Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Carcinogenic exposures in racial/ethnic groups
(USC Thesis Other)
Carcinogenic exposures in racial/ethnic groups
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Carcinogenic Exposures in Racial/Ethnic Groups
by
Sarah Elizabeth Cole
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA,
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(EPIDEMIOLOGY)
August 2022
Copyright 2022 Sarah Elizabeth Cole
ii
Acknowledgements
I would like to acknowledge the guidance, expertise, and patience of my committee members
when assisting me in this research. Specifically, Myles Cockburn, PhD, for his years of frequent meetings
with lots of laughter that kept me going through the hard times, insight into analyses, his confidence in
my writing skills and especially his amazing knack for making really complex things seem simple and
pumping up my confidence. Anna Wu, PhD, for giving me the opportunity to conduct a very important
analysis using an outstanding dataset and teaching me the value of tables, which I have found myself
making every time I am stumped in an analysis; Roberta McKean-Cowdin, PhD, for her candid feedback
and invaluable guidance in difficult situations; Kimberly Miller, PhD, for her insightful comments and
direction in improving my writing skills; Martin Allen, PhD for bolstering my self-esteem and teaching me
all the technical details about UV and dosimetry I would never have mastered otherwise.
I additionally want to thank department investigators who have been critical in my development
and success in the program. A thanks to Victoria Cortessis, PhD who patiently spent hours and hours
answering all my epidemiologic and life questions; Kimberly Siegmund, PhD who graciously provided
mathematical support on numerous occasions, and always did it in a way that uplifted me.
I would like to thank my classmates that helped me through the program (we made it!! ). Charlie
Zhong, PhD, for providing moral and academic support and listening and giving valuable feedback to my
presentations; David Bogumil, PhD for being my sounding board while working through issues that hurt
my brain, providing hours upon hours of emotional support, invaluable feedback and teaching me things
about epidemiology I had never even heard of before.
I want to thank my friends and family who were there for me throughout this journey. First my
father, Roger Wm. Cole. I wish so much you were here to see me finish this degree, I know you would be
so proud. I hate the lung cancer that took you from me. You were always my biggest believer, telling
iii
me I could do anything I wanted and teaching me so many of the skills I needed to get through this
program, including the value of hard work and thinking for myself. To Gayle Bellman, my Mommy, who
was there for me every time I needed to cry, helped me persevere every time I wanted to quit and who
learned very well you aren’t supposed to ask doctoral students when they are going to finish. To Brett
Cole, my brother, who when sick with a cold, packed a U-Haul with all my belongings and drove it out to
California from Texas, only to have it die in the desert in 100 degree heat, which led him to having to
transfer everything to another truck, so I could begin this program decades ago. To my sweet rescue
dog/daughter/girl Abigail, who came to me during a deep depression during covid lockdown. Your
cheerful personality and constant love gave me a reason to get up in the morning and got me working
on my dissertation again. You faithfully lay next to me providing warmth and comfort while I wrote
most of this work. I could not have done it without you. I hope to pay you back with more frequent play
time and walks now. Thanks to my boss Matt Ruchin, for all your support and kind words, allowing me
as much time off as I needed to finish this work and encouraging me to put myself first, which is a lesson
I truly needed. Thanks to my friends, Nora Ruel, for helping with hard macro/SAS questions and making
me laugh and smile; David Smith, PhD for helping me with graphs, providing cases of diet coke to get
through late nights of writing and listening to me whine. Thanks to everyone for putting up with me all
these years when I was not so much fun to be around.
iv
Table of Contents
Acknowledgements ....................................................................................................................................... ii
List of Tables ............................................................................................................................................... vii
List of Figures ............................................................................................................................................. viii
Abstract ........................................................................................................................................................ ix
Introduction ................................................................................................................................................. ix
Chapter 1: Cumulative Menstrual Months and Breast Cancer Risk by Hormone Receptor Status and
Ethnicity: The Breast Cancer Etiology in Minorities (BEM) Study ................................................................. 1
Background ................................................................................................................................................ 1
Breast Cancer Incidence and Risk ......................................................................................................... 1
Breast Cancer Subtypes ........................................................................................................................ 2
Breast Cancer Mortality ........................................................................................................................ 3
Age and Breast Cancer Risk .................................................................................................................. 3
Age at First Birth and Breast Cancer Risk.............................................................................................. 3
Parity and Breast Cancer Risk ............................................................................................................... 4
Breastfeeding and Breast Cancer Risk .................................................................................................. 4
Early Menarche and Menopause as Breast Cancer Risk Factors .......................................................... 5
Cumulative menstrual months and Breast Cancer Risk ........................................................................ 5
Exogenous Estrogens and Breast Cancer Risk ...................................................................................... 5
Oral Contraceptives and Breast Cancer Risk......................................................................................... 6
Hormone Replacement Therapy (HRT) and Breast Cancer Risk ........................................................... 6
Breast Density and Breast Cancer Risk ................................................................................................. 7
Benign Breast Disease and Breast Cancer Risk ..................................................................................... 7
Family History of Breast Cancer and Breast Cancer Risk ...................................................................... 8
Overweight/Obesity, Physical Activity, Diabetes and Breast Cancer Risk ............................................ 8
Alcohol and Dietary Factors/Tobacco Smoking and Breast Cancer Risk ............................................ 10
Racial/Ethnic Distribution of Breast Cancer Risk Factors ................................................................... 10
Abstract .................................................................................................................................................... 11
Introduction ............................................................................................................................................. 12
Materials and Methods ........................................................................................................................... 13
Study Sample ...................................................................................................................................... 13
Data Collection and Harmonization .................................................................................................... 13
Statistical Analysis ............................................................................................................................... 15
v
Results ...................................................................................................................................................... 16
Discussion ................................................................................................................................................ 18
Tables ....................................................................................................................................................... 22
Chapter 2: Correlation between Objective Measures of Sun Exposure and Self-reported Sun Protective
Behavior and Attitudes in Predominantly Hispanic Youth ......................................................................... 31
Background .............................................................................................................................................. 31
Melanoma Incidence .......................................................................................................................... 31
Age and Gender as Melanoma Risk Factors ....................................................................................... 32
Ultraviolet Light and Sunburn as Melanoma Risk Factors .................................................................. 32
Pigmentary Characteristics - Freckling, Skin, Eye and Hair Color as Melanoma Risk Factors ............ 33
Race/Ethnicity as a Melanoma Risk Factor ......................................................................................... 35
Nevi as a Melanoma Risk Factor ......................................................................................................... 36
Family History and Genetics as Melanoma Risk Factors .................................................................... 37
UVR Measurement ............................................................................................................................. 39
Bias in Assessment of Melanoma Risk Factors ................................................................................... 41
Validation of self-reported UVR exposure .......................................................................................... 42
Abstract .................................................................................................................................................... 42
Introduction ............................................................................................................................................. 43
Materials and Methods ........................................................................................................................... 45
Study Subjects ..................................................................................................................................... 45
Study Design ....................................................................................................................................... 45
Summarizing UVR data from dosimeters ........................................................................................... 46
Data Preparation for ROC Analysis ..................................................................................................... 47
ROC Analysis ....................................................................................................................................... 47
Results ...................................................................................................................................................... 48
Participation and Demographics ........................................................................................................ 48
UVR Data from Dosimeters ................................................................................................................. 49
Questionnaire Results ......................................................................................................................... 49
ROC Results ......................................................................................................................................... 50
Discussion ................................................................................................................................................ 54
Tables ....................................................................................................................................................... 59
Figures...................................................................................................................................................... 69
Chapter 3 Objectively Measured Ultraviolet Radiation Exposure as a Predictor of Sunburn in
Predominantly Hispanic Youth.................................................................................................................... 70
vi
Abstract .................................................................................................................................................... 70
Introduction ............................................................................................................................................. 71
Materials and Methods ........................................................................................................................... 73
Study Subjects ..................................................................................................................................... 73
Study Design ....................................................................................................................................... 74
Summarizing UV data from dosimeters ............................................................................................. 75
Data Preparation ................................................................................................................................. 75
Statistical Analysis ............................................................................................................................... 76
Results ...................................................................................................................................................... 78
Discussion ................................................................................................................................................ 80
Tables ....................................................................................................................................................... 86
References .................................................................................................................................................. 91
vii
List of Tables
Table 1 Prevalence of Breast Cancer Risk Factors Across Racial/Ethnic Groups ........................................ 22
Table 2 Description of Individual Studies ................................................................................................... 23
Table 3 Characteristics of Cases and Controls by Ethnicity ........................................................................ 24
Table 4 Characteristics of Cases and Controls by Ethnicity and Menopausal Status ................................. 26
Table 5 Risk of Breast Cancer Associated with Cumulative Menstrual Months (CMM) ............................. 27
Table 6 Risk of Breast Cancer Associated with Cumulative Menstrual Months (CMM) in Premenopausal
and Postmenopausal Women, by Ethnicity and Body Mass Index ............................................................. 28
Table 7 Risk of HR+ and HR- Breast Cancer associated with Cumulative Menstrual Months (CMM) by
Ethnicity and Menopausal Status ............................................................................................................... 29
Table 8 Fitzpatrick Phototyping Scale ......................................................................................................... 59
Table 9 Age Adjusted Rate of Cutaneous Melanoma per 100,000 by Race/Ethnicity (2014-2018 SEER
data) ............................................................................................................................................................ 60
Table 10 Exposure Times by UVI Categories ............................................................................................... 61
Table 11 Demographics............................................................................................................................... 62
Table 12 ROC Results .................................................................................................................................. 64
Table 13 Demographics............................................................................................................................... 86
Table 14 Questionnaire Items, Responses and Associations ...................................................................... 88
Table 15 Model Comparison ....................................................................................................................... 90
viii
List of Figures
Figure 1 Benign Nevii .................................................................................................................................. 69
Figure 2 Atypical Nevi ................................................................................................................................. 69
Figure 3 ABCD Features .............................................................................................................................. 69
ix
Abstract
Introduction
Race and ethnicity can influence what we eat, our weight, how much we exercise, the amount
of time we spend outside; a myriad of exposures that together impact our lifetime risk of cancer. While
some racial/ethnic disparities in cancer rates are due to inherited genes, and therefore not modifiable,
others, like differences due to varying carcinogenic exposures may be revised. Identifying these
potentially transformable exposures and successfully altering them through targeted interventions could
lead to large public health impacts.
This dissertation focuses on identifying and accurately measuring carcinogenic exposures in two
types of cancers with known racial/ethnic disparities - breast cancer and melanoma. These cancers have
very different risk factors and the racial/ethnic disparities in them exist for vastly dissimilar reasons.
Breast cancer is the most common malignancy in women, with a lifetime risk of approximately 1
in 8, resulting in over 266,120 new cases and 40,920 deaths in the United States annually.[1]
Established risk factors for breast cancer include race/ethnicity, young age at menarche, nulliparity,
older age at menopause, use of contraceptive and menopausal replacement hormones, family history of
breast cancer, low levels of physical activity, tobacco smoking and high body mass index (BMI) among
postmenopausal women.[2] Some of these risk factors, such as BMI, tobacco smoking and alcohol use,
have been shown to influence levels of circulating estrogen.[3] Lifetime exposure to estrogen has been
shown to be one of the major determinants of breast cancer risk.[2, 4-6] Cumulative menstrual months
(CMM) has been used as a surrogate measure of aggregate endogenous estrogen exposure, and has
been related to risk of breast cancer among European women[7, 8], women in Northern Mexico[9], and
Asian women in Los Angeles County [10] and in Asia.[11] There is currently limited information on the
relationship between CMM and breast cancer risk in nonwhite populations.
x
The first project used the Breast Cancer Etiology in Minorities (BEM) study that has harmonized
extensive questionnaire data from four population-based studies of breast cancer with large numbers of
African Americans, Asian Americans, Hispanics, and non-Hispanic whites (NHWs) to address this
racial/ethnic gap.[10, 12-14] This project utilized menstrual and reproductive events to calculate CMM
uniformly across the four studies, allowing comparison of risk associations by race/ethnicity,
menopausal status, BMI category and breast cancer subtypes. Numerous risk factors were adjusted for
in the analysis (education, BMI, alcohol consumption, first-degree family history of breast cancer,
personal history of benign breast disease, and menopausal status). Odds ratios (ORs) and 95%
confidence intervals (CI) were calculated using conditional logistic regression, with matched sets defined
jointly by study, age group and race/ethnicity. Subtype-specific analyses (hormone receptor positive
(HR+), hormone receptor negative (HR-)) and BMI-specific analyses (< 30 kg/m
2
, > 30 kg/m
2
), were
stratified by race/ethnicity and by menopausal status. Differences in associations by race/ethnicity or
menopausal status were tested for by including interaction terms in the model. Results are presented
overall and by menopausal status, subtype and BMI category for both race/ethnicity-adjusted models
and race/ethnicity stratified models.
Skin cancer is the most common malignancy in the United States.[15] Melanoma, one of the
less frequently occurring skin cancers, initiates in melanocytes - cells that control pigmentation. While
melanoma accounts for fewer than 5% of all cutaneous malignancies, it is responsible for the majority of
skin cancer mortality.[15, 16] Rates of melanoma are rapidly increasing in the United States with a 31%
increase in invasive cases in the last decade and 99,780 new cases and 7,650 deaths predicted in 2022.
[15, 17] Most cases of melanoma are attributable to ultraviolet radiation (UVR) exposure [18-20], with
UVR and sunburns experienced as a child greatly increasing risk of melanoma compared to similar
exposure as an adult.[21, 22] Sunburn is the most commonly used proxy for exposure to UVR in
epidemiologic research.[23]. The heightened risk associated with UVR exposure in childhood coupled
xi
with the opportunity to influence behavior over the lifetime, makes accurate knowledge about this age
group crucial in a successful melanoma risk reduction program.
Rates of poor prognosis melanoma, tumors thicker than 1.5 mm at diagnosis, are increasing in
California among Hispanics much faster than non-Hispanic Whites, concurrent with rapid growth in the
Hispanic population.[24, 25] The majority of UVR exposure studies have been conducted in non-Hispanic
whites, with more recent work extending to Hispanics.[26] Most risk reduction studies to date have
utilized self-reported sun exposure as a surrogate for UVR exposure, however self-reported sun
exposure has been shown to have poor reproducibility, be prone to differential and non-differential
recall bias and only have moderate association with objectively measured UVR, impairing the ability to
detect meaningful associations due to imprecise estimates caused by information bias. [27-31]
Utilizing dosimeters to collect UVR data may potentially help modulate the bias seen in self-reported
UVR exposure. Dosimeters allow more accurate recording of time spent outdoors than self-report.
Additionally, self-report measures fail to obtain data on UVI, which is needed in order accurately
determine risk of sunburn and is obtained by dosimetry.
The second project utilized data from a novel study using objectively measured UVR exposure in
high risk, predominately Hispanic youth. This project was nested in the SunSmart study, a randomized
intervention aimed to elicit positive changes in sun protective attitudes, self-efficacy, knowledge and
behaviors in Title I public schools in Los Angeles County.[32-35] To determine the association of UVR
exposure categories with answers to questions obtained at baseline regarding different constructs
(acculturation, sun protective behavior and knowledge, family interventions), answers to questions were
dichotomized and receiver operating characteristics (ROC) analysis was then performed for each UVR
category. UVR cutoffs were chosen that maximized the sensitivity and specificity for a specific question
with the highest AUC for each UVR category.
xii
The third project utilized data from the SunSmart dosimeter sub-study, described in the second
project, to determine the association between self-reported sunburn and the highest day measurement
of daily cumulative UVI divided by total daily minutes at non-zero UVI (average UVI per minute outside).
Receiver operating characteristics analysis was performed using logistic regression with self-reported
sunburn in the last month as the dependent variable and average UVI per minute outside as an
independent variable, adjusted for dichotomized variables that have been previously shown to reduce
the risk of sunburn: student’s self-reported skin color and how often students reported using sunscreen
both in and out of school. The following variables that also may influence the probability of sunburn
were tested for possible confounding: gender and grade as well as each of the following separately for
both in and out of school wearing a hat, long sleeves and long pants. Possible confounders were tested
and retained in the model if they changed the estimate of the average UVI per minute outside odds ratio
by more than 10% singly or by more than 20% in combination with another variable. Once all
confounders were determined, the AUC, sensitivity, specificity and cutoff value of the final model and
the cutoff value when predicting sunburn were then described.
1
Chapter 1: Cumulative Menstrual Months and Breast Cancer Risk by Hormone
Receptor Status and Ethnicity: The Breast Cancer Etiology in Minorities (BEM)
Study
Background
Breast Cancer Incidence and Risk
Breast cancer is the most common malignancy in women, with a lifetime probability of
approximately 1 in 8, resulting in over 266,120 new cases and 40,920 deaths in the United States
annually, accounting for 30% of new cancer diagnoses in women.[1] Breast cancer is pathologically
highly heterogeneous, with some types displaying slow growth and excellent prognosis, with others
being highly aggressive and lethal.[36] Female sex and older age are both major risk factors for breast
cancer [1] Less than 10 % of breast cancer is attributable to inherited genetic mutations; the majority of
breast cancer is associated with environmental, reproductive and lifestyle factors, some of which are
potentially subject to modification.[37] Lifetime estrogen exposure has repeatedly been shown to
increase breast cancer risk. [2, 4-6] This association was first suggested in 1896 when oophorectomy
was found to cause regression of breast cancer in premenopausal women.[38] In 1958 it was
determined that the ovaries were responsible for estrogen production.[39] The estrogen receptor
protein was then found in 1967.[40] The subsequent observation that estrogen receptor positive tumors
were more frequent in postmenopausal women led to the discovery of the association between length
of estrogen exposure and breast cancer risk.[41, 42] Interestingly, other known risk factors for breast
cancer have been shown to influence levels of circulating estrogen.[3]
2
Breast Cancer Subtypes
There are four major breast cancer molecular subtypes, Luminal A, Luminal B, HER2 (human
epidermal growth factor receptor 2 [EGFR2]) overexpressing and Triple Negative Breast Cancer. Each
varies in incidence, response to therapy and prognosis. Luminal A, which is the most common breast
cancer subtype accounts for 68% of all breast cancer. Luminal A is pathologically defined as estrogen
receptor (ER) and/or progesterone receptor (PR) positive, HER2 negative and has low Ki67. It has the
best prognosis of all subtypes and is responsive to endocrine therapy.[43] Luminal B is the second most
common subtype, accounting for 10% of all breast cancers, with a worse prognosis than Luminal A, but
is also responsive to endocrine therapy. Luminal B is pathologically defined as ER+ and/or PR+, HER2+
(or HER2− with high Ki67). Triple negative breast cancer (TNBC), which is ER-, PR- and HER2- accounts
for 10% of all breast cancer and has a younger average age at diagnosis than the other breast cancer
subtypes, occurring more frequently in the pre-menopausal setting. TNBC lacks targeted therapy,
relying on chemotherapy and immunotherapy for systemic treatment and has a poor prognosis. HER2
overexpressing, the least common subtype, comprises 4% of all breast cancer and is HER2+ and ER-, PR-.
HER2 overexpressing breast cancers are aggressive and have poor short term survival, in spite of existing
targeted therapies. [43] Seven percent of diagnosed breast cancers have unknown types. [43]
Lifetime risk of the more indolent luminal subtypes is highest in non-Hispanic white (NHW)
women, 8.10 (95% CI 7.94, 8.20), and much lower in Asian Americans 5.06 (95% CI 4.81, 5.34), African
Americans 4.70 (95% CI 4.41, 5.02) and Hispanics 4.60 (95% CI 4.41, 4.80). [44] The risk pattern,
however, looks very different among the racial/ethnic groups for the more aggressive TNBC subtype.
African Americans have the highest risk of TNBC 1.98 (95% CI 1.80, 2.17), followed by NHW 1.25 (95% CI
1.20, 1.30), Hispanics 1.04 (95% CI 0.96, 1.13) and Asian Americans 0.77 (95% CI 0.67, 0.88).[44]
Analysis of data from the Women's Health Initiative has shown that African Americans have
approximately a fivefold greater chance than NHW of having high-grade (poorly differentiated, faster
3
growing) ER- tumors.[45] Other research has shown that African Americans, Hispanics and Asian
Americans all have increased risk of poor prognosis receptor status tumors (ER- and PR-) compared to
NHW.[46] The percentage of each breast cancer subtype varies with age and menopausal status.
Luminal A breast cancer is seen more frequently among postmenopausal women while Luminal B,
TNBC, and HER2 overexpressed are more common among premenopausal women .[47]
Breast Cancer Mortality
While African American women have lower overall incidence rates of breast cancer than NHW,
their mortality rates are 37% higher.[48] African American women have the lowest 5-year breast cancer
survival rate (77.5 %) of all the racial/ethnic groups, while Asian Americans have the highest (90.3 %)
followed by NHW (88.8%) and Hispanics (83.8%).[49] Differences among the racial/ethnic groups in
lifetime risk of the breast cancer subtypes account for some of the racial/ethnic disparities seen in
breast cancer mortality rates, but other factors include inequities in access to medical care, breast
cancer screening, accurate communication about screening results, timeliness of treatment and
diagnosis, as well as appropriateness of treatment.[50]
Age and Breast Cancer Risk
Risk of breast cancer increases steadily with age. The ten year probability of a 70 year old
woman developing breast cancer is 3.74 %, almost 10 times higher than the ten year probability for a 30
year old woman (0.43 %).[51]
Age at First Birth and Breast Cancer Risk
High exposure to endogenous levels of estrogen has been associated with increased risk of
breast cancer in both premenopausal and postmenopausal women.[52, 53] Many of the risk factors
that increase a woman’s exposure to endogenous estrogen have been found to increase risk of breast
cancer, with factors that reduce exposure reducing risk. As an example, every year of increase in age at
4
first birth increases the risk of breast cancer by about 1.7%.[51] The strength of this association varies by
ER status with the odds ratio for older versus younger (age ≥30 vs. <25 years) age at first birth was 1.6
for ER+ breast cancer and 1.2 for ER- breast cancer.[54]
Parity and Breast Cancer Risk
Parity has been shown to reduce the risk of breast cancer, particularly in ER+ breast cancer, with
each birth reducing the risk of breast cancer by approximately 7%, with a possible mechanism identified
as differentiation of breast tissue after childbirth.[55, 56] However, the association between parity and
breast cancer risk varies by breast cancer subtype. Analysis of data from the Women’s Health Initiative
has shown that among postmenopausal women having two or more children decreases the risk of ER+
breast cancer by 12 %, but increases the risk of TNBC by 46 %.[57] Another study also found ER
+ breast
cancers to have a greater risk reduction for parity than ER- breast cancers, with the odds ratio for ≥3
births compared to 0 births 0.7 for ER+ breast cancer and 0.9 for ER-. [54]
Breastfeeding and Breast Cancer Risk
Accumulating evidence has shown a modification of the parity association by breastfeeding.
High parity without breastfeeding was shown to increase risk of ER-PR- breast cancer, but a similar risk
was not observed for parous women who also breastfed.[58] Additionally, studies have found parous
women who never breastfed experienced increased risk of TNBC compared to nulliparous women,
while parous women who also breastfed had decreased risk of breast cancer [59, 60]. The high risk of
TNBC among African American women may be explained in part by this modification by breast feeding.
In the Black Women’s Health study, parity was associated with an increased breast cancer risk in African
American women under 45 years of age, but decreased risk for African American women 45 and
older.[61]
5
Early Menarche and Menopause as Breast Cancer Risk Factors
Other reproductive factors that influence exposure to endogenous estrogen, like early
menarche, and late menopause also effect the risk of breast cancer. For every year menopause is
delayed breast cancer risk increases by 3%. Similarly, for every year menarche is delayed breast cancer
risk is decreased by 5%.[62] Cessation of exposure to endogenous estrogen via ovariectomy in
premenopausal women has been consistently shown to reduce breast cancer risk.[63] Women at high
risk of breast cancer due to BRCA 1 and BRCA2 mutations have been shown to have their breast cancer
risk reduced by undergoing surgical menopause with compared to undergoing natural menopause; a risk
reduction also seen in women who undergo oophorectomy after natural menopause.[64]
Cumulative menstrual months and Breast Cancer Risk
Cumulative menstrual months, a surrogate measure of aggregate endogenous estrogen
exposure, has been associated with an increased risk of breast cancer among European women [8, 65]
as well as Asian Americans in Los Angeles County [66] and in Asia [11, 67]. Previous studies have
typically calculated cumulative menstrual months as age at menopause minus age at menarche (for
postmenopausal women) or age at interview/cancer diagnosis minus age at menarche (for
premenopausal women), with some studies additionally excluding months of pregnancy [8, 65, 66, 68]
and months of oral contraceptive use [8, 65, 66]. Only three studies have investigated cumulative
menstrual months by breast cancer subtypes, none of which studied more than one racial/ethnic group.
[66, 68, 69].
Exogenous Estrogens and Breast Cancer Risk
Exposure to exogenous estrogen through use of oral contraceptives and hormone replacement
therapy (HRT) also increases breast risk, with dose, recency and length of use all effecting the strength
of the association.[70, 71]
6
Oral Contraceptives and Breast Cancer Risk
While oral contraceptive use increases breast cancer risk, a pooled analysis of 54 studies
showed that this increase in risk subsides ten years after cessation.[72, 73] Newer formulations have
lower doses of hormones than those used in the 1960’s, potentially changing the associated risk. One
analysis that found oral contraceptive use before 1975 was associated with increased risk of ER-PR-
cancer, with no association found in women who began to use oral contraceptives in 1975 or later.[58] .
