preproc_MFPCA {MJMbamlss}

Preprocessing step to create MFPCA object

Description

This function takes the data und uses the residuals of marker-specific additive models to estimate the covariance structure for a MFPCA

Usage

preproc_MFPCA(
  data,
  uni_mean = "y ~ s(obstime) + s(x2)",
  time = "obstime",
  id = "id",
  marker = "marker",
  M = NULL,
  weights = FALSE,
  remove_obs = NULL,
  method = c("fpca", "fpca.sc", "FPCA", "PACE"),
  nbasis = 10,
  nbasis_cov = nbasis,
  bs_cov = "symm",
  npc = NULL,
  fve_uni = 0.99,
  pve_uni = 0.99,
  fit = FALSE,
  max_time = NULL,
  save_uniFPCA = FALSE,
  save_uniGAM = FALSE
)

Arguments

data

Data.frame such as returned by function simMultiJM.

uni_mean

String to crate a formula for the univariate addtive models.

time

String giving the name of the longitudinal time variable.

id

String giving the name of the identifier.

marker

String giving the name of the longitudinal marker variable.

M

Number of mFPCs to compute in the MFPCA. If not supplied, it defaults to the maximum number of computable mFPCs.

weights

TRUE if inverse sum of univariate eigenvals should be used as weights in the scalar product of the MFPCA. Defaults to FALSE (weights 1).

remove_obs

Minimal number of observations per individual and marker to be included in the FPC estimation. Defaults to NULL (all observations). Not removing observations can lead to problems if the univariate variance estimate is negative and has to be truncated, then the scores for IDs with few observations cannot be estimated.

method

Which package to use for the univariate FPCA. Either function adapted function 'fpca', 'FPCA' from package fdapace, 'fpca.sc' from package refund, or function 'PACE' from package MFPCA.

nbasis

Number of B-spline basis functions for mean estimate for methods fpca, fpca.sc, PACE. For fpca.sc, PACE also bivariate smoothing of covariance estimate.

nbasis_cov

Number of basis functions used for the bivariate smoothing of the covariance surface for method fpca.

bs_cov

Type of spline for the bivariate smoothing of the covariance surface for method fpca. Default is symmetric fast covariance smoothing proposed by Cederbaum.

npc

Number of univariate principal components to use in fpca.sc, PACE.

fve_uni

Fraction of univariate variance explained for method FPCA.

pve_uni

Proportion of univariate variance explained for methods fpca, fpca.sc, PACE.

fit

MFPCA argument to return a truncated KL fit to the data. Defaults to FALSE.

max_time

If supplied, forces the evaluation of the MFPCs up to maxtime. Only implemented for method = 'fpca'.

save_uniFPCA

TRUE to attach list of univariate FPCAs as attribute to output. Defaults to FALSE.

save_uniGAM

TRUE to attach list of univariate additive models used to calculate the residuals. Defaults to FALSE.

Value

An object of class MFPCAfit with additional attributes depending on the arguments save_uniFPCA, save_uniGAM, fit.

Examples

data(pbc_subset)
mfpca <- preproc_MFPCA(pbc_subset, uni_mean = paste0(
    "logy ~ 1 + sex + drug + s(obstime, k = 10, bs = 'ps') + ",
    "s(age, k = 10, bs = 'ps')"),
    pve_uni = 0.99, nbasis = 5, weights = TRUE, save_uniFPCA = TRUE)