I have the Kaplan-Meier estimate for the survival function which I obtained using R's survival package:

data.surv <- Surv(data$days_to_event, data$event_flag)
data.fit <- survfit(data.surv ~ 1, data = data)

The data are censored at 90 days after the initial signal so the survival function has 90 points and its value at the last point is equal to the ratio of the total number of non-events at 90 days after the signal to total number of observations in the data set. This value is far from 0 since the signal causes the event to happen rather infrequently. Is that to be expected?

Now I have another data set where each observation has had the signal in the past (within 90 days) but no event so far. What is the proper way to apply the survival function estimate to the data to get expected number of events for some horizon? I thought $1-S(t_{current} + t_{horizon})/S(t_{current})$ for each observation is supposed to give me the probability, then adding them up would give the total count of the events expected from now to the horizon but even applying this to the original data set gives result which is way too low. It feels like there must be some mistake but I cannot find it. Could it be because of censoring (the majority of observations is censored - if the event does not occur within 90 days, the signal is thought to have lost its ability to cause the event)?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.