Dependent-censoring diagnostic — morie_dependent_censoring

Fits a Cox model with CENSORING as the outcome. Covariates that predict censoring are covariates whose censoring is informative, and the fitted coefficients say which ones and how strongly.

Usage

morie_dependent_censoring_hazard(time, event, X, ties = "efron")

Arguments

time: follow-up times.
event: 1 for an event, 0 for censored.
X: covariate matrix.
ties: tie correction.

Value

list with beta_censoring, se, p_value, beta_event, dependent, n_flagged.

Details

What no diagnostic can do is detect censoring that depends on the unobserved event time itself – the quantity being censored is by definition not there to correlate with. A null result here is therefore not evidence of independent censoring; it only rules out the dependence that runs through the measured covariates.

References

Robins, J. M. and Finkelstein, D. M. (2000). Correcting for noncompliance and dependent censoring. Biometrics, 56(3), 779-788.

Examples

set.seed(1)
X <- matrix(rnorm(120), ncol = 2)
tt <- rexp(60); ev <- rbinom(60, 1, 0.6)
morie_dependent_censoring_hazard(tt, ev, X)$n_flagged
#> [1] 0