However, a more recent prospective cohort study found a positive association between breast cancer
and use of hormonal contraception, with risk rising with longer duration of use and the increased risk
persisting after discontinuation among the women who had used hormonal contraceptives for 5 years
or more compared to never users.[74]
Hormone Replacement Therapy (HRT) and Breast Cancer Risk
HRT is the administration of exogenous estrogen, given alone or more commonly in combination
with progesterone or other hormones in perimenopausal or postmenopausal women.[75] The increased
risk of breast cancer associated with HRT has been observed in numerous studies.[76, 77] In the large
UK Million Women Study the relative risk (RR) of breast cancer increased 1.66 times for current users
of HRT compared to never users.[76] This risk has been found to rise with duration of use. In a cohort
study of 22,929 Chinese women in Taiwan, the hazard ratio (HR) was 1.48 after 4 years of use and 1.95
after 8 years of use compared to women who were non-users.[77] The increased risk of breast cancer
associated with HRT is higher for women taking regimens containing both estrogen and progesterone,
compared women taking estrogen alone.[75] Cessation of administration of HRT significantly decreases
risk of breast cancer after two years.[75] The deleterious effects of HRT were reported in 2003, the
results of the Women's Health Initiative randomized controlled trial of HRT, causing use of HRT to
greatly reduce, decreasing the incidence rate of breast cancer in the United States by an estimated
7%.[78]
7
Breast Density and Breast Cancer Risk
Breast density is a risk factor that has generated substantial interest.[79] Breast density
measures how much fibrous and glandular tissue there is in a breast, as compared to fat tissue. Breast
density is categorized by radiologists into the following four groups: Class A (or 1): Fatty, Class B (or 2):
Scattered fibroglandular density, Class C (or 3): Heterogeneously dense, Class D (or 4): Extremely dense.
Women classified as having the most dense breast tissue have been shown to have two to six times the
risk of breast cancer compared to women with the least dense breasts.[80] Women with dense breasts
are also more likely to have a family history of breast cancer, indicating a possible genetic
component.[81] However, a study examining the association between acculturation and breast cancer
density among Chinese immigrants in the United states found that women in the highest acculturation
category had denser breasts compared to the lowest acculturation category (OR 3.1, 95% CI 1.6-6.0),
pointing to environmental/lifestyle factors contributing to this risk.[82] Women with dense breasts in
this study were found to have fewer live births, higher age at first live birth, and higher dairy food
intake, factors previously found to be associated with breast density, but also known risk factors for
breast cancer. [82]
Benign Breast Disease and Breast Cancer Risk
Benign breast disease encompasses a large number of nonmalignant pathologic diagnoses of the
breast tissue including fibroadenomas, cysts, fibrocystic disease, papillomas, and ductal epithelial
proliferations with or without atypia.[83] Benign breast disease occurs primarily in women of child
bearing age, with the incidence highest between the ages of 30 and 50 years.[84] Diagnosis of non-
symptomatic benign breast disease is frequently found when screening for breast cancer. Women with
symptomatic benign breast disease typically present with symptoms such as pain, a palpable mass, or
nipple discharge, which are also symptoms of breast cancer (incidence of breast cancer in women
presenting with these symptoms is 2-7%, 8% and 5-21% respectively).[85] Thus, the diagnosis of benign
8
breast disease involves testing to eliminate the possibility of malignancy. Benign breast disease has
been shown to increase the risk of breast cancer. A recent meta-analysis found breast cancer risk varied
by benign breast disease type with women diagnosed with non-proliferative disease having 1.17 (95 %
CI 0.94–1.47) times the risk of breast cancer compared to women without benign breast disease, which
increased to 1.76 (95 % CI 1.58–1.95) in women with proliferative disease without atypia, and 3.93 (95 %
CI 3.24–4.76) in women diagnosed with atypical hyperplasia not otherwise specified. [85]
Family History of Breast Cancer and Breast Cancer Risk
Women with a mother or a sister diagnosed with breast cancer are at increased risk for the
disease.[86] Family history of breast cancer is related to nearly a quarter of all cases of breast cancer,
making it a significant risk factor.[86] The Generations study, conducted in the UK, found that women
with one first degree relative with breast cancer had a 1.75 times higher risk of breast cancer compared
to women with no family history of the disease, increasing to 2.5 times higher risk in women with two or
more first degree relatives with history of the disease.[86] Familial breast cancer is ascribable to
mutations of breast cancer genes, including but not limited to BRCA1 and BRCA2.[87]
Overweight/Obesity, Physical Activity, Diabetes and Breast Cancer Risk
Overweight/obesity is usually measured by body mass index (BMI), calculated as weight in
kilograms divided by height in meters squared, as it is more easily operationalized in studies than
percent body fat, for which it is a less accurate surrogate.[88] Overweight/obesity has been found to
increase the risk of breast cancer in the postmenopausal women in numerous epidemiological
studies.[89] Conversely, obesity in premenopausal women has been found to decrease breast cancer
risk. [89] A meta-analysis done by Renehan et al. quantified this risk, finding that each 5 kg/m
2
increase
in BMI increased the relative risk (RR) of postmenopausal breast cancer by 12% (95% CI 8-16) but
decreased the RR of premenopausal breast cancer by 8% (95% CI 3-12).[90] Some studies have found
the decreased breast cancer risk conferred by obesity in premenopausal women may be restricted to
9
industrialized countries and younger obese women, with no protection given to women with a family
history of breast cancer.[91-93]
Premenopausal and postmenopausal women synthesize estrogen in very different ways.[94]
The ovaries are the primary source of estrogen synthesis in premenopausal women. Postmenopausal
women synthesize estrogen at peripheral sites, with adipose tissue being the primary source in obese
postmenopausal women.[89] Aromatase, found in adipose tissue of normal and tumorous breast
tissues convert androgens into estrogen, which can cause local estrogen levels in breast tumors to be
ten times higher than levels measured in the serum.[95, 96] Due to this mechanism of estrogen
synthesis in postmenopausal women, increases in circulating estrone, estradiol, and free estradiol have
been found to be associated with increasing BMI in this population.[97] Regular physical activity and
weight loss have both been found to be associated with decreasing levels of serum estrogens in some
studies.[89, 97]
There is evidence of a reduction in breast cancer risk with increasing physical activity. A
systematic review has revealed a strong relationship for postmenopausal women with risk reductions
reported from 20% to 80%, with weaker evidence for premenopausal women.[98] For pre- and
postmenopausal women combined, physical activity is associated with a 15–20% decreased risk of
breast cancer, with a 6% decrease in risk for each additional hour of physical activity per week.[98]
Lower levels of estrogen induced by insulin resistance increases the risk of cancer in organs with
high levels of estrogen receptors such as the breast. [99] This association creates a 20% increased risk
of breast cancer among women with type 2 diabetes.[100] Among women with a breast cancer
diagnosis, women with pre-existing diabetes also have a 16% increased risk of death.[101] This
association may be modified by metformin use, as compared to non-diabetics type 2 diabetic women on
10
metformin have been shown to have an increased risk of TNBC (HR 1.74; 95% CI, 1.06-2.83), and
nonsignificant trends toward decreased ER+ breast cancer and increased ER- breast cancer.[102]
Alcohol and Dietary Factors/Tobacco Smoking and Breast Cancer Risk
Normal levels of alcohol consumption has been shown to elevate the level of estrogen in the
blood and trigger the estrogen receptor pathways.[87] An analysis of harmonized data from 53 case
control studies revealed that women that drink 35-44 grams of alcohol per day increase their risk of
breast cancer by 32%, compared to women that abstained from alcohol; risk increased 7.1% for each
additional 10 grams intake per day.[103, 104] High fat diets are calorie dense and are associated with
high levels of adiposity, a known risk factor for breast cancer.[105, 106] Therefore it is not surprising
that diets consisting of high levels of fat have also been shown to increase risk of breast cancer.[107]
Fat in the diet has also been associated with different subtypes of breast cancer, with high total and
saturated fat conferring greater risk of ER+PR+ breast cancer but not ER-PR- breast cancer and high
levels of saturated fat increasing risk of HER2- disease.[108]
Early epidemiologic cohort studies found no evidence of increased breast cancer risk among
smokers.[109, 110] However, mutagens from cigarette smoke have been detected in the breast fluid of
non-lactating women and the presence of smoking-related DNA adducts in epithelial cells of breast milk
demonstrate that cigarette components access the breast tissue samples obtained in xxx. [111] After
review of more recent evidence, both the Canadian Expert Panel on Tobacco Smoke and Breast Cancer
Risk International Agency for Research on Cancer have concluded that there is a causal link between
cigarette smoking and breast cancer.[112, 113]
Racial/Ethnic Distribution of Breast Cancer Risk Factors
Many breast cancer risk factors, including early menarche, postmenopausal obesity and delayed
childbearing confer increased breast cancer risk in all groups. However the prevalence of these risk
11
factors vary significantly across the different racial/ethnic groups, with family history of breast cancer
and history of benign biopsy most prevalent in non-Hispanic Whites (NHW) (Table 1).[50] High BMI in
the postmenopausal setting is frequently seen in all racial/ethnic groups but Asians.[45, 114]
Postmenopausal HRT use and nulliparity/delayed first birth is most frequently seen in NHW and Asians,
[114] while African Americans and Hispanics are most likely to have early menarche.[45]
Abstract
Reproductive and hormonal factors may influence breast cancer risk via endogenous estrogen
exposure. Cumulative menstrual months (CMM) can be used as a surrogate measure of this exposure.
Using harmonized data from four population-based breast cancer studies (7,284 cases and 7,242
controls), we examined ethnicity-specific associations between CMM and breast cancer risk using
logistic regression, adjusting for menopausal status and other risk factors. Higher CMM was associated
with increased breast cancer risk in non-Hispanic Whites, Hispanics and Asian Americans regardless of
menopausal status (all FDR adjusted p trends=0.0004), but not in African Americans. In premenopausal
African Americans, there was a suggestive trend of lower risk with higher CMM. Stratification by body
mass index (BMI) among premenopausal African American women showed a nonsignificant positive
association with CMM in non-obese (BMI <30 kg/m
2
) women and a significant inverse association in
obese women (OR per 50 CMM=0.56, 95% CI 0.37-0.87, P trend=0.03). Risk patterns were similar for
hormone receptor positive (HR+; ER+ or PR+) breast cancer; a positive association was found in all
premenopausal and postmenopausal ethnic groups except in African Americans. HR- (ER- and PR-)
breast cancer was not associated with CMM in all groups combined, except for a suggestive positive
association among premenopausal Asian Americans (OR per 50 CMM=1.33, P=0.07). In summary, these
results add to the accumulating evidence that established reproductive and hormonal factors impact
12
breast cancer risk differently in African American women compared to other ethnic groups, and also
differently for HR- breast cancer than HR+ breast cancer.
Introduction
Established risk factors for breast cancer include young age at menarche, nulliparity, older age
at menopause, use of menopausal hormones, and high body mass index (BMI) among postmenopausal
women.[2] Several of these risk factors have been associated with endogenous estrogen levels in cross-
sectional studies.[3] Cumulative menstrual months (CMM), a surrogate measure of aggregate
endogenous estrogen exposure, has been related to risk of breast cancer among European women,[7, 8]
women in Northern Mexico,[9] and Asian women in Los Angeles County [10] and in Asia.[11] CMM
maybe a better measure of lifetime exposure to endogenous estrogen than individual menstrual and
reproductive factors. Previous studies calculated CMM as age at menopause minus age at menarche for
postmenopausal women, and age at interview or age at cancer diagnosis minus age at menarche for
premenopausal women.[7-11, 68, 69] Some CMM calculations also excluded months of pregnancy and
months of oral contraceptive (OC) use.[7, 8, 10] There is currently limited information on the
relationship between CMM and breast cancer risk in nonwhite populations. To address this gap, we used
the Breast Cancer Etiology in Minorities (BEM) study that has harmonized extensive questionnaire data
from four population-based studies of breast cancer with large numbers of African Americans, Asian
Americans, Hispanics, and non-Hispanic Whites (NHWs) and had investigated individual menstrual and
reproductive variables and examined these risk factors by ethnicity.[10, 12-14] Consistent with previous
studies,[7, 8, 10] we included key menstrual and reproductive events to calculate CMM uniformly across
the four studies, allowing us to compare risk associations by ethnicity, menopausal status, and breast
cancer subtypes.
13
Materials and Methods
Study Sample
The BEM study[59] harmonized data from three population-based case-control studies of breast
cancer [the San Francisco Bay Area Breast Cancer Study (SFBCS)[12], the Los Angeles County Asian
American Breast Cancer Study (AABCS)[66], and the 4-Corners Breast Cancer Study (4-CBCS)[115] ] and
the Northern California Breast Cancer Family Registry (NC-BCFR[116] ).[14] Limiting the analysis to
women diagnosed with a first primary invasive breast cancer at ages 18-79 years and NC-BCFR cases
with diagnosis from 1995 to 2003 for whom population controls were available, this pooled dataset
included interview data for 7,767 control women without a prior breast cancer and 7,895 breast cancer
cases. All studies were approved by the Institutional Review Board at each study site and all participants
signed an informed consent.
Data Collection and Harmonization
Details of the study design, sources of cases and controls and participation rates in the parent
studies included in the BEM study have been reported previously (see Table 2).[59] Briefly, structured
questionnaires were administered by in-person interviews in English, Spanish (NC-BCFR, SFBCS, 4-CBCS)
or Chinese (Cantonese or Mandarin; AABCS, NC-BCFR). AABCS selected neighborhood controls using a
well-established algorithm applied in other case-control studies in Los Angeles County.[71, 117]
Controls were 1:1 individually matched to cases by specific Asian ethnicity and age (± 5 years). 4-CBCS
controls were selected from populations living in the cases’ state and frequency matched to cases by
ethnicity and age within 5 years. SFBCS controls were selected through random digit dialing and were
1:1 frequency matched to cases by race/ethnicity and 5-year age group. For Hispanic cases diagnosed
1995-1998, controls were frequency matched at a 1:1.5 ratio. NC-BCFR controls were selected through
random digit dialing and were 2:1 frequency matched to cases diagnosed between 1995-1998 by
race/ethnicity and 5-year age group. Ethnicity was based on self-identification. AABCS included women
14
who self-identified as Asian (Chinese, Japanese, Filipina, or other Asian). 4-CBCS included women who
self-identified as Hispanic, Native American or NHW. SFBCS included women who identified as African
American, Hispanic, or NHW. Up to 2 race/ethnicities could be reported. Participants who reported
NHW and a minority ethnicity (African American or Hispanic) were classified according to the minority
ethnicity. In NC-BCFR, breast cancer cases diagnosed from 1995-1998 included any self-identified
ethnicity with oversampling of cases who self-identified as Hispanic, African American or Asian
American; cases diagnosed from 1999-2009 included those who self-identified as African American,
Hispanic, Chinese, Filipina, or Japanese. Up to four ethnicities could be reported. Participants who
reported NHW and a minority ethnicity (i.e., African American, Asian American, Hispanic) were classified
according to the minority ethnicity. Information on all risk variables was collected up to the reference
year, defined as the calendar year before diagnosis for cases, or before interview (AABCS, NC-BCFR) or
selection into the study (SFBCS, 4-CBCS) for controls. Questionnaire data included age, ethnicity,
education, history of breast cancer among first-degree relatives, menstrual history, pregnancy history,
breastfeeding practices, OC use, menopausal hormone therapy (HT) use, and height and weight (self-
reported in reference year or measured at interview). Information on lifetime history of smoking was
available in all studies, except for SFBCS which included questions on smoking only in the last cycle of
data collection. All studies, except for 4-CBCS, collected data on country of birth. Details on tumor
characteristics, including estrogen receptor (ER) and progesterone receptor (PR) status were obtained
from cancer registry records and were available for 87% of cases (90% of African Americans and Asian
Americans, 86% of Hispanics, 84% of NHWs); totaling 5,457 hormone receptor positive (HR+; ER+ or
PR+) tumors and 1,435 hormone receptor negative (HR-; ER- and PR-) tumors.
As in previous studies,[7, 8] we calculated CMM by subtracting age at menarche from age at
menopause (or age at first HT use) for postmenopausal women and from reference age (age at breast
cancer diagnosis for cases or age at interview for controls) for premenopausal women, and then
15
subtracting total months of pregnancy (including those not resulting in live births), total months of
breastfeeding, and total months of OC use, presumed to be anovulatory periods. We assigned 9 months
for full-term pregnancies if the duration of pregnancy was not reported (n=9). Women were classified
as premenopausal if they still had menstrual periods or were pregnant, breastfeeding or
perimenopausal during the reference year, defined as either the calendar year before diagnosis for
cases, the calendar year before interview for controls in AABCS and NC-BCFR, or the calendar year
before selection into the study for controls in SFBCS and 4-CBCS. Women were classified as
postmenopausal if they reported that prior to the reference year their periods had stopped naturally or
due to surgery, medical treatment or other reasons. Women who still had periods when they started
using HT were classified as postmenopausal if their reference age was 55 years or older, but were
excluded from analysis if under age 55 years (158 cases, 191 controls). We also excluded from analysis
women who were not in one of the four ethnic groups (7 cases, 6 controls), or with missing data on
parity (1 case) or OC use (445 cases, 328 controls; mostly from the SFBCS). The present analysis on CMM
was based on 7,284 breast cancer cases and 7,242 control women.
Statistical Analysis
We calculated odds ratios (ORs) and 95% confidence intervals (CI) using conditional logistic
regression analysis, with matched sets defined jointly by study, age group (<35, 35-39, 40-44, 45-49, 50-
54, 55-59, 60-64, 65-69, 70-74, ≥75 years) and ethnicity (African American, Asian American, Hispanic,
NHW). All regression models also adjusted for education (less than high school, high school graduate,
some college or technical school, college graduate or higher, unknown), BMI (<18.5, 18.5-24.9, 25-29.9,
30.0-34.9, 35.0-39.9, ≥40 kg/m
2
, unknown), alcohol consumption (0, >0 to <7, ≥7 drinks per week,
unknown), first-degree family history of breast cancer (yes, no, unknown), personal history of benign
breast disease (yes, no, unknown), and menopausal status (premenopausal, postmenopausal). We
conducted analyses for all breast cancers combined and separately by ethnicity (African Americans,
16
Asian Americans, Hispanics, and NHWs) and by menopausal status. For subtype-specific analyses (HR+,
HR-), we stratified the analyses by ethnicity and by menopausal status. We tested for differences in
associations by ethnicity or menopausal status by including interaction terms in the model. Multiple
comparisons were controlled for by using the false discovery rate (FDR) method of Benjamini and
Hochberg, with an alpha of 0.05 and two-sided p values.[118] OC use was not included as a transient
risk in the model because of missing data on age start and age stop in all four studies. Statistical
analyses were performed using SAS software, version 9.4 (SAS Institute Inc, Cary, NC).
Results
The study sample consisted of NHWs (2,232 cases, 2,479 controls), African Americans (752
cases, 601 controls), Hispanics (1845 cases, 2212 controls), and Asian Americans (2,455 cases, 1,950
controls) (Table 3). African American cases and controls had the highest mean BMI and Asian Americans
the lowest. Except for Asian Americans, cases had lower mean BMI, were more likely to report a history
of benign breast disease, to be college graduates, weekly alcohol drinkers and longer OC users than
control women. Across all four ethnic groups, cases compared to control women had fewer full-term
pregnancies, but were more likely to have a family history of breast cancer, and to be current HT users.
Mean CMM was higher in cases than control women, except in African Americans. Distributions by
menopausal status are also shown (Table 4).
Table 5 shows risk associations by quartiles of CMM. In pre- and postmenopausal women
combined, we found a trend of higher breast cancer risk with increasing number of CMM in NHWs,
Hispanics, and Asian Americans (P trend=0.0004); women in the highest quartile of CMM were 50% to 73%
more likely to be diagnosed with breast cancer than those in the lowest quartile. In contrast, CMM was
not associated with risk in premenopausal and postmenopausal African Americans combined
(P trend=0.65) (P heterogeneity (ethnicity)=0.05). However, higher CMM was associated with a suggestively lower
risk among premenopausal African American women (OR=0.59, 95% CI: 0.28-1.26 for highest vs lowest
17
quartile of CMM; P trend=0.34), while there was no association in postmenopausal African American
women (P trend=0.84). CMM was positively associated with increased risk in postmenopausal NHW
women (P trend=0.0004), but not in premenopausal NHWs. In Hispanics and Asian Americans, CMM was
positively associated with risk in both premenopausal and postmenopausal women, with a stronger risk
pattern in postmenopausal Hispanics (P trend=0.0004) and in premenopausal Asian American women
(P trend=0.0004). Analyses per 50 CMM showed similar risk patterns as the analyses by quartile cut-
points.
Because obesity is associated with lower risk of breast cancer in premenopausal women
BMI,[119] we explored whether the inverse association between CMM and risk of breast cancer in
premenopausal African American women may be influenced by BMI. Stratifying the analyses by BMI
(Table 6), among non-obese (<30 kg/m
2
) premenopausal women, CMM was positively associated with
breast cancer risk in all ethnicity groups combined (OR per 50 CMM=1.21, 95% CI 1.12-1.31,
P trend=0.0004), with the association being strongest in Asian Americans (OR per 50 CMM=1.38, 95% CI
1.21-1.57, P trend=0.0004) and weakest in African American women (OR per 50 CMM=1.04, 95% CI 0.77-
1.42, P trend =0.84). In contrast, in obese (≥30 kg/m
2
) premenopausal women, CMM was associated with
lower breast cancer risk in African American women (OR per 50 CMM=0.56, 95% CI 0.37-0.87, P trend
=0.03), but not in the other ethnic groups.
In postmenopausal women, across all ethnicity groups combined, higher CMM was associated
with significantly increased risk in both non-obese (OR per 50 CMM=1.15, 95% CI 1.09-1.21) and obese
(OR per 50 CMM= 1.15, 95% CI 1.06-1.24) women. However, in postmenopausal African American
women, no positive association was observed in those who were non-obese (OR per 50 CMM=0.95, 95%
CI=0.80-1.13). In obese postmenopausal women, positive associations were strongest in Hispanics (OR
per 50 CMM=1.20, 95% CI 1.06-1.36) and NHWs (OR per 50 CMM=1.13, 95% CI 0.99-1.29), intermediate
18
in African Americans (OR per 50 CMM=1.16, 95% CI 0.93-1.44), but absent in Asian Americans (OR per
50 CMM=0.95, 95% CI 0.62-1.46).
Results by HR status showed that HR+ breast cancer was positively associated with CMM in
NHWs, Hispanics and Asian Americans, but not in African American women (P heterogeneity (ethnicity)=0.03)
(Table 7). The OR per 50 CMM ranged from 1.17 (95% CI 1.10-1.26) in NHWs to 1.24 (95% CI 1.13-1.36)
in Asian Americans. Risk associations for HR+ breast cancer were generally similar for premenopausal
and postmenopausal NHW, Hispanic and Asian American women. In contrast, risk of HR+ breast cancer
in premenopausal African American women decreased with increasing CMM; women in the highest
quartile had an OR of 0.41 (95% CI 0.17-0.98) compared to those in the lowest quartile (P trend=0.11). In
postmenopausal African American women, HR+ breast cancer risk was not associated with CMM
(P trend=0.68) (Table 7). This difference in association with HR+ breast cancer between premenopausal
and postmenopausal African American women approached statistical significance (P heterogeneity=0.09).
In premenopausal and postmenopausal women combined, CMM was not associated with risk of
HR- breast cancer in any of the four ethnic groups (Table 7). In premenopausal women, a trend of
increasing risk of HR- breast cancer with increasing CMM was limited to Asian American women (OR per
50 CMM=1.33, 95% CI 1.02-1.73). CMM was unrelated to risk of HR- breast cancer in postmenopausal
women (OR per 50 CMM =1.02, 95% CI=0.94-1.11). There were no significant differences in these HR-
risk associations by menopausal status and ethnicity except for Asian American women; the OR per 50
CMM was 0.95 (95% CI 0.80-1.11) in postmenopausal women which differed significantly from that in
premenopausal Asian American women (P heterogeneity=0.03).
Discussion
In this large population-based pooled analysis, CMM was positively associated with breast
cancer risk in premenopausal and postmenopausal NHW, Hispanic and Asian American women,
19
compatible with previous studies,[7, 8, 10, 11] but CMM was not associated with risk in African
American women. There was a suggestive inverse association between CMM and breast cancer risk
among premenopausal African American women which was stronger in those who were obese
(P trend=0.03); this inverse association also approached significance for African American women with HR+
breast cancer (P trend=0.115). CMM was not significantly associated with risk of HR- breast cancer; null
associations were found among all four ethnic groups of premenopausal and postmenopausal women,
except for a positive association with CMM among premenopausal Asian American women that was
borderline significant (P trend=0.07).
The positive association between CMM and risk of HR+ breast cancer, but not with HR- breast
cancer, is consistent with accumulating evidence that the relationships with menstrual and reproductive
factors differ by HR status.[43, 120, 121] Hormonal mechanisms predominate primarily in the
relationships between reproductive factors and HR+, but not HR- breast cancer. Consistent with
previous studies,[122-126] the proportion of HR- breast cancers was higher in both premenopausal
(31%) and postmenopausal (27%) African American women than respective NHW (24% vs 14%),
Hispanic (25% vs 20%), and Asian American (18% vs 19%) women, likely contributing to the lack of an
overall association between CMM and breast cancer risk in African American women. However, reasons
for the divergent risk association between CMM and HR+ breast cancer in premenopausal African
Americans compared to premenopausal NHW, Hispanic and Asian American women are not clear since
the same covariates were adjusted for across all the ethnic groups under study.
