American Journal of Epidemiology
ª The Author 2011. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. All rights reserved. For permissions, please e-mail: [email protected].
Vol. 174, No. 5
DOI: 10.1093/aje/kwr117
Advance Access publication:
June 30, 2011
Practice of Epidemiology
Population Attributable Fractions of Adenocarcinoma of the Esophagus and
Gastroesophageal Junction
Catherine M. Olsen*, Nirmala Pandeya, Adèle C. Green, Penelope M. Webb, and David C.
Whiteman for the Australian Cancer Study
* Correspondence to Dr. Catherine M. Olsen, Cancer Control Laboratory, Queensland Institute of Medical Research, PO Royal
Brisbane Hospital, Herston, Queensland 4029, Australia (e-mail: [email protected]).
Initially submitted October 20, 2010; accepted for publication March 16, 2011.
Obesity, gastroesophageal reflux, and smoking have repeatedly been shown to be important and independent
risk factors for adenocarcinoma of the esophagus (EAC) and of the gastroesophageal junction (GEJAC). There
have been few attempts, however, to quantify the proportion of disease associated with these potentially
modifiable factors. The authors have estimated the population attributable fraction of EAC and GEJAC attributable
to obesity, symptoms of gastroesophageal reflux, and smoking using data from a population-based case-control
study conducted in Australia between 2002 and 2005. Cases were patients with EAC (n ¼ 364) or GEJAC (n ¼ 425).
Controls (n ¼ 1,580) were randomly sampled from a population register. Combinations of smoking, body mass index
(weight in kilograms divided by height in meters squared), and gastroesophageal reflux together accounted for
76% (95% confidence interval: 66, 84) of EAC cases and 69% (95% confidence interval: 58, 78) of GEJAC cases.
Individually, high body mass index (30) and frequent acid reflux (1 time/week) accounted for the greatest
proportions of EAC (23% and 36%, respectively), and smoking and frequent symptoms of acid reflux accounted
for the greatest proportions of GEJAC (43% and 28%, respectively). The present study suggests that these
cancers may be largely prevented by maintaining healthy body mass index, avoiding smoking, and controlling
symptomatic gastroesophageal reflux.
esophageal adenocarcinoma; esophagogastric junction; esophagus; gastroesophageal reflux; obesity; smoking
Abbreviations: BMI, body mass index; CI, confidence interval; EAC, adenocarcinoma of the esophagus; GEJAC, adenocarcinoma
of the gastroesophageal junction; GER, gastroesophageal reflux; PAF, population attributable fraction.
The rates of adenocarcinomas of the esophagus (EAC) and
the gastroesophageal junction (GEJAC) are increasing in
many countries worldwide (1–4), particularly among men.
These consistent increases in incidence across populations
in the absence of notable changes in methods of detection
and diagnosis (5, 6) strongly implicate nongenetic factors as
the underlying causes of these cancers. Obesity, gastroesophageal reflux (GER), and smoking have repeatedly been shown
to be important and independent risk factors for both EAC
and GEJAC (7–10), and there is evidence of biologic interaction between obesity and GER (11) in the etiology of these
cancers. A necessary next step is to quantify the burden of
disease associated with these potentially modifiable factors to
quantify the public health impact of prevention strategies.
Population attributable fractions (PAFs) are useful in assessing the impact of disease risk factors in a given population, as they take into account both the strength of the
association and the prevalence of each factor in the population. A risk factor strongly associated with the disease but
with low prevalence in the population will have less of a public health impact than will a risk factor with a similar effect
that affects a large proportion of the population. A single risk
factor approach is most often used in the calculation of PAFs;
however, it can lead to some challenges in interpretation
because the resulting PAFs do not take into account the possible interaction of multiple competing risks (12). Partial
PAFs present the risk estimates for combinations of common
risk factors so that additive effects can be explored (13).
582
Am J Epidemiol. 2011;174(5):582–590
Population Attributable Fractions of Esophageal Cancer
583
Table 1. Distribution of Demographic and Lifestyle Characteristic of Controls and People With Esophageal Adenocarcinoma and
Gastroesophageal Adenocarcinoma in the Population-Based Australian Cancer Study, 2002–2005
Controls
Variable
Men
No.
Women
%
No.
Participants With EAC
P Valuea
%
Age, years
Men
No.
Women
%
No.
Participants With GEJAC
Men
P Valuea
%
No.
Women
%
No.
