Can I calculate median survival time and its confidence interval using the Kaplan-Meier curve? I don't have available raw data

Question

I am conducting a meta-analysis to compare the differences between Group A and Group B by measuring the pooled Hazard Ratios (HRs) for the median Progression-Free Survival (PFS) and Overall Survival (OS).

While I have accurately calculated the pooled HRs using R, I also intend to present the pooled median values for PFS and OS.

Pooled median values can be calculated with their respective medians and ranges (95% CI or IQR). However, there are limitations due to studies in which the range of median values cannot be available. I am curious if there are methods to estimate the median values and ranges using techniques such as Kaplan-Meier curves or HRs from each study.

Answer 1

I am not aware of meta-analysis of median survival. There are some approaches that meta-analyze the ratio of medians, under the (questionable) assumption that this ratio approximates the hazard ratio.

If you want to obtain a "pooled" median (for group A), you will need two ingredients:

1. reference survival function (very well characterized population with individuals treated/exposed to B - it can be a pragmatic trial, large cohort study with good representativeness and low risk of bias).
2. summary HR from your meta-analysis.

You will basically convert the survival function from the reference population into a hazard function, and then multiply the hazard function by the summary HR, and convert it back to survival function.

For more details, check this question:

Is it possible to apply a hazard ratio estimated from one survival distribution to another