![[Stable]](figures/lifecycle-stable.svg)
Pulls parameters, standard errors, OFVs and condition numbers out of ext
files, applies transformations. This function is useful when numeric values
are needed. rr is easier to read, however it returns characters. A wide
and long format is available via two different functions.
coef_wide(m, trans = TRUE)
coef_long(m, trans = TRUE)Arguments
- m
An nm object.
- trans
Logical (default =
TRUE). Transform parameters using comments in $THETA/$OMEGA/$SIGMA.
Value
data.frame of extracted model parameter values. coef_wide()
returns a data.frame in wide format. Vector valued objects m, will be
stacked vertically with one row per run. coef_long() returns a
data.frame in long format. Vector valued objects m, will be stacked
horizontally.
NONMEM coding conventions used by NMproject
The convention for $THETA comments used by NMproject is
value ; name ; unit ; transformation
e.g.
$THETA
0.1 ; KA ; h-1 ; LOG
The options for THETA transformations are: LOG, LOGIT, RATIO and
missing. LOG and LOGIT refer to log and logit transformed THETAs,
respectively where the parameters should be back-transformed for reporting.
RATIO refers to ratio data types, i.e. parameters that are positive and
have a meaningful zero. Most parameters like KA, CL, EMAX fall into this
category, but covariates effects which can go negative do not. RSEs are
calculated for ratio data. Missing transformations are suitable for all other
parameters, here no RSEs will be calculated, only raw SE values will be
reported.
The convention for $OMEGA is similar but without a unit item:
value ; name ; transformation
e.g.
$OMEGA
0.1 ; IIV_KA ; LOG
The options for OMEGA are either LOG or missing. LOG indicating that the
individual parameter distribution is log normally distributions and should be
reported as a CV% (and associated RSE%) rather than as the raw NONMEM
estimate.
The convention for $OMEGA is just : value ; name.
THETA transformations using trans = TRUE
The value of FINAL and RSE% (always accompanied with a % symbol in outputs)
in the returned tibble is the reported standard error (where applicable)
where \(\theta\) and \(se(\theta)\) are the NONMEM reported values of
parameters and standard errors, respectively:
LOG\(FINAL = exp(\theta), RSE = 100\sqrt(exp(se(\theta)^2) - 1)\)
RATIO\(FINAL = \theta, RSE = 100se(\theta)/\theta\)
LOGIT\(FINAL = 100/(1 + exp(-\theta)), SE = se(\theta)\)
- missing
\(FINAL = \theta, SE = se(\theta)\)
OMEGA transformations using trans = TRUE
The value of FINAL and RSE% (always accompanied with a % symbol in outputs)
in the returned tibble is the reported standard error (where applicable)
where \(\omega^2\) and \(se(\omega^2)\) are the NONMEM reported values of
parameters and standard errors, respectively
LOG\(FINAL = 100\sqrt(exp(\omega^2) - 1), RSE = 100(se(\omega^2)/\omega^2)/2\)
- missing
\(FINAL = \omega^2, SE = se(\omega^2)\)
SIGMA transformations using trans = TRUE
The value of FINAL and RSE% (always accompanied with a % symbol in outputs)
in the returned tibble is the reported standard error (where applicable)
where \(\sigma^2\) and \(se(\sigma^2)\) are the NONMEM reported values of
parameters and standard errors, respectively. All sigmas are reported as
standard deviations.
- all sigmas
\(FINAL = \sqrt(\sigma^2), RSE = 100se(\sigma^2) / \sigma^2\)