Fits a survival model pooled across several landmark times (superlandmarking)

Usage

fit_superlandmark(
  x,
  formula,
  landmarks,
  w,
  dynamic_covariates = c(),
  include_clusters = FALSE,
  validation_fold = 0
)

Arguments

x: An object of class LandmarkAnalysis.
formula: A formula to be used in survival submodel fitting. Must reference the LM column directly to model landmark-varying covariate effects (e.g. via an interaction with a function of LM) or a landmark-varying baseline hazard (via strata(LM)).
landmarks: Numeric vector of at least two landmark times, for which risk sets must already have been computed via compute_risk_sets.
w: Scalar prediction window. Each landmark time \(s\) in landmarks uses the (fixed) horizon \(s+w\).
dynamic_covariates: Vector of time-varying covariates to be used in the survival model.
include_clusters: Boolean indicating whether to propagate cluster membership to survival analysis.
validation_fold: If positive, cross-validation fold where model is fitted. If 0 (default), model fitting is performed on the complete dataset.

Value

An object of class LandmarkAnalysis.

Details

This function implements "superlandmarking": a single Cox model is fitted to a "superdataset" subjects are present as many times as landmarks they are at risk. The landmark time is available to formula as the LM column, so that covariate effects and/or the baseline hazard can be allowed to vary with landmark time (via an interaction term with covariates f(LM), or strata(LM)).

Because each subject contributes one row per landmark at which they are still at risk, a cluster(@ids) term is added to formula automatically (unless already present).

Examples

data(epileptic)
epileptic_dfs <- split_wide_df(
  epileptic,
  ids = "id", times = "time",
  static = c("with.time", "with.status", "treat", "age", "gender", "learn.dis"),
  dynamic = c("dose"),
  measurement_name = "value"
)
x <- LandmarkAnalysis(
  data_static = epileptic_dfs$df_static,
  data_dynamic = epileptic_dfs$df_dynamic,
  event_indicator = "with.status",
  ids = "id", event_time = "with.time",
  times = "time", measurements = "value"
) |>
  compute_risk_sets(c(90, 180, 365.25)) |>
  fit_superlandmark(
    formula = survival::Surv(event_time, event_status) ~
      treat * splines::ns(LM, 2) + age + strata(LM),
    landmarks = c(90, 180, 365.25),
    w = 2 * 365.25
  )