Exercises in Epidemiology Applying Principles and Methods
Exercises in Epidemiology Applying Principles and Methods
Exercises in Epidemiology Applying Principles and Methods
NOEL S. WEISS
1
1
Oxford University Press, Inc., publishes works that further
Oxford Universitys objective of excellence
in research, scholarship, and education.
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
9 8 7 6 5 4 3 2 1
Printed in the United States of America
on acid-free paper
CONTENTS
Introduction vii
References 237
Index 243
This page intentionally left blank
INTRODUCTION
There are a lot of texts that deal with the principles and meth-
ods of epidemiology. Ive been a coauthor of one of these myself.
All of the texts, to a greater or lesser extent, provide examples
of real or hypothetical epidemiologic studies to illustrate a given
principle or method. For many (probably most) readers of these
books, the examples help to solidify an understanding of the
topic at hand.
What the examples do not provide is the opportunity to
consider, on ones own, how a particular issue ought to be dealt
with, or how a particular question should be addressed. The
purpose of this book is to supplement the material contained in
the textbooks in such a way that the reader is forced to: (1) iden-
tify situations in which the validity or accuracy of a particular
design or analytic approach may be limited; and (2) determine
how that limitation might be overcome. Such actions are just
those that epidemiologists have to take when they are planning
research or are reviewing that of others.
The key word in the preceding paragraph is supplement. The
present book cannot stand alone as a means of learning about
epidemiology, or even as a means of being introduced to the
v i i i | INTRODUCTION
Type of tumor
Race Mucinous Other Total
Men Women
Answer 1.4
Answer 1.5
27 5,677,867
= 0.53 since it allows for the
23 2,569,639 number of driver-miles at risk.
Answer 1.8
19 181
Relative risk = = 2.3
98 2,191
RATES AND PROPORTIONS | 21
Answer 1.9
a. Relative mortality
Answer 1.12
Dead Alive
Belt 45% ?
No belt 55% ?
Total 100% 100%
Answer 1.14
M P 1.5 3.0
= = = 2.
W P 3.0 1.5
Suicides
Type of No. of men per 100 men
cancer with cancer No. of suicides (95% CI)
10 = .95 ( 8 ) + .05t
Causal Inference
Does your automobile have body damage? Fix it, and youll sig-
nificantly reduce your probability of involvement in another
traffic mishap. There is a distinct psychological advantage to
having even minor auto collision damage repaired as soon as
possible. Studies have shown that drivers of newly repaired
automobiles tend to drive more defensively than those with
unrepaired damage.
Answer 2.6
and/or
Answer 2.8
Oral # of # of Odds
Air travel contraceptives cases controls ratio 95% CI
No No 54 94 1.0 Ref.
No Yes 95 48 3.5 2.15.8
Yes No 4 5 1.4 0.36.8*
Yes Yes 20 2 17.4 3.9157.0*
*Exact confidence limits.
Answer 2.11
Answer 2.12 The bias results from the fact that the effect
being measureddifference in age at menarcheis intertwined
with the definition of the groups being comparedpre-versus
postmenarche initiation of training. Thus, if a group of girls
with age-at-menarche distributed normally about a mean of
13 years all began to train at 13, the mean age at menarche
for the postmenarche trained would be <13 years while the
mean age at menarche for the premenarche trained would be
>13 years. For example:
Confounding
1,300 1,300
Rate per 100,000 population
1,200 1,200
1,100 1,100
Crude death rate
1,000 1,000
900 900
800 800
700 700
600 600
Age-adjusted death rate
500 500
0 0
1940 1950 1960 1970 1980 1990
Answer 3.1
19881990 20002002
Incidence per Incidence
Age Population 1,000 person- Population per 1,000
(years) size years size person-years
Disability
Yes No
All Children
IVF 101 5,579
Comparison cohort 119 11,241
Singletons only
IVF 45 3,183
Comparison cohort 115 10,955
Nonsingletons
IVF 56 2,396
Comparison cohort 4 286
Answer 3.5
101 5,680
a) Crude relative incidence = = 1.70
119 11,360
Cases Controls
Number of the other
four risk factors AZT No AZT AZT No AZT
0 0 0 40 88
1 0 3 51 73
2 2 9 33 22
3, 4 6 7 7 6
0 73 2 22 6 6
+ +
Adjusted OR = 127 66 26 = 0.27
3 51 9 33 7 7
+ +
127 66 26
8 131
Crude OR = = 0.61
19 189
*Adjusted for drivers age, ethnicity, alcohol consumption in past 6 hours, seat belt use, vehicle
speed, average driving time each week, driving license status, vehicle insurance status, and weather.
