Lifetime Ambient UV Radiation Exposure and Risk of Basal Cell Carcinoma by Anatomic Site in a Nationwide U.S. Cohort, 1983–2005

Basal cell carcinoma (BCC) of the skin is the most common malignancy in populations of European ancestry, with estimates of over 1 million people diagnosed with BCC in the United States each year (1, 2). Rates of BCC have substantially increased worldwide over the last 30 years (3). The total economic costs of diagnosis and treatment in the United States exceeded $8.1 billion in 2011 (4) and costs are quite rapidly increasing in a number of countries (5, 6). The relationship of BCC with UVR exposure is suspected to vary by anatomic site, which could have important implications for recommendations aimed at primary prevention. On the basis of clinical reports and small and limited case–control studies, the majority of BCCs are thought to occur on typically sun exposed areas such as the head and neck (3, 7) supporting the important causative role of solar UV radiation (UVR; ref. 8). Other risk factors include occupation, male sex, older age, lighter skin, blue/green eye color, and lighter or red hair color, photosensitizing agents, indoor tanning, ionizing radiation, obesity, and certain medical conditions or medications, (3, 7, 9–14) in particular some leading to immune suppression (3, 15).

There has been very little assessment of ambient UVR or other risk factors for BCC by anatomic site. Evaluations have been limited to case series (16, 17), case–control studies in western Australia (18) and Italy (19), and a relatively small skin cancer screening study in the town of Nambour in Queensland, Australia (20). The findings have been inconsistent. Although there have been comprehensive case–control studies assessing UVR and other postulated risk factors in specific states in the United States [reviewed by Gossai and colleagues (3)] and in cohorts of health professionals from (at least) 11 populous U.S. states (21), limited information is available on nationwide occurrence and risk factor relationships for BCC and none have prospectively evaluated cumulative ambient UVR in relation to risk by anatomic site in the United States, in part because information on BCC is not systematically collected by U.S. population-based cancer registries (21, 22). Compared with case–control studies, cohort studies minimize recall bias (although residential location would be expected to be recalled with similar accuracy in both types of study), and enable estimation of absolute risks, which are important measures of public health impact. Data collected nationwide in the United States from a wide range of latitudes provides a large range of cumulative UVR exposures, thereby increasing the statistical power to detect UVR-associated risks relative to a more regionally defined study.

The objective of the current study is to assess the relationship between estimated ambient lifetime UVR with incidence of BCC by anatomic site using data from the U.S. Radiologic Technologists (USRT), a large nationwide prospective cohort, which has collected detailed information on anatomic sites of skin cancers in a population exposed to a wide range of lifetime ambient UVR. Features of this investigation include a primary emphasis on absolute risks and assessment of potential effect modification by constitutional and selected lifestyle factors.

The USRT cohort consists of 146,022 radiologic technologists living in the United States and its territories who were certified by the American Registry of Radiologic Technologists for at least 2 years between 1926 and 1982 and followed using yearly registry recertification. For cohort members not recertified, vital status has been obtained through linkage with tracing databases and the Social Security Administration's “presumed living” search. Linkage of those known or presumed deceased with the National Death Index was conducted to confirm that they were deceased. Methods of the study have been described previously (10, 23–25) and detailed information can be found online at www.radtechstudy.nci.nih.gov. Three surveys were undertaken between 1983 and 2005. For the first survey, questionnaires were sent during 1983 to 1989 to 132,298 surviving cohort members with a postal address; 90,305 (68%) technologists completed the questionnaire. A second questionnaire was sent to 126,628 living and located members during 1994 to 1998; 90,972 (72%) technologists responded. If a technologist answered both the first and second questionnaire, the earlier of the two was designated as the baseline questionnaire. A third questionnaire was sent in 2003 to 2005 to 105,694 participants who had answered at least one previous questionnaire and were still known living; of these, 73,838 (73%) responded.

In the current study, we evaluated participants who completed a baseline (first or second or both) questionnaire, the third questionnaire, did not report history of cancer (including BCC) prior to their baseline questionnaire, and who had sufficiently complete residential history that they could be linked to a database of UVR irradiance estimates by calendar time period and location in the continental United States. Those not completing the third questionnaire, in particular those dying before this point, were not included in the analysis. There were 63,912 White participants, both Hispanic and non-Hispanic, that met these criteria. Non-White respondents were excluded from detailed analyses due to the notably lower BCC incidence rates compared with Whites (3), the likely lower UVR sensitivity to skin cancer (26), and the relatively small number of these participants; however, summary results are presented in Table 1. Cohort members were followed from baseline questionnaire completion to the earlier of first primary cancer diagnosis, including BCC, or the completion of the third questionnaire (administrative censoring).

