Time-varying covariates

Time-varing covariates, such as creatinine values can be implemented easily as well. They just require the additional times argument:

covariates <- list(
  "WT" = new_covariate(value = 70),
  "CR" = new_covariate(
    value = c(0.8, 1, 1.2),
    times = c(0, 48, 72)
  )
)

By default, PKPDsim assumes that you want to interpolate (linearly) between measurements of the time-varying covariates. If you prefer to implement the covariate using last-observation-carried-forward (in other words a step function), specify the method = "LOCF" argument to new_covariate().

Covariates for multiple patients

A table of covariates can be supplied to sim() with covariate values per individual. It can handle both static and time-varying covariates. A covariate table could look like this:

cov_table <- data.frame(
  id  = c(1, 1, 2, 3),
  WT  = c(40, 45, 50, 60),
  SCR = c(50, 150, 90, 110),
  t   = c(0, 5, 0, 0)
)

The id and t (time) columns can be omitted when only static covariates are to be used. Again, make sure that the headers used for the covariates match exactly with the covariate names specified in the model definition.

A full example for a model with (simulated) covariates for a patient population:

parameters <- list(
  CL = 1,
  V = 10,
  KA = 0.5
)
n_ind <- 50
cov_table <- data.frame(
  'id' = 1:n_ind,
  'WT' = rnorm(n_ind, mean = 70, sd = 5)
)

model <- new_ode_model(
  code = '
     CLi = CL * pow((WT/70), 0.75)
     Vi  = V * (WT/70)
     dAdt[1] = -KA*A[1]
     dAdt[2] =  KA*A[1] -(CLi/Vi)*A[2]
   ',
   declare_variables = c('CLi', 'Vi'),
   covariates = c('WT'),
   dose = list(cmt = 1),
   obs = list(cmt = 2, scale = 'V * (WT/70)')
)

regimen <- new_regimen(
  amt  = 30,
  n = 4,
  type = 'bolus',
  interval = 12
)

dat <- sim(
  ode = model,
  parameters = parameters,
  t_obs = c(0.5, 2, 4, 8, 12, 16, 24),
  n_ind = n_ind,
  regimen = regimen,
  covariates_table = cov_table,
  output_include = list(covariates = TRUE)
)

## Simulating 50 individuals.

Note: PKPDsim does not handle missing covariate values. If you do have missing covariate data, probably the best approach would be to impute the values manually before simulation, e.g. based on the mean observed / known value, or the correlation between the covariates.