Based on results we have previously published on subtype-specific associations with menstrual
and reproductive variables including age at menarche, parity, breastfeeding, and age at
menopause,[127] we explored whether differences in risk patterns with individual menstrual and
reproductive factors contributed to divergent risk associations between CMM and HR+ breast cancer
risk in African American women compared to Hispanic, Asian American and NHW women. In the Black
20
Women’s Health Study, late menarche, high parity, and breastfeeding were not associated with reduced
risk of HR+ breast cancer,[61] whereas these factors were inversely associated with risk in the other
ethnic groups, as we have described in detail previously.[127] Our findings of differences in the
associations of parity and breastfeeding with risk of HR+ breast cancer in premenopausal African
American women are largely supportive of previous studies conducted in African Americans elsewhere
in the United States.[60, 61, 128-130]
In the present analysis, BMI appeared to modify the risk association with CMM in
premenopausal women. In obese premenopausal African American women, there was a pattern of
lower breast cancer risk among those with higher CMM; but this was not observed among obese
Hispanic women. Reasons for the divergent risk pattern in premenopausal Hispanic women are not
known; the prevalence of obesity (BMI ≥30 kg/m
2
) was high among premenopausal Hispanic cases
(26.8%) and controls (34.7%), similar to that of African American cases (32.6%) and controls (39.7%) and
considerably higher than in NHW cases (15.9%) and controls (18.6%) and Asian American cases (4.2%)
and controls (3.6%).
Some limitations and strengths of our study should be considered. While we used comparable
methods to define cumulative menstrual months in our harmonization and data analysis, the specific
questions on the relevant menstrual and reproductive events that were asked in the individual studies
were not identical. Although total months of breastfeeding were subtracted, the questions on
breastfeeding were not based on exclusive breastfeeding when anovulatory cycles may be more likely.
In addition, our analyses did not include 773 women (445 cases and 328 controls), primarily from SFBCS,
who were missing on information on OC use. Our questions on body mass index only represented data
at one time point (i.e., in the reference year). Information on life course body weight and other body
size composition measures will improve our understanding of its complex effects on breast cancer risk.
Since our sample size of African Americans was modest and the smallest of the four ethnic groups, risk
21
estimates for African Americans may be less stable. The CMM method of analysis does not account for
different trajectories of life, such as age at first birth, or transient risk due to factors such as use of OCs
or hormone replacement therapy. However, there are important study strengths, including a
population-based study design, and a diverse study sample which allowed, for the first time, estimation
of risk associations with CMM among Hispanic and African American women. Moreover, we were able
to examine risk patterns separately by menopausal status, BMI, and hormone receptor status.
In summary, we present, for the first time, a comparison between four ethnic groups of breast
cancer risk associated with CMM, as a surrogate of endogenous estrogen exposure. The divergent risk
patterns among African American women highlight the need to continue to identify opportunities to
examine breast cancer risk patterns specifically by ethnic groups.
22
Tables
Table 1 Prevalence of Breast Cancer Risk Factors Across Racial/Ethnic Groups
Family
history
(%)
Early
menarche
(%)
Late
1
or
no
parity
(%)
History
of
benign
biopsies
(%)
Postmenopausal
high BMI (%)
Alcohol
intake
2
(%)
Postmenopausal
HRT use
3
(%)
African–
American
12 24 19 19 51 2 26
Asian/Pacific
Islander
9–12 21 14–27 17 11 1 29–48
Hispanic 9 25 18 15 37 2 35
NHW 14–17 22 19–59 20 28 4 43–54
When estimates were available from multiple sources, a range of percentages is presented (data taken from
[45, 131]
1 ≥ age 30 years.
2 ≥2 drinks/day.
3 Prior to 2003.
23
Table 2 Description of Individual Studies
24
Table 3 Characteristics of Cases and Controls by Ethnicity
25
Table 3 Characteristics of Cases and Controls by Ethnicity (continued)
26
Table 4 Characteristics of Cases and Controls by Ethnicity and Menopausal
27
Table 5 Risk of Breast Cancer Associated with Cumulative Menstrual Months (CMM)
28
Table 6 Risk of Breast Cancer Associated with Cumulative Menstrual Months (CMM) in Premenopausal and Postmenopausal
Women, by Ethnicity and Body Mass Index
29
Table 7 Risk of HR+ and HR- Breast Cancer associated with Cumulative Menstrual Months (CMM) by Ethnicity and Menopausal Status
30
Table 7 (continued) Risk of HR+ and HR- Breast Cancer associated with Cumulative Menstrual Months (CMM) by Ethnicity and
31
Chapter 2: Correlation between Objective Measures of Sun Exposure and
Self-reported Sun Protective Behavior and Attitudes in Predominantly Hispanic
Youth
Background
Melanoma Incidence
Skin cancer is the most common malignancy in the United States, with melanoma responsible
for the majority of skin cancer mortality.[15] In the United States in 2022, 99,780 new cases and 7,650
deaths due to the disease are predicted, with men and women projected to have a 2.3% lifetime risk of
the disease. [15, 17] Age-adjusted rates of new cases of melanoma are rapidly increasing, rising on
average 1.4% each year over 2009–2018, while age-adjusted death rates have been falling on average
3.2% each year due to recent treatment advances, including targeted and immunotherapy.[132, 133]
As geography varies, so do factors that influence level of ultraviolet radiation, the greatest risk factor for
melanoma. Melanoma incidence has been shown to increase in lower latitudes as compared to higher
latitudes worldwide. [16] Altitude also influences melanoma incidence, with higher altitudes associated
with higher incidence.[134] Air pollution can also contribute to carcinogenesis, with components of
pollution acting synergistically with ultraviolet radiation to create large amounts of oxidative stress on
skin cells.[135-137] Melanoma can occur anywhere on the body, but it is most commonly found on
areas of skin with frequent sun exposure, such as the face, neck, hands, and arms. The distribution of
primary site varies by gender, with melanoma most frequently occurring on the back for men and leg or
arms for women.[138] Prognosis of melanoma worsens as stage at diagnosis advances, with 99.5% of
people diagnosed with localized disease surviving five years, 68.0% of people with regional disease and
only 29.8% of those with distant disease.[132] In the United States the median age at death due to
melanoma is 71.[132]
32
Age and Gender as Melanoma Risk Factors
Incidence of melanoma increases with age. In the United States the median age at diagnosis is
65, but it is the most commonly diagnosed cancer in people aged 25 to 29. [132] In ages 15 to 29 it is
the third most common cancer for men and fourth most common for women.[132] Males are at higher
risk of melanoma than females, with the rates per 100,000 in the United States 29.3 and 18.0
respectively.[132] When stratified by age group, however, women have higher incidence rates of
melanoma than men until age 40 years, but by age 75 male’s rates are almost three times higher than
females.[139] The reasons for the differences in melanoma risk between the sexes are unknown, but
preliminary research indicates the cause is most likely biologic. Animal studies have shown that sex
steroid hormones, such as estrogen receptor beta (ERβ) decrease melanoma tumor aggressiveness
while testosterone increases it.[140]
Ultraviolet Light and Sunburn as Melanoma Risk Factors
Approximately 86% of all melanomas are attributable to ultraviolet radiation from the sun.[141]
Intermittent sun exposure is the primary risk factor for melanoma, while chronic, continuous exposure
to UVR is linked to non-melanomatous skin cancers and actinic keratosis.[142, 143] Occupations
associated with being outside in the sun for long periods, such as farming, have no increased risk of
melanoma, but are associated with a heightened risk of developing other skin cancers such as squamous
cell carcinoma.[144] Intermittent sun exposure is sporadic, usually occurring during recreation
(weekends and holidays) among people whose skin has not adapted to the sun, such as those who
work primarily indoors. Multiple studies have documented the association between melanoma and
intermittent sun exposure, with estimates of increased risk, as measured by odds ratios, ranging
from 1.6 to 1.7.[142, 145, 146]
33
Sunburn history is the most commonly used surrogate marker of intermittent sun exposure in
epidemiologic research.[23] People who have had more than ten severe, painful sunburns have twice
the risk of melanoma compared to those with nine or few sunburns in their lifetime.[21] Sunburns
occurring in childhood confer higher risk of melanoma than sunburns occurring as an adult.[146, 147]
Additionally, risk of melanoma associated with a given level of sun exposure during adulthood has been
shown to increase multiplicatively with higher sun exposure during childhood.[148] This has lead
researchers to conclude that avoidance of UVR exposure during childhood will decrease risk of
melanoma more than the same behavioral modification as an adult. [148]
Artificial UVR exposure also increases the risk of developing melanoma. One tanning bed
session exposes a person to significantly higher levels of UVA than being outdoors.[149] A meta-
analysis conducted by Gallagher et al showed that artificial UVR exposure via sunbeds and sunlamps
significantly increases the risk of melanoma 1.25 times (OR; 95% CI 1.05-1.49) compared to people with
no such exposure.[150] More recent studies have concluded that one indoor UVR tanning session
increases the risk of developing melanoma by 20%, and each additional session during the same year
increases the risk almost another two percent.[151, 152]
Pigmentary Characteristics - Freckling, Skin, Eye and Hair Color as Melanoma Risk Factors
Pigmentation of skin, determined by the amount of melanin, is highly correlated with melanoma
propensity, with lighter skin conferring greater risk. Ephelides, commonly referred to as freckles, are
another form of skin pigmentation that is associated with increased risk of melanoma. Ephelides are
thought to be genetically determined by the MC1R gene, which controls the majority of pigmentation,
and are more commonly found in people with fair skin or red hair.[153] People that have ephelides
have almost double the risk of melanoma compared to those who do not [relative risk (RR) 1.99 (95% CI:
1.79–2.20)].[154] The increased melanoma risk associated with ephelides is independent of the risk
34
associated with melanocytic nevus, another well-known melanoma risk factor that will be described
later.[155]
Pigment in the eyes is a determinant of melanoma risk, with individuals having blue, green, grey
or hazel eyes at increased risk.[154, 156] Compared to individuals with darker eyes, those with
blue/blue-grey eyes have 1.57 times the risk of melanoma (RR 95% CI: 1.39–1.78) and those with
green/grey/hazel eyes have 1.51 times the risk (RR 95%CI: 1.28–1.79).[154]
Pigment in the hair can also elevate risk of melanoma, with individuals with blond, light brown
and red hair at higher risk than darker haired individuals. [154] Specifically, meta-analysis has
determined that compared to individuals with dark hair, people with red hair have 2.64 (RR 95% CI:
2.25–3.10) times the risk of melanoma, those with blonde hair have twice the risk (2.00 RR 95% CI:
1.47–2.73) and those with light brown hair have 1.46 times the risk (RR 95% CI: 1.26–1.68). [154]
The Fitzpatrick phototyping scale was developed as a way to estimate the response of different
skin types to UVR. The scale assesses an individual’s risk of sunburn based on questions regarding
phenotype (eye color, hair color, number of freckles, skin color) and dermal response to sun exposure
(tanning, burning, blisters and peeling overall and on the face). A score calculated from responses to
these questions categorizes individuals into one of six phototype groups. (Table 8).[31] The scale has
been shown to have higher accuracy when assessed by physicians than self-report, due to African
Americans, Hispanics and Asians frequently not understanding the concepts of tanning and sunburn.
[31] Consequently new questions have been proposed for the scale which incorporate descriptions of
sunburn (pink/red, irritated, tender, or itchy skin) and tanning (darker skin) into the assessment. [31]
A recent meta-analysis was undertaken to more accurately determine the contribution of the
different pigmentary characteristics to melanoma risk. This study found that relative risk of melanoma
decreased across phototypes. In comparison to phototype IV, people with phototype I have 2.27 times
35
the risk of melanoma (95% CI: 1.77–2.92), people with phototype II have 1.99 (95% CI: 1.62–2.45) times
the risk and those with phototype II have just 1.35 times the risk (95% CI: 11.12–1.63). The meta-
analysis also found that pooled relative risk estimates were lower in studies that adjusted for hair or eye
color or both compared to those that did not.[154]
Race/Ethnicity as a Melanoma Risk Factor
Melanoma occurs infrequently in darker skinned individuals, partially due to the
photoprotective effect of melanin.[157] The protective melanin barrier found in darker skin decreases
the amount of both UV A and B radiation (inducers of skin cell death and malignant transformation) that
penetrate the skin.[158, 159] 7.4% of UVB and 17.5% of UVA radiation penetrates African American skin,
compared to the substantially greater 24% of UVB and 55% of UVA in NHW skin.[160] Compared to
fairer skinned individuals, UVB radiation penetration of the epidermis is diminished by 50% in those with
darker skin.[161]
Due to the intrinsic link between skin color and race/ethnicity, the incidence of melanoma varies
greatly across racial/ethnic groups, with melanoma 10 to 12 times more common in NHW, and 6 to 7
times more common in Hispanics relative to African Americans (AA).[157] Overall. NHW have ten times
the risk of developing cutaneous melanoma compared to African Americans, Asians or Hispanics. This
risk elevation is not true for all types of melanoma with risk of plantar melanoma about equal for NHW
and African Americans, while NHW having highest risk of non-cutaneous melanomas (such
as mucosal).[162]
In the United States, the National Program of Cancer Registries and the Surveillance, Epidemiology, and
End Results (SEER) program collects and reports data on cancer by type and race/ethnicity.[132] These
data display the striking racial/ethnic disparities described above, with white males having 34.7 times
the rate of melanoma of African American males and white females having 24.6 times the rate of
36
melanoma as black females (Table 9). Compared to Hispanics, Non-Hispanics have 6.6 times higher rates
of melanoma in males and 4.14 times higher rates in females. Furthermore, rates of poor prognosis
melanoma, tumors thicker than 1.5 mm at diagnosis, are increasing in California among Hispanics much
faster than non-Hispanic Whites, concurrent with rapid growth in the Hispanic population.[24, 25]
Nevi as a Melanoma Risk Factor
Nevi, commonly referred to as moles, are benign melanocytic neoplasms of the skin that are
present in the majority of the population. Nevi typically emerge at ages 4 and 5 all over the body,
including under the nails and on the soles of feet. The presentation of nevi varies through the lifespan.
During pubescence the pigmentation in nevi darkens and in the 70’s and 80’s nevi tend to
disappear.[153] Genetics are a strong determinant of the number of nevi a person has, but sun
exposure is also a factor, with chronic sun exposure associated with nevi development rather than
sunburn.[155, 163] The physical appearance of nevi is diverse; the surface varies from raised to flat with
colors including flesh, pink and brown. Benign nevi are usually symmetrical, have smooth, regular
borders and surface, uniform color and are less than the size of a pencil eraser (6 mm) (Figure 1). [153]
Dysplastic nevi, which contain a cluster of atypical melanocytes, occur in 2-6% of people in the United
States and are considered a transition or precursor to cutaneous melanoma, comparable to the
relationship between colonic polyps and cancer of the colon.[164, 165] Atypical nevi are nevi that have
physical characteristics that increase the probability that they contain underlying dysplasia (Figure 2).
[153] The International Agency for Research on Cancer (IARC) uses the following criteria to define
atypical nevi: there must be a macular component in at least one area, additionally at least three of the
following features must be present: (a) border not well defined, (b) size 5 mm or more, (c) color
variegated, (d) contour uneven, (e) presence of erythema.[166] Congenital melanocytic nevi, also
referred to as birthmarks, are also classified as nevi.
37
Twenty five percent of melanoma occurs in pre-existing nevus, through the transitional pathway
described above (benign - dysplastic – melanoma), increasing to 29-49% for non-familial
melanoma.[153, 167] Melanoma often presents clinically with some or all of the ABCDE features
(Figure 3), specifically asymmetry (A), border and surface irregularity (B), color variability (C),
diameter greater than 6 mm (D), and evolution or change (E).[153, 168]
In general the higher a person’s total nevus count, the greater their risk of melanoma, with risk
differing by size and type of nevi.[169] Compared to people with less than 15 total nevi those with 101-
120 nevi have almost 7 times the risk of melanoma (RR = 6.89; 95% CI 4.63, 10.25), with even 16-40 nevi
conferring increased risk (RR = 1.47; 95% CI: 1.36, 1.59). [166] Having any number of atypical nevi
increases the risk of melanoma almost ten times when compared to having no atypical nevi (RR = 10.12;
95% CI: 5.04, 20.32).[166] Size of nevi also influences melanoma risk, with larger (>5 mm) and giant (>20
cm) nevi associated with significantly higher risk.[170] In congenital melanocytic nevi the lifetime risk of
melanoma is 2.6-4.9% for small and medium nevi, which increases to 6-20% for giant nevi.[170]
Dysplastic nevi are present in 34%-56% of melanoma cases with their presence also significantly
increasing melanoma risk.[166, 171]
Family History and Genetics as Melanoma Risk Factors
A family history of melanoma, defined as the presence of melanoma in one or more first-degree
relatives, greatly increases a person’s melanoma risk (RR = 1.74, 1.41–2.14).[172] However only about
ten percent of incident cases report a family history of the disease.[173] The genetic variants that
increase risk of melanoma are inherited following an autosomal dominant pattern with incomplete
penetrance.[173] CDKN2A, a tumor suppressor gene, is the main gene known to confer an increased
risk of melanoma, with 20%-40% of melanoma prone families harboring CDKN2A mutations.[173, 174]
Only approximately .2% of melanoma cases are due to CDKN2A mutations, however, in spite of these
38
mutations being present in 90% of inherited melanoma.[175] Melanomas due to genetic predisposition
usually present at a younger age (<40 years), as multiple primary melanomas or with a history of
precursor lesions such as dysplastic nevi and are more likely to be superficially invasive with a better
prognosis.[176, 177]
In a unique population-based study by Begg et al, 3,550 incident cases of melanoma in Australia,
Canada, and the United States were genotyped and 65 carriers of CDKN2A mutations identified. The
history of melanoma in first degree relatives was obtained in each of the 65 carriers and used to
calculate the lifetime risk of melanoma in this sample utilizing the kin–cohort method.[175] In carriers
of CDKN2A mutations, at 50 years of age the risk of melanoma was 14% (95% CI = 8% to 22%), increasing
to 24% (95% CI = 15% to 34%) at age 70 and 28% (95% CI = 18% to 40%) at age 80. Eighteen of the 65
carriers (28%) had three or more first-degree relatives with melanoma, but only one was a carrier of a
CDKN2A mutation. This is in contrast to the much higher rates reported in a study conducted by the
Melanoma Genetics Consortium, in which 80 families with documented CDKN2A mutations and 3 or
more cases of melanoma were examined.[178] This familial based study reported risk of melanoma to
be 30% (95% CI = 12% to 62%) by age 50 and 67% (95% CI = 31% to 96%) to age 80.[178] Thus CDKN2A
mutation carriers that do not come from multiple-case families appear to have much lower risk of
melanoma than those that do, but are still at higher risk than the general population.[175] Other
genetic mutations shown to increase risk for melanoma include variations to the MC1R gene which
prevents production of the protective pigment eumelanin in melanocytes, increasing damage from UV
radiation.[179] Mutations in the MC1R gene may also increase risk of melanoma without UVR induced
dermal damage, presenting regardless of skin phototype. These MC1R mutation, non UVR induced
melanomas appear more frequently in association with mutations in other genes involved in the
melanoma pathway such as BRAF and CDKN2A. [180] Mutations in CDK4 have also been shown to
increase risk of melanoma, albeit more rarely.[181]
39
Several family cancer syndromes have been shown to increase melanoma risk. B-K mole
syndrome, also called Familial Atypical Multiple Mole-Melanoma (FAMMM) syndrome, is associated
with the hereditary susceptibility genes CDKN2A and CDK4 and presents as large (> 5 mm), irregular, and
dysplastic nevi. These nevi frequently appear on areas of the body that are protected from UV light, like
the trunk. FAMMM syndrome increases the risk of cutaneous and ocular melanomas as well as other
malignancies. [153, 182] Although FAMMM accounts for less than 5% of melanoma cases overall,
between the ages of 20 and 59 years, 56% of people with FAMMM develop melanoma, with 100%
developing melanoma by age 76.[166, 183] Familial retinoblastoma, Li-Fraumeni cancer syndrome and
Lynch syndrome type II are also associated with a higher risk of melanoma.[184]
UVR Measurement
As UVR exposure varies by latitude, ecological studies have frequently utilized differences in
disease rates in populations living at different latitudes as evidence of an association between UVR
exposure and disease.[185] This approach is valid at the population level, as the average UVR dose that
adults receive has consistently been shown to be approximately 3–4% of the daily ambient UVR, 5% for
children, with little variance.[186] UVR data are geographically widely available as many major cities
monitor ground level radiation, with more remote areas covered by modelling, utilizing data obtained
from satellites. [187] Individuals within a population, however, have high variability in UVR exposure,
with their doses ranging from .1 to ten times the average dose, making these gross average estimates
too inaccurate for use in epidemiologic studies examining individual risk.[188]
Many methods exist to assess cumulative UVR exposure. Conjunctival ultraviolet auto-
fluorescence (CUVAF) quantifies sun damage to the eye, which has been shown to correlate with
lifetime UVR exposure, but more highly with the last six months of exposure.[189] Lifetime UVR
exposure can also be approximated by microtopographic assessment of dermal injury, in which a
silicone cast of the back of the hand is taken and skin changes graded according to a standard.[190]
40
Estimation of UVR exposure by assessment of skin damage is also done via paraocular photography
followed by computer processing and independent grading, as well as chromater grading of pigment
change. [191] Serum 25-hydroxyvitamin D levels have also been shown to correlate with UVR exposure
for the past six weeks.[187] Physical exams that result in a higher perceived age than actual are
correlated with histologic changes consistent with sun exposure (r = 0.79, P < 0.0001.[192] Dermal
elastosis is the histologic gold standard used to assess the extent of UVR induced damage, with
increasing elastosis correlating with increasing UVR induced damage to the dermis. This histologic
change however, has not been associated with increased risk for melanoma, but rather skin cancers
associated with high levels of chronic UVR exposure.[193] p53 mutations have also been shown to
increase dose dependently with both acute and chronic UVR exposures.[194]
Many of these proxies of UVR exposure are expensive, requiring staff with high levels of training
and advanced equipment, prohibiting their use in large studies. More importantly none of these
methods inform in regards to whether the UVR exposure was chronic or intense, which determines both
sunburn and melanoma risk. Dosimetry is a method for quantifying UVR exposure that records data that
informs on timing, duration and intensity. Electronic ultraviolet (EUV) dosimeters, which measure
personal erythemally weighted UVR exposures have been used in numerous behavioral studies. With
appropriate calibration, Allen et al have shown that EUV can be engineered with spectral responsivities
and cosine response errors approaching those of meteorological-grade reference instruments.[195] EUV
not only reduces device related measurement error, but has sufficient temporal resolution to allow
correlation of UVR duration and intensity data with activities recorded by diary. This allows
noncompliant data to be identified, and factors such as sunscreen use and protective clothing to be
accounted for in analyses, further reducing bias.
41
Bias in Assessment of Melanoma Risk Factors
Most epidemiological studies of risk factors for melanoma have relied on retrospective self-
report to determine association, yet these assessments have repeatedly been shown to be prone to
measurement error.[27-30] For example in an early study repeated retrospective assessments among
cases and controls regarding melanoma risk factors displayed poor reliability (Kappa coefficient .37 to
0.57).[30] In a novel study, Cockburn et al nested a case control trial in the Twin Study cohort to assess
differences in recall of the same events between matched sets of twins disparate on melanoma
diagnosis.[28] OR’s based on case reported exposures for both twins were found to be higher than OR’s
based on control reports for sunbathing (as a child and adult) as well as and mole frequency and
freckling in childhood. Stratified analysis revealed that case based OR estimates were higher for
sunbathing as a child in cases that believed their melanoma was due to sun exposure (OR 8.0; 95
percent CI: 1.4, 14.5) compared to those that did not (OR 0.8; 95 percent CI: 0.2, 3.0). [28] A case
control study nested in the Nurse’s Health Study found tendency to sunburn during childhood assessed
at baseline and during follow-up to have poor reliability for both cases and controls (Kappa 0.45 and
0.42 respectively), with the magnitude and direction of change similar, displaying non-differential
bias.[27] However, a similar nested case control study in the Norwegian Women and Cancer cohort
found differential misclassification, with cases reporting greater shifts from baseline in reported hair
color, number of nevi and change in skin color after chronic sun exposure. This recall bias resulted in
odds ratios based on the retrospective assessments to be higher than the prospective based OR. All
retrospectively assessed risk factors however, displayed increases or decreases, with some even
changing direction, which differed between categories of the same variable, indicative of inherent
measurement error.[29]
42
Validation of self-reported UVR exposure
Very few studies have attempted to validate UVR exposure information obtained via
questionnaire with objectively measured data. The first study to do so was by Dwyer et al in 1996 in
Tasmania.[196] Students were administered questionnaires regarding average sun exposure and
melanoma risk factors in November of 1992 and March of 1993 and polysulphone badges were worn for
two consecutive weekends in November of 1993. Self-reported time spent outdoors and time spent in
the sun on the two previous questionnaires showed the best correlation with the polysulphone badge
measurements of UVR exposure with correlation coefficients of .65 for girls and .43 for boys for time
outdoors and .30 for girls and .53 for boys for time in the sun. Thus, this study found only moderate
correlation between objectively measured UVR exposure and self-report.
The lack of strong correlation in the study by Dwyer may have been because of the amount of
time between the questionnaire and the measure. More recently a study was done in melanoma
survivors, who wore a UVR sensor on their chest when outside for 10 days and made a self-report
nightly of their time spent outdoors.[197] A Network Flow Alignment framework was used to align self-
report and objective UVR sensor data to correct misalignment between the two measures. Under-
reporting of sun exposure time, defined as 30 minutes or more of underreported sun exposure on a day,
occurred on 51% of the days in the analysis. Most troubling was the fact that the rates of under-
reporting of sun exposure were highest for events that began from 12-1pm, a high UVI time of day,
greatly increasing melanoma risk.
