Personal sun exposure and risk of non Hodgkin lymphoma: A pooled analysis from the Interlymph Consortium

Abstract

In 2004–2007 4 independent case-control studies reported evidence that sun exposure might protect against NHL; a fifth, in women only, found increased risks of NHL associated with a range of sun exposure measurements. These 5 studies are the first to examine the association between personal sun exposure and NHL. We report here on the relationship between sun exposure and NHL in a pooled analysis of 10 studies participating in the International Lymphoma Epidemiology Consortium (InterLymph), including the 5 published studies. Ten case-control studies covering 8,243 cases and 9,697 controls in the USA, Europe and Australia contributed original data for participants of European origin to the pooled analysis. Four kinds of measures of self-reported personal sun exposure were assessed at interview. A two-stage estimation method was used in which study-specific odds ratios (ORs) and 95% confidence intervals (CIs), adjusted for potential confounders including smoking and alcohol use, were obtained from unconditional logistic regression models and combined in random-effects models to obtain the pooled estimates. Risk of NHL fell significantly with the composite measure of increasing recreational sun exposure, pooled OR = 0.76 (95% CI 0.63–0.91) for the highest exposure category (p for trend 0.01). A downtrend in risk with increasing total sun exposure was not statistically significant. The protective effect of recreational sun exposure was statistically significant at 18–40 years of age and in the 10 years before diagnosis, and for B cell, but not T cell, lymphomas. Increased recreational sun exposure may protect against NHL. © 2007 Wiley-Liss, Inc.

It was suggested at first that sunlight might increase risk for non-Hodgkin lymphoma (NHL) because of parallel upwards trends in incidence of melanoma and NHL, a positive geographical correlation between incidence rates of NHL and non-melanocytic skin cancer and an increased risk of NHL in people with a history of skin cancer.1, 2, 3, 4 Four independent studies directly relating personal sun exposure to NHL risk have suggested the opposite however: that sunlight might protect against NHL.5, 6, 7, 8 Reduced risks of NHL were associated with increasing sun exposure on non-working days or vacations in an Australian population,5 more sunbathing, sunburns and sunlamp exposure in the large SCALE study in Sweden and Denmark,6 with greater exposure to the midday summer sun, greater residential ambient UV and greater sunlamp or tanning booth exposure in a US study,7 and with using sunbeds or sunlamps and taking holidays in a sunny climate in a German study.8 One such study, which included only women in Connecticut, has found increased risks of NHL associated with a range of sun exposure measurements including having a suntan for less than 3 months a year and a suntan history of more than 60 years, and with increasing duration of spending time in strong sunlight in summer; the increased risk was particularly for chronic lymphocytic leukaemia and small lymphocytic lymphoma.9

We have done a collaborative pooled analysis that included these 5 and 5 additional case-control studies. All studies are participants in the International Lymphoma Epidemiology Consortium (InterLymph; http://epi.grants.cancer.gov/InterLymph/) and included measures of personal sun exposure that have been related to risk of NHL. Our analysis aims to evaluate systematically the possible effects of sun exposure of different types, at different ages and in different latitudes on risk of NHL. Because of the large sample size, we are also able to separately analyse NHL subtypes.

Material and methods

Study population

Studies were eligible for the pooled analysis of sun exposure and NHL if they were participating in InterLymph and had completed data collection, studied at least 1 measure of personal sun exposure, and had an electronic dataset available for pooling. Ten case-control studies were eligible and supplied data for the pooled analysis; 9 studies included men and women, 1 (Connecticut) included women only. The individual studies in the European multicentre study called Epilymph10 were analysed in 2 groups: Germany (referred to as Epilymph 1 in the pooled analysis) used its own sun exposure questionnaire, and France, Ireland, Italy and Spain (Epilymph 2) used 2 closely related sun exposure questionnaires. Selected information about locations and study populations for the 10 studies is given in Table I. Only participants of European origin and those not known to be HIV-positive were included in the pooled analysis. Participants of non-European origin were excluded because of their comparatively small numbers and generally much lower cutaneous sensitivity to sun exposure. The studies obtained informed consent from participants and ethics approval from their local human research ethics committees.

Table I. Characteristics of Case-Control Studies Included in the Pooled Analysis

Study name	Location	Age range	year of diagnosis	N	Participation (%)	N	Source	Participation(%)	Reference
1 RDD=random digit dialling. CMMS=Centers for Medicare and Medicaid Services. 2 The UK study included cases of Hodgkin's lymphoma and pair-matched controls; this analysis included controls frequency-matched to NHL cases.
NSW	New South Wales (NSW) and the Australian Capital Territory Australia	20–74	2000–2002	653	85	666	Random selection from electoral register; frequency matched by age, sex, state or territory	61	5, 11
NCI-SEER	Detroit, MI; Iowa; Los Angeles CA; Seattle WA USA	20–74	1998–2000	519	76	436	<65 years RDD; 65+ years random selection from Centers for Medicare and Medicaid Services, stratified by study area, age, sex, race	52	7
Nebraska	Nebraska USA	20–75	1999–2002	369	74	511	RDD, frequency matched by age and sex	78	12
Mayo Clinic	Iowa, Wisconsin; Minnesota, USA	18+	2002–2005	224	63	445	Frequency matched by age, sex, county of residence	69	n/a
Connecticut	Connecticut USA	21–84 womenonly	1995–2001	570	72	667	Women only: <65 years RDD; 65+ years random selection from Centers for Medicare and Medicaid Services. Frequency matched within 5 years of age	RDD1:69; CMMS:47	9
BC	Vancouver and Victoria Canada	20–82	2000–2004	628	78	649	Random selection from Client Registry of the Ministry of Health, frequency matched by age, sex, region	46	n/a
United Kingdom (UK)	Yorkshire Lancashire, South Lakeland and parts of South West England UK	18–64	1998–2001	700	75	1011	Individually matched2 by age, sex, region of residence (north, south) from general practice lists	71	13
SCALE	Denmark and Sweden	18–74	2000–2002	3055	81	3187	Random selection from population register	71	6
Epilymph 1	Germany	18–82	1999–2002	514	87	705	Random selection from population register; individually matched by age, sex, study region	44	8
Epilymph 2	France, Italy, Ireland and Spain	17–96	1998–2004	1011	87	1420	Hospital controls in Spain, France, Ireland, random selection from population census list in Italy; all matched by age +/−5 years), sex, study region	68% overall participation: 81% hospital,: 52% population controls	10

