gcpatel126
☆    

India,
2015-12-09 19:08
(3031 d 15:40 ago)

Posting: # 15712
Views: 19,530
 

 Emax and ED50: NLME [Software]

Thread locked
Dear All

How to calculate Emax and ED50 (Fitting based on the assumption of nonlinear mixed effect model (population model) or Fitting based on nonlinear least square regression pooling individual observation (naive pooled data method) in WinNonlin.

Thanks,
Dr. G C Patel


Edit: Category changed. [Helmut]
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-12-10 03:09
(3031 d 07:39 ago)

@ gcpatel126
Posting: # 15713
Views: 17,664
 

 Emax and ED50: NLME

 
Hi G.C.,

please RTFM – the “Phoenix NLME 1.3 User’s Guide” in particular. See the two examples pp183–9 (pp191–7 of the PDF). Alternatively you can use one of WNL’s PD models (#101–108): In the Structure-tab   Set WNL  . Try it with the emax-dataset (Population fitting, default setup):

model     run option      EC50    Emax  
NLME Emax Naive pooled  49.4242  101.383
          FOCE ELS      49.3846  100.905
WNL 101   Naive pooled  49.4242  101.383
          FOCE ELS      49.3846  100.905

Here estimates are identical (generally they are at least very close). However, I prefer the genuine NLME-model (maximum likelihood instead of least squares). Use naïve pooling or FO only to get initial estimates (they are biased). Carry them over as starting values to one of the other engines.

[image]

[image]

[image]

[image]

[image]


If you have difficulties with the setup, I suggest to register at Certara’s forum. There you could upload your project-file for examination by people who get paid for that. :-D

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
gcpatel126
☆    

India,
2015-12-10 18:34
(3030 d 16:14 ago)

@ Helmut
Posting: # 15719
Views: 17,424
 

 Emax and ED50: NLME

 
Hi Helmut,

Many Thanks for your quick support. I have only one query now.

In the example data emax.dat, all AUEC values are positive.

However in our data we have both positive and negative. When we run the NLME population models, Should we convert them to all positive or keep as it is?

Thanks,
Dr. G C Patel
SDavis
★★  
Homepage
UK,
2015-12-15 12:50
(3025 d 21:58 ago)

@ gcpatel126
Posting: # 15730
Views: 17,129
 

 Emax and ED50: NLME

 
Dr. Patel,

I suppose it depends a little on what your PD measurement is at the moment, I would say conceptually at least to me, it's easier ot use raw untransformed data, if you have negative values does that mean you are looking at a change from baseline?

Simon

PS to reiterate Helmut's comment that on the Certara forum you could post your project and we can perhaps advise better then.

Simon
Senior Scientific Trainer, Certara™
[link=https://www.youtube.com/watch?v=xX-yCO5Rzag[/link]
https://www.certarauniversity.com/dashboard
https://support.certara.com/forums/
gcpatel126
☆    

India,
2015-12-29 16:49
(3011 d 17:59 ago)

@ SDavis
Posting: # 15776
Views: 16,627
 

 Emax and ED50: NLME

 
Hi Simon,
Please find below FDA guideline data for calculation ED50 and Emax using nonlinear mixed effect model (population model). Please let me know value matches with guideline or not.

Subject Dose    AUEC
        (hrs)
   1    0.25   -1.23
   1    0.50   -7.39
   1    0.75   -1.48
   1    1.00   -3.80
   1    1.50   -0.23
   1    2.00    5.77
   1    4.00   -4.74
   1    6.00   -1.53
   2    0.25   -0.02
   2    0.50   -5.13
   2    0.75   -8.92
   2    1.00  -24.56
   2    1.50  -19.21
   2    2.00   -1.80
   2    4.00  -43.07
   2    6.00  -41.56
   3    0.25  -13.87
   3    0.50  -15.03
   3    0.75  -18.39
   3    1.00  -16.25
   3    1.50  -15.44
   3    2.00  -23.74
   3    4.00  -24.80
   3    6.00  -21.79
   4    0.25  -27.27
   4    0.50   -3.71
   4    0.75  -43.82
   4    1.00  -44.39
   4    1.50  -77.04
   4    2.00  -66.80
   4    4.00  -62.96
   4    6.00  -71.60
   5    0.25  -10.65
   5    0.50    7.72
   5    0.75  -23.42
   5    1.00  -20.37
   5    1.50  -19.95
   5    2.00  -32.00
   5    4.00  -32.81
   5    6.00  -61.51
   6    0.25  -10.41
   6    0.50   -5.94
   6    0.75   -2.29
   6    1.00   -8.92
   6    1.50  -20.64
   6    2.00  -19.52
   6    4.00   -8.52
   6    6.00  -19.01
   7    0.25    4.20
   7    0.50  -12.31
   7    0.75    1.34
   7    1.00  -18.84
   7    1.50  -42.70
   7    2.00  -37.29
   7    4.00  -45.46
   7    6.00  -37.24
   8    0.25  -11.95
   8    0.50    7.45
   8    0.75    5.95
   8    1.00    8.78
   8    1.50    1.26
   8    2.00  -48.83
   8    4.00  -71.77
   8    6.00   -8.14
   9    0.25  -12.36
   9    0.50   12.95
   9    0.75    1.88
   9    1.00  -43.35
   9    1.50  -20.97
   9    2.00  -39.79
   9    4.00  -57.55
   9    6.00  -34.18
  10    0.25    1.15
  10    0.50  -39.45
  10    0.75  -40.68
  10    1.00  -16.19
  10    1.50    6.87
  10    2.00   10.75
  10    4.00  -37.64
  10    6.00  -35.01
  11    0.25  -30.03
  11    0.50  -39.56
  11    0.75  -61.06
  11    1.00  -43.58
  11    1.50  -40.73
  11    2.00  -62.01
  11    4.00  -27.82
  11    6.00  -33.60
  12    0.25   -7.25
  12    0.50   14.73
  12    0.75  -21.09
  12    1.00   10.81
  12    1.50    0.51
  12    2.00  -10.51
  12    4.00  -14.89
  12    6.00   16.14