Abstract
Melanoma incidence is rapidly increasing, with poor prognosis cases growing faster in California
Hispanics than in non-Hispanic Whites. Ultraviolet Radiation (UVR) exposure as a child has been found
to disproportionately increase the risk of melanoma. To determine correlates of UVR exposure in this
high-risk population, we conducted a study in predominately Hispanic 4th and 5th grade classrooms in
43
Los Angeles County, a high UVR environment. To address potential reporting bias, electronic UV
dosimeters were utilized to objectively measure the association between UVR and constructs
(acculturation, sun protective behavior and knowledge, family interventions) obtained on baseline
questionnaires. Tanning attitude (wanting to get a tan) was associated with lower median non-zero UVR
weekend time (1.73 minutes versus 22.17, AUC 82.08, Sensitivity 0.78, Specificity 0.73) and lower
median weekend average Erythemal UVR Dose (0.01 versus 0.18, AUC 80.16, Sensitivity 0.78, Specificity
0.70). Sun protective knowledge and family discussion of sunscreen were also inversely associated with
objectively measured UVR time. Students spent a median 30.61 (IQR 19.88) minutes outside per day,
with only 10.93 (IQR 14.75) median minutes of it occurring in non-school hours. We determined the
majority of UVR exposure in this population occurs at school, providing valuable guidance for future
interventions.
Introduction
Skin cancer is the most common malignancy in the United States, with melanoma responsible for
the majority of skin cancer mortality.[15] Rates of melanoma are rapidly increasing in the United States
with 99,780 new cases and 7,650 deaths predicted in 2022.[15, 17] Most cases of melanoma are
attributable to ultraviolet (UVR) radiation exposure [18-20], with UVR exposure and sunburns
experienced as a child greatly increasing risk of melanoma compared to similar exposure as an adult.[21,
22] California, with one of the highest rates of melanoma in the world, is a high UVR environment.[198]
Rates of poor prognosis melanoma, tumors thicker than 1.5 mm at diagnosis, are increasing in California
among Hispanics much faster than non-Hispanic Whites, concurrent with rapid growth in the Hispanic
population.[24, 25] Exacerbating this growing public health concern, survival after diagnosis of melanoma
is shorter in Hispanics compared to non-Hispanic Whites, largely due to disparate access to medical care
[199, 200]
44
The majority of UVR exposure studies have been conducted in non-Hispanic whites, with more
recent work extending to Hispanics.[26] These later studies have found evidence of disparities in sun
protective behavior, with Hispanic high school youth less likely to utilize sunscreen or sun protective
clothing and more likely to use tanning beds than non-Hispanic whites.[201] As seen in adults, high
school Hispanic youth have lower perceived risk of skin cancer and knowledge about self-screening than
non-Hispanic whites.[201, 202] Among Hispanic adults, greater acculturation, the process by which a
cultural group encounters and selectively adopts the beliefs and behaviors of another culture, has been
linked with lower utilization of sun protective clothing and increased risk of sunburn. [203, 204] Higher
levels of acculturation in Hispanics are also associated with increased sun cancer knowledge and
sunscreen use, but also less utilization of shade and more positive attitudes towards tanning, all
mediated by level of educational attainment, creating a unique challenge in this population.[203, 205-
207]
The heightened risk of UVR exposure in childhood coupled with the opportunity to influence
behavior over the lifetime makes knowledge about UVR exposure in this age group crucial to a successful
melanoma risk reduction program. Most risk reduction studies in this age group to date have utilized self-
reported sun exposure, which is only moderately associated with objectively measured UVR, impairing
the ability to detect meaningful associations. [197, 208, 209] To address this potential bias, we conducted
a novel study in predominately Hispanic youth, in a high UVR environment, objectively measuring UVR
with electronic dosimeters that have previously displayed concordance with meteorological-grade
reference instruments.[195] Extending our previous findings utilizing self-reported UVR exposure in this
population, we hypothesize that parental and peer sun protective behaviors will be inversely associated
with measured UVR exposure as will self-efficacy to sun protect, while sun protective barriers will be
positively associated with UVR exposure.[33] Similarly, we expect prior sunburns and level of
acculturation to have no association with UVR exposure.
45
Materials and Methods
Study Subjects
This study was nested in the SunSmart study, a school based, randomized intervention aimed to
elicit positive changes in sun protective attitudes, self-efficacy, knowledge and behaviors.[32] SunSmart
was conducted in 4th and 5th grade classrooms in Los Angeles County.[32-35] Classrooms were in
predominately Hispanic, Title I public schools with high percentage of low-income students. For this sub-
study, classrooms were selected as a sample of convenience from the 24 schools taking part in SunSmart.
The study was conducted in the spring semesters of 2014, 2015 and 2016. The University of Southern
California Institutional Review Board (IRB) approved the study and youth assent, as well as parental
permission, were obtained for all participants.
Study Design
As part of the main SunSmart study, baseline questionnaires were completed in classrooms to
assess the frequency of sunburn, sun protective behavior and attitudes, as well as other measures described
in detail below. Upon completion of the surveys, youths were recruited for the dosimeter sub-study.
Recruitment and Consent Parental permission and youth assent to participate in the dosimetry
component of the study were mailed to all potential recruits. Once consent and assent were obtained,
youth were mailed dosimeters and instructions for their use.
Collection of UVR data from dosimetry Students were instructed to wear dosimeters any time they
were outdoors (unless swimming) for a 2 week period, and to wear dosimeters so that they were in full
sunlight, taking care not to cover them with clothing or school bags. Dosimeters were worn on the wrist.
Parents received daily text reminders as well as phone calls every day to ensure youth compliance. At the
end of the two week period, the dosimeters were returned to study staff in the classroom, or returned by
mail, and participants were sent gift cards ($10 for a local grocery store) as compensation for participation.
Dosimeter data was collected before any in class intervention was done as part of the main SunSmart study.
46
Data on sun exposure knowledge, attitudes and behaviors, and other baseline covariates
Baseline questionnaires assessed the frequency of sunburn, sun protective behavior (use of sunscreen,
long sleeves and pants, shade seeking/peak hour sun avoidance) in self and peers, sun protective barriers,
sunscreen availability and parental communication and monitoring regarding sun protective behavior, as
previously reported.[34, 35] The questionnaires also assessed knowledge of sun protective methods and
sun exposure consequences. Estimated time spent outdoors, attitudes regarding sun protective behavior
as well as self-efficacy to this behavior were also quantified. The Acculturation, Habits, and Interests
Multicultural Scale for Adolescents (AHIMSA) scale was administered to determine the level of
acculturation to the United States [210]. Skin phototype was self-assessed by children who were given
visual images of five skin tones and their corresponding verbal descriptions. The skin phototype
descriptions ranged from 1 = very fair to 5 = very dark and were adapted from the Fitzpatrick skin
phototype scale.[211]
Summarizing UVR data from dosimeters
Personal erythemal UVR radiation exposure, as measured by the Ultraviolet Index (UVI), was
assessed using wearable electronic UV dosimeters. The lightweight battery-powered dosimeters (35 mm
diameter, 19 grams) measured total incoming solar radiation, using an erythemally-matched aluminum
gallium nitride (AlGaN) Schottky photodiode and a polytetraflouroethylene (PTFE) diffuser, as described in
detail elsewhere [195, 212]. The dosimeters were configured to continuously measure UVR irradiance (in
units of UVI) during daylight hours at 8 second sampling intervals, with the data time-stamped and stored in
on-board memory. The recorded UVR data was downloaded by study staff via a USB microport at the end of
the study period using proprietary software [195]. The data was integrated to provide two objective
measures of UVR exposure: (1) Erythemal UVR dose in units of standard erythemal dose (SED) where 1 SED
= 100 Jm
-2
of erythemally-weighted UVR; and (2) Daily outdoor minutes at different UVI thresholds, e.g. non-
zero UVI, UVI 1 to 3, and UVI greater than 3 (as shown in Table 10). The distribution of UVR minutes in each
47
category was tested for normality by visual inspection of graphed data and the Shapiro Wilk test using
alpha=.05. None of the data were normally distributed so medians and interquartile ranges (IQR) are
reported.
Data Preparation for ROC Analysis
In order to perform receiver operating characteristics analysis, answers to questionnaire data were
dichotomized. Questions using ordinal scales were split at the median of the scale. In a few rare instances
where > 70% of responses were in one category at an extreme of the scale, the high response group was
compared to all other groups. Answers to knowledge questions were dichotomized as correct versus
incorrect/don’t know. Overall acculturation to the United States and bicultural (affiliation to both the
United States and home country) scores were generated from answers to the AHIMSA scale, as previously
described, and dichotomized at the median [210]. For one question, responses on the questionnaire were
incorrectly categorized for the first year of the study, so this year was excluded from the analysis. For
another question all categories for the answer overlapped so the entire question was excluded from the
analysis.
Excluded Data If a dosimeter had no UVR exposure data for a weekday when other youth had UVR
data, it was assumed that the dosimeter had been left at home and that date was excluded from the
analysis. If no data was recorded at all for the entire two-week period, that subject was excluded from the
analysis. Observations with UVI less than .133 and greater than 13 were deleted as data in this range
appeared to be largely artifact.
ROC Analysis
Receiver operating characteristics analysis was performed for each UVR exposure category to
determine the AUC and UVR cutoff values when predicting the dichotomized data described above. When
identifying youth at high risk of excessive UVR exposure for possible low risk primary preventive
48
intervention, sensitivity is more important than specificity. Thus, UVR cutoffs were chosen that resulted in
the maximum sensitivity that had a corresponding specificity no more than 10% below it. Among cut-points
with the same sensitivity, but several specificities, the cut-point that corresponded with the maximum
specificity was chosen. Fisher’s exact test was used on two-by-two tables of dichotomized questionnaire
data when two questions were both found to have high sensitivity and specificity for a specific UVR
exposure category and the association between the two questions was of interest. All statistical analyses
were performed using SAS version 9.4 (SAS Institute, Cary, NC).
Results
Participation and Demographics
A total of 286 students were mailed participation packets. As this was a sub-study that had to occur
prior to the main study intervention to change sun protective behaviors, consent had to be obtained within
3 weeks of mailing the forms, allowing limited opportunity for follow-up on non-responders. A total of 98
(34.27%) students refused participation in the study (including non-responders). Thirty (10.49%) students
agreed to participate but their dosimeters were not received (lost by student, lost in the mail or never
returned). Seven (2.45%) students agreed to participate but their dosimeters malfunctioned resulting in no
usable data. Twenty-six (9.09%) students agreed to participate but their paperwork (consent, assent or
questionnaire) was incomplete, so their data was not included.
The final analysis included 125 (43.71%) students from four schools, who were accrued evenly over
the three years of the study, as summarized in Table 11 Demographics. The sample was predominately
Hispanic (79.2%) and equally distributed across gender and grade. Students were mostly well acculturated
(50.4%) and moderately identified (40.0%) with both the United States and their home country. Most
students self-reported having light brown skin (59.2%) and black hair (42.4%).
49
UVR Data from Dosimeters
Dosimeter data was collected over a median of 13 days (range 1-26), with the majority of data
collected on weekdays rather than weekends (median (range) 11 (1-25) versus 2 (0-9), respectively).
Dosimeter data are summarized in Table 10. Students spent a median 30.61 (IQR 19.88) minutes outside
(non-zero UVI) per day, with the majority of that time at UVI less than 1 - only a median of 4.21 (IQR 4.40)
minutes were recorded at UVI 1-3 and 0.61 (IQR 0.76) median minutes at UVI greater than 3. Most of this
sun exposure occurred at school, with a median 7.20 (IQR 7.73) minutes of it occurring during lunch, and
only 10.93 (IQR 14.75) median minutes outside of school hours. Students had much less sun exposure on
weekends with 18.27 (IQR 28.78) median recorded non-zero UVI minutes than on weekdays - 31.46 (IQR
20.97) median minutes.
Questionnaire Results
A large percentage of students (35.2%) reported having a sunburn in the last month, as well as
often or sometimes spending time outside to get a tan (87.70%). Very few (16.53%) reported having
sunscreen at home. Self-reported sun protective behavior varied. The majority of students reported often
or sometimes wearing long pants both in (84.00%) and outside of school (68.60%). Long sleeves were
often or sometimes worn in school (67.20%), but not outside of school (44.63%). Self-reported wearing
of hats and application of sunscreen, however, were infrequent. Hats were usually or often worn by only
15.20% of students in school and 28.93% outside of school, and sunscreen usually or often applied by
33.60% of students in school and 27.05% outside of school. Student recounted behavior of their friends
largely mimicked their own behavior. Students reported few parental sun protective interactions in the
last two weeks, with only 14.88% of parents speaking to them about sunscreen and 20.49% about wearing
a hat, long pants or long sleeves when it was sunny. Regarding sun protective knowledge, the majority of
students knew basic concepts like too much sunlight can cause skin cancer (62.30%) and it is not healthy
to get a tan (56.56%), but few knew more complex concepts like you can get sunburned on a cloudy day
50
(7.38%) or how often to apply sunscreen on a cloudy day (24.80). With regards to self-efficacy to practice
sun protection, the majority of students believed they could practice sun protective behavior in the future.
Students reported they could or probably could wear a hat (81.97%), play in the shade (92.56%), apply
sunscreen (65.29%) and wear long pants and sleeves (54.92%) when it was sunny.
ROC Results
Tanning Attitude Although only 8.20% of students reported wanting to get a tan, this response
was highly associated with average non-zero UVR weekend time (AUC 82.08, Sensitivity 0.78, Specificity
0.73). See Table 12 for the most interesting ROC results. Wanting to get a tan was also strongly associated
with average weekend SED (AUC 80.16, Sensitivity 0.78, Specificity 0.70). Students wanting to get a tan
had significantly less UVR exposure time on weekends than those that did not (median minutes (IQR):
1.73 (5.60) versus 22.17 (32.02)), as well as lower weekend SED (median SED (IQR): 0.01 (0.06) versus 0.18
(0.28)).
Answers to this question were also associated with average UVR time outside of school (AUC
77.23, Sensitivity 0.70, Specificity 0.76), with exposure again lower among students wanting a tan than
those that did not (median minutes (IQR): 3.31 (7.71) versus 11.49 (15.06)). Responses to this question
also showed a borderline association with time at non-zero UVI (AUC 66.96, Sensitivity 0.60, Specificity
0.62). Although exposure trends were in the same direction, the differences in exposure for time at non-
zero UVI were not as great between students wanting to get a tan and those that did not (median minutes
(IQR) 24.67 (13.12) versus 31.31 (19.99)). An association with wanting to get a tan was also seen with
weekend time spent between UVI 1-3 (AUC 73.41, Sensitivity 0.67, Specificity 0.63), once more with
students wanting to get a tan having less exposure than those that did not (median minutes (IQR) .13
(1.07) and 2.07 (3.65) respectively).
Similarly, the 8.20% of students who thought it was important to have a tan had much lower
weekend UVR exposure (median 6.20 minutes, IQR 15.60) than those that did not (median 22.13 minutes,
51
IQR 33.58), which also showed an association (AUC 71.08, Sensitivity 0.60, Specificity 0.70). They also had
a smaller SED on their lowest weekend day (median SED (IQR) 4.46 × 10
-3
(4.72 × 10
-3
) and 3.09 × 10
-2
(1.27
× 10
-1
) respectively), which also had an association (AUC 75.18, Sensitivity 0.60, Specificity 0.64).
Importance of a tan also showed borderline associations with overall time spent at UVI greater than 3
(AUC 67.28, Sensitivity 0.70, Specificity 0.60) and time spent at UVI greater than 3 during the week (AUC
66.29, Sensitivity 0.60, Specificity 0.63). The associations were also inverse, with students thinking it was
important to have a a tan having less UVI greater than 3 exposure time than those that did not, on both
weekdays (median minutes and IQR .27 (.29), .62 (.78) respectively) and weekends (median minutes and
IQR .25 (.47), .67 (.85) respectively), although exposure at this UVI level was very low overall.
In concordance with the previously mentioned tanning attitude associations, the 13.11% of
students who thought a tan made them more attractive had a smaller SED on their lowest day than
those that did not (median SED and IQR 2.00 × 10
-3
(7.66 × 10
-4
) and 2.95 × 10
-3
(4.71 × 10
-3
)
respectively). The association between belief in tanning attractiveness and lowest day SED was strong
(AUC 72.08, Sensitivity 0.69, Specificity 0.63), with a strong association also seen for lowest weekday
SED (AUC 70.84, Sensitivity 0.69, Specificity 0.62).
Sun Exposure Knowledge Students highly knowledgeable about sun exposure had lower recorded
UVR time in several categories. The 12.30% of students that knew that a UVR Index of 10 meant that
there was more light than a UVR Index of 3, had lower overall UVI greater than 3 exposure time (median
.32 minutes, IQR .45) than those that did not (median .63 minutes, IQR .88), which approached a
significant association (AUC 69.78, Sensitivity 0.66, Specificity 0.60). This signal was mostly coming from
the time spent during the week, which had a borderline association (AUC 68.19, Sensitivity 0.66, Specificity
0.60), as the students answering correctly also had lower UVI greater than 3 exposure time on the
weekdays (median .32 minutes, IQR .52) compared to those that answered incorrectly (median .69
minutes, IQR .97). Congruently, the highest weekday SED also had a borderline association (AUC 67.98,
52
Sensitivity 0.65, Specificity 0.60), with those answering correctly again limiting their exposure (median
(IQR): 0.65 (0.42) versus 0.84 (0.70)).
Analogously, the 7.38% of students that knew it was possible to get a sunburn on a cloudy day
also had lower non-zero UVI weekday exposure time (median 22.82 minutes, IQR 11.30) than those that
did not (median 32.18 minutes, IQR 21.13), which displayed a borderline association (AUC 66.76,
Sensitivity 0.66, Specificity 0.56). Knowing how often sunscreen should be applied on a sunny day was
also somewhat associated (AUC 66.03, Sensitivity 0.67, Specificity 0.58) with exposure time at UVI greater
than 3, with the 24.80% of students that had this knowledge recording less time (median .34 minutes, IQR
0.57) than those that did not (median .64 minutes, IQR .76).
Sun protective behavior, attitudes and self-efficacy The only self-reported student behavior
associated with UVR exposure time was application of sunscreen when outside of school. The 29.03% of
students reporting often or sometimes applying sunscreen during this time period had much lower UVR
exposure time on weekends (median 8.13 minutes, IQR 20.62) than those that rarely or never did (median
22.37 minutes, IQR 32.73), AUC 66.19, Sensitivity 0.65, Specificity, 0.56. Application of sunscreen outside
of school was also moderately associated with the highest weekend day SED (AUC 69.95, Sensitivity 0.70,
Specificity 0.63). The association was in the same direction, with the median SED lower for those with
more frequent sunscreen application (median 0.09, IQR 0.38 versus median 0.36, IQR 0.59).
Self-reports of predicted future sun protective behavior showed some associations with UVR
exposure. Interestingly, the 7.44% of students that said they could not or probably could not play in the
shade had lower daily cumulative UVI divided by total daily minutes at non-zero UVI on the highest day
measured (median .84, IQR .28) than those that said they could or probably could (median 1.04, IQR .43),
AUC 68.06, Sensitivity 0.67, Specificity 0.58.
53
Similarly, the 33.61% of students that reported that it was “true” or “a little true” that they did
not like the way they looked in a hat had lower daily average cumulative UVI divided by total daily minutes
at non-zero UVI (median .56, IQR .17) than those that said it was untrue or a little untrue (median .66, IQR
.17), AUC 67.75, Sensitivity 0.66, Specificity 0.58. This was also true of overall time between UVI 1-3
(median minutes and IQR 2.80 (3.78), 4.74 (4.02) respectively), AUC 66.49, Sensitivity 0.63, Specificity 0.53
and UVI time 1-3 on weekdays (median minutes and IQR 3.09 (3.78), 5.07 (3.89) respectively), AUC 66.23,
Sensitivity 0.66, Specificity 0.61 as well.
Peer and family influence Student reported peer sun protective behavior and parental sun
protective interactions also did not show any association with UVR exposure, other than a question asking
if students had spoken to anyone in their household about sunscreen in the last two weeks. Students
responding yes to this question (10.87%) had lower non-zero UVI weekend time (median minutes and IQR
4.33 (25.31), 21.43 (30.64) respectively) and lower weekend time between UVI 1-3 (median minutes and
IQR .31 (.30), 2.13 (3.80) respectively) than those responding no or maybe, which revealed borderline
associations (AUC 66.40, Sensitivity 0.68, Specificity 0.60 and AUC 68.96, Sensitivity 0.71, Specificity 0.70
respectively). In agreement with these results, children that reported parental interaction regarding
sunscreen had lower average weekend SED than those that did not (median (IQR): .03 (.17) versus .18
(.26)), which had a similar association (AUC 68.29, Sensitivity 0.72, Specificity 0.70). No association was
observed between a student’s bicultural score and UVR exposure, or their acculturation score and UVR
exposure.
Concordance of self-reported and measured UVR exposure time On the one question that asked
students to gauge their sun exposure (the time they spent outside at lunch while at school), their answers
were not associated with true recorded UVR exposure time. Students self-reporting 1-15 minutes of
exposure at lunch had 8.59 (IQR 8.35) recorded median minutes of UVR exposure during school lunch
periods, while those reporting greater exposure time (16 minutes or more) actually had lower 6.50 (IQR
54
8.73) median minutes of exposure (AUC 55.70, Sensitivity 0.50, Specificity 0.48). These answers, however,
were associated with of time spent at UVI 1-3 on weekends (AUC 72.49, Sensitivity 0.63, Specificity 0.57),
with again students reporting lower minutes outside at lunch having higher recorded UVR exposure times
on weekends than those self-reporting more time outside at lunch (median 2.20 (IQR 3.60) versus median
.23 (IQR 1.90)).
Discussion
We demonstrated that sun protective knowledge, attitude and family intervention are inversely
associated with objectively measured UVR exposure in predominately Hispanic youth living in a high UVR
urban environment. These UVR exposure associations were augmented by similar associations with SED.
In line with our previous findings, acculturation and sunburn in the last month at baseline were not
associated with objectively measured UVR exposure.[34] We observed that the majority of UVR exposure
in this population occurs at school, providing valuable guidance for future preventative interventions.
Although data are conflicting, numerous previous studies have found people to be inaccurate
reporters of their own sun exposure. [208, 209, 213, 214] In an effort to overcome this potential bias,
we objectively measured UVR exposure to more effectively determine its association with self-reported
sun protective constructs in predominantly Hispanic youth. Similar to previous studies using self-reported
sun exposure, our study found few of these sun protective constructs to be associated with actual
measured sun exposure. Our study extended prior research documenting erroneous reporting of sun
exposure to Hispanic youth, as we found students unable to accurately report their time outside at lunch.
While students overall sun exposure duration was not excessive, the preponderance of it occurred
during school hours. While it has been hypothesized many times that most sun exposure is incurred during
school hours, it has not often been objectively measured. One study using questionnaires estimated up
to 47% of children’s sun exposure occurred at school, while our objectively measured data indicate an
55
even higher proportion of 64%.[215] This novel finding presents a unique opportunity to have an
immediate impact on actual youth sun exposure by staging UVR exposure reduction interventions in
elementary schools, as opposed to the current knowledge and attitude focused counseling programs.
Adoption of the CDC guideline for school programs that recommend increasing shade structures next to
play and sports fields or even moving some physical activity to indoor areas may be a highly effective sun
exposure reduction measure given our results.[216] These changes minimizing UVR exposure time could
be implemented concurrently with other CDC recommendations such as routine sunscreen use before
going outside and education. Our study found sunscreen use and knowledge to be protective as well as
areas that need improvement and could be utilized along with reduction in UVR exposure in an evidence
based targeted approach to rapidly reduce UVR exposure in this high risk youth population. Overall, our
findings indicate these measures may have the largest public health impact and are further enforced by
meta-analysis indicating elementary school based interventions are associated with positive
outcomes.[217] .
Students that reported speaking to someone in their household about sunscreen use in the past
two weeks had less weekend low and midlevel UVI exposure time than those that did not. Similarly,
students reporting often or sometimes applying sunscreen when not in school also had lower exposure
time in these categories. This concurs with previous studies that have found parent and child sun
protective behavior to be highly associated.[34, 218, 219] Very few (16.53%) students reported having a
bottle of sunscreen at home. This information is extremely concerning given that this study was conducted
in a high UVR region and sunscreen utilization by children has been shown to be strongly associated with
frequency of parental application.[220, 221] Lack of sunscreen in homes may be due to disparate access,
as stores in predominantly Hispanic neighborhoods are half as likely to have sunscreen available for
purchase as stores in predominantly non-Hispanic White neighborhoods.[222] Based on these data,
interventions focused on the household, emphasizing sunscreen availability and use may have the most
56
influence on Hispanic children’s UVR exposure on the weekends. Hispanic households frequently involve
multiple generations, thus effective interventions and measurements of their effects should be tailored
to include all household members.[223]
We found higher sun protective knowledge led to less measured UVR exposure time at both low
and high levels during the week, as seen in an adult Danish dosimetry study.[23] Overall assessed sun
protective knowledge in our study, however, was low. Knowledge was assessed at baseline, albeit, prior
to any intervention. Numerous studies conducted by both ourselves and others have been repeatedly
demonstrated that interventions are effective in increasing sun protective knowledge. [32, 224, 225]
Of note, studies in adolescents have shown that while sun protective knowledge is necessary to
change sun protective behavior, it is unable to effect this change on its own, with attitudes towards
tanning having a greater association with self-reported UVR exposure.[226] Our study replicates these
previous findings as we found attitudes towards tanning to be the best predictor of objectively measured
UVR exposure.[23, 227, 228] There was concordance in our measures of tanning attitudes with students
reporting they wanted to get a tan statistically significantly more likely to report that it was important to
get a tan (p= 0.004). We also saw an inverse association between tanning attitudes and sun protective
knowledge, with students wanting to get a tan more likely to know how often to apply sunscreen on a
sunny day, which approached significance (p=.06). Interestingly, previous research, also found tanning
affirmative attitudes to be inversely associated with sun exposure knowledge and strongly associated with
reported UVR exposure in both college students and parents.[229-231]
Counterintuitively, children in our study that wanted to get a tan had less sun exposure. We
therefore may conclude that asking a child if they want to get a tan is not a good proxy for actual sun
exposure, perhaps due to parental oversight. Children reporting tanning affirmative attitudes in this
study had higher levels of knowledge about sun protective measures and belonged to households with
57
higher engagement in them compared to their peers not reporting these attitudes, who had more sun
exposure. As sun protective behavior has been shown to vary by level of acculturation, this may be a
finding specific to Hispanic children and should be explored in a multi-ethnic cohort.[203, 205-207]
Alternatively, the desire of these children to get a tan may be a yearning to fit in with their already
tan peers, that they are unable to act on due to high levels of parental oversight. It has been shown that
as children age, parental sun protective measures decrease and affirmative attitudes towards tanning in
these older children have been found to be predictors of higher levels of self-reported sun exposure.[230,
232, 233] Therefore, there is a strong possibility that this association may wane with increasing age. More
investigation needs to be done to determine if there indeed is an inverse relationship between tanning
attitudes and sun exposure in Hispanic youth and if this relationship varies throughout the lifespan. This
information would determine the potential benefit of targeting tanning attitudes as part of an effective
sun exposure intervention program in this high-risk population.
