From: Paul Hutson <prhutson@pharmacy.wisc.edu>

Subject: ordinal dependent models

Date: 24 Sep 1998 12:41:22 -0400

I have been asked to process data that includes analgesic concentrations and pain rating on an ordinal 0-10 scale. It is a retrospective analysis, but seems like one worth pursuing. Has anyone experience in performing NONMEM PD fits of ordinal data? Any suggestions or pitfalls in the approach?

Thanks

Paul

****

From: "Jill B. Fiedler-Kelly" <jbf@cognigencorp.com>

Subject: Re: ordinal dependent models

Date: 28 Sep 1998 14:22:37 -0400

Hi Paul,

In response to your query regarding modeling ordinal data, a model you might want to look into is the proportional odds model. I have used this model to evaluate repeated patient scores measured on a 0-4 scale (none, slight, mild, moderate, severe) over time in NONMEM. (See Alan Agresti's book, An Introduction to Categorical Data Analysis for introductory information.) There is also some very helpful information at the Palo Alto ftp site (ftp://pkpd.icon.palo-alto.med.va.gov/nonmem.dir/) from Jaap Mandema (subdirectory JWM.DIR) where I think he was developing models for analgesics and pain relief. I hope this helps.

Sincerely,

Jill Fiedler-Kelly, M.S.

Vice President, Population PK/PD

Pharmaceutical Outcomes Research, Inc.

435 Lawrence Bell Drive, Suite 7

Williamsville, NY 14221