Crude Age-adjusted
prevalence prevalence
Before After ratio ratio
fortification fortification (95% CI) (95% CI)
Length of 48 29
observation
(months)
Maternal age 30.1 30.9
(mean, SD) (0.16) (0.081)
(years)
Number of 218,977 117,986
women
Number of 248 69
women with
an affected child
Prevalence 1.13 0.58 0.52 (0.40 0.62
(per 1,000 0.67) (0.460.83)
infants)
and
Age of mother
Under 20 2024 2529 3034 3539 4049
Total years years years years years years
a. 0.36
b. Greater than 0.36
c. Less than 0.36
d. No prediction is possible.
Type OC used
Triphasic 52% 42%
Monophasic
30 microgram ethinylestradiol 27.2% 29.2%
50 microgram ethinylestradiol 7.9% 13.2%
Norethisterone 9.4% 9.0%
Progestin-only 3.5% 6.2%
80
69 72
70
63
60 Black
53
50 47
White
Percent
43
40
30
20
10
0
All women Never married Ever married
Marital status of mother at time of birth
% cigarette smokers
Age (years) Male Female
% cigarette
smokers
Marital Status Male Female
Answer 3.23
Cohort Studies
Answer 4.2
a. Not all patients in this study were followed for the same
length of time. Specifically, the 437 patients vaccinated
during the 2-year study period were at risk for hospi-
talization only for the period after immunization. Thus,
even had the hospitalization rates been the same in
vaccinated and unvaccinated groups, the cumulative inci-
dence would be lower in the former group. This would lead
to an overestimate of the vaccines efficacy.
b. The analysis should employ a person-time denominator,
so that incidence rates can be calculated. The 437 persons
vaccinated after November 15, 1993, would contribute
person-time-at-risk to the experience of the unvacci-
nated persons until their date of vaccination, and to the
experience of vaccinated persons afterward. Because
the risk of hospitalization for pneumonia and influenza
may be relatively low in the several week period immedi-
ately after November 15 (i.e., before the seasonal peak),
adjustment for calendar period may be needed. Otherwise,
confounding could arise from the different calendar dis-
tribution of person-time between vaccinated and unvac-
cinated groups.
COHORT STUDIES | 125
The results of the study for deaths from heart disease are
shown in the table below. (The SMRs presented are adjusted for
age, sex, and calendar time.)
SMR estimates with 95% confidence intervals for mortality for
heart disease for all Washington Works employees compared
to 2 external reference populations
DuPont 8-state
WW regional employee
cohort WV population population
95% 95%
Cause of death O E SMR CI E SMR CI
All heart disease 314 475.6 66.0 58.9, 284.5 110.4 98.5,
73.7 123.3
Answer 4.3 Persons with heart disease are less likely than
other persons to become employed and stay employed, and also
are more likely to die of heart disease. The proportion of WW
employees with heart disease is almost certainly smaller than in
the West Virginia population, leading to a spuriously low esti-
mate of relative mortality from heart disease when these popu-
lation death rates are used as a basis for comparison. A similar
distortion would not be expected to be present using mortality
rates of other workers as a means of determining the expected
number of deaths.
COHORT STUDIES | 127
Answer 4.8 The data produced by the authors of the letter are
misleading. A comparison of the risk of death from breast
cancer in women who do and do not receive HRT needs to incor-
porate the possible influence of HRT on both incidence and on
case-fatality. The use of mortality rates in users and nonusers
by the authors of the article does this; the use of case-fatality
alone does not.
COHORT STUDIES | 137
Duration of
employment
(years) Observed Expected SMR
Yes 14.3%
No
Delivered at 39 weeks 9.1%
Delivered after 39 weeks 15.0%
*These generally were not elective inductions, but the presence of an indication
for induction was adjusted for in the analysis.
Case-Control Studies
The fact that several relatives sometimes stutter has led others
to assert that stuttering is inherited. Yet in forty years of expe-
rience I have met more stutterers who had no close relatives
or ancestors who stuttered than those who had. It is my obser-
vation that most stutterers are hypersensitive persons, and
I believe that hypersensitivity is acquired at an early age
through the childs environment.