Classification of NHL and NHL subtypes

Most studies classified non-Hodgkin lymphoma using the WHO classification of lymphoma14; 1 study (Connecticut) used the REAL classification. The classification of subtypes has been described elsewhere.15 Although the WHO classification defines multiple myeloma as NHL, these cases had been excluded from most studies and thus from the pooled analysis. In most studies, pathologists reviewed pathology reports and, in some studies, pathology samples; the classification systems used in each study were combined in consultation with InterLymph's pathology working group.16 In addition to examining sun exposure effects in specific subtypes of NHL, we examined them in B cell subtypes as a group and T cell subtypes as a group.

Data collection and exposure definitions

Each study provided an anonymized electronic data set of sun exposure, demographic and potentially confounding variables which had been collected by in-person or telephone interviews. Original data were received for each study; we also received copies of original questionnaires and descriptions of methods. The variables provided included sun exposure, sun-related variables such as pigmentary characteristics and sun sensitivity, case or control status, NHL subtype for cases, sex, age, ethnic origin, and histories of alcohol consumption and cigarette smoking. The sun exposure data available and the age intervals in which they were collected are summarised in Table II. Detailed descriptions of the data collection methods for sun exposure information have been published for the New South Wales (NSW),5, 11 SCALE,6 the National Cancer Institute - Surveillance, Epidemiology, and End Results multicentre study (NCI-SEER),7 Connecticut,9 Germany (Epilymph 1)8 studies, and descriptions of the studies, but not sun exposure methods, for Epilymph 2,10 the UK13 and Nebraska.12 Published information is not yet available on the British Columbia (BC) and Mayo Clinic studies. All datasets were checked for internal consistency and agreement with previously published results and discrepancies resolved with the original investigators.

Table II. Sun Exposure Measures by Centre and Specific Age Groups and Time Intervals Used

Study	(1)		(2)	(3)	(4)	Age or time intervals in each study
	Outdoors and not in the shade in warmer months or summer hours on:		In the sun in leisure activities	In sun-light	Sun-bathing in summer-time,	Age intervals in analyses							10 years before
	Working days	Non-working days	Hours per day/week/month	Hours per day or week	Frequency per week or month	10–17	18–40		41–60		>60		Diagnosis or interview
1 Calculated from participants' age-specific data. 2 Asked by season. 3 No season was indicated. 4 No season and no specific days in the week were mentioned. NCI, UK, SCALE gave categories for respondents to select one. NSW = New South Wales, Australia; British Columbia, Canada; NCI-SEER = NCI-SEER Study, USA; Connecticut, USA; Nebraska, USA; MAYO = Mayo Clinic, USA; UK = United Kingdom; SCALE = Denmark & Sweden; Epilymph 1 = Germany; Epilymph 2 = France, Spain, Ireland, Italy.
NSW	Yes	Yes				10	20	30	40	50	60		Calculated1
British Columbia	Yes	Yes				10	20	30	40	50	60	70	Calculated
NCI-SEER	Yes	Yes				teens	20s	30s					In past 10 years
Connecticut	Yes	Yes				<18	18–40		41–60		>60		Calculated
Nebraska				Yes2									2 years ago
MAYO				Yes3		13–21	22–40		41+				Calculated
UK	Yes	Yes											10 years ago
SCALE					Yes		20						5–10 years ago
Epilymph 1			Yes4			10	20	30	40				Calculated
Epilymph 2	Yes3	Yes3				10	20	30	40				Calculated

Four kinds of measures of personal sun exposure had been collected in the studies (Table II). These were: hours spent outdoors and not under any shade in summer or the warmer months of the year on working days and on non-working days in NSW, British Columbia, NCI-SEER, Connecticut, UK and Epilymph 2 (season not specified); hours per day, week or month spent in the sun in leisure activities in Epilymph 1; hours per day or week in sunlight in Nebraska and Mayo Clinic; and the frequency per week or month of sunbathing in summertime in Denmark or Sweden in SCALE. Data on total sun exposure were collected in 8 studies: in 6 studies (NSW, British Columbia, NCI-SEER, Connecticut, Epilymph 2, UK) by asking separately about working and non-working days sun exposure, and in 2 studies (Mayo, Nebraska) the hours spent in sunlight were considered to be a measure of total sun exposure (they did not separately ask about working and non-working days). Working days or nonrecreational sun exposure was collected in 6 studies (NSW, British Columbia, NCI-SEER, Connecticut, Epilymph 2, UK) and recreational sun exposure in these 6 and in the 2 that asked only about recreational sun exposure (Epilymph 1 and SCALE). Seven studies obtained information on sun exposure in specific age intervals over most of life (NSW, British Columbia, NCI-SEER, Connecticut, Mayo) or just up to 40 years of age (Epilymph 1 and Epilymph 2). The remaining studies addressed more limited periods of life: at 5, 10 and 20 years before diagnosis (UK), before 2 years ago (Nebraska), and at age 20 and at 5–10 years before diagnosis (SCALE). The approach to measuring sun exposure by autobiographical enquiry in successive periods of life, used in most of the Interlymph studies, has been successful in past studies of skin cancer (see Refs.17–19) and shown good reproducibility (intraclass coefficient = 0.77) for total lifetime exposure hours;20, 21 it has also been validated against histologic changes indicating chronic sun damage.22

From these measures we constructed variables for pooling across studies for total sun exposure, non-working days or recreational sun exposure (referred to as recreational sun exposure) and working days or non-recreational sun exposure (referred to as non-recreational sun exposure). When possible, each covered the “lifetime” (see below), ages 10–17 (or <18) and 18–40, 41–60, 61–84 years—the age intervals of the Connecticut study—and 5, 10 and 20 years before diagnosis. The exposure hours reported for a specific age were assigned to the construct that contained that age; for example age 10 to the interval 10–17 years of age. The intervals 10–17 and 18–40 years and 10 years before diagnosis were common to most studies (Table II); results for them and for lifetime total sun exposure are presented in this report. Lifetime totals covered exposure from age 10 to age 40 or longer, depending on the study, and were calculated by summing responses across the available age intervals in each study; a time-weighted average was used for analysis.

