American Journal of Epidemiology
Copyright O 1999 by The Johns Hoptdns University School of Hygiene and Public Hearth
AD rights reserved
Vol. 150, No. 5
Printed in U.SA.
Risk Factors for Basal Cell Carcinoma of the Skin in Men: Results from the
Health Professionals Follow-up Study
Rob M. van D a m , u Zhiping Huang,1 Eric B. Rimm,1^4 Martin A. Weinstock,6 Donna Spiegelman,46 Graham A.
Colditz,3-4 Walter C. Willett,1-3-4 and Edward Giovannucci1*4
The authors examined the relation of constitutional factors and sun exposure to risk of basal cell carcinoma
of the skin (BCC) in a prospective cohort of 44,591 predominantly Caucasian US male health professionals,
40-75 years of age and free of cancer at enrollment in 1986. During 8 years of follow-up, 3,273 cases of selfreported BCC were documented. The following variables were each associated with an elevated risk of BCC:
having red hair; green, hazel, or blue eyes; a tendency to sunburn; and north European ancestry. The lifetime
number of blistering sunburns was also positively associated with BCC risk (p trend < 0.0001). Compared with
men who as teenagers had been outside less than once a week, men who had been outside weekly (relative
risk (RR) = 1.30; 95% confidence interval (Cl): 1.14, 1.47) and daily (RR = 1.42; 95% Cl: 1.24, 1.63) had an
elevated risk of BCC. Living in a region of residence with high solar radiation as an adult was also associated
with an increased risk of BCC (RR = 1.48; 95% Cl: 1.36,1.60), whereas living in such a region only in childhood
did not increase BCC risk. These results confirm the role of constitutional factors and suggest that adult sun
exposure increases BCC risk. Am J Epidemiol 1999; 150:459-68.
men; neoplasms, basal cell; skin neoplasms; sunburn; sunlight
In the White population, basal cell carcinoma of the
skin (BCC) is the most common form of malignancy;
it has been estimated that each year about 850,000
new cases occur in the United States (1). Most studies
suggest that the age-adjusted incidence has increased
substantially during the past two decades (2-4), which
may reflect an increase in recreational sun exposure.
A further absolute increase is expected because of the
aging of the population and greater exposure to solar
ultraviolet radiation due to depletion of the ozone
layer (5). Although the mortality due to BCC is low,
the malignancy is associated with substantial morbidity and health care costs (6).
The primary established risk factor for BCC is ultraviolet sunlight exposure. Light hair and eye color,
north European ancestry, and a skin that burns easily
and tans poorly are all associated with a higher risk of
BCC (4, 7-9). However, few studies were large
enough to elucidate the relative importance of these
factors and to estimate the magnitude of the relative
risks with reasonable precision. The association
between sunburns and BCC is less established (4,
8-11), and the importance of moles as an indicator of
BCC risk has only been investigated in one study (7).
Moreover, the effects of the pattern of sun exposure
and exposure in different periods in life are still highly
controversial. It has long been believed that the risk of
BCC was proportional to the cumulative dose of ultraviolet radiation received (10), and three case-control
studies have shown an association between total lifetime sun exposure and BCC risk (10, 12, 13).
However, this finding has not been confirmed (8, 14,
15), and some recent studies have suggested that sun
exposure during childhood (7, 9, 11) and recreational
exposure (10, 16), which is likely to be characterized
by infrequent, intense increments (10), are particularly
important for the development of BCC. In contrast to
studies of squamous cell carcinoma (17), studies of
BCC have not shown a positive association with ciga-
Received for publication July 16,1998, and accepted for publication January 22, 1999.
Abbreviations: BCC, basal cell carcinoma of the skin; Cl, confidence interval; RR, relative risk.
1
Department of Nutrition, Harvard School of Public Health,
Boston, MA.
2
Division of Human Nutrition and Epidemiology, Wageningen
Agricultural University, Wageningen, Netherlands.
3
Channing Laboratory, Department of Medicine, Brigham and
Women's Hospital and Harvard Medical School, Boston, MA.
4
Department of Epidemiology, Harvard School of Public Hearth,
Boston, MA.
5
Dermatoepidemiology Unit, VA Medical Center, and Department
of Dermatology, Rhode Island Hospital and Brown University,
Providence, Rl.
6
Department of Biostatistics, Harvard School of Public Health,
Boston, MA.
Reprint requests to Dr. Edward Giovannucci, 665 Huntington
Ave., Boston, MA 02115.
459
460
van Dam et al.
rette smoking (8, 18, 19). Most of the evidence on risk
factors for BCC has been derived from case-control
studies (4), in which recall bias is a potential problem.
Of the three prospective studies that have been conducted, only one study did not have substantial loss to
follow-up (4), and that study included only women (8).
We examined the effect of constitutional factors, sun
exposure, and smoking on the risk of BCC in a large
prospective study of predominantly Caucasian US
men.
MATERIALS AND METHODS
Study population
The Health Professionals Follow-up Study started in
1986 when 51,529 male health professionals completed
a questionnaire on medical history and known and suspected risk factors for cancer and other major illnesses
(20). The study population included 29,683 dentists,
10,098 veterinarians, 4,185 pharmacists, 3,745
optometrists, 2,218 osteopathic physicians, and 1,600
podiatrists, who were 40-75 years of age (mean, 54.7
years) in 1986. Only 1.7 percent of cohort members
were of Asian origin, and 1.0 percent were of AfricanAmerican origin. Exposure or reporting of exposure
may be affected by a diagnosis of cancer. Therefore, we
excluded men who reported any cancer (including
squamous cell carcinoma or a previous basal cell carcinoma) at baseline and at the beginning of each 2-year
time period. After exclusion, 44,591 men remained for
the analyses at baseline.
