (https://stats.stackexchange.com/q/390095/234366) – amedicalenthusiast Feb 03 '19 at 14:40

0 Answers0