Solar UV radiation and cancer in young children

Abstract

Background

Studies have shown that higher solar UV radiation exposure (UVR) may be related to lower risk of some cancers in adults. Recently an ecological study reported lower risks of some cancers among children living in higher UVR cities and countries. In a large population-based case-control study in California we tested the hypothesis that childhood cancers may be influenced by UVR.

Methods

Cancers in children ages 0 to 5 years were identified from California Cancer Registry records for 1986–2007 and linked to birth certificate data. Controls were sampled from the birth certificates at a ratio of 20:1. Based on birth address, we assigned UVR exposure in units of Watt-hours/m² using a geostatistical exposure model developed with data from the National Solar Radiation Database.

Results

For cases with UVR exposure of 5111 Watt-hrs/m² or above we estimated a reduction in odds of developing acute lymphoblastic leukemia (OR: 0.89, 95% CI: 0.81, 0.99), hepatoblastoma (OR: 0.69, 95% CI: 0.48, 1.00), and non-Hodgkin’s lymphoma (OR: 0.71, 95% CI: 0.50, 1.02) adjusting for mother’s age, mother’s race and child’s year of birth. We also observed a small increase in odds for intracranial/intraspinal embryonal tumors (OR: 1.29, 95% CI: 1.01, 1.65).

Conclusions

Our findings suggest that UVR during pregnancy may decrease the odds of some childhood cancers. Future studies should explore additional factors that may be correlated with UVR exposure and possibly include biomarkers of immune function and vitamin D.

Impact

This study shows protective associations of UVR with some childhood cancers.

INTRODUCTION

Childhood cancer is a rare disease that may be triggered prenatally. The few known causes of pediatric cancers include ionizing radiation, Down syndrome, and some genetic or chromosomal anomalies (1). Additional potential risk factors have been suggested for specific cancer types, but due to the rarity of the childhood cancers, it has been difficult to establish causes, and preventive measures are similarly lacking.

Solar UV radiation (UVR) a known risk factor for skin cancers has been identified as a potential protective factor for some cancers. UVB radiation produces vitamin D through reactions occurring in human skin (2). Recent meta-analyses of vitamin D levels and breast (3–5) and colorectal cancer (5–7) have provided some support for protective effects of vitamin D, but there have been(7) inconsistent results for other cancers (8). Also, other UV-induced mechanisms may contribute to potential protection from cancer (9).

As summarized in a review, inverse relationships between UVR and incidence or mortality of cancer of the bladder, breast, colon, esophagus, gallbladder, stomach, lung, ovary, pancreas, prostate, rectum, kidney, thyroid, uterine corpus, and vulva, as well as Hodgkin’s lymphoma, non-Hodgkin’s lymphoma, and multiple myeloma, and were found in ecological studies in the U.S. and other countries (10). An inverse association was also observed for leukemia in the U.S. (11). First, an ecological study found an inverse relationship between colon cancer mortality and annual sunlight levels in the United States, and proposed vitamin D as a possible mechanism (12). Subsequently, a landmark study in the U.S. used ground-level UVB irradiance data from NASA and age adjusted sex- and race-specific cancer mortality rates by state economic area and found inverse associations for 18 cancers in adults (13). Since this study did not adjust for other potential confounders, a later follow up to this study using state-level UVR data adjusted for alcohol consumption, Hispanic heritage, urban/rural residence, poverty level, and, as a proxy for smoking, the lung cancer mortality rate, and found inverse associations for mortality from 13 cancers in adults (14). A study in the U.S. using the NASA UVR data at the county level found inverse associations with incidence of and mortality from several cancers (11).They also found positive associations for some cancer sites (anus, cervix, melanoma, oral cavity, and other skin). Ecological studies in Europe, Australia, and Asia have also found inverse associations for several cancers in adults (10). In considering these previous studies, it should be noted that mechanisms protecting against cancer mortality may differ from those protecting against cancer incidence.

Multi-country studies, mostly using latitude as a proxy for UVR, showed lower rates of mortality or incidence of breast, lung, ovarian, kidney, brain and uterine cancer and leukemia in adults residing in countries closer to the equator where UVR levels are higher (10). A recent ecological study of childhood cancers in several countries found a protective effect of solar UVR on risks for several cancers in children ages 0–14 using rates extracted from cancer registries and adjusting for measures of economic development of the country (15). However, it may be difficult to adequately control for possible confounding factors such as smoking, alcohol use, diet, reproductive factors, infections and SES in multi-country studies.

