Approximating PO Models

[Alvin Jeffery]

Dr. Harrell, 

I have a proportional odds model I would like to approximate in order to make it more clinically useful.  If I created the PO model using lrm() and then approximate this full model with ols(), is there any way to create a nomogram of the approximated model that would include each level of the response variable (i.e., for each of the intercepts from the PO model)?  

The nomogram of the full PO model is nice and might serve the purposes of instructing clinicians, but I was hoping to parse down the variables while still keeping all of the probabilities on there, if possible.  

Thanks for your help!

Alvin

[Frank Harrell]

I wish I could remember where I have a snippet of code that might help.  I think the approach is to use ols to approximate the model for one intercept then write a series of functions such as fun=list(plogis, fun3, fun5) to translate the predicted values at the initial intercept to probabilities, where for example

fun3 <- function(lp) plogis(lp - (reference intercept used) + (new intercept))