Assessment of exposure
On the 1986 questionnaire, men were asked about
their major ancestry (southern European/Mediterranean,
Scandinavian, other Caucasian, African American,
Asian/Oriental, other origin). On a special 1987 questionnaire, men were asked to count the number of moles
on their forearms ("Please record the total number of
moles on both of your forearms between your elbow
and your wrist, of 3 mm. diameter or larger"). On the
1988 questionnaire, subjects were asked to describe
their natural hair color ("Which of the following categories most closely describes the natural color of your
hair at age 18?" (red, dark brown, blond, black, light
brown)) and their eye color ("Which of the following
categories most closely describes the color of your
eyes?" (brown/dark, hazel/green/medium, blue/light)).
The 1992 questionnaire included questions about lifetime number of blistering sunburns ("How many times
in your life have you had a sunburn that blistered?"
(never, 1-2 times, 3-5 times, 6-9 times, 10 or more
times)), sun exposure as a teenager ("During the sum-
mer as a teenager, on average, how many times per
week were you outdoors in a swimsuit?" (< I/week,
I/week, 2/week, several/week, daily)), and skin reaction
to sun as an adolescent ("As an adolescent, at the beginning of the summer, if you were out in the sun for the
first time and were to stay out for 1 hour without sunscreen, would you: painfully burn then peel, burn then
tan, or tan without burning?"). Height was assessed at
baseline, and weight and cigarette smoking habits were
assessed on each questionnaire. Body mass index
(kg/m2) was calculated for each period. The state of residence was derived from the 1987 mailing list On the
1992 questionnaire, the participants also reported thenstate of residence at birth and at 15 and 25 years of age.
We grouped states into three regions according to the
level of sun exposure, based on the annual mean daily
solar radiation (which includes ultraviolet and visible
light) in each state, which was obtained from US
Weather Bureau data (21).
Identification of basal cell carcinoma
On the questionnaires mailed to all study participants in 1988, 1990, 1992, and 1994, men were asked
whether basal cell carcinoma had been diagnosed during the previous 2 years. After repeated mailings, the
follow-up rate averaged 94 percent. Deaths were
reported by family members, coworkers, and postal
authorities, or they were identified through systematic
searches of the National Death Index. To assess the
validity (positive predictive value) of self-report of
BCC in this study population, we requested permission
to obtain medical records from a sample of 109 participants who had reported a diagnosis of BCC. For 19
men, no records were obtained: 13 did not return the
request form, one could not be contacted, and five men
did not give permission to review the record. Of the
remaining 90 men, nine denied the diagnosis, five selfreports were not confirmed by medical records, and 76
self-reports were confirmed by medical records (70
percent of the total sample, 84 percent of confirmable
self-reports).
Statistical analysis
Age-adjusted analysis was based on incidence rates
of BCC, in person-months of follow-up. Men contributed follow-up time from the date of return of the
1986 questionnaire until BCC was diagnosed, the
occurrence of death from any cause, or December 31,
1993, whichever came first. Relative risks were calculated by dividing the incidence rate of BCC among
men in a given category of exposure by the rate in a
specified reference category. Age-adjusted relative
risks were obtained by stratified analysis using the
Am J Epidemiol
Vol. 150, No. 5, 1999
Risk Factors for Basal Cell Carcinoma in Men
Mantel-Haenszel estimator (22). Linear trends were
tested with Breslow and Day's statistic (23).
To adjust for other variables, a pooled logistic
regression approach using 2-year follow-up intervals,
analogous to survival analysis, was used to calculate
odds ratios to estimate relative risks (24). In the multivariate analysis, exposure variables were modeled as
continuous variables to test for linear trends. Tests for
statistical interaction were conducted by including
cross-product terms of the variables of interest in a
multivariate logistic regression model. For men who
did not report data for a given covariate, indicator variables for missing values were included in the multivariate models. The non-Caucasian categories of the
variable "major ancestry" were combined, because the
number of men of Asian or African-American origin in
this cohort was small. Information about cigarette
smoking and body mass index was updated in each
time period of the analysis. All reported p values are
for two-sided tests.
To assess whether residual confounding due to missing values for covariates had affected the results, we
performed a separate analysis excluding all men with
missing data. In another analysis, to evaluate the possibility of detection bias, we excluded all men at the
beginning of each 2-year period who had not had a
routine physical examination in that period or in any of
the following 2-year periods.
RESULTS
During 317,672 person-years of follow-up, 3,273
cases of BCC were diagnosed. The overall crude incidence rate was 1,030 cases per 100,000 person-years.
Age was strongly related to die incidence rate of BCC,
rising from 550 cases per 100,000 person-years among
men aged 40-44 years to 2,340 cases per 100,000 person-years among those older then 70 years.
The age-adjusted and multivariate relative risks of
BCC according to constitutional variables are shown
in table 1. As compared with men with black hair,
those with lighter shades of hair were associated with
an increased risk of BCC. A multivariate-adjusted relative risk was calculated to determine whether the
observed associations were independent from other
risk factors for BCC, including other constitutional
variables. Associations with hair color were substantially attenuated by multivariate adjustment, which
was almost solely due to allowance for other constitutional factors. Only the relative risk for red hair color
remained significant (relative risk (RR) = 1.46; 95 percent confidence interval (CI): 1.15, 1.86). Relative to
men with brown eyes, men with blue, hazel, or green
eyes were at elevated risk for BCC. After allowance
for other risk factors, however, only the association
Am J Epidemiol Vol. 150, No. 5, 1999
461
with hazel or green eyes remained significant (RR =
1.19; 95 percent CI: 1.07, 1.32).
