Our World Statistics Day conversations have been a great reminder of how much statistics can inform our lives. Do you have an example of how statistics has made a difference in your life? Share your story with the Community!
Choose Language Hide Translation Bar
Level VI

What to Do When Your Data Is a Curve (2019-US-45MP-213)

Level: Advanced


Sam Gardner, Principal Research Scientist, Elanco Animal Health


In many situations in pharmaceutical product development, the most relevant data is a value that is a function of time (a curve).  Traditional approaches to handling curves often ignore the full time dependency of the result and only focus on one aspect of the curve. This often results in a loss of information. 

Three analysis approaches that utilize the entire curve will be discussed: 1) When data is aligned in time, multivariate methods can be used to characterize the data; 2) When data can be described by a parameterized model, fitting that model to each curve results in model parameters that become the new data; and, 3) When the data in the curve is not time aligned, or if it is too complex to describe with a parameterized model, then the Functional Data Explorer can be used to fit a smoothing model to each curve, and again the parameters of that smooth curve fit become the new data. DOE can be combined with these methods to understand how the curves depend on the DOE factors.This talk will show how to use JMP to apply each of these approaches, combined with DOE, using real examples from pharmaceutical product development.


Link to JMP Public Report for Example 2:  https://public.jmp.com/packages/Using-a-Parametric-Model-to-Reduce-a-Cur/js-p/f9wjDHcLJQ4C5kF3vsQSl





Very rich presentation. I appreciated the flow from graph builder to mixed model to non-linear modeling to the functional data explorer.  I am curious if there is a limit with respect to the number of cases that can be analysed in the FDE? In other words if I have 500000 objects measured over 5 time periods can I analyse this data set in FDE?  Again, great presentation!

Interesting question.  I tried it on some simulated random noise data, 500000 profiles, 5 time points per profile, and JMP 15.0 crashed after processing for about 10 minutes.  I did some step up scaling, with 1000 and 10000 profiles, and those seemed to work fine.  So somewhere between 10000 and 500000 it starts to break down, but I'm not sure why.  Maybe one of the developers or technical experts can chime in on this. @chris_gotwalt1 any guidelines on data size for FDE?  

Thanks for your response. Looking forward to the response from developers of technical experts.  The application I have in mind is learning over time and the state I am currently is quite large, this the large N.  I have specified non-linear parameterized models and also fit polynomials, but am curious to see if the FDE provides a better fit or can help me to better understand the functional form of the curve.