We also constructed composite measures of total, of recreational and of non-recreational sun exposure as time-weighted averages of all relevant age and period-specific measurements in each study (see Table II) with additional imputed values where necessary to extend the lifetime coverage to 59 years of age. The available data covering the 30s, 40s and 50s indicated relative constancy of sun exposure across these decades of age. No values were imputed for 60 years of age or older because of the likely impact of retirement on sun exposure. The composite measures were the same as the corresponding lifetime measures for NSW, British Columbia, Connecticut and Mayo. For Epilymph 1 and Epilymph 2, the values at age 40 were used to impute values up to age 59, as were the values at age 30 for NCI-SEER except for ages covered by the exposure estimate for the past 10 years. Just the average of sunbathing frequencies at age 20 and 5–10 years ago for SCALE and the hours of exposure before 2 years ago in Nebraska and 5, 10 and 20 years ago in the UK were used as composite measures, of recreational and total sun exposure respectively, for these studies.

The measures were grouped into approximate quarters for each study based on the distribution of average annual hours in their controls. The exposure levels across studies are exemplified here by the quarters of reported daily recreational exposure hours: at age 18–40, the daily hours for exposure quarters were reasonably similar in NSW and British Columbia (<3, 3, 4, 5+ hr), NCI-SEER and Connecticut (<2, 2, 3, 4+) and the UK (<2, 2–3, 4–5, 6+). Maximum daily exposure hours were somewhat higher in Epilymph 2 (5.5–8 as the cutpoint between the 3rd and 4th quarters, depending on centre) but lower overall in Epilymph 1 (<1 hr in the baseline and 2.5+ in the 4th quarter). The hours were broadly similar in each study at 10 years before diagnosis and at 18–40 years of age for the 4th exposure quarter, but the baseline (1st quarter) values for 18–40 years were less in NCI-SEER, Connecticut and the UK: <1 hr compared with <2 elsewhere. The categories from which respondents chose in SCALE (0, 1, 2–3, 4+ times a week) were used in all analyses. In the composite recreational sun exposure measure, the daily hours by exposure quarter in the controls were nearly the same in all studies.

All participating studies except Connecticut and Nebraska conducted their studies in 2 or more regional centres; each regional centre was assigned the latitude of its main town or major city. The 32 regional centres covered a wide range of latitudes and UV climates, from 59° N in Sweden to 34° N in Los Angeles and the 2 major centres in NSW, Sydney (34° S) and Canberra (35° S). The centres were grouped by latitude to cover 3 or more degrees of latitude and to provide a reasonably even distribution of participant numbers for an examination of the effects of ambient UV irradiance on the relationship between sun exposure and NHL.

Statistical analysis

We used a two-stage estimation method to obtain study-specific odds ratios (ORs) and pooled ORs and 95% confidence intervals (CIs) for all NHL in dichotomous regression models and for NHL subtypes in polychotomous regression models.23

In the first stage, each study was analysed according to its original design. For the pair-matched studies in the UK and Epilymph 1, we compared risk estimates from both conditional and unconditional logistic regression because restriction to matched pairs reduced the case numbers substantially in the Epilymph 1 study, due to missing data. We found no difference in estimates between matched and unmatched analyses for the UK and only slight differences for Epilymph 1 and so opted for the larger numbers available in the unmatched approach. All studies were designed as population-based except those in 3 of the 4 Epilymph 2 countries (Spain, France, Ireland) which were hospital-based; there was no statistically significant heterogeneity across the 4 countries on any measure of sun exposure. The population-based controls in Connecticut and NCI-SEER were drawn from 2 different sources, random digit dialling (RDD) up to 65 years and Medicare and Medicaid registers at older ages; all others came from a single source.

Unconditional logistic regression models were used to estimate ORs for quartiles of each sun exposure variable in each of the 10 studies with age (continuous), sex and dummy variables for each study's regional centres as covariates; questionnaire type was included as a covariate in the Epilymph 2 centres. There was very little difference between analyses that adjusted for age in five-year age groups, age in decades and age as a continuous variable; therefore we chose the last in the interests of minimizing the number of terms in the model. Pigmentary characteristics and the skin's reaction to sunlight, considered as likely confounders in the pooled analysis, were also included: skin colour and propensity to burn in analyses for NSW, NCI, Mayo Clinic, British Columbia, Epilymph 1 and Epilymph 2, propensity to burn for SCALE and just skin colour for the UK. No sun sensitivity variable was included as a covariate in Connecticut (skin colour was available for only 295 women) and Nebraska. Individuals with missing data for any variable were excluded from that analysis; 1.3% of participants had missing data on 1 or more covariates. Eye colour, hair colour and socioeconomic status (SES; inferred from a census indicator of deprivation in the UK and NSW and from education in all other studies) showed no evidence of confounding with sun exposure in any study and were not included as covariates. We also examined whether cigarette smoking and alcohol consumption were confounders of sun exposure in these data; adjustment for tobacco and alcohol use made minimal changes to estimates of sun exposure effects so they were not included as covariates.

