184 lines
6.9 KiB
R
Raw Normal View History

2025-01-12 00:52:51 +08:00
library(survival)
aeq <- function(x, y, ...) all.equal(as.vector(x), as.vector(y), ...)
#
# Tests of the residuals.survfit function
#
# The influence argument of survfit returns all the residuals at every time
# point, but for large data sets the result will be huge. This function uses
# a different algorithm which will be faster when the number of time
# points being reported out is small.
# Start with small data sets and work up. First simple survival.
test1 <- data.frame(time= c(9, 3,1,1,6,6,8),
status=c(1,NA,1,0,1,1,0),
x= c(0, 2,1,1,1,0,0))
indx <- order(test1$time[!is.na(test1$status)])
s1 <- survfit(Surv(time, status) ~1, test1, influence=3)
# true influence for survival and hazard, in time order
inf1 <- matrix(c(-20, rep(4,5), -10, 2, -13, -13, 17, 17,
rep(0,6))/144, ncol=3,
dimnames=list(1:6, c(1,6,9)))
inf2 <- matrix(c(10, rep(-2,5), 10, -2, 7,7, -11, -11)/72,
ncol=2)
aeq(s1$influence.surv[indx,], inf1[, c(1,2,2,3)])
aeq(s1$influence.chaz[indx,], inf2[,c(1,2,2,2)])
r1 <- resid(s1, times=c(0, 3, 5, 8, 10))
all(r1[,1] ==0)
aeq(r1[indx,2:5], inf1[,c(1,1,2,3)])
r2 <- resid(s1, times=c(0, 3, 5, 8, 10), type="cumhaz")
all(r2[,1] ==0)
aeq(r2[indx,2:5], inf2[,c(1,1,2,2)])
# AUC is a sum of rectangles, height= S, width based on time points,
# so the leverage is a weighted sum of dfbeta values for S
r3 <- resid(s1, times=c(1,4, 8, 10), type="sojourn")
inf3 <- inf1 %*% cbind(c(0,0,0), c(3,0,0), c(5,2,0), c(5,3,1))
aeq(r3[indx,], inf3)
# exp(Nelson-Aalen)
s2 <- survfit(Surv(time, status) ~1, test1, stype=2, influence=3)
r4 <- resid(s2, times=c(0, 3, 5, 8, 10), type="pstate")
inf4 <- -inf2[, c(1,2,2)] %*% diag(s2$surv[c(1,2,4)])
aeq(r4[indx,2:5], inf4[,c(1,1,2,3)])
aeq(s2$influence.surv[indx,], inf4[,c(1,2,2,3)])
r5 <- resid(s2, times=c(1,4, 8, 10), type="sojourn")
inf5 <- inf4 %*% cbind(c(0,0,0), c(3,0,0), c(5,2,0), c(5,3,1))
aeq(r5[indx,], inf5)
# Fleming-Harrington
# This one is hard, the code still fails
s3 <- survfit(Surv(time, status) ~1, test1, ctype=2, influence=2)
inf6 <- matrix(c( rep(c(5, -1), c(1, 5))/36, c(5,-1)/36,
c(21,21,-29, -29)/144), ncol=2)
# r6 <- resid(s3, times =c(0, 3, 5, 8, 10), type="cumhaz")
# Part 2: single state, with start/stop data, multiple curves,
# second curve is identical to test1
# Then put it out of order.
test2 <- data.frame(t1 =c(1, 2, 5, 2, 1, 7, 3, 4, 8, 8,
0,0,0,0,0,0),
t2 =c(2, 3, 6, 7, 8, 9, 9, 9,14, 17,
9, 1, 1, 6, 6, 8),
event=c(1, 1, 1, 1, 1, 1, 1, 0, 0, 0,
1, 1, 0, 1, 1, 0),
x = rep(1:2, c(10, 6)),
id = 1:16)
s4 <- survfit(Surv(t1, t2, event) ~ x, test2, influence=TRUE)
r6 <- resid(s4, time=c(4, 8, 10), type="surv")
aeq(r6[1:10,], s4$influence.surv[[1]][,c(2, 5, 6)])
aeq(r6[11:16,],s4$influence.surv[[2]][,c(1,3, 4)])
aeq(r6[11:16,2:3], r1[,4:5])
r7 <- resid(s4, time=c(4, 8, 10), type="cumhaz")
aeq(r7[1:10,], s4$influence.chaz[[1]][,c(2, 5, 6)])
aeq(r7[11:16,],s4$influence.chaz[[2]][,c(1,3, 4)])
aeq(r7[11:16, 2:3], r2[,4:5])
# Compute the AUC at times 8 and 10, the first is a reporting time, the
# second is in between
r8 <- resid(s4, time= c(8, 10), type="auc")
aeq(r8[11:16,], r3[,3:4])
# curve1:
inf1 <- s4$influence.surv[[1]]
d1 <- inf1[,1:4] %*% diff(s4$time[1:5])
d2 <- inf1[,1:6] %*% diff(c(s4$time[1:6], 10))
aeq(cbind(d1, d2), r8[1:10,])
# curve2:
inf2 <- s4$influence.surv[[2]]
d3 <- inf2[,1:2] %*% diff(s4$time[9:11])
d4 <- inf2[,1:4] %*% diff(c(s4$time[9:12], 10))
aeq(cbind(d3, d4), r8[11:16,])
# scramble the data
reord <- c(1,3,5,7,9,11,13, 15,2,4,6,8,10,12,14,16)
test2b <-test2[reord,]
s5 <- survfit(Surv(t1, t2, event) ~x, test2b, influence=TRUE)
r9 <- resid(s5, time=c(4, 8, 10), type="surv")
aeq(r6[reord,], r9)
