Predictions

1 R implementation for https://pclambert.net/software/stpm2/stpm2_timevar/

library(rstpm2)
library(readstata13)
suppressWarnings(rott2b <- read.dta13("https://www.pclambert.net/data/rott2b.dta"))
rott2b <- transform(rott2b, time = pmin(rf,60)/12)
rott2b <- transform(rott2b, event = (rfi==1) & (time < 60/12))
fit <- stpm2(Surv(time,event) ~ hormon + nsx(age,df=3,centre=60,stata=TRUE) + pr_1, 
             data=rott2b, df=4, stata=TRUE)
eform(fit)
Profiling...

                                                exp(beta)        2.5 %       97.5 %
(Intercept)                                  6.055607e-04 5.715701e-04 6.402896e-04
hormonyes                                    1.253597e+00 1.065865e+00 1.462097e+00
nsx(age, df = 3, centre = 60, stata = TRUE)1 6.550003e-01 5.594371e-01 7.702334e-01
nsx(age, df = 3, centre = 60, stata = TRUE)2 3.449221e-01 1.993938e-01 5.962187e-01
nsx(age, df = 3, centre = 60, stata = TRUE)3 9.626652e-01 7.189637e-01 1.275743e+00
pr_1                                         9.080228e-01 8.944134e-01 9.214805e-01
nsx(log(time), df = 4)1                      4.474345e+02 4.148646e+02 4.799754e+02
nsx(log(time), df = 4)2                      1.974220e+02 1.894423e+02 2.047062e+02
nsx(log(time), df = 4)3                      5.391834e+04 4.862060e+04 5.986957e+04
nsx(log(time), df = 4)4                      7.282589e+01 6.949526e+01 7.650186e+01
nd <- data.frame(age=60, hormon="no", pr_1=0)
plot(fit, type="surv", newdata=nd, xlab="Years from surgery")

surv_1.png

nd <- data.frame(age=60, hormon="no", pr_1=0)
plot(fit, type="haz", newdata=transform(nd, hormon="yes"), xlab="Years from surgery",
     line.col="blue", rug=FALSE)
lines(fit, type="haz", newdata=nd, col="red")

haz_1.png

s_time1 <- predict(fit, newdata=transform(rott2b, time=1), type="surv")
hist(s_time1,xlab="Time from surgery",main="")

histo_1.png

s_time5 <- predict(fit, newdata=transform(rott2b, time=5), type="surv")
hist(s_time5,xlab="Time from surgery",main="")

histo_2.png

nd <- with(rott2b, data.frame(age=seq(min(age), max(age), length=301), hormon="no", pr_1=3.43, time=1))
s <- predict(fit, newdata=nd, type="surv", se.fit=TRUE)
ci.polygon <- function(x, y, ...) polygon(c(x,rev(x)), c(y[,1],rev(y[,2])), ...)
matplot(nd$age, s, type="n", xlab="Age (years)")
ci.polygon(nd$age, s[,2:3], col="grey", border="grey")
lines(nd$age, s[,1])

surv_1_age.png

s <- predict(fit, newdata=transform(nd,time=5), type="surv", se.fit=TRUE)
matplot(nd$age, s, type="n", xlab="Age (years)")
ci.polygon(nd$age, s[,2:3], col="grey", border="grey")
lines(nd$age, s[,1])

surv_5_age.png

Author: Mark Clements

Created: 2018-08-02 Thu 11:29

Validate