Kaplan-Meier Estimate of the Censoring Distribution

Fits the "reverse" Kaplan-Meier estimator \(G(t) = P(C > t)\), treating censoring as the event, for inverse-probability-of-censoring weights. Estimate it on the data a model was trained on, and apply it to the data being scored.

Usage

censoring_km(truth)

# S3 method for class 'censoring_km'
predict(object, time, left = FALSE, trunc = 0, ...)

Arguments

truth

A right-censored [survival::Surv()] object with one element per subject, such as the `.truth` column of [surv_subject_truth()].

object

A `censoring_km` object.

time

Numeric vector of times.

left

If `TRUE`, the left limit \(G(t^-)\), i.e. censoring strictly before `time`.

trunc

Lower bound applied to \(G\), so weights stay finite.

...

Not used.

Value

An object of class `censoring_km`, with `time` and `surv`; use `predict(object, time)` for \(G(t)\).

See also

[graf_weights()]

Examples

if (requireNamespace("survival", quietly = TRUE)) {
  cens <- censoring_km(survival::Surv(c(2, 4, 6, 8), c(0, 1, 0, 1)))
  predict(cens, c(1, 2, 5, 7))
}
#> [1] 1.000 0.750 0.750 0.375