Censoring on progression free survival curve

Hi, oncology fellow here trying to understand censoring in survival analysis. Any help from the stat gurus here would be much appreciated.

I’m reading this article looking at a drug combination progress free survival (PFS). There was a comment saying:

Regeneron also provided an analysis estimating that median progression-free survival was 24 months, seemingly far superior to the 10.1 months in Opdualag’s label. However, caution must be used here: the survival curve for fianlimab plus Libtayo is heavily censored, and with just a few more early events the median could easily halve.
Article here if interested: Esmo 2022 – Regeneron stakes its Lag3 claim | Evaluate

I hope to understand what does it mean by survival curve being heavily censored, and how did it not account for early events. They are saying because this survival curve is heavily censored, the median PFS may be over-estimated and therefore the numerically higher median PFS compared to another clinical trial may not be clinically significant. But I am having difficulty conceptualizing this - any help would be greatly appreciated. Thank you all very much.

1 Like

you wouldn’t say the estimated median doesn’t account for early events; it seems they are just speculating on the impact such events would have on the estimated median if they had been seen. Reporting the confidence interval would be a better way to make the point. I wouldn’t say PFS is heavily censored (the median is at least estimable, ie the censoring rate <50% - it is just the proportion without an event). More important might be when and why censoring is occurring. If there is early censoring it is worth wondering what explains it


Please give us details on the cause of censoring.

1 Like