The skin's reaction to 1 hour of sun exposure during
adolescence, at the beginning of the summer, was
strongly associated with BCC. The age-adjusted relative risks for the skin reactions "burn then tan" and
"painfully burn then peel" compared with "tan without
burning" were 1.62 (95 percent CI: 1.47, 1.79) and
2.29 (95 percent CI: 2.06, 2.54), respectively.
Estimates were similar after multivariate adjustment.
Because sunburns reflect the skin's reaction to sun (in
combination with exposure to sun), "lifetime number
of sunburns" was not included in the multivariate
model.
As compared with men of north European origin,
men of south European origin had an age-adjusted relative risk of 0.79 (95 percent CI: 0.72, 0.86).
Participants with a non-Caucasian ancestry had a relative risk of 0.42 (95 percent CI: 0.33,0.54). These estimates were not materially altered by multivariate
adjustment.
To evaluate the usefulness of combinations of
reported skin reaction to sun, hair color, and eye color
to identify persons at high risk for BCC, relative risks
were calculated for men with several of these factors.
In general, all three variables added somewhat to the
distinction between subjects at higher and lower risk
(table 2). The relative risk for men with hazel, green,
or blue eyes, red or blond hair, and a tendency to
painfully sunburn and peel in adolescence was 3.08
(95 percent CI: 2.51, 3.79), relative to men with dark
hair, brown eyes, and no tendency to burn. We
observed a significant interaction between age and sun
sensitivity (p values for interaction < 0.05). In the
older age group (age ^ 60 years), the relative risk was
2.06 (95 percent CI: 1.52, 2.80) for highly sun-sensitive men (highest risk category in table 2) relative to
men with a low sun sensitivity (reference category in
table 2). The relative risk was 1.59 (95 percent CI:
1.29, 1.96) for men with moderate sun sensitivity
(intermediate categories in table 2). Among younger
men (age < 60 years), sun sensitivity as measured by
hair and eye color and skin reaction to sun was a
stronger predictor of BCC occurrence; compared with
men with a low sun sensitivity, highly sun-sensitive
men had a relative risk of 4.20 (95 percent CI: 3.5,
5.59), and men with a moderate sun sensitivity had a
relative risk of 2.26 (95 percent CI: 1.80, 2.84).
The risk of BCC according to variables that reflect a
combination of constitutional factors and sun exposure
is shown in table 3. The lifetime number of blistering
sunburns was related to BCC risk in a dose-response
manner. Men with a history of more than 10 sunburns
had an age-adjusted relative risk of 2.06 (95 percent
462
van Dam et aJ.
TABLE 1. Relative risk (RR) of basal cell carcinoma by constitutional variables, Hearth Professionals
Follow-up Study, 1986-1994*
Variable
Natural hair color
Black
Dark brown
Ught brown
Blond
Red
No.
of
cases
No.
of
person-years
240
1,029
914
324
106
28,754
104,961
77,814
25,417
5,600
Test for trend
Eye color
Brown
Hazel/green
Blue
746
896
967
85,988
75,404
81,042
Test for trend
Skin reaction to sun as an adolescent
Tan, not burn
Bum, then tan
Painfully bum, then peel
583
1,242
888
81,453
111,673
56,240
Test for trend
Major ancestry
North European
South European
Non-Caucasian
2,436
599
71
217,111
69,685
14,928
Test for trend
Age-adjusted
RR
1
1.31 (1.13,
1.54(1.33,
1.65(1.40,
2.37(1.89,
1.50)$
1.78)
1.96)
2.98)
MuWvariateadjusted RRt
1
1.08(0.93,
1.15(0.99,
1.13(0.94,
1.46(1.15,
1.25)
1.34)
1.35)
1.86)
p < 0.0001
p = 0.004
1
1.41 (1.28, 1.55)
1.38(1.25, 1.51)
1
1.19(1.07, 1.32)
1.10(0.98, 1.22)
p < 0.0001
p=0.11
1
1.62(1.47, 1.79)
2.29 (2.06, 2.54)
1
1.51 (1.37, 1.67)
2.13(1.90,2.38)
p < 0.0001
p < 0.0001
1
0.79 (0.72, 0.86)
0.42 (0.33, 0.54)
1
0.87 (0.79, 0.95)
0.53 (0.42, 0.68)
p < 0.0001
p < 0.0001
* Data were missing for 660 cases and 75,126 person-years of follow-up for hair color, 664 cases and 75,238
person-years of follow-up for eye color, 560 cases and 68,307 person-years of follow-up for skin reaction to sun,
and 167 cases and 15,950 person-years of follow-up for ancestry.
t Model included age, time period, hair color, eye color, skin reaction to sun, ancestry, number of moles on
forearms, region of residence, lifetime number of sunburns, exposure to sun as a teenager, and body mass index.
A separate model for the multlvariate adjustment of skin reaction to sun was used, which did not Include lifetime
number of sunburns.
X Numbers in parentheses, 95% confidence interval.