These ecological studies have provided a good foundation for this field, but may still be subject to the ecological fallacy or residual confounding. Also, several studies have used latitude, which does not capture variation in UVR due to elevation, terrain, or other factors. For example, in the United States UVB levels are higher at higher latitudes and also west of the Rocky Mountains due to a thinner stratospheric ozone layer and higher elevations (10). Other studies have used UVR measures with low spatial resolution such as state- and country-level, which may obscure important exposure differences at a smaller spatial scale. Case-control and cohort study designs have also been used to examine potential protective associations with some cancers, mainly in adults (16–23). In particular several studies investigated non-Hodgkin’s lymphoma with some prospective studies showing a protective association (18, 22, 23) with UVR and others showing a harmful (19, 20) or null association (21).

Only one previous study examined the association between UVR and cancer in children, and it was limited to ecological data at the level of cities and countries. Our objective was to assess the associations between UVR during pregnancy and childhood cancers in California in a population-based case-control study using UVR exposures based on mother’s address from the birth certificate. Due to its diverse latitudes (32°30′ – 42° North) and elevations (282 ft below sea level in Death Valley to 14,494 ft at the peak of Mt Whitney), California receives a wide range of UVR. Other studies in California have linked similar UVR measures to increased risk for melanoma (24, 25) and reduced risk of non-Hodgkin’s lymphoma (22).

MATERIALS AND METHODS

Study population

Cancer cases in children ages 0 to 5 years were identified from California Cancer Registry records for 1988–2007 and matched to their birth certificates (26, 27). Using first and last names and date of birth, we were able to match 89% of cases to a California birth certificate. Controls without a diagnosis of cancer prior to age 6 were also sampled from the California birth certificates for the same years at a ratio of 20:1, frequency matched on year of birth. Maternal address and information on potential confounding variables were obtained from the birth certificates. Using data from California death certificates, we excluded controls who died before age 6 (n=1,522). After excluding nine cases and 610 controls with home addresses outside of California, for whom we lacked UVR exposure information, our study population comprised 10,476 cases and 207,568 controls.

Outcomes were defined based on Surveillance Epidemiology and End Results (SEER) groupings. We included the following cancers in our analysis: acute lymphoblastic leukemia (SEER code 11), acute myeloid leukemia (12), Hodgkin’s lymphoma (21), non-Hodgkin’s Lymphoma (22,23), astrocytomas (32), ependymomas/choroid plexus tumors (31), other gliomas (34), intracranial/intraspinal embryonal tumors (33), other intraspinal/intracranial neoplasms (35,36), neuroblastoma (41), Wilms’ tumor (61), hepatoblastoma (71), bone tumors (81, 83–85), rhabdomyosarcoma (91), other soft tissue sarcomas (94,95), germ cell tumors (101,102,103), and retinoblastoma (050). Cases were not limited to first primary incident cancers. As a test of the validity of our exposure measure, childhood melanoma (114) was also examined even though our study only included 39 cases. This study was approved by the University of California, Los Angeles Institutional Review Board.

UV exposure assessment

UVR exposure in units of Watt-hours/m² (Wh/m²) was assigned to subjects based on a geostatistical exposure model (ANUSPLIN) that estimates ground-level UVR exposure using data from the National Solar Radiation Database from over 200 UVR measurement stations and also takes into account elevation, latitude and longitude (28). Using information from 30 years of data (1961–1990), the model predicts average daily total global solar radiation (AVGLO), which is defined as the total amount of direct and diffuse solar radiation in Wh/m² received on a horizontal surface. Annual average UVR was then calculated based on a 20 km buffer around each mother’s residential address from the child’s birth certificate to capture exposure at home and in nearby areas. These measures serve as a proxy for mothers’ exposure to UV light during pregnancy. Exposure was divided into quartiles based on its distribution among control subjects (Q1: 3133 – 4946; Q2: >4946 – 5030; Q3: >5030 – 5111; Q4: >5111 – 5804 Watt-hrs/m²).

Statistical methods

We used unconditional logistic regression to examine associations between UVR exposure and the aforementioned childhood cancers. All models were adjusted for our matching variable, child’s birth year. We also adjusted for maternal race/ethnicity since individuals with more pigmentation in their skin need more UVR to maintain appropriate vitamin D levels, and parental race/ethnicity is associated with most childhood cancers (29, 30). We also adjusted for maternal age in the model because higher maternal age is associated with a greater risk of several childhood cancers and may be related to time spent outdoors and sun protection behaviors (31–33). Finally, we evaluated parity, neighborhood socioeconomic status and payment method for prenatal care as potential confounders in our models using a 10% change in estimate criterion for inclusion in the model.

