Interactive demo · Applied Statistical Learning in R and Python · Chapter 11

Kaplan–Meier curves with censoring on the brain cancer data

Of the 88 brain tumour patients, 35 died during follow-up and 53 are censored: we only know that they were alive at their last contact (the + marks). The Kaplan–Meier curve drops only at deaths, by the fraction of the patients still at risk who die there, and a censored patient simply leaves the risk set. Move the time cursor to see the risk set, the conditional survival factor and the estimate at that time; split the patients by a predictor to compare groups; compare the curve with two naive estimates that mishandle the censored patients.

Split the patients by

Time t (months)

Try these

Everything is computed in your browser from the 88 patients: the risk set rk and the deaths qk at each death time, Ŝ(t) = ∏ (rk − qk)/rk over the deaths up to t, Greenwood's standard error of log Ŝ, and the interval Ŝ·exp(±1.96 se) cut at 1. One patient has no recorded diagnosis and is left out of the diagnosis split.

In the lecture notes