TABLE 2. Relative risk* of basal cell carcinoma by skin reaction to sun In different pigmentary traits, Health Professionals
Follow-up Study, 1986-1994
Skin
reaction
Tan, not bum
Bum, then tan
Painfully bum, then peel
Brown eyes
Dark
hair
1
1.68(1.39,2.05)
1.87(1.44,2.44)
LJghi brown
hair
1.31 (0.89, 1.93)t
1.75(1.29,2.36)
2.87(2.05,4.01)
Hazel/green/blue eyes
Red/blond
hair
Dark
hair
Light brown
hair
Red/blond
hair
1.95(0.79,4.81)
1.72(0.80,3.69)
2.55(1.54,4.23)
1.24(0.99, 1.55)
2.03(1.69,2.44)
2.60(2.10,3.22)
1.32(1.04, 1.68)
1.99(1.66,2.40)
2.96 (2.43, 3.59)
1.34(0.92, 1.96)
1.93(1.53,2.43)
3.08(2.51,3.79)
• Adjusted for age, time period, region of residence, and sun exposure as a teenager,
t Numbers In parentheses, 95% confidence Interval.
CI: 1.81, 2.36) compared with men who had never had
a sunburn. Additional adjustment for constitutional
variables attenuated this association. However, sunburns already reflect sensitivity to sun, and the ageadjusted relative risk probably gives a better indication
of the public health impact of sunburns. Other variables reflecting sun exposure were not included in the
multivariate models for number of sunburns and moles
to avoid "over-control."
The age-adjusted relative risk of having one or two
moles on the forearms was 1.28 (95 percent CI: 1.15,
1.42) compared with men who had no moles on their
forearms. Having more than two moles on the forearms was not associated with a further increase of risk.
Am J Epidemiol Vol. 150, No. 5, 1999
Risk Factors for Basal Cell Carcinoma in Men
463
TABLE 3. Relative risk (RR) of basal cell carcinoma by variables reflecting a combination of
constitutional factors and sun exposure, Health Professionals IFollow-up Study, 1986-1994*
Variable
Lifetime no. of blistering sunbums
None
1-2
3-5
6-9
£10
No.
of
cases
No.
of
person-years
Age-adjusted
RR
Muttivariateadjusted RRt
348
616
641
403
715
44,034
62,585
59,648
32,927
51,336
1
1.26(1.10, 1.44)$
1.45(1.27, 1.65)
1.72(1.49,1.99)
2.06(1.81,2.36)
1
1.14(1.00,1.30)
1.20(1.05, 1.38)
1.33(1.14,1.54)
1.49(1.30,1.71)
p < 0.0001
p < 0.0001
1
1.28(1.15,1.42)
1.33(1.15, 1.54)
1.24(1.04, 1.49)
1
1.27(1.14, 1.41)
1.29(1.11, 1.50)
1.20(1.00, 1.45)
p < 0.0001
p < 0.0001
Test for trend
No. of moles on forearms
None
1-2
3-5
1,383
491
200
129
Test for trend
140,427
39,539
14,727
10,058
* Data were missing for 550 cases and 67,142 person-years of follow-up for lifetime number of sunbums and
for 1,070 cases and 112,921 person-years of follow-up for number of moles on forearms.
t Models included age, time period, hair color, eye coior, ancestry, skin reaction to sun, and body mass index.
$ Numbers in parentheses, 95% confidence interval.
Similar relative risks were observed after allowance
for other variables, indicating that having moles on the
forearms was an independent predictor of BCC. To
evaluate further the extent to which assessing the number of moles improves the prediction of BCC occurrence, information about moles was added to the highest category of BCC risk as shown in table 2. Men with
hazel, green, or blue eyes, red or blond hair, a tendency
to painfully sunburn and peel in adolescence, and one
or more moles on their forearms had a relative risk of
4.01 (95 percent CI: 2.85, 5.64) relative to men with
brown eyes, dark hair, no tendency to sunburn in adolescence, and no moles on their forearms. The incidence rate in this low-risk reference category was 559
cases per 100,000 person-years.
In table 4, the association between sun exposure
variables and BCC is shown. Compared with men who
as teenagers had been outside less than once a week,
men who had been outside weekly (RR = 1.30; 95 percent CI: 1.14, 1.47) and daily (RR = 1.42; 95 percent
CI: 1.24, 1.63) had an elevated multivariate-adjusted
risk of BCC. The multivariate model for frequency of
having been outdoors in a swimsuit as a teenager and
potential sun exposure according to region did not
include lifetime number of sunbums and number of
moles, because these variables also reflect sun exposure. However, the region of residence and frequency
of having been outside in a swimsuit as a teenager
were corrected for each other to separate the different
natures of exposure.
Men who were living in a region of relatively high
solar radiation had a multivariate-adjusted relative risk
Am J Epidemiol
Vol. 150, No. 5, 1999
of 1.48 (95 percent CI: 1.36, 1.60) compared with men
living in a region of low solar radiation. Living in a
region with moderate solar radiation was not associated with an increased risk of BCC. To assess the relative risks of potential sun exposure according to the
region of residence in different periods of life, we created the variable "region over lifetime." Men who had
lived only in a "high sun region" in childhood (at birth
and at age 15, not currently) did not have a higher risk
of BCC compared with men who had "always" (at
birth, at age 15, at age 25, and currently) lived in a
"low sun region" (RR = 1.04; 95 percent CI: 0.77,
1.39). Modeling the same variable, with the region of
residence in adolescence instead of childhood, yielded
similar results; men who had lived in a region of high
solar radiation at age 15 and age 25 (but not currently)
had a relative risk of 1.16 (95 percent CI: 0.83, 1.63)
compared with men who had "always" lived in a
region with low solar radiation.
No clear association existed between cigarette
smoking and BCC (table 5). A decrease in risk was
observed with higher quintiles of body mass index
(table 5). Men in the highest quintile of body mass
index had an age-adjusted relative risk of 0.79 (95 percent CI: 0.70, 0.90) relative to men in the lowest quintile. Allowance for other risk factors did not appreciably attenuate this association. Body mass index at age
21 was not significantly associated with BCC {p trend
= 0.15; multivariate relative risk of the highest versus
the lowest quintile = 0.94; 95 percent CI: 0.84, 1.06).