Answer 5.2 No. The lower confidence limit of the odds ratio
associated with not breast-feeding is 7.7; thus, there is little
possibility that chance is solely responsible for the association.
There would be concern over the sample size only if it were
important to know more precisely where, within the range
of 7.736.0, the true odds ratio lies.
CASE-CONTROL STUDIES | 153
a. Method A
b. Method B
c. Neither is a better choice than the other.
+ 200 100
2.25
800 900
Using Method A, 5% 800 = 40 cases truly negative for T1
would be labeled as T1positive, as would 5% 900 = 45
controls.
Method A: Observed T1 status
+ 200 + 40 100 + 45
1.64
800 40 900 45
Using Method B, 5% 200 = 10 cases truly negative for T1,
would be labeled as T1-negative, as would 5% 100 =
5 controls.
Method B: Observed T1 status
+ 200 10 100 5
1.99
800 + 10 800 + 5
CASE-CONTROL STUDIES | 155
Answer 5.4
Do you believe the control group chosen for this study led
to bias in the estimate of the size of the association between
fatal snowmobile trauma and alcohol use? If yes, why, and would
an unbiased estimate be greater or smaller than that obtained
by the authors? If not, why not?
1 6 0 | EXERCISES IN EPIDEMIOLOGY
Answer 5.7
Insulator 47 13
Shipbuilder 31 21
Plumber 35 28
Furnace or boiler
installer or repairman 21 10
+ 15 15
259 15 134 = 110 259 15 72 = 172
15 15
Odds ratio = = 1.56
110 172
The authors interpretation of the results of their study
is incorrect.
(It turns out that assumption (c) above is almost certainly
not a valid one, i.e., a relatively high proportion of men whose
longest held job involved brake lining work or repair indeed had
been employed at other times in a high risk occupation. When
in a reanalysis attention was restricted to men who had no such
history,34 no greater proportion of mesothelioma cases (1/33)
than controls (9/171) had been employed to do brake work.)
CASE-CONTROL STUDIES | 167
Multiple Causal
Pathways and Effect
Modification
Incidence of
VT per 10,000
Cases of VT Person-years person-years
Answer 6.2
Negative
No OC 0.8
Current OC 3.0 3.7 2.2
Positive
No OC 5.4
Current 28.2 5.0 22.8
Men 10 30 20 3
Women 5 25 20 5
*Rate per 100,000 person-years.
<20 20
Vaccinated Cases Population Cases Population
Odds ratio (OR) and 95% CI were adjusted for age, race/ethnicity, education, family
history of endometrial cancer, age at menarche, full-term pregnancies, duration of oral
contraceptive use, duration of hormone therapy use, menopausal status, and height.
1 9 0 | EXERCISES IN EPIDEMIOLOGY
Answer 6.8 a.
Risk difference
Maternal history Relative risk (per 1,000)
a) Rate difference
b) Rate ratio
Explain.
1 9 4 | EXERCISES IN EPIDEMIOLOGY
Antiretroviral Rate
treatment difference
Type of HIV (per 100 Rate
infection Yes No person-years) ratio
Risk of perinatal
death (per 100)
Risk
Gestational First-born Second-born difference
age (weeks) twin twin Relative risk (per 100)
HBsAg+ HBsAg
Case Control Case Control
HBsAg+ HBsAg
Case Control Case Control
Odds ratio 20 35
History of
ulcer NSAID use Cases Controls
No No 607 15,242
No Yes 171 597
Yes No 405 1,430
Yes Yes 106 164
Screening
Answer 7.1
Yes 3 9 12
No 16 104 120
19 113 132
Ovarian cancer
LPA Yes No Total
Answer 7.3
+ A b a+b
c (0%) d (100%) c+d
Answer 7.4
Study group
Annual Biennial
screening screening Control
Answer 7.7
Answer 7.8 The design of this study does not permit an unbi-
ased estimate of the potential impact of photofluorographic
screening on mortality from stomach cancer. The problem stems
from what has been referred to as healthy screenee bias.26 The
exposed cohort in the present study has received a screening
exam in the prior year that was negative: Had it been positive
for cancer, such persons would have been removed from the
cohort. The unexposed cohort has not had such persons iden-
tified or excluded. Thus, even if screening led to completely
ineffective treatment, or to no treatment at all, the observed
relative mortality from stomach cancer would be lower in the
screened group.
SCREENING | 225
% screened 48 60 32 44
% not screened 52 40 68 56
Odds ratio 0.6 0.6