Limitations of this study include its small sample size, collection of data during one season only
and that it was conducted only in one city in an urban setting, limiting the generalizability of the findings.
Days with dosimeter readings of less than 16 seconds of non-zero UVI were considered days that
dosimeters were not worn and not included in the analysis. There is a chance that these actually were
days children did not go outside, resulting in our data reporting higher average UVR exposure times than
the true values. Additionally, our study did not measure air pollution, which has been shown to act
synergistically with sun exposure to create large amounts of oxidative stress on skin cells, contributing to
carcinogenesis.[135-137] Due to the design of the study, nested in the intervention, only three weeks
were available to obtain consent, lowering participation rates. Strengths of this study include the use of
objectively measured UVR exposure, examination of both UVR time and SED associations and a
predominately Hispanic sample in California, a location that we have previously shown is associated with
increased risk of invasive melanoma in this racial/ethnic group.[24]
58
In summary, we report the findings of our novel study determining associations between self-
reported sun protective knowledge, attitudes, self-efficacy, and peer and family influence with objectively
measured UVR exposure in Hispanic youth at high risk of invasive melanoma.
59
Tables
Table 8 Fitzpatrick Phototyping Scale
Type Features of unexposed skin Tanning and burning
1
very pale white skin, often with green or blue eyes
and fair or red hair
burns without tanning
2 white skin, often with blue eyes burns and does not tan easily
3 fair skin with brown eyes and brown hair burns first then tans
4 light brown skin, dark eyes, and dark hair burns a little and tans easily
5 brown skin, dark eyes, and dark hair
easily tans to a darker color and
rarely burns
6 dark brown or black skin, dark eyes, and dark hair never burns but tans darker
60
Table 9 Age Adjusted Rate of Cutaneous Melanoma per 100,000 by Race/Ethnicity (2014-2018 SEER
data)
Age Adjusted Rate of Cutaneous Melanoma per
100,00
Males Females
All Race/Ethnicities 29.3
18.0
White 34.7
22.1
Black 1.0 0.9
Asian /
Pacific Islander
1.5
1.2
American Indian /
Alaska Native
6.4
5.0
Hispanic 5.0
5.0
Non-Hispanic 33.1 20.7
61
Table 10 Exposure Times by UVI Categories
Time Period
Overall N=125
Median (IQR)
Weekday N=125
Median (IQR)
Weekend N=95
Median (IQR)
Ultraviolet Index (UVI) category
Average daily minutes at non-zero UVI
30.61 (19.88) 31.46 (20.97)
18.27
(28.78)
Average daily minutes at UVI 1 to 3 4.21 (4.40) 4.19 (4.48) 1.63 (3.58)
Average daily minutes at UVI greater than 3 .61 (.76) .64 (.87) .10 (.43)
Average daily minutes at non-zero UVI during lunch
1
NA 7.20 (7.73) NA
Average daily minutes at non-zero UVI outside of school
2
10.93 (14.75) NA NA
Average daily cumulative UVI divided by cumulative
minutes at non-zero UVI
0.62 (0.18)
NA NA
Daily cumulative UVI divided by cumulative minutes at
non-zero UVI –highest day measurement
1.00 (0.42)
NA NA
SED Average day 0.30 (0.20)
0.31 (0.24) 0.16 (0.26)
SED Highest day 0.87 (0.66) 0.81 (0.59) 0.30 (0.57)
SED Lowest day 2.68 x 10
-3
(4.02 x 10
-3
)
2.94 x 10
-3
(4.76 x 10
-3
)
0.02 (0.13)
1 Defined as 11:00 a.m. – 1:00 p.m. weekdays. 2 Defined as before 8 a.m. and after 3 p.m. on weekdays and any weekend time
62
Table 11 Demographics
n (%)
Gender
Female 61 (48.8%)
Male 64 (51.2%)
Race/Ethnicity
Hispanic 99 (79.2%)
Non-Hispanic 26 (20.8%)
Asian/Pacific Islander 3 (2.4%)
White 4 (3.2%)
Black/African American 12 (9.6%)
American Indian/Native American 3 (2.4%)
Other/Mixed Race 2 (1.6%)
Not Specified 2 (1.6%)
Grade
4
th
67 (53.6%)
5
th
58 (46.4%)
Age
≤ 9 years old 45 (36.0%)
10 Years Old 63 (50.4%)
11 Years Old 17 (13.6%)
Year
2014 41 (32.8%)
2015 42 (33.6%)
2016 42 (33.6%)
Hair Color
Blonde 2 (1.6%)
Light Brown 15 (12.0%)
Medium Brown 18 (14.4%)
Dark Brown 37 (29.6%)
Black 53 (42.4%)
63
Table 11 (continued)
n (%)
Skin Color
Very Fair 1 (0.8%)
Fair 14 (11.2%)
Light Brown 74 (59.2%)
Dark Brown 33 (26.4%)
Very Dark 1 (0.8%)
Missing 2 (1.6%)
Number of Sunburns in Last Month
0 times 81 (64.8%)
1 time 14 (11.2%)
2-3 times 22 (17.6%)
3-4 times 3 (2.4%)
5+ times 5 (4.0%)
US Acculturation Score (lower score is higher US acculturation)
Well 0 to 2 63 (50.4%)
Moderately 3 to 5 36 (28.8%)
Poorly 6 to 8 14 (11.2%)
Missing 12 (9.6%)
Bicultural Score (identifies with both US and home country)
0 to 2 24 (19.2%)
3 to 5 50 (40.0%)
6 to 8 39 (31.2%)
Missing 12 (9.6%)
64
Table 12 ROC Results
Question Survey Response UV Category N Median IQR AUC Sensitivity Specificity
At school in
the past
month,
about how
much time
were you
outside
each day
around
lunchtime?
1-15 minutes Average UV time
outside of school
64 12.10 15.52 66.33 0.60 0.61
16 minutes or more 20 7.20 9.32
1-15 minutes Average UV time at
lunch
64 8.59 8.35 55.70 0.50 0.48
16 minutes or more 20 6.50 8.73
1-15 minutes Average time at UV
1-3 on weekend
51 2.20 3.60 72.49 0.63 0.57
16 minutes or more 16 0.23 1.90
1-15 minutes Average weekend
SED
51 0.19 0.28 67.52 0.63 0.57
16 minutes or more 16 0.06 0.15
1-15 minutes Highest weekend
SED
51 0.32 0.58 65.81 0.63 0.59
16 minutes or more 16 0.11 0.39
When I was
outside but
not at
school in
the past
month, I
applied
sunscreen
Rarely/Never Average UV
weekend time
66 22.37 32.73 66.19 0.65 0.56
Often/Sometimes 27 8.13 20.62
Rarely/Never Average weekend
SED
66
0.20
0.29 66.55 0.70 0.63
Often/Sometimes 27 0.08 0.16
Rarely/Never Highest weekend
day SED
66 0.36 0.59 66.95 0.70 0.63
Often/Sometimes 27
0.09
0.38
You can get
a sunburn
on a cloudy
day.
Wrong Answer/Did
Not
Average UV
weekday time
113 32.18 21.13 66.76 0.66 .56
Correct Answer 9 22.82 11.30
65
Table 12 ROC Results (continued)
Question Survey Response UV Category N Median IQR AUC Sensitivity Specificity
A UV Index
of 10
means
there’s
more UV
light than a
UV Index of
3.
Wrong Answer/Did
Not
Average time at >
UV3
107 0.63 0.88 69.78 0.66 0.60
Correct Answer 15 0.32 0.45
Wrong Answer/Did
Not
Average time at >
UV3 on weekday
107 0.69 0.97 68.19 0.66 0.60
Correct Answer 15 0.32 0.52
Wrong Answer/Did
Not
Highest weekday
SED
107 0.84 0.70 67.98 0.65 0.60
Correct Answer 15 0.65 0.42
On a sunny
day, how
often
should you
put on
sunscreen?
Wrong Answer/Did
Not
Average time at >
UV3
94 0.64 0.76 66.03 0.67 0.58
Correct Answer 31 0.34 0.57
I don’t like
the way I
look in a
hat. For me,
this is:
Untrue/A little
untrue
Average daily
cumulative UVI
divided by
cumulative time
outside
81 0.66 0.17 67.75 0.66 0.58
A little true/True 41 0.56 0.17
Untrue/A little
untrue
Average time at UV
1-3
81 4.74 4.02 66.49 0.63 0.53
A little true/True 41 2.80 3.78
Untrue/A little
untrue
Average time at UV
1-3 on weekday
81 5.07 3.89 66.23 0.66 0.61
A little true/True 41 3.09 3.78
A good tan
makes me
look more
attractive.
For me, this
is:
Untrue/A little
untrue
Lowest day SED 106 2.95
× 10
-3
4.71
× 10
-3
72.08
0.69
0.63
A little true/True 16 2.00
× 10
-3
7.66
× 10
-4
Untrue/A little
untrue
Lowest weekday
SED
106 3.30
× 10
-3
5.91
× 10
-3
70.84
0.69
0.62
A little true/True 16 2.06
× 10
-3
1.03
× 10
-3
66
Table 12 ROC results (continued)
Question Survey Response UV Category N Median IQR AUC Sensitivity Specificity
It is
important
to have a
tan. For me,
this is:
Untrue/A little
untrue
Average UV
weekend time
83 22.13 33.58 71.08 0.60 0.70
A little true/True 10 6.20 15.60
Untrue/A little
untrue
Average time at >
UV3
112 0.62 0.78 67.28 0.70 0.60
A little true/True 10 0.27 0.29
Untrue/A little
untrue
Average time at >
UV3 on weekday
112 0.67 0.85 66.29 0.60 0.63
A little true/True 10 0.25 0.47
Untrue/A little
untrue
Average weekend
SED
83 0.18 0.27 66.99 0.60 0.63
A little true/True 10 0.05 0.18
Untrue/A little
untrue
Lowest weekend
day SED
83 3.09
× 10
-2
1.27
× 10
-1
75.18 0.60 0.64
A little true/True 10 4.46
× 10
-3
4.72
× 10
-3
I want to
get a tan.
For me, this
is:
Untrue/A little
untrue
Average UV time
outside of school
112 11.49 15.06 77.23 0.70 0.76
A little true/True 10 3.31 7.71
Untrue/A little
untrue
Average time at
non zero UVI
112 31.31 19.99 66.96 0.60 0.62
A little true/True 10 24.67 13.12
Untrue/A little
untrue
Average UV
weekend time
84 22.17 32.02 82.08 0.78 0.73
A little true/True 9 1.73 5.60
Untrue/A little
untrue
Average time at UV
1-3 on weekend
84 2.07 3.65 73.41 0.67 0.63
A little true/True 9 0.13 1.07
Untrue/A little
untrue
Lowest day SED 112 2.81
× 10
-3
4.54
× 10
-3
67.14 0.60 0.50
A little true/True 10 2.17
× 10
-3
8.82
× 10
-4
67
Table 12 ROC results (continued)
Question Survey Response UV Category N Median IQR AUC Sensitivity Specificity
(continued)
I want to
get a tan.
For me, this
is:
Untrue/A little
untrue
Average
weekend SED
84 0.18 0.28 80.16 0.78 0.70
A little true/True 9 0.01 0.06
Untrue/A little
untrue
Highest
weekend day
SED
84 0.32 0.59 79.63 .67 .76
A little true/True 9 0.03 0.11
In the last
two weeks,
have you
spoken
with
anyone in
your
household
about
sunscreen?
No/Maybe Average UV
weekend time
82 21.43 30.64 66.40 0.68 0.60
Yes 10 4.33 25.31
No/Maybe Average time
at UV 1-3 on
weekend
82 2.13 3.80 68.96 0.71 0.70
Yes 10 0.31 0.30
No/Maybe Average
weekend SED
82 0.18 0.26 68.29 0.72 0.70
Yes 10 0.03 0.17
I can play in
the shade.
Cannot/Probably
Cannot
Highest day
cumulative UVI
divided by
cumulative
time outside
9 0.84 0.28 68.06 0.67 0.58
Can/Probably Can 112 1.04 0.43
How many
times in the
past month
were you
sunburned?
0 times Average time
at > UV3
81 0.60
0.75
50.72 0.43 0.37
1 or more times 44 0.64
0.77
68
Table 12 ROC results (continued)
Question Survey Response UV Category N Median IQR AUC Sensitivity Specificity
How many
times in the
past month
were you
sunburned?
(continued)
0 times Highest day
cumulative UVI
divided by
cumulative time
outside
81 0.96 0.35 64.25 0.66 0.59
1 or more times 44 1.14
0.54
69
Figures
Figure 1 Benign Nevii
Figure 2 Atypical Nevi
Figure 3 ABCD Features
70
Chapter 3 Objectively Measured Ultraviolet Radiation Exposure as a Predictor of
Sunburn in Predominantly Hispanic Youth
Abstract
There is a developing epidemic of melanoma in the Hispanic population of California, with poor
prognosis cases growing faster in Hispanics than in non-Hispanic Whites. Ultraviolet Radiation (UVR)
exposure and sunburns experienced as a child incurs greater risk of melanoma when contrasted with
similar adult exposure. Current assessments of melanoma risk are primarily based on retrospective self-
report and have been shown to be biased and have poor reproducibility. Adopting a precision public
health approach by utilizing dosimeters to detect a signature (i.e. distinguishing) UVR that is associated
with sunburn could provide a less biased measure of lifetime melanoma risk. This UVR signature would
allow us to more accurately identify youth with modest UVR exposure versus those with hazardous UVR
exposure who are in need of intervention. To determine a signature UVR associated with sunburn, we
conducted a study in predominately Hispanic 4th and 5th grade classrooms in Los Angeles County, a high
UVR environment. After adjusting for confounders, the highest day’s average UVI per minute outside
was shown to have a statistically significant association with sunburn in the last month, p=.0.0009.
Students with hazardous UVR had 4.91 times the risk of reported sunburn (95% CI 1.92- 12.54)
compared to students with modest UVR. Use of sunscreen and hats were found to be associated with
higher risk of sunburn, presumably reflecting greater UVR exposure accompanying sunscreen
application and wearing hats, while wearing long pants was protective against sunburn, although none
of these associations reached statistical significance. Students with darker skin phototype had higher risk
of sunburn and lower rates of sun protective behavior, analogous with the increase of invasive
melanoma in Hispanic adults in California. In summary, we have demonstrated the potential of
precision public health by objectively determining a UVR signature that identifies youth at five times
greater risk of sunburn. This UVR signature provides a valuable tool to reduce health disparities through
71
targeted primary prevention, diminishing UVR during the most vulnerable exposure period, in those
objectively determined to be at highest risk of melanoma.
Introduction
Melanoma is responsible for the majority of skin cancer mortality.[17] In the United States in
2022, 99,780 new cases and 7,650 deaths due to the disease are estimated, with men and women
projected to have a 2.3% lifetime risk of the disease. [15, 17] Rates of incident melanoma are rapidly
increasing, rising on average 1.4% each year over 2009–2018.[132] As we previously reported, there
is a developing epidemic of melanoma in the Hispanic population of California, with rates of poor
prognosis melanoma, tumors thicker than 1.5 mm at diagnosis, increasing among Hispanics much faster
than non-Hispanic Whites, coinciding with rapid growth in this racial/ethnic group.[24, 25] Most cases
of melanoma are attributable to ultraviolet radiation (UVR) exposure. [18-20]. Similar increases in
melanoma incidence among Hispanics are being seen across the United States.[234] A high UVR
environment, California has one of the highest incidence rates of melanoma attributable to UVR
exposure in the United States. [235]
Most epidemiological studies of risk factors in melanoma have utilized retrospective self-report
to determine association, yet these self-reported data have repeatedly been shown to have
measurement error due to recall bias, with both random and systematic components, bringing into
question their validity.[27-31] Case control studies with repeated self-report assessments have shown
poor reliability. One such study assessing sunburn history had Kappa coefficients ranging from 0.37 to
0.57, while another assessing tendency to sunburn during childhood had poor reliability for both cases
and controls (Kappa 0.45 and 0.42 respectively.[27, 30]
Assessment of other melanoma risk factors, such as nevi, freckling and dermal response to sun
exposure, have also been shown to be prone to recall bias. A unique nested a case control study was
conducted to assess differences in recall of the same events between matched sets of twins discordant
72
for melanoma diagnosis.[28] Odds ratios (OR) based on case reported exposures for both twins were
higher than OR based on control reports for sunbathing (as a child and as an adult) as well as nevi
frequency and freckling in childhood. Stratified analysis revealed that case based OR estimates were
higher for sunbathing as a child for cases that believed their melanoma was due to sun exposure (OR
8.0; 95% CI: 1.4, 14.5) than for those that did not (OR 0.8; 95% CI: 0.2, 3.0).[28] A similar nested case
control study in the Norwegian Women and Cancer cohort found differential misclassification of risk
factors, with cases displaying greater shifts in reported hair color, number of nevi and change in skin
color after chronic sun exposure. This recall bias resulted in OR on these risk factors based on the
retrospective assessments to be higher than the OR based on prospective assessments. All risk factors
retrospectively assessed however, displayed accentuation or attenuation, with some changing direction,
which differed between categories of the same variable.[29]
There is little that can be done to correct recall bias in completed epidemiologic studies, unless
information is available from validation studies that can be used for adjustment. Unfortunately, very few
such studies have been done to date, with one study only finding moderate correlation between self-
reported and objectively measured UVR and another finding underreporting of time outside by 30
minutes or more on 51% of days collected, with the underreporting most frequently occurring between
noon and 1 pm, a known period of high ultraviolet index (UVI).[196, 197]
Spurious associations may be discovered when the association between risk factors and
melanoma are being inaccurately estimated, while other significant associations may be left
undiscovered. As these studies inform our public health prevention strategies, accurate assessment is
critical. We have no way of knowing if current public health primary prevention strategies will improve
melanoma rates, as they may be targeting associations that are largely due to bias and artifact and may
be missing substantial actionable associations that could have significant impact.
73
Utilizing dosimeters to collect UVR data may potentially help modulate the bias seen in self-
reported UVR exposure. Dosimeters allow more accurate recording of time spent outdoors than self-
report. Additionally, self-report measures fail to obtain data on UVI, which is needed in order accurately
determine risk of sunburn and is obtained by dosimetry. Sunburns, the most commonly used proxy for
UVR exposure in epidemiologic research, increase risk of melanoma for everyone, but sunburns
experienced as a child increase risk of melanoma more than a similar sunburn exposure as an adult.[21,
22] While we are unable to utilize dosimeters to objectively collect past UVR exposure for adult case
control studies, we have the opportunity in childhood to objectively identify youth experiencing
hazardous UVR via dosimetry and intervene. A validated UVR signature identifying exposure in a
predominantly Hispanic sample of children that is associated with sunburn, would give researchers a
valuable tool, allowing targeting of resources to diminish UVR in those objectively determined to be at
highest risk for melanoma, during their most vulnerable exposure period.[236] This is the first step in
employing precision public health, utilizing technology to more accurately assess risk and provide “the
right intervention to the right population at the right time”, potentially reducing health disparities.[237]
In a first step to discover this potentially impactful UVR signature, we conducted an analysis utilizing
data from a novel study of objectively measured UVR exposure in predominately Hispanic youth residing
in a high UVR environment.
Materials and Methods
Study Subjects
This study was nested in the SunSmart study, a school based, randomized intervention
aimed to elicit positive changes in sun protective attitudes, self-efficacy, knowledge and
behaviors.[32] SunSmart was conducted in 4th and 5th grade classrooms in Los Angeles
County.[32-35] Classrooms were in predominately Hispanic, Title I public schools with similar
74
socio-economic status. For this sub-study, classrooms were selected as a sample of convenience
from the 24 schools taking part in SunSmart. The study was conducted in the spring semesters of
2014, 2015 and 2016. The University of Southern California Institutional Review Board (IRB)
approved the study and youth assent, as well as parental permission, were obtained for all
participants.
Study Design
As part of the main SunSmart study, baseline questionnaires were completed in classrooms
to assess the frequency of sunburn and sun protective behavior and attitudes, as well as other
measures as previously reported. [34, 35] Upon completion of the surveys, youths were recruited
for the dosimeter sub-study. Parental permission and youth assent to participate in the dosimetry
component of the study were mailed to all potential recruits. Once consent and assent were
obtained, youths were mailed dosimeters and instructions for their use.
Baseline questionnaires assessed the frequency of sunburn and sun protective behavior
(use of sunscreen, hats, long sleeves and pants both in and out of school) in the past month, as
well as demographics and other measures, as previously reported.[34, 35] Skin phototype was self-
assessed by children who were given visual images of five skin tones and their corresponding verbal
descriptions. The skin phototype descriptions ranged from 1 = very fair to 5 = very dark and were adapted
from the Fitzpatrick skin phototype scale.[211]
Collection of UV data from dosimetry Students were instructed to wear dosimeters any time they
were outdoors (unless swimming) for a two week period, and to wear dosimeters so that they were in full
sunlight, taking care not to cover them with clothing or school bags. Dosimeters were worn on the wrist.
Parents received daily text reminders as well as phone calls every day to ensure youth compliance. At the
75
end of the two-week period, the dosimeters were returned to study staff in the classroom, or returned by
mail, and participants were sent gift cards ($10 for a local grocery store) as compensation for participation.
Dosimeter data was collected before any in class intervention was done as part of the main SunSmart
study.
Summarizing UV data from dosimeters
Personal erythemal UVR radiation exposure, as measured by the Ultraviolet Index (UVI), was
assessed using wearable electronic UV dosimeters. The lightweight battery-powered dosimeters (35 mm
diameter, 19 grams) measured total incoming solar radiation, using an erythemally-matched aluminum
gallium nitride (AlGaN) Schottky photodiode and a polytetraflouroethylene (PTFE) diffuser, as described in
detail elsewhere [195, 212]. The dosimeters were configured to continuously measure UVR irradiance (in
units of UVI) during daylight hours at 8 second sampling intervals, with the data time-stamped and stored in
on-board memory. The recorded UVR data was downloaded by study staff via a USB microport at the end of
the study period using proprietary software [195].
After download, the total cumulative UVI for each day was divided by the total cumulative minutes
at non-zero UVI for the day, making a variable describing the average UVI per minute when outdoors for the
day. The highest days value for each subject was then selected to determine each student’s highest day
average UVI per minute outdoors. The distribution of the variable was tested for normality by visual
inspection of graphed data and the Shapiro Wilk test using alpha=.05. The variable was not normally
distributed (<0.0001) so the median and interquartile range (IQR) are reported.
Data Preparation
In order to perform receiver operating characteristics analysis, answers to behavioral questions
were dichotomized as often/sometimes and rarely/never. If a dosimeter had no UV exposure data for a
weekday when other youth had UV data, it was assumed that the dosimeter had been left at home and that
76
date was excluded from the analysis. If no data were recorded at all for the entire two-week period, that
subject was excluded from the analysis. Observations with UVI less than .133 and greater than 13 were
deleted as data in this range appeared to be largely artifact.
Statistical Analysis
Receiver operating characteristics analysis was performed using logistic regression with self-
reported sun burn in the last month (yes/no) as the dependent variable and highest day average UVI per
minute as an independent variable, adjusted for variables that have been previously shown to reduce
the risk of sunburn: student’s self-reported skin color (very fair/fair/light brown versus dark brown/very
dark) and how often students reported using sunscreen (often/sometimes versus rarely/never) both in
and out of school. The following variables that also may influence the probability of sunburn were
tested for possible confounding: gender (boy vs. girl) and grade (fourth vs. fifth) as well as each of the
following separately for both in and out of school (often/sometimes versus rarely/never) wearing a hat,
long sleeves and long pants.
When identifying youth at high risk of excessive UV exposure for possible preventive
intervention, which carries low risk, sensitivity is more important than specificity. Thus, a cutoff for the
final model was chosen that resulted in the maximum sensitivity that had a corresponding specificity no
more than 10% below it. Among cut points with the same sensitivity, but several specificities, the cut
point that corresponded with the maximum specificity was chosen. In order to avoid potential loss of
power by using the median as the cut point to dichotomize the average UVI per minute variable, the cut
point was chosen that maximized the AUC and sensitivity in a univariate model, while maintaining an
acceptable specificity. [238] The cut point for average UVI per minute was used to classify students into
those receiving lower levels of UVR, referred to as “modest” and higher referred to as “hazardous”.
77
As it has been previously reported that utilization of the 10% change in estimate criterion is
inaccurate, with appropriate cutoffs varying by many factors, a very thorough approach to identify
confounders was undertaken .[239, 240] In order to understand the relationship between variables,
models were explored adding each confounder individually and in combination to a baseline model
consisting of average UVI per minute and known confounders, noting the change in AUC, sensitivity,
specificity and average UVI per minute estimate with each model. The best model was found using this
approach. A second analytic approach was then conducted which ran best subsets analysis and models
that contained all confounders identified in the first approach were examined to see if the addition of
any other variables improved AUC, sensitivity and specificity, to ensure no important associations were
missed. The best model utilizing this approach was selected. Finally, a third analytic method was
applied, that employed slight modifications to the standard 10% change in estimate criteria. Possible
confounders were tested by entering each variable into the initial baseline model separately to see if
they changed the estimate of the average UVI per minute odds ratio by more than 10%, in which case
they were retained in the model. Once this was done variables that were found not to be confounders
initially were tested for confounding again in the new model both individually and in pairs. If a variable
changed the estimate of the average UVI per minute odds ratio by more than 10% by itself it was
retained in the model. If two variables changed the estimate of the average UVI per minute odds ratio
by more than 20% in combination, they were both retained in the model. This iterative process
continued until no additional confounders were found and the model using this method determined.
