c. Consideringonly those variables that are candidates for
being assessed as nonconfounders,
i. How many subsets of these variables need to be
considered to address confounding?
ii. List the variables that are contained in each of the
above subsets.

d. Suppose that the following results are obtained when
comparing tables of ORs for different subsets of V
variables in the previously stated (question 2) reduced
model obtained after interaction assessment:
Model

Variables dropped

from model

Table of ORs Within 10%
of Gold Standard Table?
1 C 1 No
2 C 2 Yes
3 C 3 No
4a C 1 C 2 Yes
4b C 1 andC 3 Yes
5 C 1 andC 1 C 2 No
6 C 2 andC 1 C 2 No
7 C 3 andC 1 C 2 Yes
8 C 1 andC 2 Yes
9 C 1 andC 3 Yes
10 C 2 andC 3 Yes
11 C 1 ,C 2 andC 3 Yes
12 C 1 ,C 2 andC 1 C 2 Yes
13 C 1 ,C 3 andC 1 C 2 Yes
14 C 2 ,C 3 andC 1 C 2 Yes
15 C 1 ,C 2 ,C 3 , andC 1 C 2 No
16 None Yes (GS model)
Based on the above results, what models are eligible to
be considered as final models after confounding assess-
ment?
e. Based on your answer to part d, how would you address
precision?

3. In addition to the variablesE 1 ,E 2 ,E 3 ,C 1 ,C 2 ,C 3 consid-
   ered in questions 1 and 2, there were 25 other variables
   recorded on study subjects that were identified from
   the literature review and conceptualization of the study
   as potential control variables. These variables were
   screened out by the investigators as not being neces-
   sary to include in the multivariable modeling analyses
   that were carried out.

Assume that screening was carried out by putting all 25
variables in a logistic regression model together with the
variablesE 1 ,E 2 ,E 3 ,C 1 ,C 2 , andC 3 and then using a back-
ward elimination to remove nonsignificant variables.

Practice Exercises 291