Parity is related to some cancers and could be related to time spent outdoors (34, 35), but did not change OR estimates for any of the cancers. Neighborhood SES was calculated based on an algorithm developed by Yost et al from Census data in California using principal components analysis. This index was created from seven census indicator variables of SES at the block-group level (education index, median household income, percent living 200% below poverty level, percent blue-collar workers, percent older than 16 years in workforce without job, median rent, and median house value) (36). Only the odds ratios for melanoma changed by 10% or more with the addition of neighborhood SES to our models, but due to the small number of cases these estimates were statistically imprecise. Payment type for prenatal care (private/HMO/Blue Cross-Blue Shield vs. Medi-Cal/other/self-pay) has been found by our research group in previous studies to be a good marker of individual level socioeconomic status (37). Adding it to models changed ORs for other gliomas, Hodgkin’s lymphomas, and other intracranial and intraspinal neoplasms. However, we decided not to include this variable in the final models since it only affected estimates for a few cancers and did not meaningfully change the interpretation of results for these cancers. Based on the above considerations, our final models adjusted for maternal race/ethnicity, maternal age, and child’s birth year. Participants with missing data for any of the covariates were dropped from the regression models. We assessed trend by running the medians of the UVR quartiles as an ordinal variable in our adjusted models.

Additionally, we conducted stratified analyses to investigate effect measure modification of UVR exposure by mother’s race/ethnicity. Only cancers with an n 20 were included to ensure adequate sample size.

RESULTS

Among the childhood cancers we examined acute lymphoblastic leukemia (ALL) was most common (36%), followed by central nervous system tumors (21%) and neuroblastoma (11%). Characteristics of cases and controls are presented in Table 1. For all cancers combined, case mothers are slightly more frequently white (41%) compared to mothers of controls (37%), and racial/ethnic distributions differed by cancer type; e.g. a higher proportion of ALL and Hodgkin’s lymphoma case mothers were Hispanic. For all cancers combined, case mothers were older and of higher individual and neighborhood socioeconomic status.

Table 1.

Characteristics of cancer cases and controls in children ages 0–5 in California (1986–2007)

	Controls		All Cancers
	n	%	n	%
Mother’s race/ethnicity
White, non-hispanic	75437	37	4204	41
Hispanic, any race	93267	46	4561	44
Black	14392	7	555	5
Asian/PI	20210	10	948	9
Total	203306	100	10268	100
Mother’s age
<20	22619	11	1049	10
20–24	50838	25	2292	22
25–29	57662	28	2961	28
30–34	48036	23	2555	24
35+	28375	14	1617	15
Total	207530	100	10474	100
Parity
0	81616	39	4058	39
1 or more	125816	61	6413	61
Total	207432	100	10471	100
Payment type for prenatal care
Private/HMO/Blue Cross-Blue Shield	91467	51	5095	56
Medi-cal/Other/Selfpay/Etc	88594	49	4026	44
Total	180061	100	9121	100
Quintiles of neighborhood SESa
1	49718	24	2376	23
2	48372	23	2436	23
3	46630	22	2367	23
4	33839	16	1727	17
5	28834	14	1559	15
Total	207393	100	10465	100
UV quartiles (Watt hrs/m^2)
3133 – 4946	51973	25	2732	26
>4946 – 5030	52047	25	2597	25
>5030 – 5111	51822	25	2651	25
>5111 – 5804	51466	25	2486	24
Total	207308	100	10466	100

^aBased on Yost et al index.(36)

Odds ratios and 95% confidence intervals for quartiles of UVR exposure adjusting for mother’s age and race/ethnicity, and child’s birth year are shown in Table 2. For children whose mothers were living in areas with UVR exposure in the highest quartile (≥5111 Watt-hrs/m²) we estimated decreased odds for developing ALL (OR: 0.89, 95% CI:0.81, 0.99), hepatoblastoma (OR: 0.69, 95% CI: 0.48, 1.00), and non-Hodgkin’s lymphoma (OR: 0.71, 95% CI: 0.50, 1.02). On the other hand, odds of being diagnosed with melanoma were increased for children of mothers with annual average UVR exposures greater than 5111 Watt-hrs/m², but our estimate’s confidence interval was wide due to the small number of cases (n= 39, 13 in the highest quartile of UVR, OR: 2.34, 95% CI: 0.88, 6.21). We also observed an increase in odds for intracranial/intraspinal embryonal tumors with UV exposure of 5111 Watt-hrs/m² or above (OR: 1.29, 95% CI: 1.01, 1.65).

Table 2.

