From: "Renee Ying Hong" yinghong@pharm.usyd.edu.au
Subject: [NMusers] NONMEM dataset development 
Date: Sat, September 4, 2004 11:56 pm 

Dear all,

A few questions about developing the data set for NONMEM.

1. In NONMEM user manual V, it says that less than 50 data records are 
allowed for each individual. Does it mean that maximum 50 data records, 
which include dose records and observation records, are allowed for each 
individual? How to change this limit?

2. Is it correct that AMT and DV observed at the time when dose is given 
cannot be recorded in the same data record?

3. Some covariables are not available for some patients. Is it correct that 
dot or 0 should be used to indicate the missing values?

3. Some patients have the study drug for more than one occasion. NONMEM is 
employed to work out the BOV. Therefore, data item OCC is added in the data 
set to indicate the sequence of the occasion. Apart from that, what else 
should be included in the data set? Does the ID#  for each occasion of the 
same patient keep the same?

Many thanks! Ying

Ying Hong
Faculty of Pharmacy
University of Sydney
Tel: 61 2 9036 5025
Fax: 61 2 9351 4391
E-mail: yinghong@pharm.usyd.edu.au
_______________________________________________________

From: "Nick Holford" n.holford@auckland.ac.nz
Subject: RE: [NMusers] NONMEM dataset development 
Date: Mon, September 6, 2004 4:00 am

Renee,

> 
> Dear all,
> 
> A few questions about developing the data set for NONMEM.
> 
> 1. In NONMEM user manual V, it says that less than 50 data records are
> allowed for each individual. Does it mean that maximum 50 data records,
> which include dose records and observation records, are allowed for each
> individual? How to change this limit?

My understanding of this limit is that it refers to the number of observation
records (not all records for an individual).
You can increase the number of observations per subject by changing the PARAMETER NO
in NSIZES and TSIZES then recompiling NONMEM.

e.g. to allow 500 obs/subject

C NO: MAX NO. OF OBSERVATION RECORDS / INDIVIDUAL RECORD
      PARAMETER (NO=500)

You may also want to change these parameters in NSIZES as well. This will increase
other array sizes that depend on NO.

LADD        250500
LIM1        50000
LIM2        50000


> 2. Is it correct that AMT and DV observed at the time when dose is given
> cannot be recorded in the same data record?

If you are using PREDPP e.g. ADVAN1 you must have AMT and DV values on different
records. There are two ways that having AMT and DV on the same record can be coded. 
Case 1: If you have a DV value on the same record as an AMT value and do not define
a MDV data item then NM-TRAN will add MDV=1 for the AMT record and the DV value will
be ignored. NONMEM will run but NM-TRAN will give you q warning like this:
 (DATA WARNING  5) RECORD     1, DATA ITEM  4: 11.9
 THE DV DATA ITEM IS POSITIVE, BUT THE MDV DATA ITEM IS 1
Case 2: If you define an MDV=0 item for a record with AMT>0 then NONMEM considers
this a data error.

> 3. Some covariables are not available for some patients. Is it correct that
> dot or 0 should be used to indicate the missing values?

You may use any value you want to indicate a missing covariate value. NONMEM treats
a dot and 0 as the same thing. It is probably a good idea to use a negative value
(e.g. -99) to signal a missing value. This makes it clearer to a human reader of the
data file that this is a strange value. You must of course write your own code to
handle the case of a missing covariate. NONMEM does not do anything sensible by
default with missing covariates.

> 3. Some patients have the study drug for more than one occasion. NONMEM is
> employed to work out the BOV. Therefore, data item OCC is added in the data
> set to indicate the sequence of the occasion. Apart from that, what else
> should be included in the data set? Does the ID#  for each occasion of the
> same patient keep the same?

Estimation of BSV and BOV is a good idea. It requires that you have a data item
(e.g. called OCC) to signal different occasions. An occasion can be defined in any
way you like but often each dose is considered a different occasion. The ID data
item should be the same for any given subject who has one or more different
occasions of data. I think of the ID data item as the covariate for BSV while the
OCC data item is the covariate for BOV.
 
Nick

--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand
email:n.holford@auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/
_______________________________________________________

From: "Renee Ying Hong" yinghong@pharm.usyd.edu.au
Subject: RE: [NMusers] NONMEM dataset development 
Date: Tue, September 7, 2004 10:49 pm

Hi, Nick,

Many thanks for your detailed reply of those questions!!!

I am still not sure about coding the missing covariate. My understanding of 
your third reply is that there is no extra code if I signal the dot or zero 
for missing value. However, if -99 is to signal the missing covariate 
value, I have to write the code, which will tell NONMEM to treat the 
covariate with -99 as missing value. What code should I put in the NM-TRAN?

Kind Regards, Renee

_______________________________________________________

From: "Nick Holford" n.holford@auckland.ac.nz
Subject: RE: [NMusers] NONMEM dataset development 
Date: Tue, September 7, 2004 11:57 pm

Renee,

It does not matter what value (dot, 0, -99, 42) is used to signal a missing
covariate. If you need to use that covariate in your model then you must supply code
to impute the missing value in order for NONMEM to estimate the covariate effect
sensibly.

e.g. if AGE is missing then:

IF (AGE.LE.0) THEN AGE=60 ; 60 might be the median age in your dataset

More complex methods can be used to impute missing values e.g. you could use other
known covariates to make a better guess of the missing covariate value. 
Try searching on Google for 'nonmem missing covariate imputation' and check out some
of the links e.g.
http://www.cognigencorp.com/nonmem/nm/99sep112000.html

Nick
_______________________________________________________