transform_blq#

pharmpy.modeling.transform_blq(model, method='m4', lloq=None)[source]#

Transform for BLQ data

Transform a given model, methods available are m1, m3, m4, m5, m6 and m7 [1]. The blq information can come from the dataset, the lloq option or a combination. Both LLOQ and BLQ columns are supported. The table below explains which columns are used for the various cases:

lloq option	LLOQ column	BLQ column	Used as indicator	Used as level	Note
Available	NA	NA	DV < lloq	lloq
NA	Available	NA	DV < LLOQ	LLOQ
NA	NA	Available	BLQ	nothing	Only for M1 and M7
NA	NA	NA	NA	NA	No BLQ handling
NA	Available	Available	BLQ	LLOQ	DV column not used
Available	NA	Available	BLQ	lloq
Available	Available	NA	DV < lloq	lloq	Column overridden
Available	Available	Available	DV < lloq	lloq	Columns overridden

BLQ observations are defined as shown in the table above. If both a BLQ and an LLOQ column exist in the dataset and no lloq is specified then all dv values in rows with BLQ = 1 are counted as BLQ observations. If instead an lloq value is specified then all rows with dv values below the lloq value are counted as BLQ observations. If no lloq is specified and no BLQ column exists in the dataset then all rows with dv values below the value specified in the DV column are counted as BLQ observations.

M1 method:: All BLQ observations are discarded. This may affect the size of the dataset.
M3 method:: Including the probability that the BLQ observations are below the LLOQ as part of the maximum likelihood estimation. For more details see [1]. This method modifies the Y statement of the model (see examples below).
M4 method:: Including the probability that the BLQ observations are below the LLOQ and positive as part of the maximum likelihood estimation. For more details see [1]. This method modifies the Y statement of the model (see examples below).
M5 method:: All BLQ observations are replaced by level/2, where level = lloq if lloq is specified. Else level = value specified in LLOQ column (see table above). This method may change entries in the dataset.
M6 method:: Every BLQ observation in a consecutive series of BLQ observations is discarded except for the first one. The remaining BLQ observations are replaced by level/2, where level = lloq if lloq is specified. Else level = value specified in LLOQ column (see table above). This method may change entries in the dataset as well as the size of the dataset.
M7 method:: All BLQ observations are replaced by 0. This method may change entries in the dataset.

Current limitations of the m3 and m4 method:

Does not support covariance between epsilons
Supports additive, proportional, combined, and power error model

Parameters:

model (Model) – Pharmpy model
method (str) – Which BLQ method to use
lloq (float, optional) – LLOQ limit to use, if None Pharmpy will use the BLQ/LLOQ column in the dataset

Returns:

Model – Pharmpy model object

Examples

>>> from pharmpy.modeling import *
>>> model = load_example_model("pheno")
>>> model = transform_blq(model, method='m4', lloq=0.1)
>>> model.statements.find_assignment("Y")
    ⎧ EPS₁⋅F + F   for DV ≥ LLOQ
    ⎪
    ⎨CUMD - CUMDZ
    ⎪────────────    otherwise
Y = ⎩ 1 - CUMDZ