Detailed
Outline
I. Overview (pages 348–350)
A. Focus: how to assess discriminatory performance
(DP) of a binary logistic model.
B. Considers how well the covariates in a given model
help to predict (i.e., discriminate) which subjects
will develop the outcome (Y¼1, or thecases) and
which will not develop the outcome (Y¼0, or the
noncases).
C. One way to measureDP: consider thesensitivity
(Se) and specificity (Sp) from a classification table
that combines true and predicted outcomes over
all subjects.
D. An alternative way to measureDP: involves a plot
(i.e., ROC curve) and/or summary measure (AUC)
based on a range of cut-points chosen for a given
model.
II. Assessing Discriminatory Performance using
Sensitivity and Specificity Parameters (pages
350–354)
A. Classification Table
i. One way to assess DP.
ii. Combines true and predicted outcomes over
all subjects.
iii. Cut-point (cp) can be used withP^ðXÞto predict
whether subject is case or noncase:
If^PðXÞ>cp, then predict subjXto be case;
otherwise, predict subjXto be noncase.
B. Sensitivity (Se) and specificity (Sp)
i. Computed from classification table for fixed
cut point.
ii. Se¼proportion of truly diagnosed cases
¼Pr(true positive | true case) ¼nTP/n 1
iii. Sp¼proportion of falsely diagnosed
¼Pr(true negative | true noncase)¼nTN/n 0
iv. The closer both Se and Sp are to 1, the better is
the discrimination.
v. Sp and Se values vary withcp:
cpdecreases from 1 to 0)Se increases
from 0 to 1, and Sp decreases from 1 to 0.
Sp may change at a different rate than the
Se depending on the model considered.
vi. 1Sp more appealing than Sp:Detailed Outline 373