Adjusted odds ratios and 95% confidence intervals for the association between quartiles of UV radiation exposure based on mother’s address at birth and cancer in offspring ages 0–5 in California (1986–2007)

		Adjusted for birth year (controls n=207308)			Adjusted for mother’s age, race, and child’s birth year (controls n=203018)
Cancer	UV quartilesa (Watt hrs/m2)	Cases (n)	OR	95 %CI	Cases (n)	OR	95 %CI	Trend p-value
ALL		3396			3324
	Q1		ref			ref
	Q2		0.93	(0.85, 1.03)		0.91	(0.83, 1.01)
	Q3		1.03	(0.94, 1.13)		0.97	(0.88, 1.07)
	Q4		0.93	(0.84, 1.02)		0.89	(0.81, 0.99)	0.042
AML		565			552
	Q1		ref			ref
	Q2		0.99	(0.78, 1.24)		1.00	(0.79, 1.26)
	Q3		1.02	(0.81, 1.29)		1.00	(0.79, 1.27)
	Q4		0.87	(0.69, 1.11)		0.90	(0.70, 1.15)	0.447
Astrocytoma		801			789
	Q1		ref			ref
	Q2		0.82	(0.67, 1.00)		0.91	(0.75, 1.12)
	Q3		0.83	(0.68, 1.01)		0.92	(0.75, 1.12)
	Q4		0.92	(0.76, 1.11)		0.96	(0.79, 1.17)	0.648
Bone tumors		79			78
	Q1		ref			ref
	Q2		0.83	(0.41, 1.67)		0.80	(0.39, 1.65)
	Q3		1.47	(0.80, 2.73)		1.37	(0.73, 2.57)
	Q4		1.37	(0.73, 2.57)		1.22	(0.64, 2.32)	0.411
Other gliomas		220			217
	Q1		ref			ref
	Q2		0.93	(0.65, 1.33)		0.99	(0.68, 1.43)
	Q3		0.80	(0.55, 1.16)		0.90	(0.61, 1.32)
	Q4		0.82	(0.57, 1.19)		0.91	(0.62, 1.32)	0.568
Ependymoma and choroid plexus tumors		244			241
	Q1		ref			ref
	Q2		1.04	(0.74, 1.46)		1.08	(0.76, 1.53)
	Q3		0.89	(0.63, 1.27)		0.92	(0.64, 1.33)
	Q4		0.84	(0.59, 1.21)		0.88	(0.61, 1.27)	0.465
Hepatoblastoma		258			256
	Q1		ref			ref
	Q2		0.95	(0.69, 1.32)		0.95	(0.68, 1.32)
	Q3		0.85	(0.61, 1.19)		0.82	(0.58, 1.16)
	Q4		0.7	(0.49, 1.00)		0.69	(0.48, 1.00)	0.044
Hodgkin’s lymphoma		62			62
	Q1		ref			ref
	Q2		1.19	(0.51, 2.75)		0.90	(0.38, 2.10)
	Q3		2.33	(1.11, 4.89)		1.72	(0.81, 3.66)
	Q4		1.7	(0.78, 3.71)		1.32	(0.60, 2.92)	0.345
Non-Hodgkin’s lymphoma		271			268
	Q1		ref			ref
	Q2		0.80	(0.57, 1.12)		0.84	(0.60, 1.18)
	Q3		1.01	(0.73, 1.38)		1.05	(0.76, 1.45)
	Q4		0.72	(0.51, 1.01)		0.71	(0.50, 1.02)	0.119
Intracranial and intraspinal embryonal tumors		559			550
	Q1		ref			ref
	Q2		1.19	(0.94, 1.52)		1.26	(0.98, 1.61)
	Q3		1.14	(0.89, 1.45)		1.20	(0.93, 1.54)
	Q4		1.26	(1.00, 1.61)		1.29	(1.01, 1.65)	0.047
Other intracranial and intraspinal neoplasms		113			108
	Q1		ref			ref
	Q2		0.82	(0.48, 1.39)		0.95	(0.55, 1.66)
	Q3		0.93	(0.56, 1.55)		1.05	(0.62, 1.80)
	Q4		0.92	(0.55, 1.54)		1.05	(0.61, 1.79)	0.839
Neuroblastoma		1070			1042
	Q1		ref			ref
	Q2		0.81	(0.68, 0.96)		0.90	(0.76, 1.07)
	Q3		0.81	(0.68, 0.95)		0.91	(0.77, 1.09)
	Q4		0.87	(0.74, 1.03)		0.93	(0.79, 1.11)	0.385
Rhabdomyosarcoma		364			352
	Q1		ref			ref
	Q2		1.17	(0.89, 1.55)		1.22	(0.91, 1.63)
	Q3		0.92	(0.68, 1.23)		1.01	(0.74, 1.37)
	Q4		0.88	(0.65, 1.19)		0.97	(0.71, 1.33)	0.804
Other soft tissue sarcomas		140			136
	Q1		ref			ref
	Q2		1.14	(0.69, 1.90)		1.08	(0.64, 1.82)
	Q3		1.54	(0.96, 2.48)		1.49	(0.92, 2.43)
	Q4		1.33	(0.82, 2.18)		1.26	(0.77, 2.09)	0.260
Wilms’ tumor		824			812
	Q1		ref			ref
	Q2		0.90	(0.75, 1.09)		0.91	(0.75, 1.11)
	Q3		0.83	(0.68, 1.00)		0.87	(0.71, 1.06)
	Q4		0.91	(0.75, 1.09)		0.92	(0.76, 1.12)	0.335
Germ cell tumors		370			363
	Q1		ref			ref
	Q2		1.10	(0.83, 1.46)		1.17	(0.88, 1.57)
	Q3		1.12	(0.85, 1.49)		1.17	(0.87, 1.56)
	Q4		0.86	(0.63, 1.16)		0.9	(0.66, 1.23)	0.666
Retinoblastoma		606			591
	Q1		ref			ref
	Q2		1.02	(0.81, 1.27)		1.03	(0.82, 1.30)
	Q3		1.008	(0.81, 1.26)		1.03	(0.82, 1.30)
	Q4		0.877	(0.70, 1.10)		0.88	(0.69, 1.11)	0.338

