Hi, can anyone help me to understand the standard errors printed in the output
of survfit.coxph()?
time<-sample(1:15,100,replace=T)
status<-as.numeric(runif(100,0,1)<0.2)
x<-rnorm(100,10,2)
fit<-coxph(Surv(time,status)~x)
### method 1
survfit(fit, newdata=data.frame(time=time,status=status,x=x)[1:5,],
conf.type='log')$std.err
[,1] [,2] [,3] [,4] [,5]
[1,] 0.000000000 0.000000000 0.000000000 0.000000000 0.00000000
[2,] 0.008627644 0.008567253 0.008773699 0.009354788 0.01481819
[3,] 0.008627644 0.008567253 0.008773699 0.009354788 0.01481819
[4,] 0.013800603 0.013767977 0.013889971 0.014379928 0.02353371
[5,] 0.013800603 0.013767977 0.013889971 0.014379928 0.02353371
[6,] 0.013800603 0.013767977 0.013889971 0.014379928 0.02353371
[7,] 0.030226811 0.030423883 0.029806263 0.028918817 0.05191161
[8,] 0.030226811 0.030423883 0.029806263 0.028918817 0.05191161
[9,] 0.036852571 0.037159980 0.036186931 0.034645002 0.06485394
[10,] 0.044181716 0.044621159 0.043221145 0.040872939 0.07931028
[11,] 0.044181716 0.044621159 0.043221145 0.040872939 0.07931028
[12,] 0.055452631 0.056018832 0.054236881 0.051586391 0.10800413
[13,] 0.070665160 0.071363749 0.069208056 0.066655730 0.14976433
[14,] 0.124140400 0.125564637 0.121281571 0.118002021 0.30971860
[15,] 0.173132357 0.175309455 0.168821266 0.164860523 0.46393111
survfit(fit, newdata=data.frame(time=time,status=status,x=x)[1:5,],
conf.type='log')$time
[1] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
### method 2
summary(survfit(fit, newdata=data.frame(time=time,status=status,x=x)[1:5,],
conf.type='log'),time=10)$std.err
1 2 3 4 5
[1,] 0.04061384 0.04106186 0.03963184 0.03715246 0.06867532
By reading the help of ?survfit.object and ?summary.survfit, the standard error
provided in the output of method 1 (survfit()) was for cumulative
hazard-log(survival), while the standard error provided in the output of method
2 (summary.survfit()) was for survival itself, regardless of how you choose the
value for "conf.type" ('log', 'log-log' or
'plain'). This explains why the standard error output is different
between method 1 (10th row) and method 2.
My question is how do I get standard error estimates for log(-log(survival))?
Thanks!
John
[[alternative HTML version deleted]]