Cases included self-reported incident first primary BCC (ICD10 C44.91) among subjects who responded to the third questionnaire, and who had previously completed the first or second questionnaire (or both) as described in more detail in an earlier report (10). Follow-up stopped at the first BCC reported by a technologist, and all subsequent BCCs were ignored. Efforts to validate self-reported BCC cases, conducted for 1,355 BCCs self-reported by USRT cohort members in the second questionnaire resulted in retrieval of 50% of medical records sought among which 97% (668 BCCs) of self-reported BCCs were confirmed (27). Medical records were sought for the 4,862 BCCs self-reported on the third questionnaire and obtained for 2,058 (42.3%) of these cohort members; of these, 1,762 (85.6%) self-reports were confirmed as a BCC. Because of the high medical record confirmation of cases for which medical records were available, we include all self-reported BCC cases in this analysis. Information on behavior, in particular, time spent outdoors was also captured on the third questionnaire. However, we judge that the possibility of biased recall of activity is substantial, so that we did not use this information. There is no anatomic site information captured on the fourth questionnaire, so that these data cannot be used.

A skin map anatomic diagram was sent to 5,990 USRT cohort members (5,927 of them White) reporting BCC and other skin cancers on the third questionnaire. Cohort members were asked to indicate the anatomic site of the skin cancer on the diagram, the type of skin cancer, and the year of diagnosis. We instructed cohort members to report only skin cancers diagnosed by physicians, and not to report “pre-cancerous” skin lesions or lesions that were treated but never confirmed to be skin cancer. Each mark on the anatomic diagram was coded to anatomic location and date of diagnosis. For purposes of the current article, cases include those self-reported BCCs with recorded anatomic location. Following Scotto and colleagues (28) we grouped the BCC into four main anatomic groupings, reflecting the likely degree of UVR exposure, namely face/head/neck + supraclavicular anatomic region (henceforth referred to as the head and neck), upper extremities, trunk, and lower extremities.

We used total UVR (= UVA+UVB) as a measure of ambient solar ultraviolet exposure. A standard UVR exposure metric is cumulative UVR radiant exposure [in units of Joule (J) per square meter (J m^–2)], which is proportional to cumulative solar UVR energy deposition on a surface over a period of time. For concision we shall refer to cumulative UVR exposure rather than cumulative UVR radiant exposure from now on. This measure of cumulative ambient UVR exposure is recommended by the Commission Internationale de l'Eclairage (CIE; International Commission on Illumination; ref. 29). We used AVerage daily total GLObal (AVGLO) ground-based solar radiation measurements which are described in more detail elsewhere (30, 31) and in the Supplementary Materials and Methods. On the third questionnaire, residential location for five age periods (age <13, 13–19, 20–39, 40–64, 65+) was collected. Individual age-specific ambient UVR levels were determined by linking the residential locations with the AVGLO database (31).

The USRT has collected extensive baseline information on personal constitutional (skin, eye, and hair color), demographic (gender, age at diagnosis, racial/ethnic group), lifestyle [body mass index (BMI), smoking, and alcohol use], reproductive (age at menarche, age at menopause, use of oral contraceptives, menopausal hormone therapy), medical conditions and treatments, particularly disorders and medications associated with immune dysfunction or immune suppression (organ transplant, autoimmune diseases, immunosuppressive drugs), environmental exposures (UVR, occupational ionizing radiation), and genetic factors, which have been previously associated with increased risk of BCC (3). We assessed the effect of adjusting for these potentially confounding variables on cumulative UVR and excess absolute risk of BCC. We also assessed the extent to which this broad group of known and postulated BCC risk factors modified UVR-associated BCC risk.