^aQ1: 3133 – 4946; Q2: >4946 – 5030; Q3: >5030 – 5111; Q4: >5111 – 5804 Watt-hrs/m².

Effect estimates for models stratified by mother’s race/ethnicity are shown in Table 3. For ALL, we observed a 16% decrease in odds among Hispanic mothers and a 35% decrease in odds among African-American mothers living in counties in the highest quartile of UVR exposure. For black mothers a similar decrease in odds was observed in lower quartiles, but the confidence intervals were wide reflecting small cell sizes. The estimated effect for hepatoblastoma was strongest in the top quartile of UVR in Hispanics (OR=0.60, 95% CI: 0.35, 1.02), and children of White mothers in the top quartile of UV exposure were also protected, but by to a lesser degree, and the 95% CI included the null value (OR=0.79, 95% CI=0.44, 1.41). Effects could not be estimated in children of African-American mothers due to a small number of hepatoblastoma cases. For non-Hodgkin’s lymphoma, a 39% decrease in odds was observed for children of White mothers in the top quartile of exposure (OR=0.61, 95% CI:0.37, 1.01), while no effects were seen in children of Hispanic mothers, and our sample size was insufficient for children of African-American mothers. We also observed protective effects for neuroblastoma and germ cell tumors in Hispanic children only.

Table 3.

