Cross-validated Z-residuals for Survreg Models — CV_Zresidual_survreg

Internal function to compute cross-validated Z-residuals for parametric survival regression models fitted with survreg.

Usage

CV_Zresidual_survreg_survival(fit.survreg, data, nfolds, foldlist, n.rep, ...)

Arguments

fit.survreg: A fitted survreg model object.
data: Optional data.frame used for cross-validation. Highly recommended if the original model was fit without specifying the data argument or if foldlist is supplied.
nfolds: Integer. Number of folds for cross-validation (\(K\) in K-fold CV).
foldlist: Optional list specifying custom fold assignments. If NULL, folds are generated internally, typically stratified by the survival response.
n.rep: Integer. Number of repeated Z-residual samples to generate per observation (Monte Carlo replications for censored observations).
...: Additional arguments passed to the residual calculation function Zresidual_survreg_survival().

Value

A numeric matrix containing the cross-validated Z-residuals (\(N \times nrep\)), where \(N\) is the total number of observations. The matrix carries the following diagnostic attributes:

Survival.Prob: Out-of-sample predicted survival probabilities.
linear.pred: Out-of-sample linear predictors (on the survreg scale).
censored.status: Event indicator (1 = event, 0 = censored).
covariates: Data frame of covariates used.
object.model.frame: The full model frame used for CV residual calculation.
type: Character string, typically "survival".

Details

This function implements the K-fold cross-validation procedure. In each fold, the parametric survival model (preserving the original distribution, e.g., Weibull) is refitted on the training data. The out-of-sample randomized Z-residuals are then calculated on the held-out test data.

Data Handling Note: If data is NULL, the internal model frame must be manually reconstructed into a standard data frame (un-packing the Surv object) before the survreg model can be successfully refitted on the training subset of each fold. Failed model fits during cross-validation result in NA residuals for the corresponding test fold.