Calculating the confidence interval for median survival.     

Beginning in Prism version 10.5.0, Prism will report the median survival for each analyzed group along with its associated 95% confidence interval. While there are multiple different methods to calculate the confidence interval for median survival (see links at the end of this FAQ), Prism has implemented the "complementary log-log" method of calculation. This method involves the transformation of the survival function:

[\begin{equation}\begin{split}-\log\{ -\log\bigl(\hat{S}(t)\bigr) \}\end{split}\end{equation}]

According to Talsma (2023), the Kaplan-Meier estimator and the log-log transformation represented the optimal combination that provided a good ability to estimate the 95% CI of median survival. This method is also used in a variety of other applications and packages, including SAS, Stata, and SPSS. Additionally, while it is not the default, this calculation method is also available in the survfit package of R. Generally, this approach is more aligned with the assumption of proportional hazards than other approaches. And while the Kaplan-Meier method of survival analysis does not directly require this assumption to be true to generate a survival curve, the log-rank test (which Prism uses to compare survival curves) does rely on this assumption. For these reasons, the complementary log-log method for calculating the 95% CI of median survival is the default (and only) method available in Prism (starting in version 10.5.0).

If there are other specific methods that you would like to see added to Prism for this calculation, reach out to us and let us know.