Thanks,
Dr. G C Patel


Edit: Reformatted with BBcodes. [Helmut]
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-12-29 18:46
(3011 d 16:02 ago)

@ gcpatel126
Posting: # 15777
Views: 16,828
 

 Emax and ED50: NLME

 
Hi G.C.,

❝ Please find below FDA guideline data for calculation ED50 and Emax using nonlinear mixed effect model (population model). Please let me know value matches with guideline or not.


These data smell of FDA’s 1995 (!) Guidance on topical corticosteroids, Table AIII.4… Software: P-Pharm (Simed, France). ED50 1.89 h, Emax –48.8 h. The software doesn’t exist any more, the developer went out of business, the FDA didn’t give any details (fitting algo, weighting, constraints, …). Do you really expect that anybody without a time-machine will be able to reproduce these numbers? Was the fitted line drawn by someone with shaky hands or did the software spit it out?

[image]


These nice bars are standard errors – not standard deviations, aka cheating with statistics. At 0.75 hours the mean is –17.7 and the SD 21.4. Agreed, the SE with 6.2 looks better. The CVs range from 60 – 250%. Did you look at the individual data?

BTW, addressing Simon here is not a splendid idea. He passes by about once a month in his spare time. I hope for him that he is enjoying a Christmas / New Year vacation. Why didn’t you register at Certara’s forum and upload something there – as Simon and I suggested – instead of posting a data-cemetery full of tabs nobody could copy/paste? Please read Eric Raymond’s essay How To Ask Questions The Smart Way.

I’m half tempted to close this thread.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
gcpatel126
☆    

India,
2015-12-29 19:04
(3011 d 15:44 ago)

@ Helmut
Posting: # 15778
Views: 17,279
 

 Emax and ED50: NLME

 
Hi Helmut,
Already I have registered Certara’s forum and sent same data but value does not matched with FDA guideline.

Thanks,
Dr. G C Patel
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-12-29 20:36
(3011 d 14:12 ago)

@ gcpatel126
Posting: # 15779
Views: 16,711
 

 Emax and ED50: NLME

 
Hi G.C.,

❝ Already I have registered Certara’s forum and sent same data but value does not matched with FDA guideline.


Please give a definition of “matching” – what did you expect to achieve? The FDA fitted only the means. Don’t know why you want to deal with a Population model. Below a comparison:

              Emax   ED50
FDA         -48.8    1.89
WNL 101     -39.811  1.143
PHX/NLME    -39.764  1.139
R/nls       -39.764  1.139
XLS/solver  -39.764  1.139
Gnumeric/NL -39.764  1.139

Do you see a pattern?

[image]

Given the close results obtained in other software, I would say you should not even try to “match” the FDA’s crappy results.

Nonlinear mixed-effects modeling… The recipe is a mixture of science, experience, luck, and art. The finished product sometimes is inedible. I think that the FDA’s results are crap. Not only the poor fit, but already the squeezed model. I’m able to come up with similar crap in PHX/NLME as well.

                Emax   ED50
Naive pooled  -39.765  1.139
FOCE L-B      -62.615  3.429

IMHO, with the given data (and an obsolete software nobody could explore) hoping for a “match” is futile.

[image]



Tortured data will confess to anything.    Fredric E. Menger

The combination of some data and an aching desire
for an answer does not ensure that a reasonable answer
can be extracted from a given body of data.
     John W. Tukey

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
gcpatel126
☆    

India,
2015-12-30 15:43
(3010 d 19:05 ago)

@ Helmut
Posting: # 15780
Views: 16,466
 

 Emax and ED50: NLME

 
Hi Helmut,
Thanks for your detailed response. We also get same value using WinNonlin.

