coxph_pairs: Pairwise 'coxph' comparisons

In raredd/rawr: rawr miscellaneous

Pairwise coxph comparisons

Description

Evaluate pairwise group differences in hazard ratios with coxph. Note this function is intended to be used for uni-variable models with a factor-like predictor.

Usage

coxph_pairs(
  object,
  data = NULL,
  level = 0.95,
  ...,
  relevel = FALSE,
  method = p.adjust.methods,
  digits = getOption("digits")
)

Arguments

object	an object of class survdiff, survfit, or coxph alternatively, a formula in which case data must be provided
data	(optional) a data frame in which to interpret the variables named in object if given as a formula
level	the confidence level for intervals; see confint
...	additional arguments passed to survdiff such as na.action or rho to control the test
relevel	logical; if TRUE, the reference group is flipped
method	p-value correction method (default is 'holm'; see p.adjust
digits	integer indicating the number of decimal places to be used

Value

A list with three elements:

n	the number of subjects in each pair of groups
n_model	the number of subjects in each pair of models, i.e., excluding subjects with missing data
hr	the hazard ratio between pairs of groups
lci	the lower level confidence intervals for each hazard ratio
uci	the upper level confidence intervals for each hazard ratio
p.value	Wald p-value for each hazard ratio. The lower and upper triangles of the matrix are uncorrected and adjusted, respectively, for multiple comparisons using method (the default is the Holm correction, see p.adjust)

See Also

survdiff_pairs

Examples

library('survival')
sdif <- survdiff(Surv(time, status) ~ sex, data = lung)
sfit <- survfit(Surv(time, status) ~ sex, data = lung)
cxph <- coxph(Surv(time, status) ~ sex, data = lung)

stopifnot(
  identical(coxph_pairs(sdif), coxph_pairs(sfit)),
  identical(coxph_pairs(sdif), coxph_pairs(cxph))
)

## numeric and integer variables will be treated as factor-like
sfit <- survfit(Surv(time, status) ~ extent, data = colon)
kmplot(sfit, add = TRUE)
coxph_pairs(sfit)$n

leg <- function(ii, jj, ...) {
  hr <- coxph_pairs(sfit, ...)
  sapply(seq_along(ii), function(i) {
    j <- ii[i]
    i <- jj[i]
    sprintf(
      '%s vs %s: n=%s, HR: %.2f (%.2f - %.2f), %s',
      i, j, hr$n[i, j], hr$hr[i, j], hr$lci[i, j], hr$uci[i, j],
      pvalr(hr$p.value[i, j], show.p = TRUE)
    )
  })
}
legend(
  'bottomleft', leg(c('1', '1', '1'), c('2', '3', '4')),
  lty = 1, col = 2:4, bty = 'n'
)


## compare
f <- function(x) exp(coef(x))
f(coxph(Surv(time, status) ~ extent, data = colon[colon$extent %in% c(1, 2), ]))
f(coxph(Surv(time, status) ~ extent, data = colon[colon$extent %in% c(1, 3), ]))
f(coxph(Surv(time, status) ~ extent, data = colon[colon$extent %in% c(1, 4), ]))
## etc ...


## rawr >= 1.0.0 allows multiple variables
sfit <- survfit(Surv(time, status) ~ sex + extent, data = colon)
coxph_pairs(sfit, method = 'BH')


## also allows for a formula to be passed (data arg required)
coxph_pairs(Surv(time, status) ~ sex + extent, data = colon)


## strata will be ignored
sdif <- survdiff(Surv(time, status) ~ sex + strata(inst), data = lung)
coxph_pairs(sdif)

raredd/rawr documentation built on March 4, 2024, 1:36 a.m.