The three variables assessed on the 1992 questionnaire (i.e., lifetime number of blistering sunbums, skin
464
van Dam et al.
TABLE 4. Relative risk (RR) of basal cell carcinoma by sun exposure variables, Health Professionals
Follow-up Study, 1986-1994*
Variable
Frequency outdoors in swimsuit
as a teenager in the summer
<1 time/week
1 time/week
2 times/week
Several times/week
Daily
No.
of
cases
557
425
499
err
347
No.
of
person-years
59,338
36,779
43,124
78,265
31,307
Test for trend
Region of residence§
Low sun
Medium sun
High sun
1,146
823
1,299
124,621
90,891
101,572
Test for trend
Region over lifetime!]
Always low sun
Always high sun
Only currently high sun
High sun in childhood
Others
682
392
344
48
1,072
68,102
30,733
20,587
5,356
108,568
Age-adjusted
RR
1
1.28(1.13, 1.45)t
1.29(1.15, 1.46)
1.26(1.13, 1.40)
1.29(1.13, 1.48)
Multivarlateadjusted RFtt
1
1.30(1.14,
1.34(1.19,
1.36(1.22,
1.42(1.24,
1.47)
1.52)
1.52)
1.63)
p = 0.0001
p < 0.0001
1
1.00(0.91, 1.09)
1.44(1.33, 1.55)
1
1.01 (0.92, 1.11)
1.48(1.36, 1.60)
p < 0.0001
p < 0.0001
1
1.42(1.26,1.62)
1.55(1.36,1.76)
1.04(0.77, 1.39)
1.09(0.99, 1.20)
1
1.46(1.29, 1.66)
1.60(1.40, 1.83)
1.08(0.80, 1.45)
1.09(0.99, 1.20)
• Data were missing for 568 cases and 68,860 person-years of follow-up for exposure to sun as a teenager,
five cases and 589 person-years of follow-up for region of residence, and 735 cases and 84,327 person-years of
follow-up for region over lifetime.
f Models included age, time period, hair color, eye color, skin reaction to sun exposure, ancestry, and body
mass index. The multivariate model of exposure to the sun as a teenager also included region of residence, and
the models for region of residence and region over lifetime also included exposure to the sun as a teenager.
X Numbers in parentheses, 95% confidence interval.
§ Levels of sun exposure were categorized by annual mean daily solar radiation (D. M. Freedman, S. H. Zahm,
and M. Dosemeci. BMJ 1997;314:1451-5). "Low sun" included the following states and other areas: Alaska,
Connecticut, Maine, Massachusetts, Michigan, Minnesota, New Hampshire, New York, Ohio, Oregon,
Pennsylvania, Rhode Island, Vermont, Washington, and Wisconsin. "Moderate sun" included the following:
Arkansas, Delaware, District of Columbia, Idaho, Illinois, Indiana, Iowa, Kansas, Kentucky, Maryland, Missouri,
Montana, Nebraska, New Jersey, North Carolina, North Dakota, South Dakota, Tennessee, Virginia, and West
Virginia. "High sun" included the following: Alabama, Arizona, California, Colorado, Florida, Georgia, Hawaii,
Louisiana, Mississippi, Nevada, New Mexico, Oklahoma, South Carolina, Texas, Utah, Wyoming, Puerto Rico, and
the Virgin Islands.
Tl Region at birth, at age 15 years, and at age 25 years and current region of residence were assessed. The
"high sun in childhood" category Included men who were bom in a high sun region and lived there at age 15 years,
but whose current region of residence did not have high solar radiation.
reaction to sun in adolescence, frequency of being outside in a swimsuit in the summer) could only be studied prospectively for part of the follow-up period. To
evaluate the possibility of recall bias in the assessment
of the variables of the 1992 questionnaire, separate relative risks were calculated for the period before
(1986-1992) and after (1992-1994) the return of the
1992 questionnaire (table 6). Although the associations
were qualitatively the same for the two periods, the
relative risks for higher categories of "lifetime number
of blistering sunburns" and "skin reaction to sun in
adolescence" were higher for the period before 1992.
The relative risk for teenage sun exposure was essentially the same in the two periods.
Because the variables assessed on the 1992 questionnaire were also included in models for multivariate
adjustment, estimates for other variables might also be
affected. Only the relative risks of hair and eye color
were substantially changed by adjustment for 1992
variables. Therefore, separate multivariate relative
risks were calculated for the period after 1992. For red
hair, hazel or green eyes, and blue eyes, the resulting
estimates were somewhat higher than the relative risks
for the whole study period. The relative risks for dark
brown, light brown, blond, and red hair were 1.05 (95
percent CI: 0.79, 1.39), 1.01 (95 percent CI: 0.75,
1.36), 1.12 (95 percent CI: 0.79, 1.59), and 1.72 (95
percent CI: 1.08, 2.72), respectively, as compared with
Am J Epidemiol
Vol. 150, No. 5, 1999
Risk Factors for Basal Cell Carcinoma in Men
465
TABLE 5. Relative risk (RR) of basal cell carcinoma by cigarette smoking and body mass Index (BMI),
Health Professionals Follow-up Study, 1986-1994*
Variable
Cigarette smoking
Never
Past
1-14/day
15-24/day
^25/day
No.
of
cases
No.
