Skip to contents

Make Fake Data

simulate_pdd_data.Rd

Prepares a minimal example of data in the right format for analysis via the demcrit package functions. It is compulsory to specify FAQ summary data (mean, variance and covariance, see below) and at least one additional global cognitive measure summary data.

Usage

simulate_pdd_data(N = 203, defaults = TRUE, Mu, Sigma, crits, cens = NULL)

Arguments

N

A numeric indicating number of simulated patients. Defaults to 203.

defaults

A logical indicator whether default values derived from the observed data should be used (TRUE, default) or the user will provide their values instead (FALSE).

Mu

Named numeric vector of means. Needs to contain one component denoting FAQ. If defaults = TRUE, it is set to values from observed data.

Sigma

Named matrix of variances and covariances. If defaults = TRUE, it is set at empirical values from the observed data.

crits

A data.frame with named rows and four columns:

  • rownames indicate algorithm labels,

  • IADL indicate variable from the data used for IADL deficit calculation,

  • IADL_thres indicate threshold above which there is IADL deficit,

  • cognition indicate variable from the data used for cognitive deficit calculation,

  • cognition_thres indicate threshold below which there is cognitive deficit.

cens

A matrix or data frame with two columns, the first indicating lower bound, the second the upper bound of a scale in the rowname. If defaults == TRUE, or param = NULL which does not censor data (and prints latent scores instead).

Value

A tibble containing:

id

Character; identificator of a simulated patient.

type

Character; algorithm names specified by rownames of crits

PDD

Logical; PDD diagnosis for each patient (id) calculated via each algorithm (type),

raw scores

Numeric; Scores showing raw simulated data for each variable supplied to the data-generating proces. Defaults to columns FAQ, MMSE, MoCA, sMoCA and FAQ9.

Details

The function simulates N synthetic patients' PDD status via the following steps:

  1. generate latent FAQ and cognitive scores via MASS::mvrnorm(),

  2. generate observed FAQ item 9 data via the rbinom() following tau-equivalence assumption applied to the FAQ questionnaire,

  3. censor the scores according to cens and round to nearest integer to obtain observed data

  4. apply criteria for PDD from crits to get per algorithm/patient pair probable PDD.

See also

prepare_defaults() for correct format of simulation parameters specification.

Examples

if (FALSE) { # \dontrun{
# Simulate using defaults:
sim1 <- simulate_pdd_data()
sim2 <- simulate_pdd_data(defaults = TRUE) # the same data-generating process as sim1

# Generate using two cognitive measures not correlated with FAQ:
mu <- c(FAQ = 4.05, MMSE = 26.69, MoCA = 24.07)
sd <- c(FAQ = 4.89, MMSE = 2.22, MoCA = 3.48)
corrs <- matrix(
   c(1, 0, 0, 0, 1, 0.63, 0, 0.63, 1), nrow = 3,
   dimnames = lapply(seq_len(2), \(i) c("FAQ", "MMSE", "MoCA"))
)
cens <- matrix(
  c(rep(0, 3), rep(30, 3)), nrow = 3,
  dimnames = list(c("FAQ", "MMSE", "MoCA"))
)
sigma <- MBESS::cor2cov(corrs, sd)
crits <- data.frame(
  row.names = c("A1", "A2", "A3", "A4"),
  IADL = rep(c("FAQ", "FAQ9"), 2),
  IADL_thres = rep(c(7, 1), 2),
  cognition = c(rep("MMSE", 2), rep("MoCA", 2)),
  cognition_thres = c(rep(26, 2), rep(26, 2))
)
sim3 <- simulate_pdd_data(
  N = 2000,
  defaults = FALSE, # It is crucial to set this to FALSE
  Mu = mu, Sigma = sigma
)
} # }