Jump to content


Photo
- - - - -

Q&A from Lesson 1: Nonlinear Clearance

pml

  • Please log in to reply
No replies to this topic

#1 bwendt@certara.com

bwendt@certara.com

    Advanced Member

  • Administrators
  • 282 posts

Posted 27 October 2016 - 02:43 PM

Question: How do you obtain intial estimates for error terms­
Answer: With the additive error model, assuming the residuals have a standard normal distribution with mean 0 and standard deviation of 1., CEps=1 is a good first estimate. For the multiplicative error model, a suggested intial estimate is 0.2, corresponding to 20% CV.
Question: ­Can you explain a bit more on error(CEps=1)?­
Answer: CEps is the standard deviation of the residual error, sometimes referred to as epsilon. With the additive error model, CEps=1 is a good first estimate assuming the residuals have a standard normal distribution with mean 0 and standard deviation of 1.
Question: ­Simple question, why do you choose 3-compartment model? Would you start with 1- and 2-compartment models first?­
Answer: You are absolutely correct, the best approach is to start with a simple model and work toward a more complicated model. The 1- and 2- compartment models are readily available via the Built-In modeling options. If you find that one of the Built-In models is sufficient to describe the system that you are studying, great! However, the focus of this series is on models that require customization via the PML code; in the case of PK24, the customization is due to concentration-dependent clearance.
Question: ­You are showing these models in Phoenix 6.4. Will they carry forward into Phoenix 7.0?­
Answer: Yes, the PML models will carry forward into Phoenix 7.0.
Question: ­What would you get without using Min_Cl?­
Answer: The primary model parameters are V, V2, V3, CL, CL2, and CL3. Min_CL is a secondary parameter, that is derived from the primary parameters, and it does not impact the model fit. 
Question: ­How come the CL parameter estimate is missing for the dependent CL model?­
Answer: For the model Cl=Cl0 – a*C, Cl itself is no longer a parameter – a and Cl0 are. They were included on the thetas output.
Question: ­How 1000 was selected in the linear equation­
Answer: Cmax ~ 1000 from the observed data. If clearance decreases linearly with increasing concentration, at Cmax the Clearance will be at its minimum.
Question: ­If the model has biphasic clearance ... both linear and non-linear. I have been trying an antibody PK model which has biphasic clearance­
Answer: If you start with a nonlinear Cl (Hill eqn) model, you can easily add a parallel linear Cl route via the graphical editor. Note that TMDD models will be covered on Dec. 15th.
Question: ­ What is the minimum data point requirement for a model like this (3-compartment, concentration-dependent clearance)?­
Answer: This depends on number of parameters – at a minimum, nobs = nparm +1. Nobs – nparm is referred to as degrees of freedom (df), which must be greater than or equal to 1. For a 6-parameter model, we need at least 7 datapoints. Try in general to have at least 4 or more df for the error term if possible.
Question: ­Is the 4th edition of Gabrielsson and Weiner be used­ for PML?
Answer: No, the 4th edition presents examples in WNL5 ASCII code. The 5th edition is encouraged because it ships with the completed Phoenix projects for all of the examples in PML code.
Question: ­Can you expand a bit on secondary parameters
Answer: These are custom parameters derived from the primary model parameters. In this example, the secondary parameter "Min_Cl" is defined using the statement:
secondary(Min_Cl=Cl0 - a * 1000)
Where Cl is a fitted parameter that relates to Cl0 by a linear model: Cl = Cl0 - a * C. A suite of secondary paramerters can be automatically generated for built in PK models, but not for non-PK or user model. There you must define your own.
Question: ­You might want to point out that there is no stparm statement for a­
Answer: Structural parameters make more sense in population models, if they are typical values that get adjusted by individual etas. For individual modeling, it is possible to only use fixed effects. This question has been extensively discussed in the forum:
https://support.cert...near-clearance/
Question: ­In the structural parameters, we only have CL, but in the fixed effects, we have a, Cl0 and CL, why?­
Answer: In the non-static clearance model, Cl is defined as a function of the fixed effects Cl0. It is not a parameter in itself – it is derived from other parameters.
Question: ­How to select initial value from NCA, for 3 cmt­
Answer: Please see pages 591-594 in the G&W text, 5th ed.
Question: ­How does one can select V1=?­
Answer: V1 is the volume of the central compartment. It is created in PML automatically by using the Built-in options on the Structural tab using Clearance parameterization. The initial estimate for V1 can be set in one of several ways:
1) Initial Estimates tab: here you can enter initial estimates for all parameters, and see an overlay of the predicted concentrations from those estimates compared to the observed concentrations.
2) Parameters tab, Fixed Effects sub-tab: here you can enter Initial Estimates, lower and upper boundaries for all parameters.
3) Text model: enter initial estimate values in the Fixed Effects statements. For example:
Fixef(V1 = c(0, 100, 1000))
This defines V1 as a fixed effect with initial estimate of 100, lower bound = 0, upper bound = 1000


Edited by bwendt@certara.com, 28 October 2016 - 03:14 PM.






Also tagged with one or more of these keywords: pml

0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users