library(survival) # Make sure that the old-style and new-style calls both work # new style vet2 <- survSplit(Surv(time, status) ~ ., data= veteran, cut=c(90, 180), episode= "tgroup", id="id") vet2[1:7, c("id", "tstart", "time", "status", "tgroup", "age", "karno")] # old style vet3 <- survSplit(veteran, end='time', event='status', cut=c(90,180), episode="tgroup", id="id") all.equal(vet2, vet3) all.equal(nrow(vet2), nrow(veteran) + sum(veteran$time >90) + sum(veteran$time > 180)) # Do a parallel computation using survSplit, and pyears/tcut. We should get # the same answer. # Break subjects up by year of entry and current time on study. Most of # the deaths are within 3 months # pyears complains (justifiably) about the obs with 0 days of fu, so we add 1 # data1 <- jasa data1$ayear <- as.numeric(substring(as.character(jasa$accept.dt), 1,4)) temp <- round(c(0, .25, 1,2,5)*365.25) # years ftime <- tcut(rep(0, nrow(jasa)), temp, labels=paste(c(0, .25, 1:2), c(.25, 1,2,5), sep='-')) pfit <- pyears(Surv(futime +1,fustat) ~ ayear + ftime, data1, scale=1) data2 <- survSplit(Surv(futime+1, fustat) ~ ., cut=temp, data=data1, episode = "tgroup") tab1 <- with(data2, tapply(fustat, list(ayear, tgroup), sum)) tab1 <- ifelse(is.na(tab1), 0, tab1) all.equal(as.vector(tab1), as.vector(pfit$event)) # ignore dimnames tab2 <- with(data2, tapply(tstop-tstart, list(ayear, tgroup), sum)) tab2 <- ifelse(is.na(tab2), 0, tab2) all.equal(as.vector(tab2), as.vector(pfit$pyears)) # double check that the "data" option gives the same values pfit2 <- pyears(Surv(futime +1,fustat) ~ ayear + ftime, data1, scale=1, data.frame=TRUE)$data all.equal(pfit2$pyears, pfit$pyears[pfit$pyears >0]) all.equal(pfit2$event, pfit$event[pfit$pyears >0]) # and that the rows of data2 have the right labels keep <- which(pfit$pyears >0) # these are not in the data rname <- rownames(pfit$pyears)[row(pfit$pyears)[keep]] all.equal(rname, as.character(pfit2$ayear)) cname <- colnames(pfit$pyears)[col(pfit$pyears)[keep]] all.equal(cname, as.character(pfit2$ftime))