Multi-Source Integration for KL-Integrated Cox Elastic-Net Models

Fits multiple KL-integrated Cox elastic-net models on the full data using multiple external sources, and combines the fitted coefficient vectors across sources to produce a single aggregated estimate.

Usage

coxkl_enet.multi(
  z,
  delta,
  time,
  stratum = NULL,
  beta_list = NULL,
  RS_list = NULL,
  etas,
  combine = c("mean", "median"),
  message = FALSE,
  seed = NULL,
  ...
)

Arguments

z: Matrix/data.frame of predictors of dimension n x p.
delta: Event indicator vector.
time: Survival time vector.
stratum: Optional stratum indicator vector for stratified Cox models.
beta_list: A list of external coefficient vectors. Each element must have length p. If provided, RS_list should be NULL.
RS_list: Optional list of external risk score vectors/matrices. Each element should be conformable with n. If provided, beta_list is ignored.
etas: Vector of eta values for transfer-learning shrinkage.
combine: How to combine coefficients across sources. Either "mean" (default) or "median".
message: Logical indicating whether to print progress.
seed: Optional seed for reproducibility (passed to each CV run with an offset).
...: Additional arguments passed to cv.coxkl_enet() (e.g., alpha, lambda, nlambda, lambda.min.ratio, nfolds, cv.criteria, c_index_stratum, etc.).

Value

An object of class "coxkl_enet.multi", which is a list containing:

best_beta — combined coefficient estimate across sources.
all_betas — matrix of dimension p x K_valid of coefficient vectors from each successful fit.
K — total number of external sources provided.
valid_sources — number of successful (non-error) fits used in aggregation.
combine — combination rule used.
seed — seed used (if any).

Details

Unlike coxkl_enet_bagging(), this function does not bootstrap the data. Instead, it runs cv.coxkl_enet() once per external source on the full dataset. The resulting coefficient vectors are then aggregated (by default, averaged) to obtain a combined estimate.