Models using all three methods were compared and a final model chosen, ensuring it contained only
true confounders while maximizing AUC, sensitivity and specificity. . Fisher’s exact test was used in two-
by-two tables of dichotomized questionnaire data when the association between two confounders was
of interest. All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC).
78
Results
A large percentage (35.20) of students reported sunburn in the last month. The majority
(66.40%) of students reported rarely or never applying sunscreen while in school in the last month, with
even more (71.20%) reporting they rarely or never applied sunscreen outside of school. Gender, grade
and wearing long pants or a hat outside of school did not confound the relationship between average
UVI per minute and sunburn and therefore are not included in the final model (Tables 13 and 14).
Variables that have been previously shown to reduce the risk of sunburn, use of sunscreen and skin
phototype, were included in the final model. Students with fairer skin had decreased risk of sunburn
when compared to those with darker skin and students that rarely used sunscreen both in and outside
of school also had decreased risk of sunburn, although none of these variables reached statistical
significance. Wearing a hat or long pants in school both negatively confounded the relationship
between average UVI per minute and sunburn, thus both were included in the final model. Rarely or
never wearing a hat at school decreased the risk of sunburn, odds ratio (OR) 0.46 (95% CI 0.17-1.28),
compared to often or sometimes wearing one, while rarely long wearing pants at school increased the
risk of sunburn OR 2.26 (95% CI 0.65-7.88), compared to often or sometimes wearing them, although
neither reached statistical significance.
Students were fairly equally split between how often they wore long sleeves at school
(rarely/never 50.75%, often/sometimes 49.25%), but those that rarely/never wore them at school were
much more likely to rarely/never wear them outside of school than to often/sometimes wear them
(85.00% and 15.00% respectively). Students that rarely or never wore long sleeves outside of school
were 8 times (OR 8.24 (95% CI 3.11-21.84) more likely to rarely or never wear them in school when
compared to students that often or sometimes wore long sleeves outside of school. No similar
association was seen in students that reported often/sometimes wearing long sleeves at school, with
40.74% reporting rarely/never wearing them outside of school and 59.26% often/sometimes. The
79
association between these two variables was statistically significant, p<.0001. These variables did not
change the estimate of the average UVI per minute odds ratio by more than 10% when entered
univariately but changed the estimate by more than 20% when entered jointly so they were retained in
the model. In the final model these were the only two odds ratios that had significant associations,
other than average UVI per minute, the association of interest. After adjusting for student’s self-
reported skin color, use of sunscreen (separately for in school and out of school), wearing a hat in school
and wearing long pants in school and the highest day measurement of daily cumulative UVI divided by
total daily minutes at non-zero UVI, students that reported rarely or never wearing long sleeves when
outdoors and not at school had 3.37 times the risk of sunburn compared to those that often or
sometimes did (95% CI 1.18-9.60). Additionally, students that reported rarely or never wearing long
sleeves at school had .27 times the risk of sunburn compared to those that often or sometimes did
(95% CI 0.09-0.84).
After adjusting for student’s self-reported skin color and the following behaviors in the past
month: use of sunscreen (separately for in school and out of school), wearing long sleeves (separately
for in school and out of school), wearing a hat in school and wearing long pants in school, the highest
day measurement of daily cumulative UVI divided by total daily minutes at non-zero UVI was shown to
have a statistically significant association with whether or not students reported having a sunburn in the
last month, p=.0.0009. Students with a score of greater than 1.04 on this summary measure had 4.91
times the risk of sunburn (95% CI 1.92- 12.54) compared to students with a score of 1.04 or less. The
AUC for the final model was 80.15 with sensitivity of 76.20 and specificity of 67.50 using a .30 cut point.
When compared to using the median (1.00) as a cut point for average UVI per minute variable, using the
value that optimized AUC and sensitivity while maintaining a specificity not more than 10% below the
sensitivity in a univariate model (1.04) augmented the final model (Table 15). This method improved the
AUC of the final model from 78.99 to 80.15 and the sensitivity and specificity from 71.4 and 63.6 to
80
76.20 and 67.50, respectively, with the OR estimate increasing by more than one, 3.79 (95% CI 1.51-
9.47) versus 4.91 (1.92- 12.54). Our rigorous analytic approach resulted in even greater improvement
when compared to a model using the standard approach of a median cut point and the 10% change in
estimate criteria, which resulted in AUC of 76.24, sensitivity 73.80, specificity 67.90, OR 3.03 (95% CI
1.29-7.13).
Discussion
Using objectively collected UVR in a racial/ethnic group developing an epidemic of melanoma, we
determined a UVR signature capable of identifying youth at five times greater risk of sunburn.[24]. We
also demonstrated that sunscreen and hat use increased sunburn risk, while use of long sleeves outside
of school decreased risk of sunburn, providing valuable guidance for future preventative interventions.
Surprisingly, students that rarely or never used sunscreen, both at home and at school, had
lower risk of sunburn than students that often or sometimes did. This may be explained by lower
measured UVR exposure, as assessed by average UVI per minute in students with rare sunscreen use
when compared to students with more frequent use. These results are to be expected. To be effective
sunscreens must be reapplied at 2 hour intervals, all areas of skin covered and sufficient quantity used,
but studies have shown people generally use only 20–50% of an adequate amount of sunscreen and do
not reapply sunscreen frequently enough, resulting in sunburn.[241, 242]
Our findings bring in to question the efficacy of sunscreen. Sunburns have previously been
found to occur more often on days sunscreen is used, [243, 244] which we have observed in our case
control study of melanoma in twins, where use of sunscreen was associated with higher risk of sunburn
and time spent outside (unpublished data from published work).[245] Concordantly, a recent analysis
data from the 2015 National Health Interview Survey focusing on older adults found that the only sun
protective behavior significantly associated with sunburn was sunscreen use (adjusted prevalence ratio
81
aPR = 1.27; CI = 1.05, 1.52).[246] Reliance on protective clothing may be more efficacious in sunburn
prevention.
We found that rarely wearing a hat at school also decreased risk of sunburn. Hats cover a very
small percentage of the body’s skin, the top of the head only, not the neck or the ears in the case of
baseball caps which we observed were most frequently worn in this study. Furthermore, people apply
sunscreen and put hats on hats only when they are planning to go outdoors, expecting exposure to UVR,
thus people using hats and sunscreen are already clearly at increased risk of sunburn compared to those
that do not. Use of hats and sunscreen, may cause students to falsely believe they are well protected
from the sun, leading to overexposure and resulting sunburn.
Students with darker skin phototype (dark brown or very dark brown) had higher risk of sunburn
when compared to students with lighter skin phototype (very fair, fair or light brown), as well as more
sun exposure as measured by average UVI per minute. This was true in both univariate and multivariate
models. Students with darker skin phototype were more likely than students with fairer skin to rarely
wear long sleeves both in (41.18 vs 29.21) and outside of school (59.38 vs 52.87), as well as long pants
outside of school (37.50 vs 28.74). In line with our results, it has previously been reported that children
with lighter skin phototypes take more sun protective measures than children with darker skin
phototype.[247] Students with darker skin phototype may overestimate protection from sunburn due to
their skin pigmentation, leading to less frequent use of sun protective measures and overexposure to
UVR. Their overexposure to UVR is evidenced by higher rates of reported sunburn in this group, which
by default indicates an excess of UV exposure, independent of skin phototype or use of protective
measures. Concordantly, 2015 National Health Interview Survey data recorded a surprisingly high
prevalence of sunburn among Black (13.2%; 95% CI, 11.6%-15.1%) and Hispanic (29.7%; 95% CI, 27.6%-
31.9%) adults in the United States. .[248] These findings may partially explain the growing epidemic of
melanoma in California among Hispanics.[24]
82
While rarely wearing long sleeves in school was protective for sunburn, rarely wearing long
sleeves outside of school increased risk of sunburn. Regarding use of long sleeves, children with the
lowest rates of sunburn were those that rarely wore long sleeves in school and often wore them outside
of school (16.67% sunburn rate) while children that wore long sleeves often in school and rarely outside
of school had the highest sunburn rates (51.52%). Students who reported frequently wearing long
sleeves in school were more likely to report they spoke to someone in their household about wearing
sun protective clothing, with 25.39% of students often wearing long sleeves at school reporting speaking
to their parents compared to 12.50% of students that rarely did. Students wearing long sleeves at
school had higher levels of measured UVR exposure, as assessed by average UVI per minute. Differences
in reported parental interactions were even more pronounced for wearing long sleeves outside of
school with children often wearing them reporting a parental inquiry rate of 33.33% and students rarely
wearing them a 10.45% rate. Students that rarely wore long sleeves outside of school also had higher
levels of the more dangerous UVR - highest day standard erythemal dose and average time at UVI
greater than 3. When it came to use of long sleeves, children with the lowest rates of sunburn (rarely
wore long sleeves in school and often wore them outside of school) had a parental inquiry rate of
33.33% while those with the highest rates of sunburn (long sleeves often in school and rarely never
outside of school) had a significantly lower inquiry rate of 12.12%. Thus, wearing long sleeves, both in
and outside of school, appears to be associated with and may be a proxy for parental sun protective
interactions. Due to the moderate Los Angeles climate, the choice to wear long sleeves outside of
school, when you may have longer periods of sun exposure, is likely to reflect sun protective behavior.
UVR exposure, as measured by average UVI per minute was highly associated with sunburn.
Students with hazardous UVR exposure (above the 52.5 percentile) have almost 5 times the risk of
sunburn as students with more modest exposure. UVR exposure, therefore, is the best predictor of
sunburn risk and measures taken to reduce UVR exposure may be the most effective intervention. People
83
who never go out in the sun are not at risk of sunburn, despite lack of sun protective clothing and
sunscreen. Therefore, acquiring information about individual’s sun protective clothing and sunscreen use,
without obtaining concurrent UVI exposure information is not informative for risk of sunburn. This
interpretation is reinforced by findings in other studies. Analysis of NHANES data, restricted to those most
easily sunburned, non-Hispanic Whites, revealed the odds of multiple sunburns were significantly lower
in individuals who frequently avoided the sun by seeking shade (OR = 0.70, p < 0.001) [249] Additionally,
analysis of 2015 National Health Interview Survey data found that the only sun protective measures that
effectively reduced sunburn were sun avoidance behaviors (seeking shade and not going in the sun, p<
.001).[248]
Application of a precision public health approach by utilizing technology to collect UVR data with
a higher level of granularity than traditional self-report, combined with a very thorough analytic method
resulted in the ability to identify youth at much higher risk of melanoma than our previous studies. Using
the 52% percentile of average UVI per minute to dichotomize the variable improved the AUC, sensitivity
and specificity of the model compared to using a median cut point, as well as substantially increased the
risk estimate. This highlights the importance of thorough exploration of data and not routinely defaulting
to a median cut point. As previously observed, use of quantiles, such as the median, as cut points is
problematic as it assumes homogeneity of risk within groups, resulting in loss of power and inaccurate
estimation.[238] Furthermore, use of quantiles makes it difficult to compare results across studies as the
cut points used to define categories are data driven. Our use of an innovative method of identifying
confounders resulted in a better performing model. It has been previously shown that the standard 10%
cutoff for the change in estimate criterion used to control for confounders is flawed, with appropriate
cutoffs varying by the effect size of the exposure–outcome relationship, sample size, standard deviation
of the regression error, and exposure–confounder correlation, thus it behooves us to explore alternative
methodology.[239, 240]
84
Additionally, it is important to think of the context of the ultimate application of your analysis
when picking cut points for your final model. Routine methods for selection of cut points when conducting
ROC analysis, such as the point on the ROC curve closest to (0,1) and Youden’s index J, are appropriate
when sensitivity and specificity of tests are equally important.[250] This is not the case in a lot of potential
applications. When screening individuals to determine candidates for a primary prevention intervention,
sensitivity is much more important than specificity and should be weighted more highly when determining
the cut point of a predictive model.
This study has some limitations. While adequate for our primary endpoint, our sample size was
not large, resulting in wide confidence intervals and uncertainty in results regarding skin color, use of
pants and hats. We collected data only during one season, spring, which may result in determining a
higher or lower hazardous UVI signature when compared to results obtained in other seasons. Our
findings have limited generalization as this study was conducted in only one metropolis, with no rural
setting. People living in rural settings may spend more time outside and generally are exposed lower levels
air pollution, which has been shown to act synergistically with sun exposure to create large amounts of
oxidative stress on skin cells, creating a different risk profile.[135-137] We did not include days with
dosimeter readings of less than 16 seconds of non-zero UVI in the analysis, as we considered those days
that dosimeters were not worn. However, children may just not have gone outside these days, which
would result in our data overestimating UVR exposure times compared to the true values. To accurately
record UVR dosimeters must be worn correctly. We have no way of knowing when the dosimeters were
correctly worn by students and when they were not, thus the data we are reporting may be
underestimating UVR exposure time compared to the true value. Additionally, we did not account for use
of other sun protective methods in our analysis, such as use of sunscreen and sun protective clothing,
resulting in a possible overestimation of UVR exposure. Due to the design of the study, nested in the
intervention, only three weeks were available to obtain consent, lowering participation rates, potentially
85
limiting external validity. Our sunburn data was retrospectively collected, making it subject to recall bias
and inaccurate estimates of association. Strengths of this study include the use of dosimeters to
objectively measure UVR exposure, and a predominately Hispanic sample in a high UVR environment,
which we have shown is undergoing a developing epidemic of invasive melanoma.[24]
In summary, we found in this high-risk population that having darker skin phototype, using
sunscreen and wearing hats was associated with higher risk of sunburn. These sun protective measures
and skin phototype may lead to greater UVR exposure, due to exaggerated estimates of protection,
resulting in sunburn. Although not statistically significant, use of long pants was found to consistently
reduce risk of sunburn, perhaps due to more extensive skin coverage. Higher levels of objectively
determined UVR exposure were associated with elevated risk of sunburn. Our data indicate that future
sun protective interventions should focus on sun avoidance and reinforce the need for sun protective
education in individuals with darker skin phototypes, while more information is needed regarding
utilization of long pants. We demonstrated the potential of precision public health by objectively
determining a UVR signature via dosimetry that allows us to accurately identify youth at 5 times greater
risk of sunburn, providing a valuable tool to potentially reduce health disparities using targeted primary
prevention to diminish UVR during the most vulnerable exposure period, in youth objectively
determined to be at highest risk of melanoma. This finding should be validated in future prospective
studies that include a primary prevention component.
86
Tables
Table 13 Demographics
n (%)
Gender
Female 61 (48.8%)
Male 64 (51.2%)
Race/Ethnicity
Hispanic 99 (79.2%)
Non-Hispanic 26 (20.8%)
Asian/Pacific Islander 3 (2.4%)
White 4 (3.2%)
Black/African American 12 (9.6%)
American Indian/Native American 3 (2.4%)
Other/Mixed Race 2 (1.6%)
Not Specified 2 (1.6%)
Grade
4
th
67 (53.6%)
5
th
58 (46.4%)
Age
≤ 9 years old 45 (36.0%)
10 Years Old 63 (50.4%)
11 Years Old 17 (13.6%)
Year
2014 41 (32.8%)
2015 42 (33.6%)
2016 42 (33.6%)
Hair Color
Blonde 2 (1.6%)
Light Brown 15 (12.0%)
Medium Brown 18 (14.4%)
Dark Brown 37 (29.6%)
Black 53 (42.4%)
Skin Color
Very Fair 1 (0.8%)
Fair 14 (11.2%)
Light Brown 74 (59.2%)
Dark Brown/Very Dark 34 (27.2%)
Missing 2 (1.6%)
87
Table 13 (continued)
Number of Sunburns in Last Month n (%)
0 times 81 (64.8%)
1 time 14 (11.2%)
2-3 times 22 (17.6%)
3 or more 8 (6.4%)
US Acculturation Score (lower score is higher
US acculturation)
Well 0 to 2 63 (50.4%)
Moderately 3 to 5 36 (28.8%)
Poorly 6 to 8 14 (11.2%)
Missing 12 (9.6%)
Bicultural Score (higher score indicates
identifies with both US and home country)
Low 0 to 2 24 (19.2%)
Moderate 3 to 5 50 (40.0%)
High 6 to 8 39 (31.2%)
Missing 12 (9.6%)
88
Table 14 Questionnaire Items, Responses and Associations
Variable/Question Measurement/
Response
Count
(Percent)
Highest Day
Average UVI per
minute outside
median (IQR)
Odds Ratio (95%
CI)
Univariate
Odds Ratio (95% CI)
in final model
How many times in the PAST
MONTH were you
SUNBURNED?
1 or more times 44 (35.20) 1.14 (0.54) outcome variable outcome variable
0 times* 81 (64.80) 0.96 (0.35)
Highest day measurement of
daily cumulative UVI divided by
total daily minutes at non-zero
UVI
> 1.04 59 (47.20) 1.32 (0.42) 3.29 (1.52-7.09)
4.91 (1.92-12.54)
< 1.04* 66 (52.80) 0.84 (0.24)
Skin Color Very Fair, Fair or
Light Brown
89 (71.20) 0.98 (0.45)
0.55 (0.24-1.25) 0.42 (0.16-1.12)
Dark Brown or Very
Dark Brown*
34 (27.20) 1.08 (0.43)
Missing 2 (01.60) 1.00 (0.40)
At SCHOOL in the PAST
MONTH, I applied sunscreen
with SPF 15 or higher.
Rarely/Never 83 (66.40) 0.99 (0.39) 0.33 (0.15-0.71) 0.46 (0.17-1.28)
Often/Sometimes* 42 (33.60) 1.04 (0.45)
When I was outside but NOT at
school in the PAST MONTH, I
applied sunscreen.
Rarely/Never 89 (71.20) 1.00 (0.43) 0.34 (0.15-0.78) 0.44 (0.15-1.27)
Often/Sometimes* 33 (26.40) 1.02 (0.35)
Missing 3 (2.40) 0.99 (0.51)
At SCHOOL in the PAST
MONTH, I wore a hat.
Rarely/Never
106 (84.80) 1.03 (0.42)
0.55 (0.20-1.47) 0.46 (0.17-1.28)
Often/Sometimes* 19 (15.20) 0.92 (0.45)
When I was outside but NOT at
school in the PAST MONTH, I
wore a hat.
Rarely/Never 86 (68.80) 0.98 (0.39) 0.64 (0.29-1.44) NA – not in final
model
Often/Sometimes* 35 (28.00) 1.02 (0.61)
Missing 4 (3.20) 1.04 (0.36)
At SCHOOL in the PAST
MONTH, I wore long sleeves to
protect myself from the sun.
Rarely/Never 41 (32.80) 0.94 (0.58) 0.67 (0.30-1.50) 0.27 (0.09-0.84)
Often/Sometimes* 84 (67.20) 1.03 (0.37)
When I was outside but NOT at
school in the PAST MONTH, I
wore long sleeves to protect
myself from the sun.
Rarely/Never 67 (53.60) 0.99 (0.52) 1.71 (0.80-3.64) 3.37 (1.18-9.60)
Often/Sometimes* 54 (43.20) 1.04 (0.33)
Missing 4 (3.20) 1.04 (0.33)
* reference value **Gender and Grade were also tested for confounding but were not in the final model
89
Table 14 Questionnaire Items, Responses and Associations (continued)
Variable/Question Measurement/
Response
Count
(Percent)
Highest Day
Average UVI per
minute outside
median (IQR)
Odds Ratio (95%
CI)
Univariate
Odds Ratio (95% CI)
in final model
At SCHOOL in the PAST
MONTH, I wore long pants to
protect myself from the sun.
Rarely/Never 20 (16.00) 0.94 (0.45) 1.28 (0.48-3.41) 2.26 (0.65-7.88)
Often/Sometimes* 105 (84.00) 1.05 (0.41)
When I was outside but NOT at
school in the PAST MONTH, I
wore long pants to protect
myself from the sun.
Rarely/Never 1 38 (30.40) 1.02 (0.44) 1.43 (0.65-3.14) NA – not in final
model
Often/Sometimes* 83 (66.40) 1.02 (0.48)
Missing 4 (3.20) 0.85 (0.40)
* reference value **Gender and Grade were also tested for confounding but were not in the final model
90
Table 15 Model Comparison
Model name Variables
included
AUC Sensitivity Specificity
Odds Ratio
(95% CI)
Univariate model
using median cut
point
Highest day UVI
per minute
outside
62.58 73.80 67.90 2.81 (1.31-6.04)
Model using
median cut point
with known
confounders
The above plus
skin phototype,
sunscreen use in
school,
sunscreen use
outside of
school.
74.30 66.70 74.40 2.74 (1.20-6.27)
Model using
median cut point
and standard
epidemiologic
10% change
criteria
The above plus
long pants at
school
76.24 73.80 67.90 3.03 (1.29-7.13)
New analytic
method model
using median cut
point
The above plus
long sleeves in
school and long
sleeves outside
of school and
hat at school
78.99 71.40 63.60 3.79 (1.51-9.47)
New model
analytic method
model using cut
point that
maximizes AUC in
univariate model
Same as above 80.15 76.20 67.50 4.91 (1.92-12.54)
91
References
1. Siegel, R.L., K.D. Miller, and A. Jemal, Cancer statistics, 2018. CA Cancer J Clin, 2018. 68(1): p. 7-
30.
2. Key, T.J., P.K. Verkasalo, and E. Banks, Epidemiology of breast cancer. Lancet Oncol, 2001. 2(3):
p. 133-40.
3. Key, T.J., et al., Circulating sex hormones and breast cancer risk factors in postmenopausal
women: reanalysis of 13 studies. Br J Cancer, 2011. 105(5): p. 709-22.
4. Setiawan, V.W., et al., Racial/ethnic differences in postmenopausal endogenous hormones: the
multiethnic cohort study. Cancer Epidemiol Biomarkers Prev, 2006. 15(10): p. 1849-55.
5. Travis, R.C. and T.J. Key, Oestrogen exposure and breast cancer risk. Breast cancer research :
BCR, 2003. 5(5): p. 239-247.
6. Key, T.J., Endogenous oestrogens and breast cancer risk in premenopausal and postmenopausal
women. Steroids, 2011. 76(8): p. 812-5.
7. Clavel-Chapelon, F., Cumulative number of menstrual cycles and breast cancer risk: results from
the E3N cohort study of French women. Cancer Causes Control, 2002. 13(9): p. 831-8.
8. Chavez-MacGregor, M., et al., Postmenopausal breast cancer risk and cumulative number of
menstrual cycles. Cancer Epidemiol Biomarkers Prev, 2005. 14(4): p. 799-804.
9. Rojas-Lima, E., et al., A cumulative index of exposure to endogenous estrogens and breast cancer
by molecular subtypes in northern Mexican women. Breast Cancer Res Treat, 2020. 180(3): p.
791-800.
10. Wu, A.H., et al., Traditional breast cancer risk factors in Filipina Americans compared to Chinese
and Japanese Americans in Los Angeles County. Cancer Epidemiol Biomarkers Prev, 2016.
11. Bao, P.P., et al., Association of hormone-related characteristics and breast cancer risk by
estrogen receptor/progesterone receptor status in the shanghai breast cancer study. Am J
Epidemiol, 2011. 174(6): p. 661-71.
12. John, E.M., et al., Migration history, acculturation, and breast cancer risk in Hispanic women.
Cancer Epidemiol Biomarkers Prev, 2005. 14(12): p. 2905-13.
13. Sweeney, C., et al., Reproductive history in relation to breast cancer risk among Hispanic and
non-Hispanic white women. Cancer Causes Control, 2008. 19(4): p. 391-401.
14. John, E.M., et al., Enrollment and biospecimen collection in a multiethnic family cohort: the
Northern California site of the Breast Cancer Family Registry. Cancer Causes Control, 2019.
30(4): p. 395-408.
15. Guy, G.P., Jr., et al., Vital signs: melanoma incidence and mortality trends and projections -
United States, 1982-2030. MMWR Morb Mortal Wkly Rep, 2015. 64(21): p. 591-6.
16. Matthews, N.H., et al., Epidemiology of Melanoma, in Cutaneous Melanoma: Etiology and
Therapy, W.H. Ward and J.M. Farma, Editors. 2017: Brisbane (AU).
17. Siegel, R.L., et al., Cancer statistics, 2022. CA Cancer J Clin, 2022. 72(1): p. 7-33.
18. Arnold, M., et al., Cutaneous melanoma in France in 2015 attributable to solar ultraviolet
radiation and the use of sunbeds. J Eur Acad Dermatol Venereol, 2018. 32(10): p. 1681-1686.
19. Arnold, M., et al., Global burden of cutaneous melanoma attributable to ultraviolet radiation in
2012. Int J Cancer, 2018. 143(6): p. 1305-1314.
20. Parkin, D.M., D. Mesher, and P. Sasieni, 13. Cancers attributable to solar (ultraviolet) radiation
exposure in the UK in 2010. Br J Cancer, 2011. 105 Suppl 2: p. S66-9.
21. Dennis, L.K., et al., Sunburns and risk of cutaneous melanoma: does age matter? A
comprehensive meta-analysis. Ann Epidemiol, 2008. 18(8): p. 614-27.
92
22. Gandini, S., et al., Meta-analysis of risk factors for cutaneous melanoma: II. Sun exposure. Eur J
Cancer, 2005. 41(1): p. 45-60.
23. Køster, B., et al., Knowledge deficit, attitude and behavior scales association to objective
measures of sun exposure and sunburn in a Danish population based sample. PLoS One, 2017.
