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

Presentation

I. Overview

Focus

Modeling issues not considered

in previous chapters

Goal: determine “best” model

Binary logistic model

Apply to any regression analysis

Issues:

1. Modeling strategy when there
   are two or more exposure
   variables


2. Screening variables when
   modeling


3. Collinearity diagnostics


4. Influential observations


5. Multiple testing

II. Modeling Strategy for
Several Exposure
Variables

Extend modeling strategy

 Outome:D(0,1)

 Exposures:E 1 ,E 2 ,...,Eq
 Control variables:C 1 ,C 2 ,...,Cp

EXAMPLE

Example: TwoEs

Cross-sectional study

Grady Hospital, Atlanta, GA

297 adult patients

Diagnosis:Staph. aureusinfection

Concern: potential predictors of

MRSA

This presentation addresses several modeling

strategy issues not considered in the previous

two chapters (6 and 7). These issues represent

important features ofanyregression analysis

that typically require attention when going

about the process of determining a “best”

model, although our specific focus concerns a

binary logistic regression model.

We consider five issues, listed here at the left,

each of which will be described and illustrated

in the sections that follow.

In this section, we extend the modeling strat-

egy guidelines described in the previous two

chapters to consider two or more exposure

variables, controlling for covariates that are

potential confounders and/or effect modifiers.

We begin with an example involving exactly

two exposure variables.

A cross-sectional study carried out at Grady

Hospital in Atlanta, Georgia involved 297

adult patients seen in an emergency department

whose blood cultures taken within 24 hours of

admission were found to haveStaphylococcus

aureusinfection (Rezende et al., 2002). Infor-

mation was obtained on several variables that

were considered as potential predictors of

methicillin-resistance infection (MRSA).

244 8. Additional Modeling Strategy Issues