Comparison of Dose Response Curves

[#1]

I have dose-response data for two different compounds.  I know how to generate the individual dose response curves in JMP for each dose-response dataset.  However, I am unclear on is how to compare the dose-response curves - i.e. are they different from another and what is an appropriate statistical test to show this?  I realize that "compare the dose-response curves" may be too vague, as there are many potential parameters to compare -- i.e. LD10/50/90s, probit slopes.  Any guidance would be greatly appreciated.

[#2 · Accepted solution]

You did not say how you fit these curves in JMP so I will start from the beginning.

Select Analyze > Specialized Modeling > Fit Curve. Enter your response data column in the Y role. Enter your dose data column in the X role. JMP assumes that you applied the log, base 10, transform to the dose if you use the command to test parallelism (below). Enter your grouping data column (indicates reference standard or test article) in the Group role. Click OK.

Click the red triangle next to Fit Curve and select Sigmoid Curves > Logistic Curves and then the model for your curve fit. Click the red triangle next to the model and see the commands for assessing parallelism (three methods) and other things. The command for testing parallelism also computes the relative potency.

[#3]

Hi Mark,

Thank you for the reply.  I must be using a different version of JMP than you as my menus are different, and I don't see the structure you describe (I have v11.2 64-bit). Do you have any idea where the menu might be located in this version?

Previously, I had been performing dose response analysis using a generalized linear model with a probit link function, under the fit model dialogue box.

Best regards,

Paul

[#4]

I don't have JMP 11 installed anymore. but you can check. See Help > Books > Predictive and Specialized Modeling. There should be a chapter devoted to the Fit Curve platform, right before the chapter about the Nonlinear platform. If not, then JMP 11 did not include this recently added platform.

I think that the GLM could be modified with cross terms (interaction effects) to test some of the hypotheses. For example, in determining expiry, time is crossed with batch to test poolability. If this term is found to be significant, then a unique slope (degradation path) is required for each batch. If not, then a common slope can be used.

[#5]

Chapter 6 in the JMP 11 book "Specialized Models" documents how to "Analyze Models with the Fit Curve Platform"