12(5): p. e0178190.
24. Cockburn, M.G., J. Zadnick, and D. Deapen, Developing epidemic of melanoma in the Hispanic
population of California. Cancer, 2006. 106(5): p. 1162-8.
25. US Census Bureau, P.D. Table 16. Projected Rates for Components of Change by Race and
Hispanic Origin for the United States: 2015 to 2060 (NP2014-T16). 2014; Available from:
http://www.census.gov/population/projections/files/summary/NP2014-T16.xls.
26. Santiago-Rivas, M., C. Wang, and L. Jandorf, Sun Protection Beliefs among Hispanics in the US. J
Skin Cancer, 2014. 2014: p. 161960.
27. Han, J., G.A. Colditz, and D.J. Hunter, Risk factors for skin cancers: a nested case-control study
within the Nurses' Health Study. Int J Epidemiol, 2006. 35(6): p. 1514-21.
28. Cockburn, M., A. Hamilton, and T. Mack, Recall bias in self-reported melanoma risk factors. Am J
Epidemiol, 2001. 153(10): p. 1021-6.
29. Parr, C.L., et al., Recall bias in melanoma risk factors and measurement error effects: a nested
case-control study within the Norwegian Women and Cancer Study. Am J Epidemiol, 2009.
169(3): p. 257-66.
30. Berwick, M. and Y.T. Chen, Reliability of reported sunburn history in a case-control study of
cutaneous malignant melanoma. Am J Epidemiol, 1995. 141(11): p. 1033-7.
31. Eilers, S., et al., Accuracy of Self-report in Assessing Fitzpatrick Skin Phototypes I Through VI.
JAMA Dermatology, 2013. 149(11): p. 1289-1294.
32. Miller, K.A., et al., SunSmart: evaluation of a pilot school-based sun protection intervention in
Hispanic early adolescents. Health Educ Res, 2015. 30(3): p. 371-9.
33. Miller, K.A., et al., Correlates of sun protection behaviors among Hispanic children residing in a
high UVR environment. Photodermatol Photoimmunol Photomed, 2017. 33(2): p. 75-83.
34. Miller, K.A., et al., Patterns of sun protective behaviors among Hispanic children in a skin cancer
prevention intervention. Prev Med, 2015. 81: p. 303-8.
35. Altieri, L., et al., Prevalence of sun protection behaviors in Hispanic youth residing in a high
ultraviolet light environment. Pediatr Dermatol, 2018. 35(1): p. e52-e54.
36. Tao, Z., et al., Breast Cancer: Epidemiology and Etiology. Cell Biochem Biophys, 2015. 72(2): p.
333-8.
37. Rojas, K. and A. Stuckey, Breast Cancer Epidemiology and Risk Factors. Clin Obstet Gynecol,
2016. 59(4): p. 651-672.
38. Beatson, G.T., On the Treatment of Inoperable Cases of Carcinoma of the Mamma: Suggestions
for a New Method of Treatment, with Illustrative Cases. Trans Med Chir Soc Edinb, 1896. 15: p.
153-179.
39. Block, G.E., A.B. Vial, and F.W. Pullen, Estrogen excretion following operative and irradiation
castration in cases of mammary cancer; a preliminary report. Surgery, 1958. 43(3): p. 415-22.
40. Toft, D., G. Shyamala, and J. Gorski, A receptor molecule for estrogens: studies using a cell-free
system. Proc Natl Acad Sci U S A, 1967. 57(6): p. 1740-3.
41. Dorgan, J.F., et al., Relation of prediagnostic serum estrogen and androgen levels to breast
cancer risk. Cancer Epidemiol Biomarkers Prev, 1996. 5(7): p. 533-9.
42. Pike, M.C., et al., Estrogens, progestogens, normal breast cell proliferation, and breast cancer
risk. Epidemiol Rev, 1993. 15(1): p. 17-35.
43. Anderson, K.N., R.B. Schwab, and M.E. Martinez, Reproductive risk factors and breast cancer
subtypes: a review of the literature. Breast Cancer Res Treat, 2014. 144(1): p. 1-10.
93
44. Kurian, A.W., et al., Lifetime risks of specific breast cancer subtypes among women in four
racial/ethnic groups. Breast Cancer Res, 2010. 12(6): p. R99.
45. Chlebowski, R.T., et al., Ethnicity and breast cancer: factors influencing differences in incidence
and outcome. J Natl Cancer Inst, 2005. 97(6): p. 439-48.
46. Li, C.I., K.E. Malone, and J.R. Daling, Differences in breast cancer hormone receptor status and
histology by race and ethnicity among women 50 years of age and older. Cancer Epidemiol
Biomarkers Prev, 2002. 11(7): p. 601-7.
47. Clarke, C.A., et al., Age-Specific Incidence of Breast Cancer Subtypes: Understanding the Black–
White Crossover. JNCI: Journal of the National Cancer Institute, 2012. 104(14): p. 1094-1101.
48. Smigal, C., et al., Trends in breast cancer by race and ethnicity: update 2006. CA Cancer J Clin,
2006. 56(3): p. 168-83.
49. DeSantis, C., et al., Breast cancer statistics, 2011. CA Cancer J Clin, 2011. 61(6): p. 409-18.
50. Karliner, L.S. and K. Kerlikowske, Ethnic disparities in breast cancer. Womens Health (Lond),
2007. 3(6): p. 679-88.
51. Toriola, A.T. and G.A. Colditz, Trends in breast cancer incidence and mortality in the United
States: implications for prevention. Breast Cancer Res Treat, 2013. 138(3): p. 665-73.
52. Key, T., et al., Endogenous sex hormones and breast cancer in postmenopausal women:
reanalysis of nine prospective studies. J Natl Cancer Inst, 2002. 94(8): p. 606-16.
53. Key, T.J., et al., Sex hormones and risk of breast cancer in premenopausal women: a
collaborative reanalysis of individual participant data from seven prospective studies. Lancet
Oncol, 2013. 14(10): p. 1009-19.
54. Rosato, V., et al., Reproductive and hormonal factors, family history, and breast cancer
according to the hormonal receptor status. Eur J Cancer Prev, 2014. 23(5): p. 412-7.
55. Nelson, H.D., et al., Risk factors for breast cancer for women aged 40 to 49 years: a systematic
review and meta-analysis. Ann Intern Med, 2012. 156(9): p. 635-48.
56. Verma, R., et al., Pathological and epidemiological factors associated with advanced stage at
diagnosis of breast cancer. Br Med Bull, 2012. 103(1): p. 129-45.
57. Phipps, A.I., et al., Reproductive history and oral contraceptive use in relation to risk of triple-
negative breast cancer. J Natl Cancer Inst, 2011. 103(6): p. 470-7.
58. Work, M.E., et al., Reproductive risk factors and oestrogen/progesterone receptor-negative
breast cancer in the Breast Cancer Family Registry. Br J Cancer, 2014. 110(5): p. 1367-77.
59. John, E.M., et al., Reproductive history, breast-feeding and risk of triple negative breast cancer:
The Breast Cancer Etiology in Minorities (BEM) study. Int J Cancer, 2018. 142(11): p. 2273-2285.
60. Palmer, J.R., et al., Parity, lactation, and breast cancer subtypes in African American women:
results from the AMBER Consortium. J Natl Cancer Inst, 2014. 106(10).
61. Palmer, J.R., et al., Dual effect of parity on breast cancer risk in African-American women. J Natl
Cancer Inst, 2003. 95(6): p. 478-83.
62. Dall, G.V. and K.L. Britt, Estrogen Effects on the Mammary Gland in Early and Late Life and
Breast Cancer Risk. Front Oncol, 2017. 7: p. 110.
63. Endogenous, H., et al., Sex hormones and risk of breast cancer in premenopausal women: a
collaborative reanalysis of individual participant data from seven prospective studies. Lancet
Oncol, 2013. 14(10): p. 1009-19.
64. Kotsopoulos, J., et al., Oophorectomy after menopause and the risk of breast cancer in BRCA1
and BRCA2 mutation carriers. Cancer Epidemiol Biomarkers Prev, 2012. 21(7): p. 1089-96.
65. Clavel-Chapelon, F. and E.N. Group, Cumulative number of menstrual cycles and breast cancer
risk: results from the E3N cohort study of French women. Cancer Causes Control, 2002. 13(9): p.
831-8.
94
66. Wu, A.H., et al., Traditional Breast Cancer Risk Factors in Filipina Americans Compared with
Chinese and Japanese Americans in Los Angeles County. Cancer Epidemiol Biomarkers Prev,
2016. 25(12): p. 1572-1586.
67. Kawai, M., et al., Reproductive factors, exogenous female hormone use and breast cancer risk in
Japanese: the Miyagi Cohort Study. Cancer Causes Control, 2010. 21(1): p. 135-45.
68. Martinez, M.E., et al., Reproductive factors, heterogeneity, and breast tumor subtypes in women
of mexican descent. Cancer Epidemiol Biomarkers Prev, 2013. 22(10): p. 1853-61.
69. Song, N., et al., Heterogeneity of epidemiological factors by breast tumor subtypes in Korean
women: a case-case study. Int J Cancer, 2014. 135(3): p. 669-81.
70. Beaber, E.F., et al., Oral contraceptives and breast cancer risk overall and by molecular subtype
among young women. Cancer Epidemiol Biomarkers Prev, 2014. 23(5): p. 755-64.
71. Ross, R.K., et al., Effect of hormone replacement therapy on breast cancer risk: estrogen versus
estrogen plus progestin. J Natl Cancer Inst, 2000. 92(4): p. 328-32.
72. Ellingjord-Dale, M., et al., Parity, hormones and breast cancer subtypes - results from a large
nested case-control study in a national screening program. Breast Cancer Res, 2017. 19(1): p. 10.
73. Collaborative Group on Hormonal Factors in Breast, C., Breast cancer and hormonal
contraceptives: collaborative reanalysis of individual data on 53 297 women with breast cancer
and 100 239 women without breast cancer from 54 epidemiological studies. Lancet, 1996.
347(9017): p. 1713-27.
74. Mørch, L.S., et al., Contemporary Hormonal Contraception and the Risk of Breast Cancer. N Engl J
Med, 2017. 377(23): p. 2228-2239.
75. Narod, S.A., Hormone replacement therapy and the risk of breast cancer. Nat Rev Clin Oncol,
2011. 8(11): p. 669-76.
76. Beral, V., Breast cancer and hormone-replacement therapy in the Million Women Study. Lancet,
2003. 362(9382): p. 419-27.
77. Liu, J.Y., T.J. Chen, and S.J. Hwang, The Risk of Breast Cancer in Women Using Menopausal
Hormone Replacement Therapy in Taiwan. Int J Environ Res Public Health, 2016. 13(5).
78. Ravdin, P.M., et al., The decrease in breast-cancer incidence in 2003 in the United States. N Engl J
Med, 2007. 356(16): p. 1670-4.
79. Duffy, S.W., et al., Mammographic density and breast cancer risk in breast screening assessment
cases and women with a family history of breast cancer. Eur J Cancer, 2018. 88: p. 48-56.
80. Titus-Ernstoff, L., et al., Breast cancer risk factors in relation to breast density (United States).
Cancer Causes Control, 2006. 17(10): p. 1281-90.
81. Ziv, E., et al., Mammographic breast density and family history of breast cancer. J Natl Cancer
Inst, 2003. 95(7): p. 556-8.
82. Tseng, M., et al., Acculturation and breast density in foreign-born, U.S. Chinese women. Cancer
Epidemiol Biomarkers Prev, 2006. 15(7): p. 1301-5.
83. Dyrstad, S.W., et al., Breast cancer risk associated with benign breast disease: systematic review
and meta-analysis. Breast Cancer Research and Treatment, 2015. 149(3): p. 569-575.
84. Orr, B. and J.L. Kelley, 3rd, Benign Breast Diseases: Evaluation and Management. Clin Obstet
Gynecol, 2016. 59(4): p. 710-726.
85. Stachs, A., et al., Benign Breast Disease in Women. Dtsch Arztebl Int, 2019. 116(33-34): p. 565-
574.
86. Brewer, H.R., et al., Family history and risk of breast cancer: an analysis accounting for family
structure. Breast Cancer Res Treat, 2017. 165(1): p. 193-200.
87. Sun, Y.S., et al., Risk Factors and Preventions of Breast Cancer. Int J Biol Sci, 2017. 13(11): p.
1387-1397.
95
88. Lahmann, P.H., et al., A prospective study of adiposity and postmenopausal breast cancer risk:
the Malmö Diet and Cancer Study. Int J Cancer, 2003. 103(2): p. 246-52.
89. Cleary, M.P. and M.E. Grossmann, Obesity and Breast Cancer: The Estrogen Connection.
Endocrinology, 2009. 150(6): p. 2537-2542.
90. Renehan, A.G., et al., Body-mass index and incidence of cancer: a systematic review and meta-
analysis of prospective observational studies. Lancet, 2008. 371(9612): p. 569-78.
91. Pathak, D.R. and A.S. Whittemore, Combined effects of body size, parity, and menstrual events
on breast cancer incidence in seven countries. Am J Epidemiol, 1992. 135(2): p. 153-68.
92. Peacock, S.L., et al., Relation between obesity and breast cancer in young women. Am J
Epidemiol, 1999. 149(4): p. 339-46.
93. Weiderpass, E., et al., A prospective study of body size in different periods of life and risk of
premenopausal breast cancer. Cancer Epidemiol Biomarkers Prev, 2004. 13(7): p. 1121-7.
94. Lorincz, A.M. and S. Sukumar, Molecular links between obesity and breast cancer. Endocr Relat
Cancer, 2006. 13(2): p. 279-92.
95. Judd, H.L., et al., Endocrine function of the postmenopausal ovary: concentration of androgens
and estrogens in ovarian and peripheral vein blood. J Clin Endocrinol Metab, 1974. 39(6): p.
1020-4.
96. van Landeghem, A.A., et al., Endogenous concentration and subcellular distribution of androgens
in normal and malignant human breast tissue. Cancer Res, 1985. 45(6): p. 2907-12.
97. McTiernan, A., et al., Relation of BMI and physical activity to sex hormones in postmenopausal
women. Obesity (Silver Spring), 2006. 14(9): p. 1662-77.
98. Monninkhof, E.M., et al., Physical activity and breast cancer: a systematic review. Epidemiology,
2007. 18(1): p. 137-57.
99. Eketunde, A.O., Diabetes as a Risk Factor for Breast Cancer. Cureus, 2020. 12(5): p. e8010.
100. Hardefeldt, P.J., S. Edirimanne, and G.D. Eslick, Diabetes increases the risk of breast cancer: a
meta-analysis. Endocrine-Related Cancer, 2012. 19(6): p. 793-803.
101. Durrani, I.A., A. Bhatti, and P. John, The prognostic outcome of ‘type 2 diabetes mellitus and
breast cancer’ association pivots on hypoxia-hyperglycemia axis. Cancer Cell International, 2021.
21(1): p. 351.
102. Park, Y.M.M., et al., A prospective study of type 2 diabetes, metformin use, and risk of breast
cancer. Annals of Oncology, 2021. 32(3): p. 351-359.
103. Hamajima, N., et al., Alcohol, tobacco and breast cancer--collaborative reanalysis of individual
data from 53 epidemiological studies, including 58,515 women with breast cancer and 95,067
women without the disease. Br J Cancer, 2002. 87(11): p. 1234-45.
104. Jung, S., et al., Alcohol consumption and breast cancer risk by estrogen receptor status: in a
pooled analysis of 20 studies. Int J Epidemiol, 2016. 45(3): p. 916-28.
105. Golay, A. and E. Bobbioni, The role of dietary fat in obesity. Int J Obes Relat Metab Disord, 1997.
21 Suppl 3: p. S2-11.
106. Stephenson, G.D. and D.P. Rose, Breast Cancer and Obesity: An Update. Nutrition and Cancer,
2003. 45(1): p. 1-16.
107. Makarem, N., et al., Dietary fat in breast cancer survival. Annu Rev Nutr, 2013. 33: p. 319-48.
108. Sieri, S., et al., Dietary fat intake and development of specific breast cancer subtypes. J Natl
Cancer Inst, 2014. 106(5).
109. Palmer, J.R. and L. Rosenberg, Cigarette smoking and the risk of breast cancer. Epidemiol Rev,
1993. 15(1): p. 145-56.
110. Terry, P.D. and T.E. Rohan, Cigarette smoking and the risk of breast cancer in women: a review
of the literature. Cancer Epidemiol Biomarkers Prev, 2002. 11(10 Pt 1): p. 953-71.
96
111. Kispert, S. and J. McHowat, Recent insights into cigarette smoking as a lifestyle risk factor for
breast cancer. Breast Cancer (Dove Med Press), 2017. 9: p. 127-132.
112. Johnson, K.C., et al., Active smoking and secondhand smoke increase breast cancer risk: the
report of the Canadian Expert Panel on Tobacco Smoke and Breast Cancer Risk (2009). Tob
Control, 2011. 20(1): p. e2.
113. Secretan, B., et al., A review of human carcinogens--Part E: tobacco, areca nut, alcohol, coal
smoke, and salted fish. Lancet Oncol, 2009. 10(11): p. 1033-4.
114. Reinier, K.S., P.M. Vacek, and B.M. Geller, Risk factors for breast carcinoma in situ versus
invasive breast cancer in a prospective study of pre- and post-menopausal women. Breast Cancer
Research and Treatment, 2007. 103(3): p. 343-348.
115. Slattery, M.L., et al., Body size, weight change, fat distribution and breast cancer risk in Hispanic
and non-Hispanic white women. Breast Cancer Res Treat, 2007. 102(1): p. 85-101.
116. John, E.M., et al., Prevalence of pathogenic BRCA1 mutation carriers in 5 US racial/ethnic groups.
JAMA, 2007. 298(24): p. 2869-76.
117. Pike, M.C., et al., Estrogen-progestin replacement therapy and endometrial cancer. J Natl Cancer
Inst, 1997. 89(15): p. 1110-6.
118. Benjamini, Y. and Y. Hochberg, Controlling the False Discovery Rate - a Practical and Powerful
Approach to Multiple Testing. Journal of the Royal Statistical Society Series B-Statistical
Methodology, 1995. 57(1): p. 289-300.
119. Amadou, A., et al., Overweight, obesity and risk of premenopausal breast cancer according to
ethnicity: a systematic review and dose-response meta-analysis. Obes Rev, 2013. 14(8): p. 665-
78.
120. Ma, H., et al., Reproductive factors and breast cancer risk according to joint estrogen and
progesterone receptor status: a meta-analysis of epidemiological studies. Breast Cancer Res,
2006. 8(4): p. R43.
121. Gaudet, M.M., et al., Pooled Analysis of Nine Cohorts Reveals Breast Cancer Risk Factors by
Tumor Molecular Subtype. Cancer Res, 2018. 78(20): p. 6011-6021.
122. Newman, L.A., et al., Ethnicity related differences in the survival of young breast carcinoma
patients. Cancer, 2002. 95(1): p. 21-7.
123. Shavers, V.L., L.C. Harlan, and J.L. Stevens, Racial/ethnic variation in clinical presentation,
treatment, and survival among breast cancer patients under age 35. Cancer, 2003. 97(1): p. 134-
47.
124. Elledge, R.M., et al., Tumor biologic factors and breast cancer prognosis among white, Hispanic,
and black women in the United States. J Natl Cancer Inst, 1994. 86(9): p. 705-12.
125. Palmer, J.R., et al., Genetic susceptibility loci for subtypes of breast cancer in an African American
population. Cancer Epidemiol Biomarkers Prev, 2013. 22(1): p. 127-34.
126. Warner, E.T., et al., Estrogen receptor positive tumors: do reproductive factors explain
differences in incidence between black and white women? Cancer Causes Control, 2013. 24(4): p.
731-9.
127. John, E.M., et al., Menstrual and reproductive characteristics and breast cancer risk by hormone
receptor status and ethnicity: The Breast Cancer Etiology in Minorities study. Int J Cancer, 2020.
147(7): p. 1808-1822.
128. Hall, I.J., et al., Comparative analysis of breast cancer risk factors among African-American
women and White women. Am J Epidemiol, 2005. 161(1): p. 40-51.
129. Cui, Y., et al., Associations of Hormone-Related Factors With Breast Cancer Risk According to
Hormone Receptor Status Among White and African American Women. Clin Breast Cancer, 2014.
130. Bertrand, K.A., et al., Differential Patterns of Risk Factors for Early-Onset Breast Cancer by ER
Status in African American Women. Cancer Epidemiol Biomarkers Prev, 2017. 26(2): p. 270-277.
97
131. Kerlikowske, K., et al., Differences in screening mammography outcomes among White, Chinese,
and Filipino women. Arch Intern Med, 2005. 165(16): p. 1862-8.
132. Siegel, R.L., et al., Cancer Statistics, 2021. CA: a cancer journal for clinicians, 2021. 71(1): p. 7-33.
133. Curti, B.D. and M.B. Faries, Recent Advances in the Treatment of Melanoma. N Engl J Med, 2021.
384(23): p. 2229-2240.
134. Aceituno-Madera, P., et al., [Melanoma, altitude, and UV-B radiation]. Actas Dermosifiliogr,
2011. 102(3): p. 199-205.
135. Marrot, L., Pollution and Sun Exposure: A Deleterious Synergy. Mechanisms and Opportunities
for Skin Protection. Curr Med Chem, 2018. 25(40): p. 5469-5486.
136. Klaunig, J.E., Oxidative Stress and Cancer. Curr Pharm Des, 2018. 24(40): p. 4771-4778.
137. Burke, K.E. and H. Wei, Synergistic damage by UVA radiation and pollutants. Toxicol Ind Health,
2009. 25(4-5): p. 219-24.
138. Markovic, S.N., et al., Malignant melanoma in the 21st century, part 1: epidemiology, risk
factors, screening, prevention, and diagnosis. Mayo Clin Proc, 2007. 82(3): p. 364-80.
139. Rigel, D.S., Epidemiology of melanoma. Semin Cutan Med Surg, 2010. 29(4): p. 204-9.
140. Schwartz, M.R., L. Luo, and M. Berwick, Sex Differences in Melanoma. Curr Epidemiol Rep, 2019.
6(2): p. 112-118.
141. Parkin, D.M., D. Mesher, and P. Sasieni, 13. Cancers attributable to solar (ultraviolet) radiation
exposure in the UK in 2010. Br J Cancer, 2011. 105 Suppl 2(Suppl 2): p. S66-9.
142. Gandini, S., et al., Meta-analysis of risk factors for cutaneous melanoma: II. Sun exposure.
European Journal of Cancer, 2005. 41(1): p. 45-60.
143. Elwood, J.M., et al., Etiological differences between subtypes of cutaneous malignant melanoma:
Western Canada Melanoma Study. J Natl Cancer Inst, 1987. 78(1): p. 37-44.
144. Alavanja, M.C., et al., Cancer incidence in the agricultural health study. Scand J Work Environ
Health, 2005. 31 Suppl 1: p. 39-45; discussion 5-7.
145. Nelemans, P.J., et al., An addition to the controversy on sunlight exposure and melanoma risk: a
meta-analytical approach. J Clin Epidemiol, 1995. 48(11): p. 1331-42.
146. Elwood, J.M. and J. Jopson, Melanoma and sun exposure: an overview of published studies. Int J
Cancer, 1997. 73(2): p. 198-203.
147. Veierød, M.B., et al., A prospective study of pigmentation, sun exposure, and risk of cutaneous
malignant melanoma in women. J Natl Cancer Inst, 2003. 95(20): p. 1530-8.
148. Autier, P. and J.F. Doré, Influence of sun exposures during childhood and during adulthood on
melanoma risk. EPIMEL and EORTC Melanoma Cooperative Group. European Organisation for
Research and Treatment of Cancer. Int J Cancer, 1998. 77(4): p. 533-7.
149. The association of use of sunbeds with cutaneous malignant melanoma and other skin cancers: A
systematic review. Int J Cancer, 2007. 120(5): p. 1116-22.
150. Gallagher, R.P., J.J. Spinelli, and T.K. Lee, Tanning beds, sunlamps, and risk of cutaneous
malignant melanoma. Cancer Epidemiol Biomarkers Prev, 2005. 14(3): p. 562-6.
151. Wehner, M.R., et al., Indoor tanning and non-melanoma skin cancer: systematic review and
meta-analysis. BMJ, 2012. 345: p. e5909.
152. Boniol, M., et al., Cutaneous melanoma attributable to sunbed use: systematic review and meta-
analysis. BMJ, 2012. 345: p. e4757.
153. Arrangoiz, R., et al., Melanoma Review: Epidemiology, Risk Factors, Diagnosis and Staging.
Journal of Cancer Treatment and Research, 2016. 4(1): p. 1-15.
154. Olsen, C.M., H.J. Carroll, and D.C. Whiteman, Estimating the attributable fraction for melanoma:
a meta-analysis of pigmentary characteristics and freckling. Int J Cancer, 2010. 127(10): p. 2430-
45.
98
155. Bauer, J. and C. Garbe, Acquired melanocytic nevi as risk factor for melanoma development. A
comprehensive review of epidemiological data. Pigment Cell Res, 2003. 16(3): p. 297-306.
156. Gallagher, R.P., et al., Sunlight exposure, pigmentary factors, and risk of nonmelanocytic skin
cancer. I. Basal cell carcinoma. Arch Dermatol, 1995. 131(2): p. 157-63.
157. Brenner, M. and V.J. Hearing, The protective role of melanin against UV damage in human skin.
Photochem Photobiol, 2008. 84(3): p. 539-49.
158. Everett, M.A., et al., Penetration of epidermis by ultraviolet rays. Photochem Photobiol, 1966.
5(7): p. 533-42.
159. Seebode, C., J. Lehmann, and S. Emmert, Photocarcinogenesis and Skin Cancer Prevention
Strategies. Anticancer Res, 2016. 36(3): p. 1371-8.