In the second stage, the adjusted study-specific ORs and standard errors (SE) were combined in random-effects models. Pooled random effects estimates were weighted by the inverse marginal variances: the sum of the individual study-specific variances and the variance of the random study effect (the extent of heterogeneity between studies). Confidence intervals for the random effects estimates were based on the marginal variances, thus the greater the random study effect, the wider the confidence intervals. Tests for linear trend were done by fitting each categorical sun exposure variable as a single ordinal variable in the logistic regression model; p-tests for trend were based on the Wald test. We assessed inter-study variability in the same model by applying Greenland's test of heterogeneity24 and using a χ² test with degrees of freedom 1 less than the number of studies to estimate p-values. In the presence of heterogeneity, we evaluated whether individual studies contributed to the heterogeneity; when the p-value for an individual study indicated it was a possible source of heterogeneity, we assessed the impact of the study's exclusion from the analysis on the pooled estimates. All statistical tests were 2-sided with a significance level of 0ñ05. SAS software version 9.1 was used for the statistical analyses.

Results

There were 17,940 eligible participants, 8,243 cases and 9,697 controls, in the 10 case-control studies: 9,291 men and 8,649 women with a median age of 59 years (range 16–96) – cases 60 years (range 18–89) and controls 59 years (range 16–96). Ten percent of participants in NSW, British Columbia, UK, Epilymph 1 and Epilymph 2 had very fair skin, nearly half (48%) had fair skin, 34% medium and 9% dark skin. In studies that had measured propensity to burn (NSW, British Columbia, Epilymph 1, Epilymph 2, NCI-SEER, Mayo, SCALE), 13% rated themselves as likely to burn with blisters when acutely exposed to the sun, 28% as likely to burn then peel, 34% to burn then tan and 24% to never burn.

Total sun exposure

Estimated average annual total sun exposure was not significantly associated with NHL, whether over the lifetime, at 10–17 or 18–40 years of age, or in the 10 years before diagnosis. There was, though, weak evidence for a protective effect in each case (Table III). The pooled OR for the highest category of composite total sun exposure was 0.87 (95% CI 0.71–1.05) and the corresponding ORs in individual studies ranged from 0.45 to 1.32 (p for heterogeneity = 0.12) (Fig. 1). There was significant heterogeneity among studies in the effect of total sun exposure at 10–17 years of age (P for heterogeneity <0.001). Removal of the 2 studies mainly responsible for this heterogeneity, Connecticut (increased ORs for all levels above the baseline) and Epilymph 2 (all ORs close to 1.0), reduced the risk estimate in the highest exposure quarter to OR = 0.61 (95% CI 0.50–0.74) (p for trend <0.001); the p-value for heterogeneity was then 0.13.

Figure 1. Composite total sun exposure and risk of NHL.

Table III. Average Total, Recreational and Nonrecreational Sun Exposure and Risk Of NHL.

Total sun exposure				Recreational sun exposure				Nonrecreational sun exposure
Exposure quarter	Casesn = 4,499	Controls n = 5,583	OR1 (95% CI)	Exposure quarter	Cases n = 7,284	Controlsn = 8,380	OR1 (95% CI)	Exposure quarter	Cases n = 3,936	Controls n = 4,680	OR1 (95% CI)
1 Odds ratios (ORs) adjusted for covariates in analyses of individual studies. 2 Five studies had comprehensive coverage of sun exposure throughout life and were included in each model: NSW, British Columbia, NCI-SEER, Connecticut, Epilymph 2. Not all studies had data for each sun exposure measure (see Table II); maximum numbers of studies and participants are in the analyses of composite sun exposure. Studies having data for a sun exposure measurement are listed with the results for the model. 3 p for trend in the ORs. 4 p for heterogeneity among studies.
Composite sun exposure
5 studies2 + Mayo, Nebraska, UK				5 studies + Epilymph 1, UK, SCALE				5 studies + UK
1	1,214	1,437	1.00	1	2,042	2,017	1.00	1	991	1,209	1.00
2	1,118	1,392	0.92 (0.71–1.20)	2	1,789	2,142	0.88 (0.74–1.04)	2	1,037	1,226	1.06 (0.94–1.19)
3	1,080	1,390	0.90 (0.76–1.07)	3	1,869	2,202	0.85 (0.69–1.04)	3	896	1,081	0.98 (0.81–1.20)
4	1,087	1,364	0.87 (0.71–1.05)	4	1,584	2,019	0.76 (0.63–0.91)	4	1,012	1,164	1.01 (0.89–1.15)
	p for trend3		0.08				0.005				0.98
p for heterogeneity4			0.12				0.001				0.62
Lifetime
5 studies + Mayo				5 studies + Epilymph 1				5 studies
1	730	760	1.00	1	760	785	1.00	1	662	744	1.00
2	640	777	0.89 (0.66–1.20)	2	702	830	0.89 (0.67–1.17)	2	656	731	0.98 (0.84–1.14)
3	652	773	0.88 (0.70–1.10)	3	704	818	0.89 (0.65–1.22)	3	637	741	0.95 (0.81–1.10)
4	686	766	0.88 (0.69–1.11)	4	636	802	0.79 (0.54–1.14)	4	666	711	0.98 (0.84–1.15)
	p for trend3		0.21				0.21				0.72
p for heterogeneity4			0.14				<0.001				0.94
Age 10–17
5 studies + Mayo				5 studies + Epilymph 1				5 studies
1	904	971	1.00	1	847	910	1.00	1	809	933	1.00
2	821	932	0.92 (0.75–1.11)	2	801	955	0.88 (0.76–1.02)	2	989	1033	1.03 (0.89–1.21)
3	963	1,203	0.83 (0.68–1.02)	3	935	1,157	0.82 (0.67–1.01)	3	719	883	0.85 (0.69–1.05)
4	816	1,035	0.76 (0.56–1.03)	4	1,162	1,370	0.85 (0.62–1.16)	4	782	893	0.87 (0.66–1.15)
	p for trend3		0.06				0.32				0.17
p for heterogeneity4			<0.001				<0.001				0.009
Age 18–40
5 studies + Mayo, UK				5 studies + Epilymph 1, UK, SCALE				5 studies + UK
1	987	1,138	1.00	1	1,755	1,705	1.00	1	1021	1203	1.00
2	1,052	1,279	0.99 (0.83–1.19)	2	2,044	2,434	0.85 (0.75–0.95)	2	1096	1274	0.99 (0.83–1.17)
3	976	1,292	0.89 (0.72–1.09)	3	1,615	1,908	0.87 (0.73–1.03)	3	817	1024	0.95 (0.83–1.08)
4	1,006	1,226	0.93 (0.82–1.06)	4	1,737	2,082	0.80 (0.68–0.94)	4	883	1021	0.97 (0.85–1.10)
	p for trend3		0.09				0.01				0.43
p for heterogeneity4			0.52				0.01				0.93
10 years before diagnosis
5 studies + Mayo, Nebraska, UK				5 studies + Epilymph 1, UK, SCALE				5 studies + UK
1	1,005	1,171	1.00	1	1,747	1,602	1.00	1	1105	1319	1.00
2	1,002	1,305	0.90 (0.80–1.02)	2	1,853	2,057	0.87 (0.79–0.96)	2	926	1080	1.00 (0.88–1.13)
3	1,030	1,254	0.94 (0.81–1.08)	3	1,551	1,935	0.79 (0.70–0.90)	3	681	810	0.96 (0.84–1.11)
4	880	1,140	0.84 (0.72–0.97)	4	1,338	1,732	0.74 (0.62–0.88)	4	625	728	0.98 (0.85–1.14)
	p for trend3		0.06				0.001				0.73
p for heterogeneity4			0.23				0.02				0.52

