Skip to contents

Fit an AFT model with a natural-cubic-spline exposure

Source: R/fit.R
fit_aft_spline.Rd

Thin wrapper around survival::survreg() that builds a formula of the form Surv(outcome_var, status_var) ~ ns(x_var, df = ...) + covariates from the column names supplied. Used three ways in the simulation: as the oracle (on the true exposure), as the naive estimator (on the GLS-combined surrogate without ME correction), and as the inner workhorse of simex_aft_spline().

Usage

fit_aft_spline(
  data,
  x_var,
  covariates = character(0),
  df = 4,
  knots = NULL,
  outcome_var = "T_obs",
  status_var = "delta",
  dist = "lognormal",
  control = NULL
)

Arguments

data

A data frame containing the outcome, status, exposure, and covariate columns named below.

x_var

Name of the exposure column.

covariates

Character vector of additional covariate column names. Pass character(0) (the default) for an exposure-only model.

df

Spline degrees of freedom (ignored if knots is supplied).

knots

Optional numeric vector of interior knot locations.

outcome_var

Name of the time-to-event column. Default "T_obs".

status_var

Name of the event indicator column. Default "delta".

dist

Parametric AFT error distribution, passed directly to the dist argument of survival::survreg(). Any distribution survreg accepts is valid here, including "weibull", "exponential", "loglogistic", "lognormal", and "gaussian"; see survival::survreg() for the full, authoritative list. The default "lognormal" matches the DGP in generate_aft_data(). Select another family to fit your data under, or to run a sensitivity analysis against, an alternative AFT distribution.

control

Optional survival::survreg.control() object.

Value

A survreg fitted-model object.

Examples

sim <- generate_aft_data(n = 200, n_val = 100, seed = 1)
fit <- fit_aft_spline(sim$survival, x_var = "X_true",
                      covariates = c("V1","V2","V3","V4"))
coef(fit)
#>         (Intercept) ns(X_true, df = 4)1 ns(X_true, df = 4)2 ns(X_true, df = 4)3 
#>          2.95825639          1.14551074          0.79396526          1.45436664 
#> ns(X_true, df = 4)4                  V1                  V2                  V3 
#>          0.38171431          0.11330179         -0.05156747          0.50880591 
#>                  V4 
#>         -0.11017939