RE: Prediction with parameter estimates on new data
Dear Lib,
I don't think I fully understand what you are going for - do you want to use
the final parameter estimates from your model, in tandem with early timepoint
data from individuals to predict the later effects for that specific
individual? If that is the case you could simply change the EVID for all later
time points to EVID = 2 and use MAXEVAL = 0.
This would use your initial estimates as final estimates for the model and
allow you to obtain EBE's and IPRED values for individuals at all time points,
though you'd only be using early time point data for obtaining the EBEs that
help drive the rest of the predicted values.
Perhaps expanding a little more on what you are trying to accomplish can help
us lead you in the right direction more easily.
Quoted reply history
From: [email protected] [mailto:[email protected]] On
Behalf Of Lib Gray
Sent: Wednesday, March 19, 2014 11:33 AM
To: Mats Karlsson
Cc: [email protected]
Subject: Re: [NMusers] Prediction with parameter estimates on new data
That is good to know, but in this case I want to use everything for parameter
estimation, and then get predictions using those parameter estimates on a
subset of the data. Is there a way to use EVID data item for that?
On Wed, Mar 19, 2014 at 10:29 AM, Mats Karlsson
<[email protected]<mailto:[email protected]>> wrote:
Dear Lib,
If you use the EVID data item (0 for observations 1 for doses) in your data
set, you can denoted the observations for which you want to make a prediction
by EVID=2 (other type event). That way they will not contribute to parameter
estimation, but you will get predictions (both population and individual).
Best regards,
Mats
Mats Karlsson, PhD
Professor of Pharmacometrics
Dept of Pharmaceutical Biosciences
Faculty of Pharmacy
Uppsala University
Box 591
75124 Uppsala
Phone: +46 18 4714105<tel:%2B46%2018%204714105>
Fax + 46 18 4714003<tel:%2B%2046%2018%204714003>
http://www.farmbio.uu.se/research/researchgroups/pharmacometrics/
From: [email protected]<mailto:[email protected]>
[mailto:[email protected]<mailto:[email protected]>] On
Behalf Of Lib Gray
Sent: 19 March 2014 15:36
To: [email protected]<mailto:[email protected]>
Subject: [NMusers] Prediction with parameter estimates on new data
Hello NMusers,
I am new to NONMEM, and I am unsure how to use parameter estimates for
prediction with new data. More specifically, I have a K/PD nonlinear mixed
effects model developed in NONMEM, and I want to check its predictive power in
a specific way. The data I've been using has about 500 patients, each with
multiple dosing (AMT) and effect (DV) measurements. Most patients have many
measurements, spanning several months.
What I want to do is use the population estimates (and potentially also the
between-subject variability estimates) from fitting the model to my entire data
set to generate the predicted effects using only early data (for example, the
first 3 months). I have the population estimates from fitting the model to all
data, and I want to generate the model predicted effects using only early
effect data (though with all dosing data), so that I can asses the model
predicted effects for later time points.
I am new to NONMEM, and am unsure if there is a way with NONMEM or PsN of using
previous estimates on new data, in this specific way. I am looking into using
R's deSolve package, since my model contains two differential equations, but I
wanted to know if there was a more direct way.
Thanks for any help,
Lib Gray