Recreational sun exposure

Risk of NHL fell with increases in all measures of recreational sun exposure; the downtrends in ORs with increasing composite exposure and exposure at 18–40 years of age and 10 years before diagnosis were statistically significant (Table III). There was significant inter-study heterogeneity in all measures mostly due to Connecticut but also to Epilymph 1 in the lifetime, and age 10–17 measures. On their removal, the pooled estimates of ORs for exposure above the baseline fell by 20% in the 2nd to 4th quarters of the lifetime measure (p for heterogeneity 0.13) and in the 3rd and 4th quarters of exposure at 10–17 years of age (p for heterogeneity 0.03). Connecticut and Epilymph 2 were an apparent source of heterogeneity in the composite measure and when removed, the pooled OR fell by 11% in the highest exposure category (p for heterogeneity 0.06). On removing Connecticut alone, the ORs fell by <10% at age 18–40 (p for heterogeneity 0.24) and <5% in the 10 years before diagnosis (p for heterogeneity 0.20) in the highest exposure category. The 8 studies that contributed information on composite recreational sun exposure had ORs between 0.48 and 1.25 for the highest quarter of exposure and a pooled odds ratio of 0.76 (95% CI 0.63–0.91, p for heterogeneity = 0.001) (Fig. 2).

Figure 2. Composite recreational sun exposure and risk of NHL.

Nonrecreational sun exposure

Nonrecreational sun exposure was not associated either way with risk of NHL (Table III).

Heterogeneity and effect modification

Heterogeneity in the age- or period-specific and the lifetime or composite measures for total and recreational sun exposure could not be explained by differences between studies in sources of controls (population or hospital), questionnaire type, sun sensitivity, SES (data not shown), or sex distribution (Table IV). Specifically with respect to sex distribution, for composite total sun exposure the OR was 0.82 (95% CI 0.66–1.02) in the highest exposure quarter in men and 0.93 (95% CI 0.73–1.17) in women and for composite recreational exposure 0.77 (95% CI 0.63–0.93) in men and 0.77 (95% CI 0.59–1.01) in women (Table IV). We also calculated ORs in 2 categories of age at diagnosis since Connecticut and NCI-SEER had different control populations by age—<65 years RDD and 65+ years Medicare and Medicaid registers—but found little evidence of any difference in the effects of recreational sun exposure by age at diagnosis: for the highest exposure category, the OR was 0.74 (95% CI 0.62–0.88) at <65 years and 0.80 (95% CI 0.61–1.05) at 65+ years.

Table IV. Composite Total and Recreational Sun Exposure and Risk of NHL by Sex

Exposure quarter	Cases (n = 4,499)	Controls (n = 5,583)	OR1 (95% CI)	Exposure quarter	Cases (n = 7,284)	Controls (n = 8,380)	OR1 (95% CI)
Composite total sun exposure				Composite recreational sun exposure
1 Odds ratios (ORs) adjusted for covariates in analyses of individual studies 2 p for trend in the ORs. 3 p for heterogeneity by sex among studies.
NSW, BC, NCI-SEER, CT, MAYO, NEB, Epilymph 2, UK				NSW, BC, NCI-SEER, CT, Epilymph 1, Epilymph 2, UK, SCALE
Men
1	614	669	ref	1	1072	1010	ref
2	507	638	0.84 (0.66−1.08)	2	978	1119	0.87 (0.76–0.98)
3	478	624	0.82 (0.69–0.98)	3	943	1044	0.85 (0.75–0.97)
4	578	727	0.82 (0.66–1.02)	4	863	1053	0.77 (0.63–0.93)
		p trend2	0.05				0.004
p for heterogeneity among studies			0.28				0.17
Women
1	585	707	ref	1	919	957	ref
2	628	753	1.06 (0.88–1.27)	2	840	985	0.90 (0.71–1.14)
3	550	745	0.96 (0.75–1.21)	3	889	1159	0.81 (0.60–1.10)
4	559	720	0.93 (0.73–1.17)	4	780	1053	0.77 (0.59–1.01)
		p trend2	0.40				0.05
p for heterogeneity among studies			0.04				<0.001
p for heterogeneity by sex3			0.60				0.86

We looked for evidence of any modification of the effect of personal sun exposure by ambient UV irradiance by analysing the relationship of composite recreational sun exposure with risk of NHL in 5 latitude bands (Table V). The fall in risk for NHL with increasing recreational exposure was greatest in the lowest latitude band, which included the NSW centres and the lower latitude centres of NCI-SEER and Epilymph 2. There were, however, strong downtrends also in the middle and highest latitude bands. Epilymph 2 centres were a source of heterogeneity in analyses at 34–40° N, Connecticut at 41–43° N and Epilymph 1 at 50–54° N.