0.14
<0.001
0.77
30–39
29
2.8
52
9.6
5
1.5
1
2.9
4
1.1
1
1.8
40–49
90
8.7
110
20.4
22
6.7
2
5.7
32
8.7
7
12.5
50–59
267
25.7
142
26.3
91
27.7
8
22.9
97
26.3
16
28.6
60–69
389
37.4
147
27.2
125
38.0
8
22.9
132
35.8
16
28.6
70
265
25.5
89
16.5
86
26.1
16
45.7
104
28.2
16
28.6
37.1
77
21.0
20
35.7
Smoking status
Never smoker
0.04
<0.001
387
37.9
323
60.1
79
24.2
13
0.01
Ex-smoker
498
48.8
142
26.4
185
56.6
12
34.3
192
52.5
18
32.1
Current smoker
136
13.3
72
13.4
63
19.3
10
28.6
97
26.5
18
32.1
Body mass indexb
1 year prior
18–24.9
336
25–29.9
502
30–34.9
155
15.0
38
3.7
35
0.01
<0.001
32.6
234
43.7
59
171
32.0
142
44.7
71
13.3
83
26.1
59
11.0
34
10.7
6
19.4
Frequency of heartburn
or acid reflux
18.6
12
0.45
38.7
91
7
22.6
150
42.5
18
32.1
6
19.4
83
23.5
15
26.8
29
8.2
7
12.5
0.08
P Valuea
%
25.8
16
28.6
0.29
0.34
Never
438
42.5
260
48.3
70
21.4
10
30.3
99
27.0
20
36.4
Less than weekly
469
45.5
217
40.3
120
36.7
8
24.2
132
36.0
18
32.7
Once a week or more
123
11.9
61
11.3
137
41.9
15
45.5
136
37.1
17
30.9
Abbreviations: EAC, adenocarcinoma of the esophagus; GEJAC, adenocarcinoma of the gastroesophageal junction.
P value for significant difference between men and women.
b
Weight (kg)/height (m)2.
a
To our knowledge, only 1 study has examined the PAFs
for risk factors associated with EAC and GEJAC (14), and
these PAFs were calculated from data from a populationbased case-control study conducted in the United States
from 1993 to 1995. Engel et al. (14) reported high PAFs
for former and current smoking, high body mass index
(BMI), and high frequency of GER symptoms. They reported PAFs for single risk factors that were adjusted for
all other factors; however, they did not calculate PAFs for
combinations of these factors, some of which are known to
be highly correlated. For diseases with multifactorial
causes, analyses that do not examine risk-factor combinations could overestimate or underestimate the PAFs associated with individual risk factors (15), as causative factors
often co-occur and interact to greater or lesser degrees.
With an aim to quantify the theoretical scope for prevention of these diseases in the population, we computed PAFs
of EAC and GEJAC associated with BMI, smoking, and
GER by using data from a case-control study conducted in
Australia between 2002 and 2005 (11). We extended previous analyses by calculating partial PAFs for combinations
of the major risk factors. Partial PAFs can indicate the
relative importance of risk-factor combinations and help
set priorities for public health prevention efforts.
Am J Epidemiol. 2011;174(5):582–590
MATERIALS AND METHODS
Study participants
The methods used in the Australia-wide population-based
case-control study have been described previously (11). Eligible cases were people aged 18–79 years who lived in
Australia and had a histologically confirmed primary
invasive cancer of the esophagus or gastroesophageal junction diagnosed between July 1, 2002, and June 30, 2005, in
Queensland. Cases were recruited through either major treatment centers or state cancer registries. Of the 1,577 patients
with esophageal cancer invited to participate in the study
(1,191 through clinics and 386 through cancer registries),
1,102 patients (70%) returned a completed questionnaire
(367 EAC cases and 426 GEJAC cases; 309 esophageal squamous cell carcinoma cases were excluded from these analyses). Three EAC cases and 1 GEJAC case were deemed
ineligible on review and were also excluded from these
analyses, leaving a total of 364 EAC cases and 425 GEJAC
cases.
Controls were randomly selected from the national electoral roll and were frequency-matched by age (in 5-year age
bands) and state of residence to the case group. Of 3,258
584 Olsen et al.
Table 2. Adjusted Odds Ratios, Population Attributable Fractions, and 95% Confidence Intervals for Smoking,
Body Mass Index, and Heartburn or Acid Reflux Symptoms 10 Years Before the Study in Patients With
Adenocarcinoma of the Esophagus, Australia, 2002–2005
Variable
Odds
Ratio
All Participants
95% CI
PAFa, %
95% CI
Men
PAFa, %
Women
95% CI
PAFa, %
95% CI
Smoking status
Never smoker (reference)
1.0
Ex-smoker
1.5
1.1, 2.0
17
Current smoker
2.4
1.6, 3.5
11
7, 17
Ever smoker
1.7
1.2, 2.2
29
16, 45
8, 33
16
6, 36
19
6, 48
9
5, 17
27
13, 47
26
12, 45
46
23, 71
13
Body mass indexb 1 year prior
18–24.9 (reference)
1.0
25–29.9
1.4
1.0, 2.0
13
5, 28
17
8, 31
30–34.9
2.5
1.7, 3.6
15
10, 23
17
11, 24
6
35
3.7
2.2, 6.2
8
5, 13
8
5, 13
11
2, 42
Overweight or obese
1.8
1.3, 2.5
36
23, 53
41
27, 57
4
0, 100
40, 15c
0.2, 69
Heartburn/acid reflux in the
past 10 years
Never (reference)
1.0
Less than once a week
1.6
1.1, 2.2
13
6, 25
15
8, 27
13
46, 20
Once a week or more
6.4
4.5, 9.0
36
30, 42
35
29, 42
38
19, 61
Any reflux
2.6
1.9, 3.4
49
38, 60
50
39, 62
26
3, 78
Abbreviations: CI, confidence interval; PAF, population attributable fraction.
