bayes_update {tci}R Documentation

Update PK-PD model parameters using observed data values

Description

Function will update parameters of 'pkmod' object based on available data using a Laplace approximation to the posterior. Parameters and "Omega" values from 'pkmod' are used as prior point estimates and variance terms, respectively. Parameters fixed within the Omega matrix are assumed to be fixed and are not updated.

Usage

bayes_update(lpars, pkmod, inf, tms, obs, update_init = FALSE, ...)

Arguments

lpars

Logged parameter values. Can be a subset of the full set of PK or PK-PD parameter values.

pkmod

'pkmod' object. Mean values are a subset of log(pars_pk), log(pars_pd), log(sigma_add), log(sigma_mult). PK-PD parameter values not specified in 'lpars' will be inferred from 'pkmod'.

inf

Infusion schedule

tms

Times associated with observations

obs

Observed values (concentrations or PD response values)

update_init

Logical. Should initial values be updated in addition to parameter values?

...

Arguments passed to update.pkmod

Value

'pkmod' object with PK/PK-PD/sigma parameters updated based on minimizing negative logged posterior.

Examples

# evaluate negative log posterior at a new set of parameters
lpars = log(c(cl=11,q2=3,q3=25,v=20,v2=40,v3=80,ke0=1.15,sigma_add=0.3))
prior_vcov <- matrix(diag(c(0.265,0.346,0.209,0.610,0.565,0.597,0.702,0.463)),
8,8, dimnames = list(NULL,names(lpars)))
my_mod <- pkmod(pars_pk = c(cl = 10, q2 = 2, q3 =20, v = 15, v2 = 30, v3 = 50, ke0 = 1.2),
sigma_add = 0.2, log_response = TRUE, Omega = prior_vcov)
inf <- inf_manual(inf_tms = 0, inf_rate = 80, duration = 2)
tms <- c(1,2,4,8,12)
obs <- simulate(my_mod, inf = inf, tms = tms)
bayes_update(lpars, my_mod, inf, tms, obs)

# evaluate log-prior for subset of parameters (remove volume parameters)
lpars_sub = log(c(cl=11,q2=3,q3=25,ke0=1.15,sigma_add=0.15))
my_mod <- update(my_mod, Omega = matrix(diag(c(0.265,0.346,0.209,0.702,0.463)),5,5,
 dimnames = list(NULL,names(lpars_sub))))
bayes_update(lpars_sub, my_mod, inf, tms, obs)

# add a pd response and replace multiplicative error with additive error
my_mod_pd <- update(my_mod, pars_pd = c(c50 = 2.8, gamma = 1.47, e0 = 93, emx = 93),
pdfn = emax, pdinv = emax_inv, ecmpt = 4, sigma_mult = 0, sigma_add = 4)
# simulate observations
obs_pd <- simulate(my_mod_pd, inf = inf, tms = seq(0,12,0.5))
# evaluate likelihood at new parameters
lpars_pd_eval = log(c(cl=11,q3=25,v=15,ke0=1.15,sigma_add=4,c50=5,gamma=1))
prior_vcov_pd <- matrix(diag(c(0.265,0.209,0.610,0.702,0.230,0.242,0.1)),7,7,
dimnames = list(NULL,names(lpars_pd_eval)))
my_mod_pd <- update(my_mod_pd, Omega = prior_vcov_pd)
bayes_update(lpars_pd_eval, my_mod_pd, inf, tms, obs_pd)

[Package tci version 0.2.0 Index]