PKPD

Running

The code to initiate structsearch for a PKPD model in Python/R is stated below:

PythonR

from pharmpy.modeling import read_model
from pharmpy.tools read_modelfit_results, run_structsearch

start_model = read_model('path/to/model')
start_model_results = read_modelfit_results('path/to/model')

res = run_structsearch(type='pkpd',
                        search_space="DIRECTEFFECT(*)",
                        model=start_model,
                        b_init = 0.1,
                        emax_init = 0.1,
                        ec50_init = 0.7,
                        met_init = 0.3,
                        results=start_model_results)

This will take an input model model with a search_space that includes all direct effect PKPD models.

Note

For PKPD models the input model has to be a PK model with a PKPD dataset.

Arguments

The arguments of the structsearch tool for PKPD models are listed below.

Mandatory

Argument	Description
type	Type of model. In this case “pkpd”.
model	PK start model
results	ModelfitResults of the start model

Optional

Argument	Description
b_init	Initial estimate for baseline effect. Optional. Default is 0.1
emax_init	Initial estimate for E max parameter. Default is 0.1
ec50_init	Initial estimate for EC 50 parameter. Default is 0.1
met_init	Initial estimate for mean equilibration time. Default is 0.1
search_space	Search space of models to test. Optional. If None all implemented models are used.
strictness	Strictness criteria for model selection. Default is “minimization_successful or (rounding_errors and sigdigs>= 0.1)”.

Models

Currently implemented PKPD models are:

• Direct effect models.

• Effect compartment models.

• Indirect effect models.

Structsearch workflow

PKPD candidate models will be derived from the input PK model. Additionaly to the candidate models a baseline model is created, which serves as a reference model when the models are ranked.

Note : The figure above is only showing a subset of all candidate models created, indicated by “…”

Regarding DVID, DVID=1 is representing PK observations while DVID=2 is connected to PD observations.

Search space

MFL support the following model features:

Category	Options	Description
DIRECTEFFECT	model	Direct effect PD models.
EFFECTCOMP	model	Effect comprtment PD models.
INDIRECTEFFECT	model, option	Indirect effect PD models. option can be either production or degradation.

The option model describes a PKPD model, such as E _max. For more details check model types.

To test all direct effect models the search space looks as follows:

DIRECTEFFECT(*)

Search space for testing linear and emax models for direct effect and effect compartment models:

DIRECTEFFECT([linear, emax])
EFFECTCOMP([linear, emax])

Search space for testing linear indirect effect degradation models:

INDIRECTEFFECT(linear,DEGRADATION)

Results

The results object contains various summary tables which can be accessed in the results object, as well as files in .csv/.json format. The name of the selected best model (based on the input selection criteria) is also included.

Below is an example for a PKPD run.

PythonR

res = run_structsearch(type='pkpd',
                        search_space="DIRECTEFFECT(emax);EFFECTCOMP([linear,emax])",
                        model=start_model,
                        results=start_model_results)

The summary_tool table contains information such as which feature each model candidate has, the difference to the start model (in this case comparing BIC), and final ranking:

	description	n_params	d_params	dbic	bic	rank	parent_model
model
structsearch_run2	EFFECTCOMP_LINEAR	6	3	1.215004	27.745013	1	baseline_model
structsearch_run1	DIRECT_EMAX	6	3	1.212603	27.747413	2	baseline_model
baseline_model	baseline_model	3	0	0.000000	28.960017	3	baseline_model
structsearch_run3	EFFECTCOMP_EMAX	7	4	-0.984380	29.944396	4	baseline_model

Examples

Minimum required arguments to run structsearch for PKPD models:

PythonR

from pharmpy.modeling import read_model
from pharmpy.tools read_modelfit_results, run_structsearch

start_model = read_model('path/to/model')
start_model_results = read_modelfit_results('path/to/model')

res = run_structsearch(type='pkpd',
                        model=start_model,
                        results=start_model_results)

Specifying initial parameters:

PythonR

from pharmpy.modeling import read_model
from pharmpy.tools read_modelfit_results, run_structsearch

start_model = read_model('path/to/model')
start_model_results = read_modelfit_results('path/to/model')

res = run_structsearch(type='pkpd',
                        model=start_model,
                        results=start_model_results,
                        b_init = 0.09, e_max_init = 3, ec50_init = 1.5)

Run structsearch with initial estimates for all direct effect models and all indirect effect models with production:

PythonR

from pharmpy.modeling import read_model
from pharmpy.tools read_modelfit_results, run_structsearch

start_model = read_model('path/to/model')
start_model_results = read_modelfit_results('path/to/model')

res = run_structsearch(type='pkpd',
                        model=start_model,
                        results=start_model_results,
                        b_init = 0.09, e_max_init = 3, ec50_init = 1.5,
                        search_space = "DIRECTEFFECT(*);INDIRECTEFFECT(*,PRODUCTION)")