Make predictions for time-varying covariates at specified landmark times

Make predictions for time-varying covariates at specified landmark times

Usage

# S4 method for class 'LandmarkAnalysis'
predict_longitudinal(
  x,
  landmarks,
  method,
  dynamic_covariates,
  validation_fold = 0,
  ...
)

Arguments

x

An object of class LandmarkAnalysis.

landmarks

A numeric vector of landmark times.

method

Longitudinal data analysis method used to make predictions. Either "lcmm", "lme4", "locf", or a function, which can be one of two kinds:

• A summary measure, like "locf", computed directly from the raw longitudinal data and not requiring a model to have been previously fit with fit_longitudinal. Such a function must have the arguments data, id, time, value and landmark (and optionally further arguments passed through ...), and must return a named vector (or a two-column data frame) with one summary value per individual in the risk set.

• A prediction function for a model previously fit with fit_longitudinal (as is the case for "lcmm" and "lme4"), where the first argument is the fitted model object, and which also has newdata and subject arguments.

dynamic_covariates

Vector of time-varying covariates to be modelled as the outcome of a longitudinal model.

validation_fold

If positive, cross-validation fold where model is fitted. If 0 (default), model fitting is performed in the complete dataset.

...

Additional arguments passed to the prediction function (e.g. number of classes/clusters for lcmm).

Value

An object of class LandmarkAnalysis.