Odds ratios and 95% confidence intervals stratified by mother’s race/ethnicity^a

		White (controls n=75262)			Hispanic (controls n=93187)			Black (controls n=14381)
Cancer	UV quartilesb (Watt hrs/m2)	Cases (n)	ORc	95 %CI	Cases (n)	ORc	95 %CI	Case s (n)	ORc	95 %CI
ALL		1235			1675			96
	Q1		ref			ref			ref
	Q2		0.95	(0.81, 1.12)		0.91	(0.78, 1.06)		0.63	(0.39, 1.02)
	Q3		1.05	(0.90, 1.23)		0.95	(0.82, 1.11)		0.67	(0.34, 1.31)
	Q4		0.98	(0.84, 1.13)		0.84	(0.72, 0.98)		0.66	(0.37, 1.16)
AML		196			254			36
	Q1		ref			ref			ref
	Q2		0.72	(0.47, 1.09)		1.30	(0.86, 1.96)		0.94	(0.42, 2.07)
	Q3		1.00	(0.69, 1.44)		1.34	(0.90, 2.01)		0.40	(0.09, 1.81)
	Q4		0.87	(0.60, 1.26)		1.07	(0.70, 1.63)		1.09	(0.45, 2.63)
Astrocytoma		416			269			48
	Q1		ref			ref			ref
	Q2		0.87	(0.66, 1.14)		1.19	(0.80, 1.78)		0.78	(0.39, 1.58)
	Q3		0.79	(0.60, 1.04)		1.22	(0.82, 1.81)		1.30	(0.57, 2.98)
	Q4		0.95	(0.74, 1.22)		1.26	(0.85, 1.88)		0.72	(0.31, 1.70)
Bone Tumors		30			39			<20
	Q1		ref			ref			ref
	Q2		0.46	(0.13, 1.65)		1.13	(0.34, 3.77)		~	~
	Q3		1.01	(0.38, 2.64)		2.13	(0.71, 6.37)		~	~
	Q4		1.14	(0.47, 2.77)		1.56	(0.50, 4.91)		~	~
Other Gliomas		103			73			23
	Q1		ref			ref			ref
	Q2		1.08	(0.63, 1.84)		1.42	(0.66, 3.07)		0.532	(0.21, 1.33)
	Q3		0.74	(0.41, 1.36)		1.50	(0.70, 3.21)		0.406	(0.09, 1.84)
	Q4		1.23	(0.75, 2.01)		0.95	(0.42, 2.17)		0.251	(0.06, 1.14)
Ependymoma/Choroid Plexus Tumors		99			102			<20
	Q1		ref			ref			ref
	Q2		0.85	(0.50, 1.46)		1.59	(0.82, 3.07)		~	~
	Q3		0.79	(0.46, 1.36)		1.52	(0.79, 2.95)		~	~
	Q4		0.72	(0.42, 1.23)		1.11	(0.55, 2.23)		~	~
Hepatoblastoma		96			127			<20
	Q1		ref			ref			ref
	Q2		1.26	(0.75, 2.12)		0.76	(0.46, 1.25)		~	~
	Q3		0.95	(0.54, 1.67)		0.71	(0.43, 1.16)		~	~
	Q4		0.79	(0.44, 1.41)		0.60	(0.35, 1.02)		~	~
Hodgkin’s Lymphoma		<20			43			<20
	Q1		ref			ref			ref
	Q2		~	~		2.58	(0.56,11.76)		~	~
	Q3		~	~		4.16	(0.96,18.02)		~	~
	Q4		~	~		3.91	(0.89,17.21)		~	~
Non-Hodgkin’s Lymphoma		120			104			<20
	Q1		ref			ref			ref
	Q2		0.82	(0.50, 1.33)		0.63	(0.34, 1.19)		~	~
	Q3		0.83	(0.52, 1.35)		1.14	(0.65, 1.98)		~	~
	Q4		0.61	(0.37, 1.01)		0.79	(0.43, 1.44)		~	~
Intracranial/intraspinal embryonal tumors		244			237			36
	Q1		ref			ref			ref
	Q2		1.02	(0.69, 1.50)		1.18	(0.78, 1.77)		1.47	(0.66, 3.27)
	Q3		1.32	(0.93, 1.88)		0.97	(0.64, 1.47)		~	~
	Q4		1.40	(1.00, 1.95)		1.10	(0.73, 1.67)		1.35	(0.53, 3.40)
Other Intracranial/intraspinal Neoplasms		46			47			<20
	Q1		ref			ref			ref
	Q2		0.92	(0.38, 2.23)		0.85	(0.36, 2.00)		~	~
	Q3		1.34	(0.61, 2.94)		0.73	(0.31, 1.74)		~	~
	Q4		1.29	(0.59, 2.82)		0.83	(0.35, 1.94)		~	~
Neuroblastoma		537			357			67
	Q1		ref			ref			ref
	Q2		0.99	(0.78, 1.26)		0.73	(0.54, 1.00)		0.77	(0.42, 1.40)
	Q3		1.12	(0.89, 1.41)		0.65	(0.48, 0.89)		1.04	(0.49, 2.21)
	Q4		0.94	(0.75, 1.19)		0.81	(0.60, 1.10)		0.87	(0.44, 1.71)
Rhabdomyosarcoma		141			147			29
	Q1		ref			ref			ref
	Q2		1.00	(0.64, 1.57)		1.66	(0.96, 2.88)		1.472	(0.59, 3.65)
	Q3		0.95	(0.61, 1.50)		1.18	(0.66, 2.09)		1.567	(0.50, 4.94)
	Q4		0.78	(0.49, 1.23)		1.43	(0.81, 2.52)		0.562	(0.15, 2.18)
Other Soft Tissue Sarcomas		57			60			<20
	Q1		ref			ref			ref
	Q2		1.03	(0.48, 2.23)		1.03	(0.44, 2.44)		~	~
	Q3		1.09	(0.52, 2.31)		1.49	(0.66, 3.32)		~	~
	Q4		1.38	(0.70, 2.72)		1.00	(0.42, 2.38)		~	~
Wilms’ Tumor		355			337			68
	Q1		ref			ref			ref
	Q2		0.81	(0.61, 1.07)		1.03	(0.72, 1.46)		1.22	(0.64, 2.32)
	Q3		0.58	(0.42, 0.79)		1.20	(0.86, 1.70)		1.45	(0.65, 3.22)
	Q4		0.73	(0.56, 0.97)		1.15	(0.81, 1.63)		1.57	(0.79, 3.13)
Germ cell tumors		123			155			21
	Q1		ref			ref			ref
	Q2		1.02	(0.61, 1.69)		0.77	(0.50, 1.20)		1.035	(0.33, 3.26)
	Q3		1.31	(0.82, 2.10)		0.62	(0.39, 0.98)		2.628	(0.80, 8.63)
	Q4		0.9	(0.55, 1.48)		0.57	(0.35, 0.91)		0.747	(0.18, 3.13)
Retinoblastoma		204			277			45
	Q1		ref			ref			ref
	Q2		1.25	(0.86, 1.80)		0.93	(0.64, 1.34)		1.14	(0.53, 2.43)
	Q3		0.92	(0.62, 1.36)		1.02	(0.71, 1.46)		1.79	(0.74, 4.34)
	Q4		0.91	(0.62, 1.32)		0.83	(0.57, 1.20)		0.97	(0.39, 2.41)

^aOnly cancers with at least 20 cases are included.

