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

Presentation

I. Overview

FOCUS

Guidelines for “best”

models

Three-stage strategy

Valid estimate of E–D

relationship (con-

founding and effect

modification)

II. Rationale for a
Modeling Strategy

Minimum information in most
study reports,
e.g., little explanation about
strategy

Information oftennotprovided:
 How variables are chosen

 How variables are selected

 How effect modifiers are
assessed

 How confounders are assessed

Guidelines are needed for the fol-
lowing:
 To assess validity of results

 To help researchers know what
information to provide

 To encourage consistency in
strategy
 For a variety of modeling
procedures

This presentation gives guidelines for deter-

mining the “best” model when carrying out

mathematical modeling using logistic regres-

sion. We focus on a strategy involving three

stages. The goal of this strategy is to obtain a

valid estimate of an exposure–disease relation-

ship that accounts for confounding and effect

modification.

We begin by explaining the rationale for a

modeling strategy.

Most epidemiologic research studies in the lit-

erature, regardless of the exposure–disease

question of interest, provide a minimum of

information about modeling methods used in

the data analysis. Typically, only the final

results from modeling are reported, with little

accompanying explanation about the strategy

used in obtaining such results.

For example, information is oftennotprovided

as to how variables are chosen for the initial

model, how variables are selected for the final

model, and how effect modifiers and confoun-

ders are assessed for their role in the final

model.

Without meaningful information about the

modeling strategy used, it is difficult to assess

the validity of the results provided. Thus, there

is a need for guidelines regarding modeling

strategy to help researchers know what infor-

mation to provide.

In practice, most modeling strategies are ad hoc;

in other words, researchers often make up a strat-

egy as they go along in their analysis. The general

guidelines that we recommend here encourage

more consistency in the strategy used by different

researchers.

168 6. Modeling Strategy Guidelines