Evaluate Survival Model Performance — test

Computes predictive performance metrics for stratified or unstratified Cox models. Supports Loss, C-index, Integrated Brier Score (IBS), and Time-Dependent AUC (tdAUC).

Usage

test_eval(
  test_z,
  test_delta,
  test_time,
  betahat,
  test_stratum = NULL,
  train_baseline_obj = NULL,
  criteria = c("loss", "CIndex", "IBS", "tdAUC")
)

Arguments

test_z: Matrix of predictors for the test set.
test_delta: Numeric vector of event indicators (1 for event, 0 for censored).
test_time: Numeric vector of observed times.
betahat: Numeric vector of estimated coefficients.
test_stratum: Vector indicating strata for test subjects. Defaults to NULL (single stratum).
train_baseline_obj: A list containing the baseline hazard function (typically from get_baseline_hazard). Required only when criteria = "IBS".
criteria: Metric to calculate: "loss" (Log-Partial Likelihood), "CIndex" (Concordance Index), "IBS" (Integrated Brier Score), or "tdAUC" (Integrated Time-Dependent AUC).

Value

A numeric value representing the performance metric. Returns NA if the metric cannot be computed (e.g., no events in test set).

Details

For "IBS", the function predicts survival probabilities and converts them to risk (1 - S). If riskRegression fails to provide a pre-computed IBS, the function manually integrates the Brier score using the trapezoidal rule.