of
person-years
Age-adjusted
RR
MuttivartateadJustedRRt
1,424
1,519
98
70
67
147,185
134,220
8,736
8,783
8,127
1
1.03(0.96, 1.11)$
1.12 (0.92, 1.38)
0.80 (0.63, 1.02)
0.86(0.67, 1.10)
1
1.03(0.97, 1.11)
1.14(0.93, 1.40)
0.82 (0.65, 1.05)
0.86(0.67,1.10)
p=0.30
p = 0.38
Test for trend
Quintiles of BMI§
Low
2
3
4
High
590
585
573
543
481
51,659
52,546
52,900
52,542
54,066
Test for trend
1
0.99(0.88,
0.95(0.84,
0.91 (0.81,
0.79 (0.70,
1.11)
1.06)
1.02)
0.90)
p = 0.0001
1
0.99 (0.88, 1.11)
0.95(0.84, 1.07)
0.91 (0.81,1.03)
0.80(0.71,0.91)
p = 0.0002
* Data were missing for 95 cases and 10,621 person-years of follow-up for cigarette smoking and for 501
cases and 53,960 person-years of follow-up for body mass index.
t Models included age, time period, period of residence, hair color, eye color, skin reaction to the sun, ancestry, number of moles on forearms, lifetime number of sunburns, exposure to the sun as a teenager, and body mass
index.
t Numbers in parentheses, 95% confidence interval.
§ At baseline, the median values (in kg/m2) for quintile groups of body mass index were as follows: 22.1, 23.7,
25.1,26.6, and 29.5.
TABLE 6. Relative risk (RR) of basal cell carcinoma by sun
exposure and sun sensitivity variables as measured on the
1992 questionnaire, for the period before and after 1992,
Health Professionals Follow-up Study, 1986-1994
Variable
Skin reaction to sun In
adolescence
Tan, not bum
Bum, then tan
Painfully bum, then peel
Test for trend
Lifetime no. of blistering
sunburns
0
1-2
3-5
6-9
£10
Test for trend
Frequency outdoors in
swimsjjii as a teenager
In the summer
<1 time/week
1 time/week
2 times/week
Several times/week
Daily
Test for trend
Age-adjusted RR
for the
196&-1992
period
Age-adjusted RR
1.68(1.50, 1.89)*
2.36 (2.09, 2.67)
p < 0.0001
1.46(151,1.76)
2.10(1.72,2.57)
1
1
1.06(0.83, 1.37)
1.34(1.05, 1.72)
1.37(1.04, 1.82)
1.69(1.32,2.18)
p < 0.0001
1.34(1.15, 1.56)
1.50(1.28, 1.75)
1.88(1.58,2.23)
2.22(1.90,2.60)
p< 0.0001
for the
1992-1994
period
p < 0.0001
black hair. As compared with men with brown eyes,
men with hazel or green eyes had a relative risk of 1.36
(95 percent CI: 1.11, 1.68), and men with blue eyes
had a relative risk of 1.30 (95 percent CI: 1.04, 1.61).
The results after exclusion of participants with missing data were essentially unchanged, with the exception of the associations observed between ancestry and
BCC, which were substantially weakened. Compared
with men of north European origin, men of south
European origin had a relative risk of 0.96 (95 percent
CI: 0.84, 1.11), and men of non-Caucasian origin had
a relative risk of 0.74 (95 percent CI: 0.50, 1.10).
Exclusion at the beginning of each 2-year period of
men who had not had a routine physical examination
in that period or any of the following 2-year periods
did not materially alter any of the results.
DISCUSSION
1
1.23(1.06, 1.43)
1.31 (1.14, 1.51)
1.27(1.12, 1.43)
1.27(1.08, 1.48)
p = 0.0005
* Numbers in parentheses, 95% confidence interval.
Am J Epidemiol Vol. 150, No. 5, 1999
1
1.40(1.10, 1.78)
154(0.97, 1.57)
156(1.02, 1.56)
1.36(1.05, 1.77)
p<0.05
In this prospective study of 44,591 men and 3,273
newly diagnosed cases of BCC, we confirm the role of
red hair, lightly pigmented eyes, north European
ancestry, and a tendency to sunburn easily in predisposing to BCC. In addition, moles on the forearms and
lifetime number of blistering sunburns were independent risk factors for BCC. Men with a low recreational
sun exposure as a teenager, measured by the frequency
466
van Dam et al.
of being outside in a swimsuit in the summer, had a
reduced risk of BCC. High potential sun exposure
according to region of residence was associated with
an elevated risk. However, men who had only lived in
a region of high solar radiation in childhood did not
have a substantially increased risk of BCC.
As the exposure variables and occurrence of disease
were assessed by self-reports, the possible effect of
misclassification has to be considered. The comparison of the self-reports of BCC with medical records
indicated that self-report of BCC is a valid measure in
this medically knowledgeable population. Similar
questions about constitutional factors have been
shown to correlate significantly with the minimal erythema dose of ultraviolet B radiation required to produce a visibly reddened skin, the most widely accepted
objective measure of sun sensitivity (25). In addition,
a high reproducibility of these questions was observed
in a study of 386 women (26). Nondifferential misclassification of exposure and BCC is likely to have
resulted in somewhat attenuated relative risks, but in a
study of this size it seems unlikely that substantial
associations would have been missed. The high rate of
follow-up in this study reduced the potential bias due
to loss to follow-up.
The variables "lifetime number of sunburns" and
"skin tendency to sunburn," which were only studied
prospectively for part of the study period, appear to be
prone to some recall bias. The higher relative risk estimates for the period in which recall bias could have
occurred seem to indicate that cases tended to overestimate their number of sunburns and their tendency to
sunburn. This agrees with results from Weinstock et al.