The population attributable fraction was adjusted for age, sex, educational level, aspirin or nonsteroidal
antiinflammatory drug use, and the other factors listed in the table.
b
Weight (kg)/height (m)2.
c
P < 0.05 comparing population attributable fractions for men and women using an independent t test.
a
potentially eligible control participants who were contacted
and invited to participate (646 were uncontactable and
deemed ineligible), 175 were excluded because they were
deceased (16), too ill (61), or unable to communicate in
English (98), and 41 were lost to follow-up between initial
contact and participation. Of the 3,042 remaining controls,
1,680 (55%) accepted. Completed questionnaires were returned by 1,580 controls (49% of all potentially eligible
controls). The study was approved by the ethics committees
of the Queensland Institute of Medical Research and all
participating hospitals and cancer registries.
Exposure measurement
Information was collected using a self-administered questionnaire, which included questions about demographic,
medical, hormonal, reproductive, dietary, family history,
and other potential risk factors. Exposures were assessed
before a reference date, defined as 1 year before the date
of diagnosis (or date of first approach for controls), because
more recent exposures in cases could have been influenced
by the presence of subclinical disease. Participants selfreported height and weight 1 year before the date of diagnosis. Participants were asked about frequency of heartburn
(described as ‘‘a burning pain behind the breastbone after
eating’’) or acid reflux (described as ‘‘a sour taste from acid
or bile rising up into the mouth or throat’’) in the 10 years
before diagnosis and frequency of aspirin or nonsteroidal
antiinflammatory drug use in the past 5 years. Detailed
questions about past and current smoking habits asked participants whether, over their whole lives, they had ever
smoked more than 100 cigarettes, cigars, or pipes; positive
responses elicited further questions about the ages at which
they started and stopped smoking and about typical daily
consumption.
BMI, calculated as weight in kilograms divided by height
in meters squared, was classified using the World Health
Organization definitions of obesity (underweight: <18.5;
normal weight: 18.5–24.9; overweight: 25–29.9; class I
obesity: 30–34.9; class II obesity: 35–39.9; and class II
obesity: 40) (16). In the present analysis, smokers were
categorized as never smokers or ever smokers. Frequency
of acid reflux symptoms was defined as the highest reported
frequency of either heartburn or acid reflux and was categorized as never, less than weekly, and once or more per week.
Statistical methods
To calculate adjusted PAFs with 95% confidence intervals,
we used the method of Bruzzi et al. (12), which uses adjusted
odds ratios from unconditional logistic regression and prevalence of the risk factors in the study cases. This method
takes into account multiple levels of exposure and controls
for confounding. Confidence intervals for the model-based
Am J Epidemiol. 2011;174(5):582–590
Population Attributable Fractions of Esophageal Cancer
585
Table 3. Adjusted Odds Ratios, Population Attributable Fractions, and 95% Confidence Intervals for Smoking,
Body Mass Index, and Heartburn or Acid Reflux Symptoms 10 Years Before the Study for Adenocarcinoma of the
Gastroesophageal Junction, Australia, 2002–2005
Variable
Odds
Ratio
All Participants
95% CI
Men
Women
PAFa, %
95% CI
PAFa, %
95% CI
PAFa, %
95% CI
Smoking status
Never smoker (reference)
1.0
Ex-smoker
1.8
1.3, 2.4
21
16, 27
21
12, 35
18
7, 40
Current smoker
4.0
2.9, 5.7
21
13, 33
20
15, 26
28
16, 43
Ever smoker
2.3
1.7, 3.0
43
32, 54
41
28, 55
46
27, 66
1, 52
Body mass indexb
1 year prior
18–24.9 (reference)
1.0
25–29.9
1.1
0.8, 1.5
4
0.3, 39
2
0, 95
10
30–34.9
2.0
1.4, 2.9
12
7, 19
11
6, 19
18
7, 37
35
2.5
1.5, 4.1
5
3, 10
5
3, 10
6
0.8, 31
Overweight or obese
1.8
1.3, 2.5
22
10, 41
18
6, 42
33
11, 66
Heartburn/acid reflux in the
past 10 years
Never (reference)
1.0
Less than once a week
1.2
0.9, 1.6
5
1, 25
7
2, 25
6
31, 20
Once a week or more
4.4
3.2, 6.1
28
23, 34
30
24, 36
22
10, 40
Any reflux
2.6
1.9, 3.4
34
23, 46
36
25, 50
16
2, 67
Abbreviations: CI, confidence interval; PAF, population attributable fraction.
The population attributable fraction was adjusted for age, sex, educational level, aspirin or nonsteroidal
antiinflammatory drug use, and the other factors listed in the table.
b
Weight (kg)/height (m)2.
a
PAFs were calculated using logit transformation as described
by Benichou and Gail (17), except in cases in which the PAF
was negative, in which case the confidence intervals were
calculated as PAF 6 1.96 3 standard error, using the same
approach as Engel et al. (14). First, PAFs were calculated for
individual risk factors for all cases and separately for men
and women. We estimated relative risks associated with
each stratum of smoking, BMI, and acid reflux symptoms
and adjusted for age, sex, educational level, and the other 2
main variables of interest. For smoking and acid reflux
symptoms, the ‘‘never’’ category was used as the reference
category; for BMI, normal weight was used as the reference
category. We computed partial PAFs for combinations of
smoking status, BMI, and frequency of acid reflux symptoms
using 2 strata of smoking (never smokers vs. ever smokers)
and BMI (<25 vs. 25) and 3 strata of frequency of acid
reflux symptoms (never, less than weekly, or once or more per
week). We computed PAFs for all cases and adjusted them for
age, sex, educational level, and nonsteriodal antiinflammatory drug use; we then computed them separately for men
and women. The reference group for comparison was people
in the normal weight range who had never smoked and had
not reported symptoms of acid reflux in the relevant time
period. The partial PAF presented here is the reduction in
risk that would result from setting 1 exposure category in the
cross-classification of exposure factors to the lowest level of
risk while holding confounders and other joint exposure
Am J Epidemiol. 2011;174(5):582–590
levels constant. The sum of the partial attributable risks
across all joint cross-classifications of the exposure variables would be the PAF.
All analyses were conducted using the Interactive Risk
Assessment Program, version 2.2 (available from the National Institutes of Health; http://dceg.cancer.gov/bb/tools/
irap). Statistical significance was determined at a ¼ 0.05,
and all tests for statistical significance were 2-sided. Comparison of PAFs between men and women in each subset
was performed using an independent t test.
RESULTS
The characteristics of study cases and controls are presented
in Table 1. Of the 364 patients with EAC, 329 (90%) were men
and 35 (10%) were women. Of the 425 patients with GEJAC,
369 (87%) were men and 56 (13%) were women. As reported
previously, both case groups were more likely than controls to
be ever smokers, to be in the overweight or obese range of
BMI, and to report frequent symptoms of acid reflux (11, 18).
PAFs for individual risk factors
EAC. The PAFs of EAC for the individual risk factors but
adjusted for the other factors are presented in Table 2. For all
cases combined, we found that symptoms of acid reflux contributed the highest PAF (49%, 95% confidence interval (CI):
586 Olsen et al.
Table 4. Partial Population Attributable Fractions and 95% Confidence Intervals for Combinations of Exposure to
Smoking, Body Mass Index, and Heartburn or Acid Reflux Symptoms 10 Years Before the Study for
Adenocarcinoma of the Esophagus, Australia, 2002–2005
Smoking Status, Body Mass
Indexa 1 Year Prior, and Reflux
All Participants
PAFb
Men
95% CI
PAFb
Women
95% CI
PAFb
95% CI
Never smoker
Normal weight
No reflux
Reference
Reference
Reference
Reflux less than once a week
0.9
0.4, 2
0.9
0.3, 2
3
Reflux once a week or more
1
0.5, 3
1
0.4, 3
3
13, 6
0.4, 17
No reflux
2
0.7, 3
2
0.8, 3
0.1
3, 3
Reflux less than once a week
5
3, 8
5
3, 8
2
7, 4c
Reflux once a week or more
8
6, 11
8
5, 11
11
3, 29
Overweight or obese
Ever smoker
Normal weight
No reflux
1
0.6, 3
1
0.4, 3
5
1, 19
Reflux less than once a week
3
2, 5
3
2, 5
4
0.6, 23
Reflux once a week or more
5
3, 8
5
3, 7
9
3, 26
Overweight or obese
No reflux
8
6, 12
8
5, 13
10
3, 28
Reflux less than once a week
17
12, 23
18
13, 24
4
0.5, 24c
Reflux once a week or more
25
21, 30
26
21, 31
19
8, 38
Total PAF attributed to combinations
of the 3 factors
76
66, 84
78
67, 85
59
26, 85
Abbreviations: CI, confidence interval; PAF, population attributable fraction.
Weight (kg)/height (m)2.
b
Population attributable fractions were adjusted for age, sex, educational level, and aspirin or nonsteroidal
antiinflammatory drug use.
c
P < 0.05 comparing population attributable fractions for men and women using an independent t test.
a
38, 60), which was greater for men than for women (50% and
26%, respectively), although not significantly so (P ¼ 0.15).
A high proportion of EAC was attributable to overweight and
obesity in men but not in women (PAF ¼ 41% (95%
CI: 27, 57) for men vs. 4% (95% CI: 0, 100) for women),
although again this failed to reach statistical significance
(P ¼ 0.07). In contrast, a greater proportion of EAC cases
was attributable to a history of ever smoking in women (46%,
95% CI: 23, 71) than in men (26%, 95% CI: 12, 45), but
again not significantly so (P ¼ 0.10). The negative PAFs
observed for overweight women and women with low levels
of GER are based on small numbers of cases (n ¼ 7 and
n ¼ 8, respectively), and the imprecision is reflected in the
wide confidence intervals. These estimates are unlikely to
have an impact on disease at the population level.
GEJAC. In patients with GEJAC, we found that more
cancers were attributable to ever smoking (43%, 95% CI:
32, 54) than to any other single factor, and the proportions
were similar for men and women (41% and 46%, respectively) (Table 3). Similar to the pattern seen for EAC, a nonsignificantly higher proportion of GEJAC cases was
attributable to GER for men (PAF ¼ 36%, 95% CI: 25, 50)
than for women (PAF ¼ 16%, 95% CI: 2, 67) (P ¼ 0.12). In
contrast, a higher proportion of GEJAC was attributable to
overweight and obesity in women (PAF ¼ 33%, 95% CI:
11, 66) than in men (PAF ¼ 18%, 95% CI: 6, 42), but again
this was not a significant difference (P ¼ 0.21).
Partial PAFs for combinations of risk factors
EAC. Partial PAFs for combinations of exposure to
smoking, BMI, and symptoms of GER are presented in
Table 4. The total PAF associated with all combinations of
these 3 factors in the population was 76% (95% CI: 66, 84),
and it was higher for men (78% for men vs. 59% for women),
although the difference was not significant (P ¼ 0.15). The
highest partial PAF observed was among those with all
3 factors co-occurring, that is, being overweight or obese,
having a history of smoking, and reporting acid reflux symptoms as occurring at least weekly (25%, 95% CI: 21, 30).
Among cases who were never smokers, the highest PAFs
were noted for overweight or obese people with frequent
GER symptoms (8%, 95% CI: 6, 11).
GEJAC. The total PAF of GEJAC associated with all
combinations of smoking, BMI, and symptoms of GER was
of a magnitude similar to that of EAC (69%, 95% CI: 58, 78)
Am J Epidemiol. 2011;174(5):582–590
Population Attributable Fractions of Esophageal Cancer
587
Table 5. Partial Population Attributable Fractions and 95% Confidence Intervals for Combinations of Exposure to
Smoking, Body Mass Index, and Heartburn or Acid Reflux Symptoms 10 Years Before the Study for
Adenocarcinoma of the Gastroesophageal Junction, Australia, 2002–2005
Smoking Status, Body Mass
Indexa 1 Year Prior, and Reflux
All Participants
PAFb
95% CI
Men
PAFb
Women
95% CI
PAFb
95% CI
Never smoker
Normal weight
No reflux
Reference
Reference
Reference
Reflux less than once a week
1
0.1, 3
1
0.1, 3
1
5, 3
Reflux once a week or more
1
0.6, 3
1
0.6, 3
1
0.2, 9
No reflux
1
0.4, 3
0.7
0.2, 3
4
0.9, 14
Reflux less than once a week
1
0.6, 4
1
0.4, 4
2
0.2, 17
Reflux once a week or more
4
3, 7
4
2, 6
8
3, 18
Overweight or obese
Ever smoker
Normal weight
No reflux
4
2, 6
3
2, 6
5
2, 14
Reflux less than once a week
4
2, 6
4
2, 7
4
1, 13
Reflux once a week or more
4
3, 7
5
3, 7
3
0.8, 13
Overweight or obese
No reflux
9
7, 13
9
6, 13
14
7, 26
Reflux less than once a week
16
11, 21
15
11, 22
13
6, 27
Reflux once a week or more
23
19, 28
24
19, 29
16
8, 28
Total PAF attributed to combinations
of 3 factors
69
58, 78
68
56, 78
68
44, 85
Abbreviations: CI, confidence interval; PAF, population attributable fraction.
Weight (kg)/height (m)2.
b
Population attributable fractions were adjusted for age, sex, educational level, and aspirin or nonsteroidal
antiinflammatory drug use.
a
and was equal for men and women (Table 5). Individual
contributions of the 3 factors to the total PAF followed patterns similar to those seen for EAC, although among nonsmokers, high BMI had a lower PAF for GEJAC than for
EAC.
DISCUSSION
We have estimated the PAFs of EAC and GEJAC for 3 risk
factors individually and in combination. The PAF estimates
the proportion of disease that can be attributed to a risk factor
or groups of risk factors and thus the proportion of cases that
might be avoided if those factors were eliminated from the
population. Because many risk factors for multifactorial disease co-occur and/or interact, it is important to examine the
combined effects of those factors. Overall, overweight and
obesity, smoking, and frequent GER present in various combinations in people accounted for 76% of EAC cases and 69%
of GEJAC cases, with high BMI and frequent symptoms of
GER individually accounting for the greatest proportion of
EAC, and smoking and frequent symptoms of GER individually accounting for the greatest proportion of GEJAC. When
we examined the factors in different combinations, the
Am J Epidemiol. 2011;174(5):582–590
highest proportion of disease was associated with being
simultaneously overweight or obese and a smoker with frequent symptoms of acid reflux. The co-occurrence of these 3
factors may explain 25% of all cases of EAC. All 3 factors
accounted for the same proportion of GEJAC disease in men
and women (68%), but for EAC, the proportion of disease
explained by these factors was lower for women than for
men (59% vs. 76%). However, this difference was not statistically significant. The apparent differences in PAFs
between the sexes may reflect real differences in esophageal
cancer biology between men and women, although our ability to infer is constrained by the small number of affected
women in our sample. It is possible that other sources of bias
could be operating, such as systematic misclassification and
selection bias, but why these should introduce sex-specific
differences in effect is not clear. Combining these data with
those from other, similar studies would be one means of
obtaining more precise sex-specific estimates of PAF.
To our knowledge, no investigators have previously
reported PAFs of EAC for combinations of risk factors. In
a previous study, Engel et al. (14) estimated PAFs due to
smoking, high BMI, and high frequency of GER symptoms
in a US population. Those investigators reported proportions
of disease attributable to high BMI of similar magnitude as
588 Olsen et al.
those reported here; however, frequent symptoms of GER
had PAFs that were lower for men and higher for women
than in our Australian population (28% in Australia vs. 50%
in the United States for men; 40% in Australia vs. 26% in
the United States for women). The proportion of disease
attributable to both current and former smoking was higher
for women in our Australian study. The PAF considers both
the strength of association between risk factor and outcome
and the prevalence of the factor in the community, which
may vary over time and between populations. Thus, differences in PAFs between populations will reflect the different
prevalence rates of risk factors in the populations. Although
sex-specific prevalence data were not available for the study
by Engel et al. (14), the prevalence of smoking for men and
women overall (current or past) was lower in our Australian
control population than in the US study (14) (55% vs. 67%),
mostly because of a lower prevalence of current smoking
(13% vs. 22%). The prevalences of overweight/obesity and
symptoms of GER, however, were higher in our Australian
population: 64% were overweight or obese compared with
51% in the US study population, and 56% reported any
symptoms of GER compared with 47% in the US study
population. The data used for the US study were collected
more than 15 years ago, however, and the prevalences of
obesity and reflux have increased since that time (19, 20),
whereas the prevalence of smoking has declined (21).
Although the prevalence of smoking in Australia has declined since the 1950s and this trend is expected to continue
(22), concomitant increases in the prevalence of obesity,
particularly the higher levels of obesity (23), suggest that
the benefits gained from declining smoking rates may be
outweighed by the negative effects of increasing levels of
obesity. During the period 1980–2007, the prevalence of
current smoking in Australia decreased by approximately
19% for men and 11% for women (22). Over the same
period (1980–2000), the prevalence of obesity (BMI 30)
doubled and the prevalence of class III obesity (BMI 40)
increased 4-fold (24). These increases parallel increases in
the prevalence of GER (25). Thus, if current trends continue
unabated, we might expect that the incidence of these cancers in Australia will continue to rise.
Strengths of our study included the population-based design, large number of cases, and detailed information on
multiple exposures. A limitation was the relatively low participation rate among controls (49%), which could have
resulted in selection bias and an overrepresentation of more
health-conscious control participants who were less likely to
be overweight or obese. However, the demographic characteristics of our participating controls were similar to those of
the participants in the Australian National Health Survey,
a representative survey of the Australian adult population
conducted in 2004 (26). We have also previously explored
the potential for selection bias due to nonparticipation on
risk estimates for smoking and BMI. We compared odds
ratios derived from self-reported data from participating
controls with odds ratios derived using imputed data for
nonparticipating controls; we observed only a modest level
of attenuation in risk estimates among current smokers and
no evidence of bias for associations with BMI (27). The
prevalence of at least weekly symptoms of reflux in our
population sample of controls (12% among men and 11%
among women; Table 1) was similar to prevalence estimates
from other population surveys in Australia (28), the United
Kingdom (29), and Sweden (30). We therefore consider the
likelihood of biased selection on the basis of this symptom
to be no greater than for previous studies.
Our analyses relied on retrospective self-reported height
and weight. Recall of body weight and height can be subject
to misclassification, although we observed high levels of
repeatability for these measures in this study (31). Both
men and women tend to overreport their height; men tend
to overestimate their weight, but women, particularly younger women, underreport their weight (32). Thus, BMI is
more often underestimated for women than for men. Such
nondifferential misclassification would lead to an attenuation of the true association with obesity and an underestimate of the PAF. Lastly, these estimates of PAF assume that
there is a causal association between each factor and esophageal cancer and that removal of exposure to these factors
alone will lead to a decline in incidence of esophageal cancer. It is possible, however, that there are other causal factors
that are correlated with the factors examined here and that
would continue to exert a force of morbidity even if these
primary factors were removed. Although that is possible,
our prior analyses suggest that the potential role of other
factors is much less than the factors examined here.
In summary, 3 largely modifiable risk factors occurring in
various combinations explained more than two-thirds of the
cases of EAC and GEJAC in this Australian population,
making these diseases largely preventable through maintenance of a healthy BMI, avoidance of smoking, and elimination symptomatic GER from the population.
ACKNOWLEDGMENTS
Author affiliations: Division of Population Health,
Queensland Institute of Medical Research, Brisbane, Australia
(Catherine M. Olsen, Adèle C. Green, Penelope M. Webb,
David C. Whiteman); and School of Population Health,
University of Queensland, Brisbane, Australia (Nirmala
Pandeya).
This study was supported by the Queensland Cancer Fund
and the National Health and Medical Research Council of
Australia (program 199600). David Whiteman was supported by a Future Fellowship from the Australian Research
Council. Penelope Webb and Nirmala Pandeya were supported by Research Fellowships from the National Health
and Medical Research Council of Australia. Catherine
Olsen was supported by a grant from the Xstrata Community Partnership Program, Australia.
The authors thank Dr. Harish Babu for his assistance with
pathology abstractions and Dr. Peter Baker and Prof. Gail
Williams for statistical advice.
Catherine M. Olsen and Nirmala Pandeya contributed
equally to this work.
The Australian Cancer Study: Esophageal Cancer comprised the following people—investigators: Dr. David C.
Whiteman, Dr. Penelope M. Webb, Dr. Adele C. Green,
Am J Epidemiol. 2011;174(5):582–590
Population Attributable Fractions of Esophageal Cancer
Dr. Nicholas K. Hayward, Dr. Peter G. Parsons, and
Dr. David M. Purdie; clinical collaborators: B. Mark
Smithers, Dr. David Gotley, Dr. Andrew Clouston, and Ian
Brown; project manager: Suzanne Moore; database administrators: Karen Harrap, Troy Sadkowski; research nurses:
Suzanne O’Brien, Ellen Minehan, Deborah Roffe, Sue
O’Keefe, Suzanne Lipshut, Gabby Connor, Hayley Berry,
Frances Walker, Teresa Barnes, Janine Thomas, Linda
Terry, Michael Connard, Leanne Bowes, MaryRose Malt,
and Jo White; clinical contributors, Australian Capital
Territory: Charles Moss and Noel Tait; clinical contributors,
New South Wales: Chris Bambach, Andrew Biankan, Roy
Brancatisano, Max Coleman, Michael Cox, Stephen Deane,
Gregory L. Falk, James Gallagher, Mike Hollands, Tom
Hugh, David Hunt, John Jorgensen, Christopher Martin,
Mark Richardson, Garrett Smith, Ross Smith, and David
Storey; clinical contributors, Queensland: John Avramovic,
John Croese, Justin D’Arcy, Stephen Fairley, John Hansen,
John Masson, Les Nathanson, Barry O’Loughlin, Leigh
Rutherford, Richard Turner, and Morgan Windsor; clinical
contributors, South Australia: Justin Bessell, Peter Devitt,
Glyn Jamieson, and David Watson; clinical contributors,
Victoria: Stephen Blamey, Alex Boussioutas, Richard Cade,
Gary Crosthwaite, Ian Faragher, John Gribbin, Geoff
Hebbard, George Kiroff, Bruce Mann, Bob Millar, Paul
O’Brien, Robert Thomas, and Simon Wood; clinical
contributors, Western Australia: Steve Archer, Kingsley
Faulkner, and Jeff Hamdorf.
The funding bodies played no role in the design or conduct of the study; the collection, management, analysis, or
interpretation of the data; or preparation, review, or approval
of the manuscript.
Conflict of interest: none declared.
REFERENCES
1. Armstrong RW, Borman B. Trends in incidence rates of
adenocarcinoma of the oesophagus and gastric cardia in New
Zealand, 1978–1992. Int J Epidemiol. 1996;25(5):941–947.
2. Botterweck AA, Schouten LJ, Volovics A, et al. Trends in incidence of adenocarcinoma of the oesophagus and gastric cardia
in ten European countries. Int J Epidemiol. 2000;29(4):645–654.
3. Brown LM, Devesa SS, Chow WH. Incidence of adenocarcinoma of the esophagus among white Americans by sex, stage,
and age. J Natl Cancer Inst. 2008;100(16):1184–1187.
4. van Blankenstein M, Looman CW, Siersema PD, et al. Trends
in the incidence of adenocarcinoma of the oesophagus and
cardia in the Netherlands 1989–2003. Br J Cancer. 2007;
96(11):1767–1771.
5. Hurschler D, Borovicka J, Neuweiler J, et al. Increased detection rates of Barrett’s oesophagus without rise in incidence
of oesophageal adenocarcinoma. Swiss Med Wkly. 2003;
133(37-38):507–514.
6. Pohl H, Welch HG. The role of overdiagnosis and reclassification in the marked increase of esophageal adenocarcinoma
incidence. J Natl Cancer Inst. 2005;97(2):142–146.
7. Vaughan TL, Davis S, Kristal A, et al. Obesity, alcohol, and
tobacco as risk factors for cancers of the esophagus and gastric
cardia: adenocarcinoma versus squamous cell carcinoma.
Cancer Epidemiol Biomarkers Prev. 1995;4(2):85–92.
Am J Epidemiol. 2011;174(5):582–590
589
8. Kubo A, Corley DA. Body mass index and adenocarcinomas of
the esophagus or gastric cardia: a systematic review and metaanalysis. Cancer Epidemiol Biomarkers Prev. 2006;
15(5):872–878.
9. Lagergren J, Bergström R, Lindgren A, et al. Symptomatic
gastroesophageal reflux as a risk factor for esophageal adenocarcinoma. N Engl J Med. 1999;340(11):825–831.
10. Cook MB, Kamangar F, Whiteman DC, et al. Cigarette smoking
and adenocarcinomas of the esophagus and esophagogastric
junction: a pooled analysis from the international BEACON
consortium. J Natl Cancer Inst. 2010;102(17):1344–1353.
11. Whiteman DC, Sadeghi S, Pandeya N, et al. Combined effects
of obesity, acid reflux and smoking on the risk of adenocarcinomas of the oesophagus. Gut. 2008;57(2):173–180.
12. Bruzzi P, Green SB, Byar DP, et al. Estimating the population
attributable risk for multiple risk factors using case-control
data. Am J Epidemiol. 1985;122(5):904–914.
13. Walter SD. Effects of interaction, confounding and observational error on attributable risk estimation. Am J Epidemiol.
1983;117(5):598–604.
14. Engel LS, Chow WH, Vaughan TL, et al. Population attributable risks of esophageal and gastric cancers. J Natl Cancer
Inst. 2003;95(18):1404–1413.
15. Walter SD. Attributable risk in practice. Am J Epidemiol.
1998;148(5):411–413.
16. Physical status: the use and interpretation of anthropometry.
Report of a WHO Expert Committee. World Health Organ
Tech Rep Ser. 1995;854:1–452.
17. Benichou J, Gail MH. Variance calculations and confidence
intervals for estimates of the attributable risk based on logistic
models. Biometrics. 1990;46(4):991–1003.
18. Pandeya N, Williams GM, Sadhegi S, et al. Associations of
duration, intensity, and quantity of smoking with adenocarcinoma and squamous cell carcinoma of the esophagus. Am
J Epidemiol. 2008;168(1):105–114.
19. Baskin ML, Ard J, Franklin F, et al. Prevalence of obesity in
the United States. Obes Rev. 2005;6(1):5–7.
20. Friedenberg FK, Hanlon A, Vanar V, et al. Trends in
gastroesophageal reflux disease as measured by the National
Ambulatory Medical Care Survey. Dig Dis Sci. 2010;55(7):
1911–1917.
21. Farrelly MC, Pechacek TF, Thomas KY, et al. The impact
of tobacco control programs on adult smoking. Am J Public
Health. 2008;98(2):304–309.
22. Gartner CE, Barendregt JJ, Hall WD. Predicting the future
prevalence of cigarette smoking in Australia: how low can we
go and by when? Tob Control. 2009;18(3):183–189.
23. Bjørge T, Engeland A, Tretli S, et al. Body size in relation to
cancer of the uterine corpus in 1 million Norwegian women.
Int J Cancer. 2007;120(2):378–383.
24. Walls HL, Wolfe R, Haby MM, et al. Trends in BMI of urban
Australian adults, 1980–2000. Public Health Nutr. 2010;13(5):
631–638.
25. El-Serag HB. Time trends of gastroesophageal reflux disease:
a systematic review. Clin Gastroenterol Hepatol. 2007;5(1):17–26.
26. Australian Bureau of Statistics. 2004–05 National Health
Survey: Summary of Results. (ABS publication no. 4364).
Canberra, Australia: Australian Bureau of Statistics; 2006.
27. Pandeya N, Williams GM, Green AC, et al. Do low control
response rates always affect the findings? Assessments of
smoking and obesity in two Australian case-control studies
of cancer. Aust N Z J Public Health. 2009;33(4):312–319.
28. Talley NJ, Quan C, Jones MP, et al. Association of upper
and lower gastrointestinal tract symptoms with body mass
590 Olsen et al.
index in an Australian cohort. Neurogastroenterol Motil.
2004;16(4):413–419.
29. Murray L, Johnston B, Lane A, et al. Relationship between body
mass and gastro-oesophageal reflux symptoms: The Bristol
Helicobacter Project. Int J Epidemiol. 2003;32(4):645–650.
30. Nilsson M, Johnsen R, Ye W, et al. Prevalence of gastrooesophageal reflux symptoms and the influence of age and sex.
Scand J Gastroenterol. 2004;39(11):1040–1045.
31. Karatela S, Purdie DM, Green AC, et al. Repeatability of selfreported information for population-based studies of cancer.
Asian Pac J Cancer Prev. 2006;7(2):303–308.
32. Merrill RM, Richardson JS. Validity of self-reported height,
weight, and body mass index: findings from the National Health
and Nutrition Examination Survey, 2001–2006. Prev Chronic
Dis. 2009;6(4):A121.
Am J Epidemiol. 2011;174(5):582–590