Jump to content


Photo

charting capabilities fro scatter plots


  • Please log in to reply
12 replies to this topic

#1 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 01 June 2016 - 03:30 PM

I am wondering whether I can make a scatter plot of subjects's Cmax data: I have the Cmax data  and I need a semilog plot and with the geometric mean and 90% confidence intervals.

 

 

It is along the lines of the attached image with 95% confidence intervals from Graph Pad Prism, which does not offer 90% confidence intervals.

Attached Thumbnails

  • scatter_dot_plot.png


#2 Ana Henry

Ana Henry

    Advanced Member

  • Members
  • PipPipPip
  • 232 posts

Posted 01 June 2016 - 04:14 PM

Hi Angus,
Can you explain how you have spread in the x-axis for individual Cmax values?

Thanks,
Ana

#3 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 01 June 2016 - 04:42 PM

Hi Angus,
Can you explain how you have spread in the x-axis for individual Cmax values?

Thanks,
Ana

Anna:  I think Prism does this to prevent data points being too close to each other so that you get a better view of the data.



#4 Ana Henry

Ana Henry

    Advanced Member

  • Members
  • PipPipPip
  • 232 posts

Posted 01 June 2016 - 04:48 PM

I can trick Phoenix to create something similar to your plot but without spread on the x-axis.
See attached project for some tips.

AnaAttached File  FinalPlot.bmp   3MB   484 downloadsAttached File  Graphic.phxproj   414.13KB   424 downloads

#5 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 01 June 2016 - 04:51 PM

Thank You Ana; I wonder it the range on the y axis can be expanded to prevent overlap of data points.?



#6 Ana Henry

Ana Henry

    Advanced Member

  • Members
  • PipPipPip
  • 232 posts

Posted 01 June 2016 - 04:53 PM

Thank You Ana; I wonder it the range on the y axis can be expanded to prevent overlap of data points.?


Yes, that is data dependent. In this 'made-up' example the spread was not very large but you
can always change the y-axis by customizing your own range.

Hope this helps,
Ana

#7 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 01 June 2016 - 04:59 PM

Yes; you can expand the range some what.  I must look at how you did this.  I have the series of Cmax values onluy need this data.

 

What is step 1 , step 2, step 3 for me?

 

Please can you say?

 

Angus



#8 Ana Henry

Ana Henry

    Advanced Member

  • Members
  • PipPipPip
  • 232 posts

Posted 01 June 2016 - 05:20 PM

Yes; you can expand the range some what.  I must look at how you did this.  I have the series of Cmax values onluy need this data.
 
What is step 1 , step 2, step 3 for me?
 
Please can you say?
 
Angus

Hi Angus,
I tricked the program by assigning some arbitrary values: Cmax values had a new column added with a value 3,
GeoMean has a column with two values 1 and 5, and the 90%CI Geo Mean has values two values 2 and 4. I enter
this in a worksheet and then join them with the NCA and Descriptive stats results. Then in the
plot you overlay the data (3 sets) and have those arbitrary values plotted in the X-axis.
You change titles, remove markers from the CIs and plot formats as you wish. It is hard to
explain so better take a look at the project.

Ana

Edited by Ana Henry, 01 June 2016 - 05:21 PM.


#9 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 01 June 2016 - 06:45 PM

Hi Angus,
I tricked the program by assigning some arbitrary values: Cmax values had a new column added with a value 3,
GeoMean has a column with two values 1 and 5, and the 90%CI Geo Mean has values two values 2 and 4. I enter
this in a worksheet and then join them with the NCA and Descriptive stats results. Then in the
plot you overlay the data (3 sets) and have those arbitrary values plotted in the X-axis.
You change titles, remove markers from the CIs and plot formats as you wish. It is hard to
explain so better take a look at the project.

Ana

Yes;  Thank you I will look at it first thing early tomorrow.  I am a little  oversubscribed at the moment.



#10 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 02 June 2016 - 01:26 PM

Ana:  the reason I am exploring this is that in a regulatory submission document usually the PK parameters e.g. Cmax are presented in terms of descriptive statistics (mean, standard deviation, CV(%).  For a dataset treatment within the same study (with different populations). I have also used the bioequivalence criteria for one comparison in a large stuy with different populations.  I have received a comment saying that the descriptive statistics parameter data in analogous way to the one  BE  comparison should be presented as geometric means (90% CL) using semi-log plots.  The justification for this statement is that the BE Guidance (Statistical one..) points out that the PK parameter data be treated is  log normally distributed.   (IMHOThe BE guidance does not give any reference to why this is so).

 

I have in fact presented the Cmax parameter e.t.c. data as a scatter plot with a mean and a standard deviation:  that is why I am looking at this option that you have helped me with.

 

Angus



#11 Helmut Schütz

Helmut Schütz

    Advanced Member

  • Members
  • PipPipPip
  • 316 posts
  • LocationVienna, Austria

Posted 04 June 2016 - 10:25 PM

Hi Angus,

… the one BE comparison should be presented as geometric means (90% CL) using semi-log plots. The justification for this statement is that the BE Guidance (Statistical one..) points out that the PK parameter data be treated is log normally distributed.


Makes sense. PK-metrics follow a lognormal distribution indeed (well, except tmax). See this thread (including a stupid example from the FDA).
The attached project is a starter. Don’t forget to change the specification of the error-bars to 'Absolute'. I will meet Simon next week and have a chat.

PS: It would be more elegant to use a categorial XY-Plot for the PE&CI (map Dependent to X), but it does not support error bars. Hence, my workaround.

Attached Files


 Best regards,
Helmut

https://forum.bebac.at/


#12 Ana Henry

Ana Henry

    Advanced Member

  • Members
  • PipPipPip
  • 232 posts

Posted 07 June 2016 - 03:11 PM

Hi Angus,

Makes sense. PK-metrics follow a lognormal distribution indeed (well, except tmax). See this thread (including a stupid example from the FDA).
The attached project is a starter. Don’t forget to change the specification of the error-bars to 'Absolute'. I will meet Simon next week and have a chat.

PS: It would be more elegant to use a categorial XY-Plot for the PE&CI (map Dependent to X), but it does not support error bars. Hence, my workaround.


All,
Note that the next version of Phoenix coming out at the end of this month supports error bars for x-categorical plot. Here is the image of the same plot created with this new version.X-Cat_Errors.JPG

#13 Angus McLean

Angus McLean

    Advanced Member

  • Members
  • PipPipPip
  • 223 posts

Posted 07 June 2016 - 04:04 PM

Here is my chart as a snapshot done in Graph Pad Prism:  I include the individual data points and Geometric mean with 90% CI of Geometric mean.  In this case I had to make two charts and superimpose the two of them using a clear plotting area.

 

 

Attached Thumbnails

  • GEOMETRIC_MEAN_WITH_INDIVIDUAL_POINTS.png





0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users