NONMEM Users Network Archive

Hosted by Cognigen

Re: [NMusers] Additive plus proportional error model for log-transform data

From: Rong Chen <>
Date: Thu, 2 Jun 2016 07:07:05 +0000 (UTC)

Hi Ahmad,

This issue hasbeen discussed a lot and I'm afraid there's no consensus yet.=

To your question:

1.       Is there away to solve this pro=
blem for the indicated formulas?

As you said, thisproblem occurs at the early/later time points. In oth=
er words, it happenswhen prediction is relatively low. This is because ther=
e's an underlyingapproximation in the derivation of the two formulas:


Innon-transformed terms


Assuming thatEXP(x)=1+x (for small x), you get


 Variance of thisexpression


               = F**2*THETA(x)**2=

 On the otherhand, for the error model


variance isequal to


 Thus, thesemodels are similar if not identical with




So when F is rathersmall, the approximation of exp(x)=1+x doesn=
't workanymore. This may not be a problem when fitting your data.It may onl=
y occurwhen the prediction is extremely low, let say 10^-4. In opinion it's=
 safe touse this formula in most cases. But if you are seeking a perfect an=
swer,Jacob's suggestion may be the one.

 2.       Are the twoformulas below=
 equally valid?

 As far asI'm concerned, these two are equally valid. I question the r=
esult that there'sa difference in OFV and estimates between the two.

 3.       Is there analternative fo=
rmula that I can use which does not have this numerical problem?

 Maybe youcan try Stu's "double exponential error model": Y = LOG(F+=
M) +(F/(F+M))*ERR(1) + (M/(F+M))*ERR(2). 

Best regards,

Rong Chen

School ofPharmaceutical Science

Peking University

Beijing, China

      From: "Abu Helwa, Ahmad Yousef Mohammad - abuay010" <ahmad.abuhelwa_at_m=>
 To: "" <>
 Sent: Thursday, 2 June 2016, 10:27
 Subject: [NMusers] Additive plus proportional error model for log-transfor=
m data
 #yiv5739009609 #yiv5739009609 -- filtered {panose-1:2 4 5 3 5 4 6 3 2 4;}#=
yiv5739009609 filtered {font-family:Calibri;panose-1:2 15 5 2 2 2 4 3 2 4;}=
#yiv5739009609 p.yiv5739009609MsoNormal, #yiv5739009609 li.yiv5739009609Mso=
Normal, #yiv5739009609 div.yiv5739009609MsoNormal {margin:0cm;margin-bottom=
:.0001pt;font-size:11.0pt;}#yiv5739009609 a:link, #yiv5739009609 span.yiv57=
39009609MsoHyperlink {color:#0563C1;text-decoration:underline;}#yiv57390096=
09 a:visited, #yiv5739009609 span.yiv5739009609MsoHyperlinkFollowed {color:=
#954F72;text-decoration:underline;}#yiv5739009609 p.yiv5739009609MsoListPar=
agraph, #yiv5739009609 li.yiv5739009609MsoListParagraph, #yiv5739009609 div=
.yiv5739009609MsoListParagraph {margin-top:0cm;margin-right:0cm;margin-bott=
09609 span.yiv5739009609EmailStyle17 {color:windowtext;}#yiv5739009609 .yiv=
5739009609MsoChpDefault {font-size:10.0pt;}#yiv5739009609 filtered {margin:=
72.0pt 72.0pt 72.0pt 72.0pt;}#yiv5739009609 div.yiv5739009609WordSection1 {=
}#yiv5739009609 filtered {}#yiv5739009609 filtered {}#yiv5739009609 filtere=
d {}#yiv5739009609 filtered {}#yiv5739009609 filtered {}#yiv5739009609 filt=
ered {}#yiv5739009609 filtered {}#yiv5739009609 filtered {}#yiv5739009609 f=
iltered {}#yiv5739009609 filtered {}#yiv5739009609 filtered {}#yiv573900960=
9 filtered {}#yiv5739009609 filtered {}#yiv5739009609 filtered {}#yiv573900=
9609 filtered {}#yiv5739009609 filtered {}#yiv5739009609 filtered {}#yiv573=
9009609 filtered {}#yiv5739009609 filtered {}#yiv5739009609 filtered {}#yiv=
5739009609 ol {margin-bottom:0cm;}#yiv5739009609 ul {margin-bottom:0cm;}#yi=
v5739009609 Dear NMusers,   I am developing a PK model using log-tran=
sformed single-dose oral data. My question relates to using combined error =
model for log-transform data.   I have read few previous discussions =
on NMusers regarding this, which were really helpful, and I came across two=
 suggested formulas (below) that I tested in my PK models.  Both formu=
las had similar model fits in terms of OFV (OFV using Formula 2 was one uni=
t less than OFV using Formula1) with slightly changed PK parameter estimate=
s. My issue with these formulas is that the model simulates very extreme co=
ncentrations (e.g. upon generating VPCs) at the early time points (when dru=
g concentrations are low) and at later time points when the concentrations =
are troughs. These simulated extreme concentrations are not representative =
of the model but a result of the residual error model structure.   My=
 questions: 1.      Is there a way to solve t=
his problem for the indicated formulas? 2.     =
Are the two formulas below equally valid? 3.    =
 Is there an alternative formula that I can use which does not have=
 this numerical problem? 4.      Any referenc=
e paper that discusses this subject?   Here are the two formulas: 1.=
      Formula 1: suggested by Mats Karlsson w=
ith fixing SIGMA to 1: W=SQRT(THETA(16)**2+THETA(17)**2/EXP(IPRE)**2) =
  2.      Formula 2: suggested by Leonid=
 Gibiansky with fixing SIGMA to 1: W = SQRT(THETA(16)+ (THETA(17)/EXP(IPR=
E))**2  )   The way I apply it in my model is this:   FLAG=
=0           =
st regards,   Ahmad Abuhelwa School of Pharmacy and Medical Sciences =
University of South Australia- City East Campus Adelaide, South Australia A=


Received on Thu Jun 02 2016 - 03:07:05 EDT

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to: Once subscribed, you may contribute to the discussion by emailing: