Confidence bands in survival analysis | British Journal of Cancer
https://www.nature.com/articles/s41416-022-01920-5 “the appropriate type of uncertainty estimate is a confidence band constructed to account for the correlation between points on the curve, we will call this a simultaneous confidence band. This, is not the type of confidence band provided in standard software, which is constructed by joining the confidence intervals at given time points.” #epidemiology #biostatistics
@kaz_yos @PausalZ Interesting. At a skim, it looks like a pointwise interval targets nominal coverage for that particular point while simultaneous confidence bands target nominal coverage across the whole curve.
Is that your understanding too?
If not, please correct me. If so, it seems very close to a multiple testing discussion and it's not clear that either should be the automatic choice.
@Tim_P_Morris yeah that's my understanding. basically estimating a point versus the function
my understanding is that it's also related to multiple testing as well (there is a 'Bonferroni' approach to get confidence bands that is conservative)
@PausalZ Seems like if the target is difference in S(t) at a specified t then the pointwise intervals are correct. Same thing for difference in RMST?
@Tim_P_Morris that would makes sense to me. RMST is a single quantity, so pointwise should be fine (AFAIK)