Table V. Composite Recreational Sun Exposure by Latitude and Risk of NHL

Exposure quarter	Cases	Controls	OR1 (95% CI)
1 Odds ratios (ORs) adjusted for covariates in analyses of individual studies. 2 p for trend in the ORs. The 32 regional study centres were: 34°–40° − NSW (ACT, NSW); NCI-SEER (Los Angeles); Epilymph 2: Italy (Cagliari, Nuoro/Oristano), Spain (Madrid). Mayo Clinic (Iowa) and Nebraska had no data on recreational sun exposure. 41°–43° − Connecticut, USA; NCI-SEER (Iowa, Detroit); Spain (Barcelona, Tarragona); France (Montpellier). Mayo Clinic (Wisconsin, Minnesota) had no data on recreational sun exposure. 47°–49° − Epilymph 1 (Munich, Wurzberg, Heidelberg, Ludwigshafen); Epilymph 2: France (Dijon); NCI-SEER (Seattle); British Columbia (Victoria, Vancouver). 50°–54° − Epilymph 1 (Bielefeld, Hamburg); Epilymph 2: France (Amiens); Ireland (Dublin); UK (Yorkshire, Southwest). 56°–62°− SCALE (Denmark, Sweden).
NSW, BC, NCI-SEER, CT, Epilymph1, Epilymph 2, UK, SCALE
34^° S & 34^° − 40° N
1	404	312	1.00
2	243	279	0.64 (0.51–0.81)
3	284	315	0.65 (0.49–0.86)
4	278	344	0.61 (0.40–0.94)
	p for trend2		0.08
p for heterogeneity among studies			0.02
41° − 43° N
1	224	312	1.00
2	269	310	1.11 (0.68–1.80)
3	284	312	0.98 (0.56–1.74)
4	237	335	0.68 (0.32–1.42)
	p for trend2		0.27
p for heterogeneity among studies			0.005
47° − 49° N
1	330	349	1.00
2	270	339	0.80 (0.63–1.00)
3	259	344	0.74 (0.59–0.94)
4	238	284	0.73 (0.57–0.94)
	p for trend2		0.01
p for heterogeneity among studies			0.88
50° − 54° N
1	292	387	1.00
2	291	402	1.17 (0.74–1.83)
3	198	296	1.40 (0.69–2.83)
4	232	367	1.11 (0.57–2.15)
	p for trend2		0.80
p for heterogeneity among studies			0.02
56° − 62° N
1	792	657	1.00
2	716	812	0.82 (0.71–0.95)
3	844	935	0.78 (0.67–0.90)
4	599	689	0.72 (0.62–0.85)
	p for trend2		<0.001
p for heterogeneity among studies			−
p for heterogeneity of trend effect among latitude groups			0.74

NHL subtypes

Risks of both B cell and T cell lymphomas were reduced with increased recreational sun exposure, although only B cell lymphomas had a consistent downwards trend in risk as sun exposure increased (p for trend = 0.006; Table VI). For composite average annual recreational sun exposure and B cell and T cell subtypes, the strongest and most consistent evidence of reduced risk with increasing exposure in the B cell lymphomas was for follicular, diffuse large B cell, mantle cell and marginal zone lymphomas, and in the T cell lymphomas for those classified as other types (Table VII). Interstudy heterogeneity was reduced by removal of Connecticut and Epilymph 1 from the model for other types of B cell lymphomas (ORs fell by 24%). Most of the cutaneous T cell lymphomas were mycosis fungoides (N= 129), for which the ORs were increased only in the third quarter of sun exposure, OR = 1.27 (95% CI 0.73–2.20) (p for trend 0.86).

Table VI. Composite Total And Recreational Sun Exposure and Risk of B Cell and T Cell Lymphomas

Exposure quarter	Cases (n = 4,304)	Controls (n = 5,583)	OR1 (95% CI)	Exposure quarter	Cases (n = 7,087)	Controls (n = 8,380)	OR1 (95% CI)
Composite total sun exposure				Composite recreational sun exposure
1 Odds ratios (ORs) adjusted for covariates in analyses of individual studies. 2 p for trend in the ORs.
NSW, BC, NCI-SEER, CT, MAYO, NEB, Epilymph 2, UK				NSW, BC, NCI, CT, Epilymph 1, Epilymph 2, UK, SCALE
B cell lymphomas				B cell lymphomas
1	1079	1437	1.00	1	1870	2017	1.00
2	1003	1392	0.94 (0.72–1.23)	2	1610	2142	0.88 (0.75–1.02)
3	980	1390	0.93 (0.79–1.09)	3	1703	2202	0.85 (0.70–1.03)
4	972	1364	0.88 (0.72–1.06)	4	1434	2019	0.75 (0.62–0.91)
		p trend2	0.10				0.006
p for heterogeneity among studies			0.24				0.001
T cell lymphomas				T cell lymphomas
1	81	1437	1.00	1	121	2017	1.00
2	58	1392	0.75 (0.52–1.07)	2	127	2142	0.86 (0.55–1.34)
3	62	1390	0.72 (0.47–1.09)	3	112	2202	0.80 (0.61–1.06)
4	69	1364	0.86 (0.59–1.24)	4	110	2019	0.83 (0.62–1.10)
		p trend2	0.28				0.12
p for heterogeneity among studies			0.27				0.99

