Competing Risks with Relative Entropy Integration (Composite Prior, XGBoost)

This function implements the Relative Entropy (RE) framework for discrete-time competing risks models where the prior model is specified in a composite failure time formulation and both prior/local models are fit using XGBoost.

Usage

CompRiskRE_XGBoost_CP(
  prior_model,
  train_data,
  test_data,
  eta,
  xgb_params = list(objective = "multi:softprob", eval_metric = "mlogloss", num_class =
    3, max.depth = 4),
  nrounds = 100
)

Arguments

prior_model

A fitted prior XGBoost model.

train_data

A data.frame containing training data with columns: covariates, discrete follow-up time (time_y), and event indicator (event).

test_data

A data.frame containing test data with the same structure as train_data.

eta

Numeric vector of RE regularization parameters.

xgb_params

A list of XGBoost parameters. Defaults to list(objective = "multi:softprob", eval_metric = "mlogloss", num_class = 3, max.depth = 4).

nrounds

Integer, number of boosting rounds for XGBoost (default: 100).

Value

A list with components:

models

A named list of fitted XGBoost models for \(\eta = 0\) and each supplied eta.

PD

A numeric vector of predictive deviance values on the test data.

prior_PD

Predictive deviance of the prior model on the test data.

eta

The vector of eta values, including \(0\) for the baseline model.