^bQ1: 3133 – 4946; Q2: >4946 – 5030; Q3: >5030 – 5111; Q4: >5111 – 5804 Watt-hrs/m².

^cAdjusted for mother’s age and child’s year of birth.

DISCUSSION

Our results suggest a possible protective association between UVR and ALL, hepatoblastoma, and non-Hodgkin’s lymphoma in children diagnosed with any of these cancers through age five. Most estimated effect sizes were strongest in the top quartile of exposure (>5111 Watt hrs/m²). An exposure-response relationship with increasing quartiles of UVR exposure was observed for ALL, hepatoblastoma and intracranial/intraspinal embryonal tumors (p-value for trend p<.05), but not for other cancers possibly because the effect of UVR is only present at higher levels. Even though our estimates were based on a very small number of children with melanoma in this age group, our data suggested a positive association with melanoma development as would be expected if our exposure assessment for UVR was indeed valid; interestingly, with adjustment for neighborhood SES the OR for the top quartile of UVR exposure was 3.17 (95% CI: 1.16, 8.70).

The only previous study examining the association between UVR and multiple cancers in children found protective associations for lymphoid leukemia, acute non-lymphoblastic leukemia, Hodgkin’s lymphoma, brain/spinal neoplasms, sympathetic nervous system tumors, retinoblastoma, renal tumors, hepatic tumors, bone tumors, and germ cell/gonadal tumors (15). This was an ecologic study based on solar radiation data from NASA relying on age- and sex- stratified rates of cancer from the International Incidence of Childhood Cancer, Vol. II, which includes data provided by 75 registries in 57 countries adjusting for economic inequality (GINI index and gross domestic product). These findings support our results for ALL, Wilms’ tumor, hepatoblastoma, neuroblastoma, and retinoblastoma. However, our study did not replicate inverse associations they reported for brain and spinal neoplasms, and counter to this previous study, we observed a positive association for intracranial/intraspinal embryonal tumors. Since we saw no biologic explanation for this result, we interpreted it as a chance finding that needs to be replicated. For germ cell and gonadal tumors we observed a protective association only among Hispanics. Additionally we found a decreased risk for non-Hodgkin’s lymphomas, which was not observed by Musselman et al.

Of the cancers for which we found protective associations, NHL has been studied the most with regard to sun exposure effects, and a number of studies corroborate our finding of a protective effect. A case-control study of Greek children relied on reports of >15 days per year spent at a seaside resort to define high levels of sunlight exposure and found a protective association with childhood NHL (38). A large pooled analysis of ten case-control studies from several countries participating in the Interlymph Consortium showed a protective effect of recreational sun exposure assessed by questionnaire (17). The California Teachers Study (CTS) prospective cohort relied on the same UVR exposure model as our study and similarly found a reduction in Non-Hodgkin’s lymphoma risk in areas with higher UVR (22). Interestingly, the CTS study did not find any association with dietary vitamin D estimated from a validated food frequency questionnaire, causing speculations about the observed associations being due to non-vitamin D mechanisms such as immunosuppression through regulatory T cells. Recently, another prospective study of adults in six states in the U.S. found a protective association for UVR exposure for NHL incidence as well (18).

Contrary to these studies, three cohorts did not find protective associations (19–21). Also, the Vitamin D Pooling Project did not find a protective association for NHL (39). Whether or not risk or protective factors for adult NHL pertain also to childhood NHL is uncertain since the most common histopathologic types in childhood are different from those in adulthood.

With regard to leukemia, an ecological study using cancer incidence rates from the International Agency for Research on Cancer’s (IARC) GLOBOCAN database and UVR calculated using latitude and cloud cover estimates from NASA found an inverse association between leukemia incidence and UVR (40). Another ecological study using UVB data from NASA found inverse associations with leukemia incidence rates at the county level in the U.S. (11). Both of these studies focused on adults and grouped all types of leukemia together, thus subtypes such as ALL could not be investigated. The relationship between retinoblastoma and UVR was examined in two ecological studies. The first found higher incidence in countries and cities with higher annual ambient UVB (41), while the second study, building upon the first, used more cases from U.S. SEER data and found a null association (42). In a separate analysis of international data, they also found no association after adjusting for race, climate, and an indicator of economic development (42). These ecological studies used ambient annual UVR averages for cities, states or countries, in contrast to our case-control study using UVR measures based on a 20-km radius around a mother’s address. For hepatoblastoma, our study is only the second one to show a protective association from UVR exposure and these findings need to be confirmed in other studies (15).