Table VII. Composite Recreational Sun Exposure and Risk of NHL Subtypes

Exposure quarter	B cell lymphomas							T cell lymphomas
	Follicular (n = 1,642)	Diffuse (n = 2,176)	Marginal Zone (n = 492)	CLL/SLL (n = 1,204)	Burkitt's (n = 65)	Mantle Cell (n = 315)	Other types (n = 723)	Cutaneous(n = 146)	Peripheral(n = 107)	Other types (n = 217)
	OR1 (95% CI)	OR1 (95% CI)	OR1 (95% CI)	OR1 (95% CI)	OR1 (95% CI)	OR1 (95% CI)	OR1 (95% CI)	OR1 (95% CI)	OR1 (95% CI)	OR1 (95% CI)
1 Odds ratios (ORs) adjusted for covariates in analyses of individual studies. 2 p for trend. 3 p for heterogeneity among studies.
NSW, BC, NCI-SEER, CT, UK, Epilymph 1, Epilymph 2, SCALE
composite recreational sun exposure
1	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
2	0.83	0.79	0.87	0.96	1.11	0.84	1.07	1.14	0.86	0.88
	(0.71–0.96)	(0.61-1.01)	(0.64–1.17)	(0.66–1.40)	(0.54–2.29)	(0.60–1.18)	(0.75–1.52)	(0.68–1.91)	(0.45–1.64)	(0.60–1.29)
3	0.87	0.75	0.91	0.94	1.10	0.84	0.93	1.24	0.85	0.64
	(0.68–1.11)	(0.61-0.93)	(0.63–1.32)	(0.71–1.25)	(0.48–2.54)	(0.61–1.16)	(0.63–1.38)	(0.74–2.09)	(0.48–1.52)	(0.39–1.04)
4	0.73	0.69	0.74	0.83	0.78	0.71	0.93	1.07	0.83	0.69
	(0.62–0.86)	(0.55-0.87)	(0.49–1.12)	(0.59–1.17)	(0.27–2.20)	(0.50–1.01)	(0.63–1.38)	(0.62–1.84)	(0.46–1.49)	(0.46–1.05)
p trend2	<0.001	<0.001	0.17	0.29	0.22	0.08	0.51	0.86	0.67	0.02
p heterogeneity3	0.52	0.06	0.07	0.09	0.28	0.88	0.01	0.26	0.66	0.63

Discussion

Our pooled analysis of 10 InterLymph case-control studies of sun exposure and NHL shows a statistically significant fall in risk of NHL with increasing recreational sun exposure. There was a weaker downtrend in risk with increasing total sun exposure, which was not statistically significant, and no trend in either direction for non-recreational exposure. The downtrend with recreational exposure was observed with exposure at 18–40 years of age and at 10 years before diagnosis in the 8 studies for which these measures were available. The trend was less evident and not statistically significant for exposure at 10–17 years of age. There was no clear evidence of modification of the effect of recreational exposure by sex, age, sun sensitivity or latitude of residence. The trend was no more evident in B cell than T cell lymphomas but perhaps more evident in follicular, diffuse large B cell, marginal zone and mantle cell lymphomas than other types of B cell lymphomas.

This pooled analysis has the advantage of a large sample size, inclusion of studies that had information on potentially confounding factors, and a consistent classification of NHL: diagnostic accuracy is 80+% with the WHO NHL classification.25, 26 Participation rates were reasonable, mainly 70% or higher in cases and 50% or more in controls (see Table I). While low participation could cause bias if participation was related to sun exposure, we have no reason to believe it was. When these studies were first proposed in the 1990s, thought of a role for sun exposure in lymphoma etiology was rare even in scientific circles and it was certainly not a salient issue in the wider community from which the participants came.

Pooling was not as straightforward as it might be with a more readily quantifiable exposure. While questions were directed at the same exposure (sun or sun-related behaviour), not all studies covered identical exposure constructs or the same age or time intervals. Nevertheless, sun exposure measurement mostly followed a common approach of asking about daily hours in the sun at specific ages, which were the decade years in NSW, British Columbia, Epilymph 2 and broad age groups in Connecticut, NCI-SEER and Mayo Clinic, or time periods before diagnosis in NCI-SEER and the UK. This approach gave a solid framework for calculating total and recreational sun exposure hours into which we fitted the sun exposure inferred from leisure activity hours (Epilymph 1), sunbathing frequency (SCALE), or general sun exposure (Nebraska). Different numbers of studies contributed to each measure although 5 studies (NSW, British Columbia, NCI-SEER, Connecticut and Epilymph 2) were in most. When we repeated the models for lifetime total and for recreational sun exposure at age 18–40 (includes 8 studies in Table III) in the same 6 studies as for models of lifetime and age 10–17 sun exposure, the estimates hardly changed. Whatever the approach to measuring sun-related behaviour, a reasonably consistent pattern emerged: risk of NHL was less in people with increased recreational sun exposure.

The heterogeneity in the association between sun exposure and NHL was limited mainly to 2 studies, Connecticut9 and Epilymph 1.8 Both addressed their questions to particular ages in life and asked about hours in the sun on working and non-working days (Connecticut) or hours of leisure activities (Epilymph 1). Epilymph 1 had measured recreational exposure only and the Connecticut data contributed mainly to heterogeneity in the association with recreational sun exposure. The Connecticut study recruited women only; however, recreational sun exposure carried a very similar risk for NHL in men and women in our analyses (see Table IV). We cannot point to any obvious causes for the observed heterogeneity.