Thanks,
Dr. G C Patel
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-12-30 15:56
(3010 d 18:52 ago)

@ gcpatel126
Posting: # 15781
Views: 16,412
 

 Emax and ED50: NLME

 
Hi G.C.,

❝ Thanks for your detailed response.


Welcome.

❝ We also get same value using WinNonlin.


Are you satisfied now or even more worried? Why didn’t you bother to answer any of my questions? I still think that the model is wrong. If I compare FDA’s results to the others its relative error (bias, if you want) is +22.7% for Emax and +65.8% for ED50. Now what?

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
gcpatel126
☆    

India,
2015-12-30 16:19
(3010 d 18:29 ago)

@ Helmut
Posting: # 15782
Views: 16,326
 

 Emax and ED50: NLME

 
Hi Helmut,

As we all know, for most of the cases we are validating our methods and software tools based on regulatory's guideline data. The same we are doing for this ED50 calculation. We tried many tools but none of them match exactly or come too closer to guideline data result.

You have written that model is wrong, which model are you talking ? Is it mentioned in guideline ? OR any other ?

You have compared the results and there is difference also, our question is also the same if values are different compare to guideline than is it acceptable to use that tool ?

Finally do you have any preference for software to use for this calculation ?

Many thanks,
Dr. G C Patel
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-12-30 17:02
(3010 d 17:46 ago)

@ gcpatel126
Posting: # 15783
Views: 16,459
 

 Validation?

 
Hi G.C.,

❝ […] for most of the cases we are validating our methods and software tools based on regulatory's guideline data.


Not a particularly good idea. What’s written in GLs is not necessarily correct. F.i. the CVintra in Canada’s BE guidance example data set was wrong for decades. Not only the value, but also the formula. See this presentation about validation of software.

❝ The same we are doing for this ED50 calculation. We tried many tools but none of them match exactly or come too closer to guideline data result.


❝ You have written that model is wrong, which model are you talking ? Is it mentioned in guideline ? OR any other ?


Since the FDA reported only Emax and ED50 at p29 I used the simple model

(1)    E = (Emax × D) / (ED50 + D)

The GL mentions at p4 also the baseline model

(2)    E = E0 + (Emax × D) / (ED50 + D)

Let’s compare them (fitting mean data)

      Emax     ED50     E0     AIC
(1)   1.139  -39.764    –     47.85
(2)   1.183  -39.523  -0.449  49.85

Since the AIC of the simple model is lower than the one of the baseline model, (1) is the winner. Selection based on the minimum AIC is the easiest approach in comparing models. Alternatively you could run an F-test.

❝ You have compared the results and there is difference also, our question is also the same if values are different compare to guideline than is it acceptable to use that tool ?


Which tool? I cannot imagine that you will be able to find any software which comes even close to the FDA’s results. Not by any chance the software vendor went out of business. In the same guidance the FDA asked for data submitted on a diskette without specifying details. 8", 5¼", 3½"? If you want to be fully compliant, visit the next waste disposal site and search for a floppy drive. You could also ask the FDA for a copy of the famous P-Pharm software.

❝ Finally do you have any preference for software to use for this calculation ?


Phoenix/NLME. In R (nls/nlme/lme4) it is not that easy to guess suitable starting values and specify weighting schemes.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
gcpatel126
☆    

India,
2015-12-30 18:47
(3010 d 16:01 ago)

@ Helmut
Posting: # 15784
Views: 16,279
 

 Validation?

 
Hi Helmut,

Thanks for clarification on guideline's example correction and the link provided by you is really useful.

You have compared two methods and based on AIC value you have concluded the result. Which software you have used to calculate these two cases ?

My meaning for tool is that.. if my tool's result value is different then FDA's tool's result values then will it be acceptable for them ?

I know that P-Pharm software is no longer available and I also do not want to use the software which do not have any support existence.

Thanks again for investing your valuable time in this discussion.

- Dr. G C Patel
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-12-30 22:15
(3010 d 12:34 ago)

@ gcpatel126
Posting: # 15787
Views: 16,505
 

 Validation?

 
Hi G. C.,

❝ You have compared two methods and based on AIC value you have concluded the result. Which software you have used to calculate these two cases ?


Phoenix NLME 1.3. The AIC is available in all software I know (except Excel, of course).

❝ My meaning for tool is that.. if my tool's result value is different then FDA's tool's result values then will it be acceptable for them ?