(26), who observed recall bias among melanoma
patients in the assessment of tanning ability. Hence,
results from case-control studies concerning these
variables should be interpreted with caution. Hair and
eye color were assessed in 1988, and the number of
moles on the forearms was assessed in 1987. However,
for these variables, results yielded by analyses
restricted to the period of prospective assessment were
not appreciably different from the presented results
(data not shown).
Southern European ancestry was clearly protective
against BCC, as has been observed previously (7).
Although the initial analysis suggested that this association was independent from other constitutional factors, the analysis with exclusion of men with missing
data did not show an independent association.
Associations with hair color, eye color, and skin sensitivity to the sun were similar to those observed in other
studies (4, 7-9, 11); each of the factors contributed to
the prediction of BCC. This agrees with the observation that a combination of hair color and self-reported
skin reaction to sun is a better predictor of the minimal
erythema dose of ultraviolet B radiation than selfreported skin reaction alone (25). We are unaware of
prior data on the interaction between sun sensitivity
and age with respect to BCC risk. Our results indicate
that sun sensitivity, according to hair and eye color and
skin reaction to sun, is a substantially stronger predictor of BCC risk in men younger than 60 years of age
than in older men. Still, incidence rates among
Caucasian men with a low risk of BCC according to
constitutional factors (i.e., dark hair, brown eyes, no
tendency to sunburn) were high, and preventive intervention seems justified for the whole population.
The association between moles and BCC has been
studied previously only once: Kricker et al. (7)
observed an increased risk of BCC in subjects with
more than four moles on the back, hi the present data,
the association between moles on the forearms and
BCC was independent of other risk factors. Although
we cannot exclude the possibility that solar lentigines
were misclassified as moles, serf-reported information
about moles is potentially useful for the identification
of men at high risk of BCC.
Although recall bias seemed to lead to an overestimation of BCC risk according to number of sunburns,
the prospective estimates we obtained were qualitatively the same as those for the whole period and
remained statistically significant. This confirms the
results from the prospective Nurses' Health Study, in
which the lifetime number of severe and painful sunburns on the arms or face was positively associated
with BCC in a dose-response manner; the age-adjusted
relative risk for six or more burns was 2.91 (95 percent
CI: 2.37, 3.58) compared with women without sunburns (8). In several case-control studies, sunburns
were also associated with an increased risk of BCC
(9-11), but this association was less clear and was possibly affected by recall bias.
A remarkable finding of our study was that recreational sun exposure in childhood and potential sun
exposure according to region of residence in adulthood
were associated with BCC, whereas living in a high
solar radiation region in childhood did not substantially increase risk. One may argue that the association
of BCC with region of residence in adulthood is due to
a greater emphasis on detection and accurate diagnosis
of BCC in the part of the United States with a higher
solar radiation compared with regions with a lower
solar radiation (4). This is plausible because BCC is
typically a slow-growing asymptomatic tumor.
However, there are no data supporting the existence of
actual regional differences in the detection of BCC
within the United States. Furthermore, results were not
appreciably different after excluding men without rouAm J Epidemiol
Vol. 150, No. 5, 1999
Risk Factors for Basal Cell Carcinoma in Men
tine physical examinations, suggesting that detection
bias is not a likely explanation of the observed association. Associations between potential sun exposure
according to region and BCC risk were similar in studies including other populations and other definitions of
areas within the United States (8, 27), reducing the
likelihood that the observed association was due to a
confounding factor specifically for our study population or categorization of states.
An alternative explanation is that the pattern of sun
exposure during childhood is particularly important. It
has been hypothesized that intermittent sun exposure
characterized by infrequent, intense increments
increases BCC risk more than a similar dose delivered
more continuously (10). In earlier studies, recreational
sun exposure (11, 16), sunburns (9, 11), and the proportion of weekly sun exposure obtained in the weekend (10) during childhood were associated with
increased risk of BCC. These variables probably
reflect intermittent sun exposure, whereas potential
sun exposure according to US region is likely to reflect
a higher cumulative exposure. During childhood only
intermittent sun exposure may be relevant to BCC,
whereas in adulthood a higher cumulative exposure
might also independently increase risk, as these periods of life may reflect different stages of carcinogenesis. However, this does not seem to be consistent with
a case-control study in which persons who migrated
from a country with low ambient solar radiation to
Australia after 10 years of age were at reduced risk of
BCC compared with persons who were born in
Australia (7). Research with a more detailed assessment of sun exposure is required to further elucidate
how exposure in different periods in life and the nature
of sun exposure are related to BCC.
We observed an inverse association between current
body mass index and BCC. This agrees with results
from an Australian case-control study with 51 cases
(28). The difference in sun exposure and clothing
habits between persons with a lower and a higher body
mass index may be an explanation for this finding. The
lack of association between body mass index at age 21
and BCC could be another indication for the importance of adult sun exposure. Cigarette smoking did not
increase the risk of BCC, which is consistent with the
findings of earlier studies (8, 18, 19).
Men with red hair, hazel, green, or blue eyes, moles,
and a tendency to sunburn should be aware of their
increased risk of BCC. However, our results suggest that
all Caucasian US men should be advised to limit their
sun exposure and avoid sunburns. Although our findings
seem to stress the relative importance of sun exposure
during adulthood, reducing recreational sun exposure as
a teenager may also be relevant for risk of BCC.
Am J Epidemiol Vol. 150, No. 5, 1999
467
ACKNOWLEDGMENTS
Supported by research grants CA 55075 and HL 35464
from the National Institutes of Health and special institution
grant 18 from the American Cancer Society.
