Hi: My result using Kaplan-Meier estimate in survival package was inconsistent with that from Minitab. The survival probabilities are same, but their 95% CI are different from other calculation. Could you help me find what is wrong with the code? Thanks in advance. library(survival) raw.a <- c(80,24,22,12,86,-74,48,41,39,38, 27,24,22,19,18,16,11,10,6.3,76, -21,17,12,11,10,9.8,9.5,8.2,6.4, 4.7,4.3,4.3,4.2,4.1,1.6,31) data.a <- data.frame(time = abs(raw.a), event = raw.a >= 0) fit.a <- survfit(Surv(time, event) ~ 1, data=data.a) summary(fit.a) Call: survfit(formula = Surv(time, event) ~ 1, data = data.a) time n.risk n.event survival std.err lower 95% CI upper 95% CI 1.6 36 1 0.9722 0.0274 0.92000 1.000 4.1 35 1 0.9444 0.0382 0.87251 1.000 4.2 34 1 0.9167 0.0461 0.83069 1.000 4.3 33 2 0.8611 0.0576 0.75524 0.982 4.7 31 1 0.8333 0.0621 0.72007 0.964 6.3 30 1 0.8056 0.0660 0.68611 0.946 6.4 29 1 0.7778 0.0693 0.65317 0.926 8.2 28 1 0.7500 0.0722 0.62109 0.906 9.5 27 1 0.7222 0.0747 0.58978 0.884 9.8 26 1 0.6944 0.0768 0.55916 0.862 10.0 25 2 0.6389 0.0801 0.49977 0.817 11.0 23 2 0.5833 0.0822 0.44261 0.769 12.0 21 2 0.5278 0.0832 0.38749 0.719 16.0 19 1 0.5000 0.0833 0.36066 0.693 17.0 18 1 0.4722 0.0832 0.33432 0.667 18.0 17 1 0.4444 0.0828 0.30846 0.640 19.0 16 1 0.4167 0.0822 0.28309 0.613 22.0 14 2 0.3571 0.0805 0.22962 0.555 24.0 12 2 0.2976 0.0773 0.17889 0.495 27.0 10 1 0.2679 0.0751 0.15463 0.464 31.0 9 1 0.2381 0.0724 0.13120 0.432 38.0 8 1 0.2083 0.0692 0.10865 0.399 39.0 7 1 0.1786 0.0654 0.08711 0.366 41.0 6 1 0.1488 0.0609 0.06673 0.332 48.0 5 1 0.1190 0.0555 0.04773 0.297 76.0 3 1 0.0794 0.0492 0.02355 0.267 80.0 2 1 0.0397 0.0373 0.00628 0.251 86.0 1 1 0.0000 NaN NA NA ----- A R learner. -- View this message in context: http://r.789695.n4.nabble.com/95-CI-of-Kaplan-Meier-survival-in-package-survival-tp2553696p2553696.html Sent from the R help mailing list archive at Nabble.com.
Terry Therneau
2010-Sep-27 12:29 UTC
[R] 95% CI of Kaplan-Meier survival in package survival
> My result using Kaplan-Meier estimate in survival package was > inconsistent with that from Minitab. The survival probabilities are > same, but their 95% are different.Confidence intervals for a survival curve may be calcualted on the linear (poor performance), log (good), log-log (good), or logit (good) scales. I expect that Minitab and S default to different choices. See the conf.type argument in help(survfit.formula). Terry Therneau