#
# For multistate use the same data set as mstate.R, where results have been
# worked out by hand. Except, make it harder by adding an initial state.
#
tdata <- data.frame(id= LETTERS[3*c(1, 1, 1, 2, 3, 4, 4, 4, 5, 5)],
t1= c(0, 4, 9, 1, 2, 0, 2, 8, 1, 3),
t2= c(4, 9, 10, 5, 9, 2, 8, 9, 3, 11),
st= c(1, 2, 1, 2, 3, 1, 3, 0, 3, 0),
i0= c(1, 2, 3, 2, 1, 1, 2, 4, 3, 4),
wt= 1:10)
tdata$st <- factor(tdata$st, c(0:3),
labels=c("censor", "a", "b", "c"))
tdata$i0 <- factor(tdata$i0, 1:4,
labels=c("entry", "a", "b", "c"))
if (FALSE) {
#useful picture
check <- survcheck(Surv(t1, t2, st) ~1, tdata, istate=i0, id=id)
plot(c(0,11), c(1,5.5), type='n', xlab="Time", ylab= "Subject")
tdata$idx <- as.numeric(factor(tdata$id))
with(tdata, segments(t1+.1, idx, t2, idx, col=as.numeric(check$istate)))
with(subset(tdata, st!= "censor"),
text(t2, idx+.15, as.character(st)))
with(tdata, text((t1+t2)/2, idx+.25, wt))
with(subset(tdata, !duplicated(id)),
text(t1, idx+.15, as.character(i0)))
#segments are colored by current state, case weight in center, events at ends
abline(v=c(2:5, 8:11), lty=3, col='gray')
}
tfun <- function(data=tdata) {
reorder <- c(10, 9, 1, 2, 5, 4, 3, 7, 8, 6)
new <- data[reorder,]
new
}
mtest2 <- tfun(tdata) # scrambled version
mfit1 <- survfit(Surv(t1, t2, st) ~ 1, tdata, id=id, istate=i0,
influence=1)
test1 <- resid(mfit1, times= mfit1$time, collapse=TRUE)
aeq(test1, aperm(mfit1$influence, c(1,3,2)))
aeq(sqrt(apply(test1^2, 2:3, sum)), t(mfit1$std.err))
test1a <- resid(mfit1, times=c(3, 7, 9), method=1, collapse=TRUE)
minf <- aperm(mfit1$influence, c(1,3,2)) # influence has time second, resid third
aeq(test1a, minf[,,c(2,4,6)]) # interpolated times work
test2 <- resid(mfit1, times= mfit1$time, collapse=TRUE, type="cumhaz")
aeq(sqrt(apply(test2^2, 2:3, sum)), t(mfit1$std.chaz))
test3 <- resid(mfit1, times= mfit1$time, collapse=TRUE, type="auc")
aeq(sqrt(apply(test3^2, 2:3, sum)), t(mfit1$std.auc))
# Do a couple AUC by hand
atime <- c(1, 5.6, 8.1, 15)
test4 <- resid(mfit1, times=atime, type="auc", collapse=TRUE)
all(test4[,,1] ==0) # before the first time
# 5.6 covers rectangles of widths 1,1,1, and .6 after times 2, 3,4 and 5
temp <- apply(test1, 1:2, function(x) sum(x*c(1,1,1, .6, 0,0,0,0)))
aeq(temp, test4[,,2])
temp <- apply(test1, 1:2, function(x) sum(x*c(1,1,1, 3, .1, 0, 0, 0)))
aeq(temp, test4[,,3])
temp <- apply(test1, 1:2, function(x) sum(x*c(1,1,1, 3, 1, 1, 1, 4)))
aeq(temp, test4[,,4])
#
# competing risks
#
mdata <- mgus2
mdata$etime <- with(mdata, ifelse(pstat==1, ptime, futime))
temp <- with(mdata, ifelse(pstat==1, 1, 2*death))
mdata$event <- factor(temp, 0:2, c("censor", "PCM", "Death"))
mfit <- survfit(Surv(etime, event) ~1, mdata, influence=1)
rr <- resid(mfit, time=360)
index <- sum(mfit$time <= 360)
aeq(mfit$influence.pstate[,index,], rr)