RE: [NMusers] PRED for BLQ-like observations

From: Pavel Belo <nonmem_at_optonline.net>
Date: Fri, 20 Nov 2015 14:59:51 -0500 (EST)

Thank you Bill,
 
In my case it exactly doubles the number of records...  The records ar=
e
daily measures and the code is running slow enough.  I'll split the co=
de
into estimation part and one that that is redundant, but uses a larger
file and creates an output.  It will be something like
 
$EST MAXEVALS=9999 SIG=3 NOABORT PRINT=1 SORT CONSTRAIN=5
  METHOD=SAEM NBURN=0 NITER=0 POSTHOC INTERACTION
  LAPLACIAN GRD=TG(1-7):TS(8-9) CTYPE=3 CINTERVAL=10 
 
I guess the best future way is modify something in NONMEM so there is an
option to provide only PRED in the PRED column (version 7.4?).=

 
Thanks!
Pavel

 
 
 
On Fri, Nov 20, 2015 at 01:06 PM, Denney, William S. wrote:
 
 




Hi Pavel,

 

The easiest way that I know is to generate your data file with one set
of rows for estimation with M3 and another row just above or below with
MDV=1.  NONMEM
  will then provide PRED and IPRED in the rows with MDV=1.

 

Thanks,

 

Bill

 

From: owner-nmusers_at_globomaxnm.com [mailto:owner-nmusers_at_globomaxnm.com]
On Behalf Of Pavel Belo

Sent: Friday, November 20, 2015 11:47 AM

To: nmusers_at_globomaxnm.com

Subject: [NMusers] PRED for BLQ-like observations

 





Hello The NONMEM Users,





 





When we use M3-like approach, the outputs has PRED for non-missing
observations and something else for BLQ (is that PRED=CUMD?).  As in=
 the
diagnostic figures PRED for BLQs looks like noise,
  I remove them.  It is not always perfect, but OK in for most frequen=
t
cases.  





 





When we use count data such as a scale with few possible values (for=
 
example,  0, 1, 2, 3, 4, 5), it makes more sense to use PHI funct=
ion
(home-made likelihood) for all observations rather
  than to treat the count as a continuous variable an apply M3-like
approach to 1 and 5 while only (as we know, they are like LLOQ an=
d
ULOQ).  In this case, all PRED values look like noise.  A hard wa=
y
to replace the noise with PRED value is to simulate PRED
  for each point and merge them with the DV and IPRED data. Is th=
ere an
easy way? 






 





(The model runs well and better than when the count is treated as=
 a
continuous variable.)





 





Thanks!





Pavel     








Received on Fri Nov 20 2015 - 14:59:51 EST

This archive was generated by hypermail 2.3.0 : Fri Sep 27 2019 - 16:47:18 EDT