UVR modulates the immune system through vitamin D and other pathways, and it is known to cause local and systemic immunosuppression (9). The role of vitamin D during pregnancy in the health of the child has not been well characterized aside from the documented increased risk of rickets among children born to vitamin D deficient mothers (43). Vitamin D modulates the developing immune system and regulates cytokines related to IgE-mediated allergy (44). Adverse child health outcomes related to immune function, including asthma and wheezing, have been associated with low maternal vitamin D status during pregnancy (43, 44). Given its effects on the developing immune system and its potential anti-cancer properties it has been hypothesized that maternal vitamin D status may be related to childhood cancer (44).

Both UVR and vitamin D were shown to be protective against tuberculosis and influenza infections (9), and maternal influenza infection was associated with higher odds of ALL in offspring (45). This provides support for a potential role of UVR in reducing cancer risk via reducing susceptibility to viral infections. If this is the mechanism then we might expect to see seasonality in the effect of UVR. To investigate the issue of seasonality we conducted stratified analyses for ALL cases by season of birth, comparing the sunny season in California (Apr-Sep) to the less sunny season (Oct-Mar). Indeed, for ALL we observed a slightly stronger protective effect for births occurring in the April-September period (results not shown). Month of birth was related to ALL diagnosis in previous studies (46, 47).

We also investigated associations by race/ethnicity (white, Hispanic, and African-American) to examine the potential influence of skin pigmentation. UVR appeared to be protective for ALL among children of Black and Hispanic mothers, though the confidence intervals were very wide for children of Black mothers due to a small sample size. The negative associations for NHL and Wilms’ tumor were observed mainly among white children, while for hepatoblastoma the effect of UVR seems stronger among children of Hispanic than White mothers. These differences in UVR effects may not only be due to skin pigmentation but also to time spent outdoors, sun protection or other behaviors that may affect UVR exposures in these women and their children or other race/ethnicity specific cultural or behavioral factors that interact with UVR exposure effects. A nationally representative survey reported lower use of sunscreen, but higher use of shade and long sleeves for sun protection in Hispanics compared to non-Hispanic whites (48). Studies have also found that sun protection behaviors in Hispanics are related to acculturation, and thus are changing over time (49, 50).

The measure of UVR employed in this study is a composite of several years of data and therefore does not allow us to look at trimester specific exposures. This is problematic if there are narrow windows of susceptibility when a mother’s UVR exposure may be particularly important for protection against cancer in their offspring. Also, we were not able to assess the relative importance of prenatal and postnatal UVR exposures in childhood cancer etiology. However, in our stratified analyses for ALL cases by season of birth, the stronger protective effect for births occurring in the April-September period might indicate the importance of late pregnancy or early life UVR exposures in the etiology of ALL.

If a mother moved during pregnancy we may have misclassified her exposure if UVR levels for the new residence were different. A recent review found that in seven studies conducted in the U.S., 14–32% of the mothers moved during pregnancy (51), but the median distance of moving was <10km. Since our UVR exposure metric represents a 20 km buffer around the mother’s home we would not expect moves to be a strong source of exposure misclassification in this study and we would expect it to be nondifferential with respect to the outcome and likely biased estimates toward the null. Also, the increased melanoma risk in our study supports the validity of our exposure measure, as does the increased risk for melanoma in adults based on the same UVR exposure data for California in previous studies (24, 25).

Beyond the variables we were able to control for, unmeasured risk factors for childhood cancer that vary by region similarly to UVR may be causing residual confounding. To investigate possible differences by region for ALL, we conducted stratified analyses based on statewide UVR quartiles separately for Southern California, which in general has higher UVR, and Northern California (results not shown). In Southern California UVR exposure was found to decrease odds of ALL in the top three quartiles, while in Northern California only the top quartile of exposure was found to be protective. This suggests that in Northern California, only those living in areas with the highest exposures receive enough UVR for a protective effect. There may also be confounding from other risk factors that vary regionally, such as diet or health behaviors that account for theobserved associations. Also, we did not adjust for multiple comparisons.

CONCLUSION

These preliminary findings suggest that UVR during pregnancy is related to lower likelihood of some childhood cancers. The mechanism may be through vitamin D production or through other UVR-mediated immune pathways. It is also possible that the observed associations are due to residual spatial confounding from yet unknown protective factors that we should try to investigate. Further studies are needed before any specific public health recommendations can be made, and any prevention messages must be carefully tailored to balance the possible benefits of UVR with skin cancer prevention (52). Future studies should collect residential history, explore additional factors that may be correlated with UVR exposure, investigate trimester-specific effects and possibly include biomarkers of immune function and vitamin D to further explore possible pathways for the observed associations. Distinguishing the effects of UVR and vitamin D will be necessary to identify the best manner in which to protect children from these cancers.