Chapter 9 Exercises
Part I
1. An intrepid epidemiology student was examining statistics from the National Highway Traffic
Safety Administration when she made a startling discovery: smoking among drivers appeared to
be associated with fatal car accidents. She gathered 2009 data from Florida and arranged it in
the 2×2 table below:
Fatal Accident
Smoking
1289
No Smoking 1269
Total
2558
Nonfatal Accident
164723
688421
853144
Total
166012
689690
855702
Calculate and interpret the risk ratio and risk difference based on the above table.
2. With such compelling results, the student immediately began thinking of reasons for the
observed association. Could it be that people who smoked were more distracted? That they
tended to drive with only one hand on the wheel and were at greater risk of losing control?
That they had reduced visibility because of the smoky car interior? Her mentor had another
idea, however, and suggested that she consider whether a third variable, such as alcohol
consumption, might be driving the results. The student first considered whether the drivers
testing positive for blood alcohol at the scene of the accident was associated with smoking in
her data:
Smoking
Alcohol
124337
No Alcohol 41675
Total
166012
No Smoking
335475
354215
689690
Total
459812
395890
855702
Calculate and interpret the risk ratio and risk difference based on the above table.
3. She then considered whether testing positive for alcohol was associated with being in a fatal car
crash in her data:
Fatal Accident
Alcohol
1639
No Alcohol 919
Total
2558
Nonfatal Accident
458173
394971
853144
Total
459812
395890
855702
Calculate and interpret the risk ratio and risk difference based on the above table.
4. Do you think that the association between smoking and fatal car accidents originally observed
by the student may have been confounded by alcohol? Explain.
Part II
You undertake an investigation to assess the effects of hormone replacement therapy (HRT) on coronary
heart disease (CHD). You conduct a cohort study whereby you follow women with no history of CHD for
ten years. Assume complete follow-up on all women.
We will assume that the data table below is the TRUTH (i.e., not what you measured in your study, but
what you would have measured if you had conducted the study perfectly and there were no noncomparability). In the absence of any misclassification of exposure status, this is what you would have
observed.
CHD+
CHDTotal
HRT+
200
5110
5310
HRT170
5259
5429
Total
370
10369
10739
1. Calculate and interpret the risk ratio and risk difference based on these data.
Misclassification Scenario 1:
Women were asked to recall their HRT exposure experience; recall is almost never perfect. Twenty
percent of women who actually had taken HRT said that they hadnt, whereas ten percent of women
who actually had not taken HRT said that they had. This occurred regardless of disease status.
2. Use the table below to calculate the 2×2 table with misclassification.
CHD+
CHD-
Total
HRT+
HRTFinal table
Total
with misclassification:
CHD+
CHD-
Total
HRT+
HRTTotal
3. What are the risk ratio and risk difference in the misclassified table? Calculate and interpret.
4. Compare the risk ratio in question 3 to the true risk ratio in question 1. How did misclassification of
the exposure independent of disease status (non-differential misclassification of exposure) affect the
estimates in the study?
Misclassification Scenario 2:
Twenty percent of the nonusers with CHD were categorized as users. Everyone else was classified
accurately according to the TRUTH table above.
5. Is this an example of misclassification of exposure or disease?
6. Fill in the table below with the misclassified 2×2 table.
CHD+
HRT+
HRTTotal
Table with misclassification:
CHD-
Total
CHD+
CHD-
Total
HRT+
HRTTotal
7. What are the risk ratio and risk difference in the misclassified table? Calculate and interpret.
8. Compare the risk ratio in question 7 to the true risk ratio in question 1. How did this
misclassification affect the estimates in the study?
Part III
1. In a prospective study of depression and dementia, you recruit participants who are between
ages 65 and 75 and not suffering dementia. At intake, you give them a screening test for
depression. Five years later, you give them the Mini Mental State Examination, which assesses
cognitive function. There is no loss to follow-up. Your observed risk ratio is 2.40: elderly people
who are depressed have 2.40 times the risk of dementia compared with elderly people who are
not depressed over 5 years.
a. If, instead of perfect follow-up, you ended up with 20% fewer participants after 5 years. It turns
out that depressed people were more likely to drop out than non-depressed. What would the
effect be on your risk ratio?
b. Alternatively, what would be the effect on the risk ratio if the 20% of participants lost to followup were all people who were depressed and demented?
Part IV
What are some strategies that may be deployed in the design and data collection phases of
epidemiologic studies to minimize misclassification?