160. Halder, R.M. and S. Bridgeman-Shah, Skin cancer in African Americans. Cancer, 1995. 75(2
Suppl): p. 667-73.
161. Elmets, C.A., et al., Analysis of the mechanism of unresponsiveness produced by haptens painted
on skin exposed to low dose ultraviolet radiation. J Exp Med, 1983. 158(3): p. 781-94.
162. Parrish, J.A., K.F. Jaenicke, and R.R. Anderson, Erythema and melanogenesis action spectra of
normal human skin. Photochem Photobiol, 1982. 36(2): p. 187-91.
163. Dulon, M., et al., Sun exposure and number of nevi in 5- to 6-year-old European children. J Clin
Epidemiol, 2002. 55(11): p. 1075-81.
164. Rhodes, A.R., et al., Risk factors for cutaneous melanoma. A practical method of recognizing
predisposed individuals. JAMA, 1987. 258(21): p. 3146-54.
165. Bataille, V., et al., Risk of cutaneous melanoma in relation to the numbers, types and sites of
naevi: a case-control study. Br J Cancer, 1996. 73(12): p. 1605-11.
166. Gandini, S., et al., Meta-analysis of risk factors for cutaneous melanoma: I. Common and atypical
naevi. Eur J Cancer, 2005. 41(1): p. 28-44.
167. IARC monographs on the evaluation of carcinogenic risks to humans. Solar and ultraviolet
radiation. IARC Monogr Eval Carcinog Risks Hum, 1992. 55: p. 1-316.
168. Abbasi, N.R., et al., Early diagnosis of cutaneous melanoma: revisiting the ABCD criteria. JAMA,
2004. 292(22): p. 2771-6.
169. Thompson, J.F., R.A. Scolyer, and R.F. Kefford, Cutaneous melanoma. Lancet, 2005. 365(9460): p.
687-701.
170. Watt, A.J., S.V. Kotsis, and K.C. Chung, Risk of melanoma arising in large congenital melanocytic
nevi: a systematic review. Plast Reconstr Surg, 2004. 113(7): p. 1968-74.
171. Tucker, M.A., Melanoma epidemiology. Hematol Oncol Clin North Am, 2009. 23(3): p. 383-95,
vii.
172. Gandini, S., et al., Meta-analysis of risk factors for cutaneous melanoma: III. Family history,
actinic damage and phenotypic factors. European Journal of Cancer, 2005. 41(14): p. 2040-2059.
173. Potrony, M., et al., POT1 germline mutations but not TERT promoter mutations are implicated in
melanoma susceptibility in a large cohort of Spanish melanoma families. The British journal of
dermatology, 2019. 181(1): p. 105-113.
174. Goldstein, A.M., et al., Features associated with germline CDKN2A mutations: a GenoMEL study
of melanoma-prone families from three continents. J Med Genet, 2007. 44(2): p. 99-106.
175. Begg, C.B., et al., Lifetime risk of melanoma in CDKN2A mutation carriers in a population-based
sample. J Natl Cancer Inst, 2005. 97(20): p. 1507-15.
176. Hussussian, C.J., et al., Germline p16 mutations in familial melanoma. Nat Genet, 1994. 8(1): p.
15-21.
177. Aoude, L.G., et al., Genetics of familial melanoma: 20 years after CDKN2A. Pigment Cell
Melanoma Res, 2015. 28(2): p. 148-60.
99
178. Bishop, D.T., et al., Geographical Variation in the Penetrance of CDKN2A Mutations for
Melanoma. JNCI: Journal of the National Cancer Institute, 2002. 94(12): p. 894-903.
179. Healy, E., Melanocortin 1 receptor variants, pigmentation, and skin cancer susceptibility.
Photodermatol Photoimmunol Photomed, 2004. 20(6): p. 283-8.
180. Landi, M.T., et al., MC1R germline variants confer risk for BRAF-mutant melanoma. Science,
2006. 313(5786): p. 521-2.
181. Puntervoll, H.E., et al., Melanoma prone families with CDK4 germline mutation: phenotypic
profile and associations with MC1R variants. J Med Genet, 2013. 50(4): p. 264-70.
182. Clark, W.H., Jr., et al., Origin of familial malignant melanomas from heritable melanocytic
lesions. 'The B-K mole syndrome'. Arch Dermatol, 1978. 114(5): p. 732-8.
183. Lynch, H.T., T.G. Shaw, and J.F. Lynch, Inherited predisposition to cancer: a historical overview.
Am J Med Genet C Semin Med Genet, 2004. 129c(1): p. 5-22.
184. Potrony, M., et al., Update in genetic susceptibility in melanoma. Ann Transl Med, 2015. 3(15): p.
210.
185. Grant, W.B. and S.B. Mohr, Ecological studies of ultraviolet B, vitamin D and cancer since 2000.
Ann Epidemiol, 2009. 19(7): p. 446-54.
186. Godar, D.E., UV doses worldwide. Photochem Photobiol, 2005. 81(4): p. 736-49.
187. King, L., et al., Measuring sun exposure in epidemiological studies: Matching the method to the
research question. J Photochem Photobiol B, 2015. 153: p. 373-9.
188. Diffey, B., A behavioral model for estimating population exposure to solar ultraviolet radiation.
Photochem Photobiol, 2008. 84(2): p. 371-5.
189. Sherwin, J.C., et al., Reliability and validity of conjunctival ultraviolet autofluorescence
measurement. Br J Ophthalmol, 2012. 96(6): p. 801-5.
190. Battistutta, D., et al., Skin surface topography grading is a valid measure of skin photoaging.
Photodermatol Photoimmunol Photomed, 2006. 22(1): p. 39-45.
191. Worswick, S.D., M. Cockburn, and D. Peng, Measurement of ultraviolet exposure in
epidemiological studies of skin and skin cancers. Photochem Photobiol, 2008. 84(6): p. 1462-72.
192. Warren, R., et al., Age, sunlight, and facial skin: a histologic and quantitative study. J Am Acad
Dermatol, 1991. 25(5 Pt 1): p. 751-60.
193. Moon, J.S. and C.H. Oh, Solar damage in skin tumors: quantification of elastotic material.
Dermatology, 2001. 202(4): p. 289-92.
194. Nakazawa, H., et al., UV and skin cancer: specific p53 gene mutation in normal skin as a
biologically relevant exposure measurement. Proc Natl Acad Sci U S A, 1994. 91(1): p. 360-4.
195. Allen, M.W., et al., Use of Electronic UV Dosimeters in Measuring Personal UV Exposures and
Public Health Education. Atmosphere, 2020. 11(7).
196. Dwyer, T., et al., Assessment of habitual sun exposure in adolescents via questionnaire--a
comparison with objective measurement using polysulphone badges. Melanoma Res, 1996. 6(3):
p. 231-9.
197. Alshurafa, N., et al., Assessing recall of personal sun exposure by integrating UV dosimeter and
self-reported data with a network flow framework. PLoS One, 2019. 14(12): p. e0225371.
198. Cancer incidence in five continents. Volume VIII. IARC Sci Publ, 2002(155): p. 1-781.
199. Cormier, J.N., et al., Ethnic differences among patients with cutaneous melanoma. Arch Intern
Med, 2006. 166(17): p. 1907-14.
200. Wu, X.C., et al., Racial and ethnic variations in incidence and survival of cutaneous melanoma in
the United States, 1999-2006. J Am Acad Dermatol, 2011. 65(5 Suppl 1): p. S26-37.
201. Ma, F., et al., Skin cancer awareness and sun protection behaviors in white Hispanic and white
non-Hispanic high school students in Miami, Florida. Arch Dermatol, 2007. 143(8): p. 983-8.
100
202. Pipitone, M., et al., Skin cancer awareness in suburban employees: a Hispanic perspective. J Am
Acad Dermatol, 2002. 47(1): p. 118-23.
203. Coups, E.J., et al., Sun protection and exposure behaviors among Hispanic adults in the United
States: differences according to acculturation and among Hispanic subgroups. BMC Public
Health, 2012. 12: p. 985.
204. Negy, C. and D.J. Woods, The Importance of Acculturation in Understanding Research with
Hispanic-Americans. Hispanic Journal of Behavioral Sciences, 1992. 14(2): p. 224-247.
205. Andreeva, V.A., et al., Acculturation and sun-safe behaviors among US Latinos: findings from the
2005 Health Information National Trends Survey. Am J Public Health, 2009. 99(4): p. 734-41.
206. Viola, A.S., J.L. Stapleton, and E.J. Coups, Associations between linguistic acculturation and skin
cancer knowledge and beliefs among U.S. Hispanic adults. Prev Med Rep, 2019. 15: p. 100943.
207. Andreeva, V.A., et al., Preliminary evidence for mediation of the association between
acculturation and sun-safe behaviors. Arch Dermatol, 2011. 147(7): p. 814-9.
208. Cust, A.E., et al., Validation of Questionnaire and Diary Measures of Time Outdoors Against an
Objective Measure of Personal Ultraviolet Radiation Exposure. Photochem Photobiol, 2018.
94(4): p. 815-820.
209. English, D.R., B.K. Armstrong, and A. Kricker, Reproducibility of reported measurements of sun
exposure in a case-control study. Cancer Epidemiol Biomarkers Prev, 1998. 7(10): p. 857-63.
210. Unger, J.B., et al., The AHIMSA Acculturation Scale: A New Measure of Acculturation for
Adolescents in a Multicultural Society. Journal of Early Adolescence, 2002. 22(3): p. 225-251.
211. Fitzpatrick, T.B., The validity and practicality of sun-reactive skin types I through VI. Arch
Dermatol, 1988. 124(6): p. 869-71.
212. Allen, M. and R. McKenzie, Enhanced UV exposure on a ski-field compared with exposures at sea
level. Photochemical & Photobiological Sciences, 2005. 4(5): p. 429-37.
213. McCarty, C.A., Sunlight exposure assessment: can we accurately assess vitamin D exposure from
sunlight questionnaires? Am J Clin Nutr, 2008. 87(4): p. 1097s-101s.
214. Koster, B., et al., Reliability and consistency of a validated sun exposure questionnaire in a
population-based Danish sample. Prev Med Rep, 2018. 10: p. 43-48.
215. Moise, A.F., P.G. Büttner, and S.L. Harrison, Sun exposure at school. Photochem Photobiol, 1999.
70(2): p. 269-74.
216. Glanz, K., M. Saraiya, and H. Wechsle, Guidelines for School Programs To Prevent Skin Cancer, in
Morbidity and Mortality Weekly Report. 2002. p. 1-16.
217. Reyes-Marcelino, G., et al., School-based interventions to improve sun-safe knowledge, attitudes
and behaviors in childhood and adolescence: A systematic review. Prev Med, 2021. 146: p.
106459.
218. Robinson, J.K., D.S. Rigel, and R.A. Amonette, Summertime sun protection used by adults for
their children. J Am Acad Dermatol, 2000. 42(5 Pt 1): p. 746-53.
219. Turner, L.R. and R.J. Mermelstein, Psychosocial characteristics associated with sun protection
practices among parents of young children. J Behav Med, 2005. 28(1): p. 77-90.
220. Foley, P., et al., The frequency of reactions to sunscreens: results of a longitudinal population-
based study on the regular use of sunscreens in Australia. Br J Dermatol, 1993. 128(5): p. 512-8.
221. Glanz, K., et al., Factors associated with skin cancer prevention practices in a multiethnic
population. Health Educ Behav, 1999. 26(3): p. 344-59.
222. Hernandez, C., et al., Comparison of sunscreen availability in Chicago Hispanic and non-Hispanic
neighborhoods. Photodermatol Photoimmunol Photomed, 2012. 28(5): p. 244-9.
223. Pilkauskas, N.V., M. Amorim, and R.E. Dunifon, Historical Trends in Children Living in
Multigenerational Households in the United States: 1870-2018. Demography, 2020. 57(6): p.
2269-2296.
101
224. Myers, L.B. and M.S. Horswill, Social cognitive predictors of sun protection intention and
behavior. Behav Med, 2006. 32(2): p. 57-63.
225. Reynolds, K.D., et al., Mediation of a middle school skin cancer prevention program. Health
Psychol, 2006. 25(5): p. 616-25.
226. Arthey, S. and V.A. Clarke, Suntanning and sun protection: a review of the psychological
literature. Soc Sci Med, 1995. 40(2): p. 265-74.
227. Fernández-Morano, T., et al., Adolescents' Attitudes to Sun Exposure and Sun Protection. J
Cancer Educ, 2017. 32(3): p. 596-603.
228. Venning, V.L., et al., Risk Perception Plays Minimal Role in Sun Exposure Behaviours. J Cancer
Educ, 2020. 35(1): p. 125-130.
229. Dennis, L.K., J.B. Lowe, and L.G. Snetselaar, Tanning behavior among young frequent tanners is
related to attitudes and not lack of knowledge about the dangers. Health Educ J, 2009. 68(3): p.
232-243.
230. Gefeller, O., et al., The impact of parental knowledge and tanning attitudes on sun protection
practice for young children in Germany. Int J Environ Res Public Health, 2014. 11(5): p. 4768-81.
231. Andreeva, V.A., et al., Concurrent psychosocial predictors of sun safety among middle school
youth. J Sch Health, 2008. 78(7): p. 374-81; quiz 408-10.
232. Reynolds, K.D., et al., Predictors of sun exposure in adolescents in a southeastern U.S.
population. J Adolesc Health, 1996. 19(6): p. 409-15.
233. Cokkinides, V.E., et al., Sun exposure and sun-protection behaviors and attitudes among U.S.
youth, 11 to 18 years of age. Prev Med, 2001. 33(3): p. 141-51.
234. Perez, M.I., Skin Cancer in Hispanics in the United States. J Drugs Dermatol, 2019. 18(3): p. s117-
120.
235. Islami, F., et al., Cutaneous melanomas attributable to ultraviolet radiation exposure by state. Int
J Cancer, 2020. 147(5): p. 1385-1390.
236. Arnold, C., Is precision public health the future - or a contradiction? Nature, 2022. 601(7891): p.
18-20.
237. Khoury, M.J., M.F. Iademarco, and W.T. Riley, Precision Public Health for the Era of Precision
Medicine. Am J Prev Med, 2016. 50(3): p. 398-401.
238. Bennette, C. and A. Vickers, Against quantiles: categorization of continuous variables in
epidemiologic research, and its discontents. BMC Medical Research Methodology, 2012. 12(1):
p. 21.
239. Lee, P.H., Is a cutoff of 10% appropriate for the change-in-estimate criterion of confounder
identification? Journal of epidemiology, 2014. 24(2): p. 161-167.
240. Maldonado, G. and S. Greenland, Simulation study of confounder-selection strategies. Am J
Epidemiol, 1993. 138(11): p. 923-36.
241. Petersen, B. and H.C. Wulf, Application of sunscreen--theory and reality. Photodermatol
Photoimmunol Photomed, 2014. 30(2-3): p. 96-101.
242. O'Hara, M., et al., Unintended sunburn after sunscreen application: An exploratory study of sun
protection. Health Promot J Austr, 2020. 31(3): p. 533-539.
243. Thieden, E., et al., Sunscreen use related to UV exposure, age, sex, and occupation based on
personal dosimeter readings and sun-exposure behavior diaries. Arch Dermatol, 2005. 141(8): p.
967-73.
244. Wulf, H.C., I.M. Stender, and J. Lock-Andersen, Sunscreens used at the beach do not protect
against erythema: a new definition of SPF is proposed. Photodermatol Photoimmunol
Photomed, 1997. 13(4): p. 129-32.
245. Cockburn, M., et al., Determinants of melanoma in a case-control study of twins (United States).
Cancer Causes Control, 2001. 12(7): p. 615-25.
102
246. Holman, D.M., et al., Association Between Sun Protection Behaviors and Sunburn Among U.S.
Older Adults. Gerontologist, 2019. 59(Suppl 1): p. S17-s27.
247. Bonilla, C., et al., Skin pigmentation, sun exposure and vitamin D levels in children of the Avon
Longitudinal Study of Parents and Children. BMC Public Health, 2014. 14: p. 597.
248. Holman, D.M., et al., Prevalence of Sun Protection Use and Sunburn and Association of
Demographic and Behaviorial Characteristics With Sunburn Among US Adults. JAMA
dermatology, 2018. 154(5): p. 561-568.
249. Linos, E., et al., Hat, shade, long sleeves, or sunscreen? Rethinking US sun protection messages
based on their relative effectiveness. Cancer Causes Control, 2011. 22(7): p. 1067-71.
250. Youden, W.J., Index for rating diagnostic tests. Cancer, 1950. 3(1): p. 32-5.
Abstract (if available)
Abstract
Race and ethnicity can influence what we eat, our weight, how much we exercise, the amount of time we spend outside; a myriad of exposures that together impact our lifetime risk of cancer. While some racial/ethnic disparities in cancer rates are due to inherited genes, and therefore not modifiable, others, like differences due to varying carcinogenic exposures may be revised. Identifying these potentially transformable exposures and successfully altering them through targeted interventions could lead to large public health impacts.
This dissertation focuses on identifying and accurately measuring carcinogenic exposures in two types of cancers with known racial/ethnic disparities - breast cancer and melanoma. These cancers have very different risk factors and the racial/ethnic disparities in them exist for vastly dissimilar reasons.
Breast cancer is the most common malignancy in women, with a lifetime risk of approximately 1 in 8, resulting in over 266,120 new cases and 40,920 deaths in the United States annually.[1] Established risk factors for breast cancer include race/ethnicity, young age at menarche, nulliparity, older age at menopause, use of contraceptive and menopausal replacement hormones, family history of breast cancer, low levels of physical activity, tobacco smoking and high body mass index (BMI) among postmenopausal women.[2] Some of these risk factors, such as BMI, tobacco smoking and alcohol use, have been shown to influence levels of circulating estrogen.[3] Lifetime exposure to estrogen has been shown to be one of the major determinants of breast cancer risk.[2, 4-6] Cumulative menstrual months (CMM) has been used as a surrogate measure of aggregate endogenous estrogen exposure, and has been related to risk of breast cancer among European women[7, 8], women in Northern Mexico[9], and Asian women in Los Angeles County [10] and in Asia.[11] There is currently limited information on the relationship between CMM and breast cancer risk in nonwhite populations.
The first project used the Breast Cancer Etiology in Minorities (BEM) study that has harmonized extensive questionnaire data from four population-based studies of breast cancer with large numbers of African Americans, Asian Americans, Hispanics, and non-Hispanic whites (NHWs) to address this racial/ethnic gap.[10, 12-14] This project utilized menstrual and reproductive events to calculate CMM uniformly across the four studies, allowing comparison of risk associations by race/ethnicity, menopausal status, BMI category and breast cancer subtypes. Numerous risk factors were adjusted for in the analysis (education, BMI, alcohol consumption, first-degree family history of breast cancer, personal history of benign breast disease, and menopausal status). Odds ratios (ORs) and 95% confidence intervals (CI) were calculated using conditional logistic regression, with matched sets defined jointly by study, age group and race/ethnicity. Subtype-specific analyses (hormone receptor positive (HR+), hormone receptor negative (HR-)) and BMI-specific analyses (< 30 kg/m2, > 30 kg/m2), were stratified by race/ethnicity and by menopausal status. Differences in associations by race/ethnicity or menopausal status were tested for by including interaction terms in the model. Results are presented overall and by menopausal status, subtype and BMI category for both race/ethnicity-adjusted models and race/ethnicity stratified models.
Skin cancer is the most common malignancy in the United States.[15] Melanoma, one of the less frequently occurring skin cancers, initiates in melanocytes - cells that control pigmentation. While melanoma accounts for fewer than 5% of all cutaneous malignancies, it is responsible for the majority of skin cancer mortality.[15, 16] Rates of melanoma are rapidly increasing in the United States with a 31% increase in invasive cases in the last decade and 99,780 new cases and 7,650 deaths predicted in 2022. [15, 17] Most cases of melanoma are attributable to ultraviolet radiation (UVR) exposure [18-20], with UVR and sunburns experienced as a child greatly increasing risk of melanoma compared to similar exposure as an adult.[21, 22] Sunburn is the most commonly used proxy for exposure to UVR in epidemiologic research.[23]. The heightened risk associated with UVR exposure in childhood coupled with the opportunity to influence behavior over the lifetime, makes accurate knowledge about this age group crucial in a successful melanoma risk reduction program.
Rates of poor prognosis melanoma, tumors thicker than 1.5 mm at diagnosis, are increasing in California among Hispanics much faster than non-Hispanic Whites, concurrent with rapid growth in the Hispanic population.[24, 25] The majority of UVR exposure studies have been conducted in non-Hispanic whites, with more recent work extending to Hispanics.[26] Most risk reduction studies to date have utilized self-reported sun exposure as a surrogate for UVR exposure, however self-reported sun exposure has been shown to have poor reproducibility, be prone to differential and non-differential recall bias and only have moderate association with objectively measured UVR, impairing the ability to detect meaningful associations due to imprecise estimates caused by information bias. [27-31]
Utilizing dosimeters to collect UVR data may potentially help modulate the bias seen in self-reported UVR exposure. Dosimeters allow more accurate recording of time spent outdoors than self-report. Additionally, self-report measures fail to obtain data on UVI, which is needed in order accurately determine risk of sunburn and is obtained by dosimetry.
The second project utilized data from a novel study using objectively measured UVR exposure in high risk, predominately Hispanic youth. This project was nested in the SunSmart study, a randomized intervention aimed to elicit positive changes in sun protective attitudes, self-efficacy, knowledge and behaviors in Title I public schools in Los Angeles County.[32-35] To determine the association of UVR exposure categories with answers to questions obtained at baseline regarding different constructs (acculturation, sun protective behavior and knowledge, family interventions), answers to questions were dichotomized and receiver operating characteristics (ROC) analysis was then performed for each UVR category. UVR cutoffs were chosen that maximized the sensitivity and specificity for a specific question with the highest AUC for each UVR category.
The third project utilized data from the SunSmart dosimeter sub-study, described in the second project, to determine the association between self-reported sunburn and the highest day measurement of daily cumulative UVI divided by total daily minutes at non-zero UVI (average UVI per minute outside). Receiver operating characteristics analysis was performed using logistic regression with self-reported sunburn in the last month as the dependent variable and average UVI per minute outside as an independent variable, adjusted for dichotomized variables that have been previously shown to reduce the risk of sunburn: student’s self-reported skin color and how often students reported using sunscreen both in and out of school. The following variables that also may influence the probability of sunburn were tested for possible confounding: gender and grade as well as each of the following separately for both in and out of school wearing a hat, long sleeves and long pants. Possible confounders were tested and retained in the model if they changed the estimate of the average UVI per minute outside odds ratio by more than 10% singly or by more than 20% in combination with another variable. Once all confounders were determined, the AUC, sensitivity, specificity and cutoff value of the final model and the cutoff value when predicting sunburn were then described.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Predictive factors of breast cancer survival: a population-based study
PDF
The effects of tobacco exposure on hormone levels and breast cancer risk among young women
PDF
The role of pesticide exposure in breast cancer
PDF
Psychosocial and cultural factors in the primary prevention of melanoma targeted to multiethnic children
PDF
Screening and association testing of coding variation in steroid hormone coactivator and corepressor genes in relationship with breast cancer risk in multiple populations
PDF
Identifying genetic, environmental, and lifestyle determinants of ethnic variation in risk of pancreatic cancer
PDF
Arm lymphedema in a multi-ethnic cohort of female breast cancer survivors
PDF
The effects of hormonal exposures on ovarian and breast cancer risk
PDF
Genetic epidemiological approaches in the study of risk factors for hematologic malignancies
PDF
Native American ancestry among Hispanic Whites is associated with higher risk of childhood obesity: a longitudinal analysis of Children’s Health Study data
PDF
Genes and hormonal factors involved in the development or recurrence of breast cancer
PDF
Associations between isoflavone soy protein (ISP) supplementation and breast cancer in postmenopausal women in the Women’s Isoflavone Soy Health (WISH) clinical trial
PDF
Genetic and environmental risk factors for childhood cancer
PDF
Disparities in gallbladder, intra-hepatic bile duct, and other biliary cancers among multi-ethnic populations: a California Cancer Registry study
PDF
Effects of post-menopausal hormone therapy on arterial stiffness in the ELITE trial
PDF
Air pollution and breast cancer survival in California teachers: using address histories and individual-level data
PDF
Diet quality and pancreatic cancer incidence in the multiethnic cohort
PDF
Associations between ambient air pollution and hypertensive disorders of pregnancy
PDF
Investigating racial and ethnic disparities in patient experiences with care and health services use following colorectal cancer diagnosis among older adults with comorbid chronic conditions
PDF
Genes and environment in prostate cancer risk and prognosis
Asset Metadata
Creator
Cole, Sarah Elizabeth
(author)
Core Title
Carcinogenic exposures in racial/ethnic groups
School
Keck School of Medicine
Degree
Doctor of Philosophy
Degree Program
Epidemiology
Degree Conferral Date
2022-08
Publication Date
07/15/2022
Defense Date
05/12/2022
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
breast cancer,childhood,Ethnicity,Hispanic,hormone receptor negative,hormone receptor positive,Melanoma,menopausal status,menstrual months,OAI-PMH Harvest,Prevention,standard erythemal dose,Sunburn,ultraviolet radiation
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Cockburn, Myles (
committee chair
), Allen, Martin (
committee member
), Miller, Kimberly (
committee member
), Wu, Anna (
committee member
)
Creator Email
scole@usc.edu,scoleemail@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-oUC111371442
Unique identifier
UC111371442
Legacy Identifier
etd-ColeSarahE-10834
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Cole, Sarah Elizabeth
Type
texts
Source
20220715-usctheses-batch-953
(batch),
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Repository Email
cisadmin@lib.usc.edu
Tags
breast cancer
childhood
Hispanic
hormone receptor negative
hormone receptor positive
menopausal status
menstrual months
standard erythemal dose
ultraviolet radiation