Deck 5: A Dichotomous Outcome: Confounding and Regression

Question 1

Which (A, B, or C) represents the best screening test? Why? What does curve C tell about the screening test?

Answer 1

Curve A represents the best screening test because it has the highest sensitivity and specificity. Curve C has an area under the curve of 0.5. This implies that the probability of a correct diagnosis is like flipping a coin. Thus, the screening test of curve C is not useful.

Question 2

Which one of the following is implied by a screening test with a low sensitivity?

A) The test yields a large number of false positives.
B) The test yields a large number of false negatives.
C) The test yields a large number of true positives.
D) The test yields a large number of true negatives.

Answer 2

The test yields a large number of false negatives.

Question 3

When the area under the ROC curve is 0.5

A) The diagnostic test is very helpful.
B) The diagnostic test is somewhat helpful.
C) The diagnostic test is not helpful.
D) It cannot be determined from the information given.

Answer 3

The diagnostic test is not helpful.

Question 4

Define sensitivity and specificity.

Question 5

Which of the following characterizes a good screening test?

A) high sensitivity and high specificity
B) high sensitivity and low specificity
C) low sensitivity and high specificity
D) low sensitivity and low specificity

Question 6

In a body mass index (BMI) test, a cutoff point is established such that an individual is classified as obese when test score is above the cutoff point. As the cutoff score increases, how will the sensitivity and specificity of the test change?

Question 7

How does the magnitude of the parameter estimate (b) in a logistic regression influence the interpretation of the risk factor?

Question 8

Which of the following is an appropriate way to handle a confounder?

A) adjustment
B) stratification
C) group matching
D) all of the above

Question 9

Given that an interaction effect is found significant in the logistic regression model, the main effects associated with the interaction should NOT be interpreted independently. Why? What should be reported/interpreted when an interaction is present?

Question 10

Three diagnostic tests (Pearson, Deviance, Hosmer-Lemeshow) can be used to evaluate the model fit of logistic regression. Provide the null hypothesis for a test of model fit. What would you conclude if the hypothesis is not rejected?

Question 11

Which of the following is a logistic regression model? (p=probability of an event)

A) p=a+b1x1+ · · · +bkxk
B) p/(1-p)= a+b1x1+ · · · +bkxk
C) log(p)= a+b1x1+ · · · +bkxk
D) log(p/(1-p))= a+b1x1+ · · · +bkxk

Question 12

What is the rationale behind using logit (log(odds)) and not odds or probability of the event as the outcome variable in logistic regression?

Question 13

When the Breslow-Day test results in a significant p-value

A) The estimated odds ratios are different.
B) The estimated odds are different.
C) The estimated relative risks are different.
D) The estimated risks are different.

Question 14

The Mantel-Haenszel's odds ratio is often used to account for confounding. When a confounding variable is found to be an effect modifier, why is it not appropriate to report Mantel-Haenszel odds ratios?

Question 15

Briefly explain what a positive and negative parameter estimate (b) implies in a logistic regression in terms of the probability of the outcome.

Question 16

In a logistic regression model, explain how to determine whether a confounder is an effect modifier.

Question 17

Which of the following yields the correct estimated odds ratio for a 5-unit increase in an explanatory variable?

A) 5b
B) exp(5b)
C) 5exp(b)
D) exp(5)(exp(b))

Question 18

Which of the following represents an odds ratio in a logistic regression?

A) estimate of b
B) logit difference (logit=log(odds))
C) exp(b)
D) exp(logit)

Question 19

Suppose that the estimated odds ratio of asthma is reported to be 0.45 (95% CI: 0.1, 0.8) for children living near farms (measured in kilometers, km), adjusting for age, race, and sex. Interpret.

Question 20

List situations when Mantel-Haenszel's method for calculating an adjusted odds ratio is no longer appropriate. What type of analysis is an alternative in these situations?

Question 21

If the global null hypothesis "All of the true slopes are 0" is rejected, what can be concluded?

Question 22

ROC curves are often used to evaluate model fit for a logistic regression. Why is this appropriate?

Question 23

The plot below presents the odds ratio estimates and 95% confidence intervals from three different studies. Interpret.