Logistic Regression: A Self-learning Text, Third Edition (Statistics in the Health Sciences)

Presentation

I. Overview

Modeling

outcomes with

more than two

levels

FOCUS

Examples of multilevel outcomes:

1. Absent, mild, moderate, severe


2. In situ, locally invasive,
   metastatic


3. Choice of treatment regimen

01

to

Change

One approach: dichotomize outcome

2

01 2

This presentation and the presentation that

follows describe approaches for extending the

standard logistic regression model to accom-

modate a disease, or outcome, variable that has

more than two categories. Up to this point, our

focus has been on models that involve a dicho-

tomous outcome variable, such as disease pres-

ent/absent. However, there may be situations in

which the investigator has collected data on

multiple levels of a single outcome. We describe

theformand keycharacteristicsof one model

for such multilevel outcome variables: the poly-

tomous logistic regression model.

Examples of outcome variables with more than

two levels might include (1) disease symptoms

that have been classified by subjects as being

absent, mild, moderate, or severe, (2) invasive-

ness of a tumor classified as in situ, locally

invasive, or metastatic, or (3) patients’ preferred

treatment regimen, selected from among three

or more options.

One possible approach to the analysis of data

with a polytomous outcome would be to

choose an appropriate cut-point, dichotomize

the multilevel outcome variable, and then sim-

ply utilize the logistic modeling techniques dis-

cussed in previous chapters.

For example, if the outcome symptom severity

has four categories of severity, one might com-

pare subjects with none or only mild symptoms to

those with either moderate or severe symptoms.

EXAMPLE

Change

to

None Mild Moderate Severe

None or

mild

Moderate or

severe

432 12. Polytomous Logistic Regression