Generate one Monte Carlo replicate of the AFT-spline-SIMEX scenario

Produces a main-study survival sample and an external validation sample from the same data-generating process described in the manuscript appendix. The latent log-exposure is normal, the dose-response is the saturating curve f_true(), the survival times follow a log-normal AFT model with four confounders (V1-V4), and the surrogate measurements are multivariate-normal around the latent truth with covariance Sigma_u.

Usage

generate_aft_data(
  n = 2000,
  n_val = 500,
  Sigma_u = build_sigma_u(),
  x_mean = 10,
  x_sd = sqrt(5),
  f_args = list(),
  mu = 4,
  sigma_T = 0.8,
  gamma = c(0.1, -0.05, 0.3, -0.2),
  cens_rate = 5e-04,
  seed = NULL
)

Arguments

n: Main-study sample size.
n_val: Validation sample size.
Sigma_u: J x J measurement-error covariance (default 3x3 with marginal variances (2.0, 2.5, 3.0) and rho = 0.4). Use sigma_u_for_reliability() to target a given reliability.
x_mean, x_sd: Mean and standard deviation of the latent exposure X_true (shared by the main and validation samples). Defaults 10 and sqrt(5) reproduce the original arbitrary-unit scale; set e.g. x_mean = 300, x_sd = 75 for a daily-minutes scale.
f_args: Named list of extra arguments forwarded to f_true() for the dose-response (e.g. list(x_min = 150, alpha = 0.01) on a minutes scale). Default list() uses the f_true() defaults.
mu: AFT intercept.
sigma_T: AFT log-scale standard deviation.
gamma: Numeric vector of length 4: confounder coefficients in the linear predictor (V1, V2, V3, V4).
cens_rate: Rate of the exponential censoring distribution. Smaller values give heavier event experience. The default 0.0005 yields ~76% events under the other defaults.
seed: Optional integer seed.

Value

A list with elements survival (main-study data frame), validation (validation data frame with paired truth + surrogates), and truth (a list of the parameter values used).

Details

The returned survival data frame is in the shape expected by fit_aft_spline() and two_stage_bootstrap() with their defaults (outcome_var = "T_obs", status_var = "delta", covariates = c("V1", "V2", "V3", "V4"), surrogate_cols = NULL -> auto-detect via extract_surrogate_cols()).

Examples

sim <- generate_aft_data(n = 200, n_val = 100, seed = 1)
names(sim)
#> [1] "survival"   "validation" "truth"     
head(sim$survival)
#>        T_obs delta    X_true        W1        W2        W3       V1       V2 V3
#> 1  156.71416     1  8.599207  9.042557  9.435666  8.025849 32.04701 35.37220  1
#> 2  110.70489     1 10.410639  9.409792 10.765985 10.894197 38.44437 39.47827  0
#> 3  333.80655     0  8.131478  8.497433  9.839325  8.418137 37.93294 26.98501  0
#> 4  443.05466     1 13.567156 14.752046 15.172707 13.659899 28.34546 28.04566  1
#> 5   73.07738     1 10.736802  9.950293 10.977064  8.352338 18.57382 27.91889  0
#> 6 1150.60870     1  8.165377  8.262311  9.389844  6.764756 42.48831 28.12171  0
#>   V4
#> 1  1
#> 2  1
#> 3  0
#> 4  1
#> 5  1
#> 6  0