The authors are indebted to Al Wing, Mira Kaufman,
Mildred Wolff, Elizabeth Frost-Hawes, Kathleen Markham,
Kerry Demers, and Jill Arnold for expert help.
REFERENCES
1. Miller DL, Weinstock MA. Nonmelanoma skin cancer in the
United States: incidence. J Am Acad Dermatol 1994;30:774-8.
2. Gallagher RP, Ma B, McLean DI, et al. Trends in basal cell carcinoma, squamous cell carcinoma, and melanoma of the skin
from 1973 through 1987. J Am Acad Dermatol 1990;23:
413-21.
3. Fears TR, Scotto J. Changes in skin cancer morbidity between
1971-72 and 1977-78. J Natl Cancer Inst 1982;69:365-70.
4. Kricker A, Armstrong BK, English DR. Sun exposure and nonmelanocytic skin cancer. Cancer Causes Control 1994;5:
367-92.
5. Slaper H, Velders GJM, Daniel JS, et al. Estimates of ozone
depletion and skin cancer incidence to examine the Vienna
Convention achievements. Nature 1996;384:256-8.
6. Preston DS, Stern RS. Nonmelanoma cancers of the skin. N
Engl J Med 1992;327:1649-62.
7. Kricker A, Armstrong BK, English DR, et al. Pigmentary and
cutaneous risk factors for non-melanocytic skin cancer. A casecontrol study. Int J Cancer 1991;48:650-62.
8. Hunter DJ, Colditz GA, Stampfer MJ, et al. Risk factors for
basal cell carcinoma in a prospective cohort of women. Ann
Epidemiol 1990;l:13-23.
9. Zanetti R, Rosso S, Martinez C, et al. The multicentre south
European study 'Helios'. I: Skin characteristics and sunburns
in basal cell and squamous cell carcinomas of the skin. Br J
Cancer 1996;73:1440-6.
10. Kricker A, Armstrong BK, English DR, et al. Does intermittent
sun exposure cause basal cell carcinoma? A case-control study
in Western Australia. Int J Cancer 1995;60:489-94.
11. Gallagher RP, Hill GB, Bajdik CD, et al. Sunlight exposure,
pigmentary factors, and risk of nonmelanocytic skin cancer. I.
Basal cell carcinoma. Arch Dermatol 1995;131:157—63.
12. Gellin GA, Kopf AW, Garfinkel L. Basal cell epithelioma. A
controlled study of associated factors. Arch Dermatol 1965;
91:38-45.
13. Urbach F, Rose DB, Bonnem M. Genetic and environmental
interactions in skin carcinogenesis. In: Environmental cancer.
A collection of papers presented at the 24th Annual
Symposium on Fundamental Cancer Research, 1971, published for the University of Texas at Houston. Baltimore:
Williams & Wilkins, 1972:355-71.
14. Vltasa BC, Taylor HR, Strickland PT, et al. Associations of
nonmelanoma skin cancer and actinic keratosis with cumulative solar ultraviolet exposure in Maryland watermen. Cancer
1990;65:2811-17.
15. Armstrong BK, Kricker A. Epidemiology of sun exposure and
skin cancer. Cancer Surv 1996;26:133-53.
16. Rosso S, Zanetti R, Martinez C, et al. The multicentre south
European study 'Helios'. II: Different sun exposure patterns in
the aetiology of basal cell and squamous cell carcinomas of the
skin. Br J Cancer 1996;73:1447-54.
17. Grodstein F, Speizer FE, Hunter DJ. A prospective study of
incident squamous cell carcinoma of the skin in the Nurses'
Health Study. J Natl Cancer Inst 1995;87:1061-6.
18. Griego RD, Nikolaidis G, Orengo EF, et al. The relationship
between basal cell carcinoma and cigarette smoking. Dermatol
468
van Dam et al.
Surg 1996;22:95-8.
19. Karagas MR, Stukel TA, Greenberg R, et al. Risk of subsequent basal cell carcinoma and squamous cell carcinoma of the
skin among patients with prior skin cancer. JAMA 1992;267:
3305-10.
20. Grobbee DE, Rimm EB, Giovannucci E, et al. Coffee, caffeine, and cardiovascular disease in men. N Engl J Med 1990;
323:1026-32.
21. Freedman DM, Zahm SH, Dosemeci M. Residential and occupational exposure to sunlight and mortality from nonHodgkin's lymphoma: composite (threefold) case-control
study. BMJ 1997;314:1451-5.
22. Rothman KJ. Modem epidemiology. Boston: Little, Brown,
1986.
23. Breslow NE, Day NE, eds. Statistical methods in cancer
research. Vol n. The design and analysis of cohort studies.
Lyon: International Agency for Research on Cancer, 1987.
(IARC scientific publication no. 82).
24. D'Agostino RB, Lee ML, Belanger AJ, et al. Relation of
pooled logistic regression to time dependent Cox regression
analysis: the Framingham Heart Study. Stat Med 1990;9:
1501-15.
25. Weinstock MA. Assessment of sun sensitivity by questionnaire: validity of items and formulation of a prediction rule. J
Clin Epidemiol 1992;45:547-52.
26. Weinstock MA, Colditz GA, Willett WC, et al. Recall (report)
bias and reliability in the retrospective assessment of
melanoma risk. Am J Epidemiol 1991;133:240-5.
27. Fears TR, Scotto J. Estimating increases in skin cancer morbidity due to increases in ultraviolet radiation exposure.
Cancer Invest 1983;l:119-26.
28. Gilbody JS, Aitken J, Green A. What causes basal cell carcinoma to be the commonest cancer? Aust J Public Health 1994;
18:218-21.
Am J Epidemiol Vol. 150, No. 5, 1999