Covariates

Covariates can be simple or interpolated. See the “Modeling Project Files” section for more information.


  covariate(covariateName)

This specifies that there is a covariate with the given name. For time-varying covariates, the covariate statement extrapolates backward. So, for example, if a covariate is given at time=1, 2, and 3 to be 10, 20, and 30, respectively, then the covariate value in [0,1] is 10, in [1,2] is 20, and in [2,3] is 30. If no covariate value is given at time=0, the covariate value for [0,1] is also 10, as the first value propagates forward as well as backward.


  fcovariate(covariateName)

This is the same as covariate, except that it also specifies the covariate has forward direction. fcovariate extrapolates forward. So, for example, if a covariate is given at time=1, 2, and 3 to be 10, 20, and 30, respectively, the fcovariate value for [1,2] is 10, for [2,3] is 20, and for times beyond 3 (if any) it is 30. If no covariate value is given at time=0, the fcovariate value for [0,1] is also 10, as the first value propagates backward as well as forward.

There are two ways to indicate forward direction, by using the fcovariate statement in the model text, or by using fcovr in the column definition text (or both).


  interpolate(covariateName)

This also specifies that there is a covariate with the given name, however, the value of the covariate varies linearly between time points at which it is set in time-based models. If no covariate value is given, the latest value is carried forward. If no value is given at time=0, the first available value is used.

This feature should be used with caution. One reason is that, in some cases, it makes a linear model nonlinear so it cannot use the matrix exponent ODE solver. This can happen in a simple PK model parameterized by Cl and V, if V is a function of bodyweight BW, and BW is interpolated. Alternatively, if the model is parameterized by Ke and V, it is not affected because V does not enter the differential equations. A second reason is that it can give incorrect results if such a covariate effect is present in a closed-form model.


  covariate(Form())

In a text model, if a covariate is categorical, its name must be followed by empty parentheses. This informs the UI that the covariate is categorical, and thus available for stratification.