tests/quantile.R

In therneau/survival: Survival Analysis

#
# Formal test of the quantile routine for survfit
library(survival)
aeq <- function(x, y, ...) all.equal(as.vector(x), as.vector(y), ...)

# There are 8 cases:  strata Y/N,  ncol(surv) >1,  conf.int = T/F
#  Subcase: the quantile exactly agrees with a horizontal segment of 
#    the curve or not.
# First do the 4 cases where fit$surv is a vector
#
test1 <- data.frame(time=  c(9, 3,1,1,6,6,8, 10),
                    status=c(1,NA,1,0,1,1,0,  0),
                    x=     c(0, 2,1,1,1,0,0,  0))

# True survival = (6/7) * (3/5) * (1/2) for overall
#   The q's are chosen to include a point < first jump, mid, after last jump,
#   and exact intersections with the "flats" of the curve.
#   
qq <- c(13/14, 6/7, 2/3, .5, 9/35, .1)
 
# Nothing on the right hand side, simple survival (no strata)
fit1 <- survfit(Surv(time, status) ~ 1, test1, conf.type='none')
aeq(quantile(fit1, 1-qq), c(1, 3.5, 6, 9, 9.5, NA))  #without conf.int

fit2 <-  survfit(Surv(time, status) ~ 1, test1)  #with conf.int
aeq(quantile(fit2, 1-qq), 
    list(quantile = c(1, 3.5, 6, 9, 9.5, NA),
         lower = c(1,1,1,6,6,9),
         upper = rep(as.numeric(NA), 6)), check.attributes=FALSE)
aeq(quantile(fit2, 1-qq, FALSE), c(1, 3.5, 6, 9, 9.5, NA)) 


# Now a variable on the right (strata in the result)
#  curve 0: (t=6, S=3/4),  (t=9, S=3/8)
#  curve 1: (t=1, S=2/3),  (t=6, S= 0)
fit1 <- survfit(Surv(time, status) ~ x, test1, conf.type='none') 
aeq(quantile(fit1, 1-qq),
    matrix(c(6,6,9,9,NA,NA,  1,1,3.5, 6,6,6), nrow=2, byrow=T))

fit2 <- survfit(Surv(time, status) ~ x, test1)
aeq(quantile(fit2, 1-qq, FALSE),
    matrix(c(6,6,9,9,NA,NA,  1,1,3.5, 6,6,6), nrow=2, byrow=T))

temp <- quantile(fit2, 1-qq)
aeq(temp$quantile, matrix(c(6,6,9,9,NA,NA,  1,1,3.5, 6,6,6), nrow=2, byrow=T))
aeq(temp$lower,    matrix(c(6,6,6,6,9,9,    1,1,1,1, NA,NA), nrow=2, byrow=T))
aeq(temp$upper,    rep(as.numeric(NA), 12))

# Second major case set -- a survfit object where fit$surv is a matrix
#  This arises from coxph models
#  There is only 1 subject with ph.ecog=3 which is a nice edge case
cfit <- coxph(Surv(time, status) ~ age + strata(ph.ecog), lung)
sfit <- survfit(cfit, newdata=data.frame(age=c(50, 70)))
qtot <- quantile(sfit, qq)
for (i in 1:4) {
    for (j in 1:2) {
        temp <- quantile(sfit[i,j], qq)
        print(c(aeq(qtot$quantile[i,j,], temp$quantile),
                aeq(qtot$upper[i,j,], temp$upper),
                aeq(qtot$lower[i,j,], temp$lower)))
    }
}
temp <- quantile(sfit, qq, conf.int=FALSE)
all.equal(qtot$quantile, temp)