In the analysis of Table 1, assessment of heterogeneity was via an analog of the Fisher exact test based on 10⁶ Monte Carlo replicates. Missing values were coded as an extra level in each classification, but were otherwise treated like all the other levels for each classification. For each classifying variable, the full dataset was analyzed as well as the subset with information on the variable under investigation. To evaluate the relationship between UVR and BCC risk, we fitted Poisson excess relative risk and excess absolute risk models, in which the expected number of cases in the stratum with person years PY, after cumulative lagged UVR exposure, H(t) (in J cm^–2), at age t, and with various other explanatory covariates, |Z = ({Z_i})$|⁠, was given by:

for the relative risk model and

for the absolute risk model. The parameter α is the log-linear excess relative risk per unit of cumulative radiant exposure, and the parameter δ is the linear excess absolute risk per unit of cumulative radiant exposure and per year of follow-up. We provide risk estimates for both relative risk and absolute risk. Relative risk was selected as the primary risk measure because this metric is often used in epidemiologic studies of UVR and most other exposures. There are also some methodologic advantages since relative risk models were less prone to instability in model fits. Absolute risk was also used because these model estimates provide a better measure of the lifetime impact for an individual or a population of UVR exposure. In addition, absolute risk provides quantitative estimates more useful for clinical and public health purposes. Fits of linear Poisson relative risk models and Cox proportional hazards models were also attempted, but both proved very unstable, and prone to problems of non-convergence. For the purposes of the current main analyses, a loglinear model of UVR cumulative radiant exposure was used for relative risk, and a linear model of cumulative radiant exposure used for absolute risk. For most purposes cumulative UVR exposures were lagged by 5 years, by analogy with the latency often assumed for solid cancers after ionizing radiation exposure (32), but we conducted sensitivity analyses using lag periods of 2 to 15 years. Because the risk factors being analyzed were only determined from the second questionnaire, for certain analyses in relation to possible modifying risk factors (as indicated by footnotes to the relevant tables) follow-up was restricted to the point of response to the second questionnaire onward, whether or not the subject also answered the first questionnaire. Further details are given in Supplementary Materials and Methods. Model fitting was performed using R (33) and Epicure (34). Confidence intervals (CI) were estimated from the profile likelihood (35), or if this did not converge, using Wald-based CIs.

The research protocol for the USRT cohort study of cancer risks has been approved annually by the NCI Special Studies Institution Review Board (SSIRB Protocol OH97-C-N053) and the University of Minnesota Human Research Protection Program Institution Review Board (Federal Wide Assurance number 8005M02489). Completion of any questionnaire was considered implied consent for study participation; a waiver of consent was approved annually by both institutions due to the minimal risks of the study and the large size of the study population.

Among the 63,912 White USRT cohort members with information on cumulative UVR irradiance and no prior cancer diagnosis at baseline, 2,124 cases of first BCC were reported over the follow-up period (Table 1). Further details on numbers of skin map questionnaires sent and received are given in Supplementary Table S1. We present summary results for non-Whites in Table 1, which demonstrate the much lower BCC risk for these (5/53,200 = 9.4 × 10^–5 per person per year) compared with Whites (2151/1,082,740 = 2.0 × 10⁻³ per person per year). A flow diagram detailing exclusions is given in Supplementary Fig. S1. Supplementary Table S2 gives a more detailed listing of BCC by anatomic site and sex. Most of the cohort members without BCC were female, born after 1950, had college or graduate school education, did not smoke and drank once or twice a week. Similar distributions were generally seen among cohort members with BCC overall and by anatomic site (Table 1). However, BCC cases tended to be born earlier than non-cases, were less likely to complete higher educational levels (college or graduate school), were somewhat less likely to be a current smoker, were slightly more likely to consume 3+ drinks per week, were more likely to have fair complexion and blue/green/gray eyes and have red/blonde hair, were more likely to have Gaelic ancestry, less likely to be Hispanic, not to be overweight, less likely to use oral contraception, were more likely to have hormone replacement therapy for 10 or more years, were more likely to use diuretics, were more likely to exercise strenuously, were more likely to have siblings with cancer and were slightly more likely to have higher occupational radiation dose. All these differences were mostly highly statistically significant (P < 0.005), irrespective of whether the full dataset was analyzed or just those with information on each classifying variable (Table 1).

Table 2 demonstrates that BCC risks normalized per unit skin area were much higher on the face and head, by about a factor of 4 compared with the neck and supraclavicular region. (Supplementary Table S3 details the stratifying variables used to generate the person-year tables.) These two anatomic sites have much higher risk per unit skin area than the upper extremities, trunk and lower extremities. Although the two sexes had generally similar patterns of variation of risk per unit area, lower extremity risk per unit area in females was 5-fold higher than that for males (Table 2).

