topic Re: Inverse Prediction in Discussions

Inverse Prediction

Johnanderson — Thu, 05 Jul 2018 23:43:48 GMT

Dear JMP experts,

I appreciate any clarification in regard to "Inverse Prediction"!

In my databse I have variable represent a biological marker, and I have two level groups (patients A and healthy B). I would like to find a cutoff value for the marker to say that this value predicts disease at specific probability (data attached).

I did logistic fit, then ROC curve, after then "Inverse Prediction" which gave me the "prediction probability" for every value of the marker (figure attached). My questions are:

1. Are the steps that I followed, correct, to find a cut off value for the marker to predict disease A.

2. What the term "Inverse Prediction" means? more specifically, what the word "inverse" stands for?

Thanks for any clarification

John

Re: Inverse Prediction

Mark_Bailey — Fri, 06 Jul 2018 13:13:31 GMT

We usually fit a model to data, generally Y = f(X). We can then use this model to predict Y for a given X. Inverse prediction is used when we want to predict the X value that corresponds to a given Y.

Re: Inverse Prediction

Johnanderson — Fri, 06 Jul 2018 15:30:35 GMT

Dear markbailey, I immensely appreciate your reponse! I would like to know the methods that JMP uses to measure "Inverse prediction" for a certain variable. Is there any reference that may help to clarify how it works?

Sincerely,

John

Re: Inverse Prediction

Peter_Bartell — Fri, 06 Jul 2018 16:36:21 GMT

John:

A great place to start learning 'how' JMP does anything is the documentation. Here's a landing page for 'inverse prediction' within the JMP documentation. Inverse prediction as a method is native to several JMP analysis platforms so the exact answer to 'How does JMP do inverse prediction' is dependent upon the analysis platform that was used.

https://www.jmp.com/support/help/14/inverse-prediction.shtml

Re: Inverse Prediction

Mark_Bailey — Fri, 06 Jul 2018 16:49:22 GMT

Start with the ROC Curve. Open the Details report. Use this table to decide the cutoff that you want that satisfies your goals for sensitivity and specificity. Peter's solution is a good one when you know the probability and want to determine the corresponding level of the biomarker.

Re: Inverse Prediction

Mark_Bailey — Fri, 06 Jul 2018 16:56:35 GMT

In addition to the link provided by Peter for the definition, you can find more information by selecting Help > Books > Fitting Linear Models. Go to the Appendix for the statistical details about the inverse prediction method.

Re: Inverse Prediction

Johnanderson — Fri, 06 Jul 2018 17:25:23 GMT

Peter_Bartell and markbailey, I appreciate your responses. Thank you!