CHAPTER 4

Gerstman, B. B. (2003). Chapter 4: Screening for Disease In Epidemiology Kept Simple (2nd ed.). New York: Wiley-Liss. 
[This page last updated: 4/24/04]

T/F refers to a true false problem. After answering T or F, justify the response.
M/C refers to a multiple choice problem. Select the best response.

[Review Questions | Exercises]

REVIEW QUESTIONS

4.1 Introduction

Distinguish between repeatability and validity.
List sources of information used to identify cases in epidemiologic studies.
Distinguish between a symptom, sign, and test.
4 Is a feeling of nausea as perceived by a patient a sign or symptom?
5 Is the observation of "rice-water stool" (water-like diarrhea with white flecks) a sign or symptom?
6 T/F: A scale that repeatedly provides the wrong weight is valid.

4.2 Reproducibility

1 What statistic can be used to quantify the repeatability of a test?
. . . more questions will be developed . . . 

4.3 Validity

1 Define: (a) true positive (b) true negative (c) false positive (d) false negative.
2 Using the notation on p. 85, match the statistic with its definition.
Statistic: SEN, SPEC, PVP, PVN
Definitions:  
    (a) Pr(T- | D-)
    (b) Pr(T+ | D+)
    (c) Pr(D+ | T+)
    (d) Pr(D- | T-)
3 What do we mean by a "gold standard?"
4 Why not always use a gold standard test?
5 M/C: This is the probability a person that shows a negative test is actually disease-free. (a) SEN (b) SPEC (c) PVP (d) PVN
6 M/C: This is the probability a disease-free person will show a negative test result. (a) SEN (b) SPEC (c) PVP (d) PVN
7 Why do most clinicians find predictive values more useful concepts than sensitivity or specificity? 

4.4 Relation Between Prevalence of Disease and Predictive Values of a Test

1 What is the difference between true prevalence and apparent prevalence?
2 What three factors determine the PVP of a test?
3 What three factors determine the PVN of a test?
4 Will PVP be higher when the prevalence is high or low? Why?
5 T/F: PVN increases as prevalence increases.

4.5 Selecting a Cutoff Point

1 T/F: In considering a test in which high values indicate presence of disease, decreasing the cut-off point for a positive result will increase sensitivity. (We are of course assuming there is an overlap of values between the diseased and non-diseased populations.)
2 T/F: In considering a test in which high values indicate presence of disease, increasing the cut-off point will  increase specificity.
3 M/C: In a two-stage screening approach intended to find all cases, we want the initial phase of screening to (a) be specific (b) be sensitive (c) have high predictive value negative (d) have high predictive value positive.

EXERCISES

4.1  - 4.6   See text.

4.7   Data from two raters on the presence or absence of a condition are shown in the table below (Table 4.20). Calculate the kappa statistic for these data and interpret your findings.

TABLE 4.20 Agreement between two raters

               RATER B
RATER A       +      -     Total
+            150     31     181
-             28    239     267
Total        178    270     448

4.8   Data in Table 4.21 (below) represent test results on 170 patients. Calculate the test's sensitivity and specificity.

TABLE 4.21 Test results for 170 patients

               DISEASE
TEST          +      -     Total
+            15      7      22
-             3    145     148
Total        18    152     170

4.9 Why is exercise 4.7 a reproducibility analysis (and not a validity analysis)? Why is exercise 4.8 a validity analysis and not a reproducibility analysis?

4.10 During four months in 1884, Sergeant J. P. Finley predicted whether or not one or more tornadoes would occur in each of eighteen areas of the United States (Goodman & Kruskal, 1959, pp. 127-128). One of Finley's summary tables for 934 prediction periods is shown below (Table 4.21). From this table we conclude  Finley's predictions were correct (11 + 906) / 934 = .9818 of the time. This score of the value of predictability is inappropriate, since a completely ignorant person could uniformly predict "no tornado" and attain a score greater than Finley's.  (With a uniform prediction of "no tornado" you would be correct 920 / 934 = .9850 of the time.) Calculate a k statistic for Finley's data and then discuss whether his predictions were better than random.

TABLE 4.21: Comparison of Finley's Tornado Predictions and Occurrences, April 1884 (Goodman & Kruskal, 1959, p. 128)

            ACTUAL OCCURRENCE
PREDICTION     +        -       Total
+             11        14         25
-              3       906        909
Total         14       920        934

4.11 Read or re-read the premise of exercise 4.10. 
(A) Calculate the SEN and SPEC of Finley's predictions (Table 4.21, above). In plain language, discuss the meaning of these results.
(B) Calculate the PVP and PVN of Finley's predictions. Discuss the meaning of these statistics. Other than the imperfect sensitivity and specificity of the prediction, why was the PVP so low?

Key for Review Questions and Odd Numbered Exercises 
Key for Even Numbered Exercises
(may not be posted)