Heck, do you really believe the FDA insists in reproducing a result from obsolete software which is 20+ years old? It’s somehow difficult to speak about “correct” or “false”. But what do we have: One result obtained by the (in)famous P-Pharm and others which are almost identical across five other software packages and six fitting algos – but different to what’s given in the guidance. There is no democracy in science but the odds are >5:1 against the FDA. I’m pretty sure you would get similar results in other software as well (hey guys: SAS, S+, Kinetica, NCSS, SPSS, MatLab, STaTa, STATISTICA, OO Calc, …)
Maybe back in ’94/95 the FDA was hit by the Pentium bug? The time window would fit just perfectly!

We could set up an experimentum crucis. Generate data from known parameters and compare results. In my example the data are exact, i.e., without error. Do you agree that we should expect almost identical results? Emax = –50, ED50 = 1.

  x        y
 1/4    -10.0
 1/3    -12.5
 1      -25.0
 1+1/2  -30.0
 2+1/3  -35.0
 4      -40.0
 5+1/3  -42.5
 9      -45.0

In all packages I provided initial estimates of –10 and 2. R/nls by design does not allow fitting error-free data. I added ε [0, 10–6].

                                   Emax        ED50
Excel solver (Newton)          -50.00001791  0.99999968
             (gradient)        -50.00009120  1.00000539
WNL PD 101 Gauß-Newton
           (Levenberg-Hartley) -50.00003052  0.99999860
WNL PD 101 Nelder-Mead
           (simplex)           -49.99991035  0.99998636
PHX/WNL 101                    -50.00000000  1.00000000
PHX/NLME                       -50.00000000  1.00000000
Gnumeric                       -50.00000000  1.00000000
R/nls                          -50.00000000  1.00000000

Of course this is not a proof. Unfortunately the Emax-model cannot be solved in closed form. Let’s simplify: a = Emax and ED50 = b. Then we can write
y = ax/(x + b), getting
(1)   a = y(x + b)/x and
(2)   b = (xyax)/y.
a depends on b and b depends on a. End of story. You could cheat. For any of the x|y above setting b = 1 you will get a = –50 from (1). Or the other way ’round: For any x|y setting a = –50 you will get b = 1 from (2). But, wait a minute. The data are exact, right? We can plug in two data pairs, two equations, two unknowns, doable:
–25 = a/(1 + b)
–40 = 4a/(4 + b)
Simsalabim:
a = –50, b = 1
At least the comparison should give you strong confidence that these software packages perform well and what the FDA reported is poo.

See also this post.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
gcpatel126
☆    

India,
2015-12-31 08:06
(3010 d 02:42 ago)

@ Helmut
Posting: # 15788
Views: 16,267
 

 Validation?

 
Hi Helmut,

Thanks for conforming the software you have used.

I know FDA will not insist to produce that result and I focused only that to determine the right method with right tool. This conversation will helps a lot in my process.

I have also verified the values using various tools and its same.

Thanks for your immediate responses.

- Dr. G C Patel
gcpatel126
☆    

India,
2016-01-11 15:54
(2998 d 18:54 ago)

@ gcpatel126
Posting: # 15813
Views: 15,226
 

 Validation?

 
Hi Helmut,

Revisitng again this discussion, I want your suggestion on one scenario.

For one dose duration response study we have calculated ED50 and Emax by population model using Phoenix® NLME™ and we got below results.

ED50 : 120
Emax : -35

On other hand, using PPHARM software the values are as below,

ED50 : 50
Emax : -26

As you can see the difference is very big.

Now what you suggest, Should we go with Phoenix® NLME™ or PPHARM values ?

If we submit Phoenix NLME results to FDA, will be there any query from FDA as they are using PPHARM software ?

Please let us know your views.

Many thanks,
Dr. G C Patel
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2016-01-11 23:13
(2998 d 11:35 ago)

@ gcpatel126
Posting: # 15815
Views: 15,295
 

 EOD

 
Hi G.C.,

❝ For one dose duration response study we have calculated ED50 and Emax by population model using Phoenix® NLME™ and we got below results.

❝ ED50 : 120

❝ Emax : -35

❝ On other hand, using PPHARM software the values are as below,

❝ ED50 : 50

❝ Emax : -26


You are responsible for validating your software in your computing environment, not I. I don’t know P-Pharm. Since it disagrees with PHX/NLME already at the first (!) significant digit, I guess it’s crap.
Have a look at the NIST’s reference datasets. The certified results were obtained in 128bit precision. The NIST considers an estimate obtained by other software acceptable if more than 4 significant digits match the certified result. I can only say, that my installation (Phoenix 6.4.0.768 / NLME 1.3 64bit) on my machine (Xeon E3-1245v3 Haswell 4×3.4GHz / 16GB RAM) with my OS (64bit Win7 SP1 all patches) gives for most datasets matches of 5-7 significant digits already with the first set of start values (sometimes even better with the second).
Do your homework. This conversation leads to nowhere. EOD from my side. Thread closed.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,639 registered users;
80 visitors (0 registered, 80 guests [including 5 identified bots]).
Forum time: 10:49 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5