All published case-control studies that have examined individual sun exposure and risk of NHL are included in this pooled analysis, thus we cannot compare our findings with those in other studies. Ecological evidence on the association between ambient UV irradiance and risk of NHL has not clearly favoured a causal or a protective association. While the positive correlation of NHL incidence and mortality with latitude, and negative correlation with ambient UV irradiance, in the USA27, 28, 29 offer some support for a protective effect, the opposite was observed in data for the UK and Europe.29, 30 That NHL risk fell with increasing residential ambient UV in the recent case-control study of Hartge et al.7 is consistent with the protective effect of the earlier ecological studies in the USA. Occupational sun exposure, inferred mainly at one point in time from job title and only in men, has generally shown close to null associations.31, 32, 33 That risk of NHL is particularly increased in people with a history of skin cancers1, 2, 34 argues against a protective effective of sun exposure, although risk of NHL is also significantly increased after diagnosis of any cancer.35 There is evidence too for a positive relationship between self-reported skin cancer risk and risk of NHL in the NSW5 and SCALE studies.6 Such an association, however, need not imply that sun exposure is causal for NHL. Other shared environmental or genetic risk factors might explain associations between skin cancer and NHL. Persistent suppressor T cell activity or other immunological change initiated by the first tumour have also been suggested as a possible mechanism for increasing risk of the second.34

While there was evidence of a protective effect of sun exposure against both B cell and T cell lymphomas, it was more consistent for B cell lymphomas. The less consistent association with T cell lymphomas might be explained by the weak positive association seen between sun exposure and cutaneous T cell lymphomas; in addition, the small number of cases has limited the statistical power to examine this relationship. That sunlight might increase risk of cutaneous lymphomas has been suggested by a high incidence of cutaneous lymphomas in Israel, for which the sunny environment is a possible explanation, although the lack of variation in risk by continent of birth argued against sun exposure as the major initiating event.36 A role for UV in causing cutaneous T cell lymphomas has also been suggested by the presence in them of the same, UV-specific mutations in the P53 gene as are observed in non-melanocytic skin cancers, for which sun exposure is the main cause.37

Could uncontrolled confounding explain the apparently protective association between recreational sun exposure and NHL? The individual study results were adjusted for potential confounding of sun exposure with ethnic origin, pigmentary characteristics and sun sensitivity, and we also examined whether socioeconomic status, tobacco and alcohol intake were confounding but found no evidence for it. Only limited attempts, however, could be made in the pooled analysis to rule out confounding as a possible explanation for the apparently protective effects in individual studies. The SCALE study results, however, were adjusted for occupational exposure to pesticides, smoking, body mass and history of autoimmune disorders,6 which have been reported to be associated with NHL. SES was reported to have no appreciable influence in the NSW study5 and education, alcohol and weekly exercise to have none in the NCI−SEER study.7 Of other potential confounders, farming exposure is probably not common enough to explain a broad effect of sun exposure, particularly recreational exposure, and hormone use and reproductive variables in women would only be plausible if the effect was particular to women, which it was not in this analysis. Potential confounders such as diet, exercise and energy balance would justify consideration as, perhaps, would indicators of infection in childhood and atopy.

We examined the components of total sun exposure and found statistical evidence for an inverse association between NHL and recreational sun exposure; the effect was strongest at 18–40 years and in the 10 years before diagnosis. These results for recreational sun exposure could be taken to suggest that if sun exposure does protect against NHL it is an intermittent pattern of sun exposure that is the most protective. Although we looked at whether latitude modified the effect, we could not make firm conclusions because of heterogeneity in the data. The effect of recreational exposure and its strong effect near to time of diagnosis suggest vitamin D production as a mechanism by which sun exposure might protect against NHL. The main source of vitamin D is cutaneous synthesis of the pro-vitamin under the influence of sun exposure. Its synthesis, however, is tightly regulated38 and net production would probably be greater with a more intermittent pattern of exposure, as in recreational exposure, than a more continuous pattern.5 Vitamin D has been shown to have anti-proliferative and pro-differentiating effects in lymphoma cell lines, although at above physiological concentrations,39 and there is some evidence that vitamin D can induce regression of low-grade NHL.40 These relatively late stage effects would be more likely than early stage effects to produce an association of sun exposure with NHL near to the time of diagnosis. One recent study found a significant 40% reduction in NHL risk in the highest third of dietary intake of vitamin D41 but the NCI-SEER study found no evidence of such an association7; another study found an inverse association with increasing levels of predicted serum 25-hydroxyvitamin D concentration.42 Analysis of the association between serum vitamin D and NHL in 1 or more of the large cohort studies will probably provide a more certain answer. Meanwhile, a pooled analysis of the interaction between vitamin D receptor polymorphisms and sun exposure in InterLymph may contribute additional evidence and perhaps encourage future conduct of vitamin D supplementation trials.

Acknowledgements

The project was conceived and developed by the InterLymph collaborators; a protocol developed for an early InterLymph pooled analysis was made available by Dr. Lindsay Morton who assisted with interpretation of it. The funding application for support of the pooled analysis was prepared by Dr. A.M. Hughes, Dr. A. Kricker, Prof. B.K. Armstrong. The co-chairs of the InterLymph consortium at the inception of this project were Dr. P. Boffetta, Dr. M. Linet, Prof. B. Armstrong. Investigators who carried out each study and provided data from the study are: NSW (Dr. A. Kricker, Prof. B.K. Armstrong, Dr. A.M. Hughes, Mr. C. Goumas); NCI-SEER (Dr. P. Hartge); Mayo Clinic (Dr. J.R. Cerhan); Connecticut (Prof. T. Zheng, Dr. Y. Zhang); British Columbia: (Dr. J.J. Spinelli); Nebraska (Dr. B. Chiu); SCALE (Dr. K.E. Smedby, Prof. M. Melbye); United Kingdom (Prof. E. Roman, Dr. E.V. Willett); EpiLymph-Spain (Dr. S. de Sanjose); EpiLymph-France (Prof. M. Maynadié); EpiLymph-Italy (Dr. P.L. Cocco); EpiLymph-Ireland (Dr. A. Staines); EpiLymph-Germany (Prof. N. Becker, Dr. A. Nieters); EpiLymph is coordinated by Dr. P. Boffeta and Dr. P. Brennan.

The sponsors of the pooled analysis and the funding sources for the case-control studies had no role in the study design, data collection and analysis, interpretation of the results, or the preparation of the manuscript.