0

I am trying to run Kaplan-Meier on a rather odd dataset and am having difficulty determining whether I should be truncating or censoring my data. I have looked at the other feeds, including this very helpful one, but I am still confused and want to make sure my line of reasoning isn't off.

The dataset is a sample of individuals with age-at-death data ranging from birth to old age. I have a few populations from different cities, effectively. I would like to evaluate overall pattern of survivorship (across the lifecourse) between the two cities. However, I would also like to identify whether there are significant differences in survivorship between the cities for demographic segments - survivors and nonsurvivors (divided at age 18).

To look at non-survivors (those under 18) it seems to me that I should right-censor since I am factoring out the adults who have survived that period. To focus on the adult period, though, would I left truncate since I am artificially cutting out those who died before 18 years of age? If it is left truncated, would I just eliminate those entirely from my analysis before I run Kaplan-Meier?

I am not a statistician by any means and this one is really throwing me. Any advice would be most appreciated!

Janth
  • 1
  • 1
    I can understand your interest in (e.g. demographic) differences between those who survived past a certain age and those who did not, but what I don't understand is why you are interested in the *survival* of those who survived past the threshold age and those who did not... what would that even mean? – Frans Rodenburg Jan 10 '20 at 06:27
  • 1
    In addition, looking at survival of those who survived past some event and those who did not, introduces an immortal time bias for the former group. This is similar to the (spurious) finding that winning an Oscar increases life expectancy. – Frans Rodenburg Jan 10 '20 at 06:51
  • When you put it this way, I can see why it would be rather odd and unclear (to say the least). I guess I was trying to exclude the individuals who were in the sample for non-survivors when testing the survivor sample so they wouldn't be influencing the results. But maybe that isn't the right way to go about it from your 2nd comment.. Given this, the way to evaluate survivorship for the survivors would be to simply include the whole sample? Is it appropriate to right-censor the data at age 18 to evaluate whether there is a significant difference between sites in 'non-survivor' survivorship? – Janth Jan 12 '20 at 04:58
  • (Just to add - ran out of space) I have seen right-censoring done in this way in publications in my field, but when I'm reading the literature it seems to not quite fit so I'm not actually sure that it's appropriate (and don't want to replicate if it isn't). – Janth Jan 12 '20 at 05:03

0 Answers0