Table 3 and Fig. 1 demonstrate the increasing excess relative risk of BCC at all anatomic sites combined associated with increasing cumulative lifetime UVR exposure. Cumulative (lifetime) UVR exposure of 1 MJ cm^–2 for both sexes combined was associated with an overall increase in the log relative risk of BCC of 1.27 (95% CI, 0.86–1.68; P < 0.001). Table 3 also shows that there was no significant difference between the relative risks by sex overall (P = 0.153) or for any specific anatomic site. Excess relative risk of BCC in relation to UVR was much the same on the head and neck as on the trunk (Table 3). The variation in excess relative risk by anatomic site was nonsignificant (P > 0.3) for males, females, or both sexes together, whether using the two-class analysis of heterogeneity (the contrast between head and neck and all other anatomic sites), or the four-class analysis of heterogeneity (the heterogeneity between head and neck vs. upper extremity vs. trunk vs. lower extremity).

Supplementary Table S4 demonstrates that adjusting for any of the potentially confounding lifestyle or constitutional variables had little effect on the risk, although there was a reduction by about 6% in risk when adjusting for BMI (adjusted log ERR per MJ/cm² = 1.14 vs. 1.21 unadjusted). Even when adjusting for all variables simultaneously in the two separate variable groups (e.g., endogenous vs. lifestyle/reproductive/medical factors), the ERRs for BCC associated with cumulative UVR exposure remain significant.

Table 4 evaluates the potential for modification of the association between cumulative UVR and risk of BCC according to constitutional, lifestyle, and other risk factors. As shown, there was significant heterogeneity of log ERR per unit exposure for skin complexion for the upper extremity (P = 0.001) and trunk (P = 0.043). There was borderline significant (P = 0.056) heterogeneity in log ERR by eye color for upper extremity BCC; log ERR for upper extremity BCC was particularly high for those with dark or unknown complexion, with overall increases in the log ERR of BCC per MJ/cm² of 5.63 (95% CI, −2.28 to 11.11) and 50.10 (95% CI, 9.20 to >100) respectively (Table 4). There was also significant heterogeneity of ERR per unit exposure by Gaelic ancestry for all sites (P < 0.001), and for head and neck BCC (P = 0.003; Table 4); log ERR was particularly low for those with Gaelic ancestry, with overall increase in the log ERR per MJ/cm² of BCC of 0.64 (95% CI, 0.04–1.23; Table 4). There were significant (P = 0.014) variations in log ERR for upper extremity BCC in relation to use of diuretics, with those using diuretics at particularly high risk, having log ERR per MJ/cm² of 3.26 (95% CI, 0.66–5.71; Table 4). There were borderline significant (P = 0.059) variations in log ERR for upper and lower extremity and trunk BCC in relation to age at menopause, with those with menopause under the age of 25 at highest risk. ERRs were borderline significant by level of occupational radiation dose overall (P = 0.088) and for head and neck BCC (P = 0.070), with log ERR per unit cumulative radiant exposure tending to be higher for higher occupational radiation doses (Table 4). Otherwise, there does not appear to be any significant modification of log ERR of BCC risk per unit UVR cumulative radiant exposure with any lifestyle, environmental or constitutional variable, either overall or for particular anatomic locations (Table 4).

Cumulative (lifetime) UVR exposure of 1 MJ cm^–2 was associated with an overall increase in the absolute risk of BCC of 9.62 (95% CI, 6.94–11.73; P < 0.001) per 10,000 persons per year (Supplementary Table S5). Supplementary Table S5 also shows that there was no significant difference between the absolute risks by sex overall (P = 0.105) or for any specific anatomic site. The variation in absolute risks by anatomic site was much more pronounced than the variation of relative risk, as shown by Fig. 2 and Supplementary Table S5. Excess absolute risk of BCC in relation to UVR was notably higher on the head and neck than on the trunk (Supplementary Table S5). The variation in excess absolute risk by anatomic site was highly significant for females (P = 0.017 for two-class analysis of heterogeneity, P = 0.006 for four-class analysis of heterogeneity) and borderline significant for males (P = 0.099 for two-class analysis of heterogeneity, P = 0.071 for four-class analysis of heterogeneity) but the difference was not significant when averaged over the two sexes (P > 0.1). Supplementary Table S4 demonstrates that adjusting for any of the potentially confounding lifestyles or constitutional variables had generally little effect on the absolute risk, although there was a notable reduction by about 30% in risk when adjusting for complexion [adjusted excess absolute risk (EAR) per MJ/cm² per 10,000 persons per year = 10.63 vs. 15.23 unadjusted]. When adjusting for all variables simultaneously in the two separate variable groups, the ERRs remain significant using all the larger group of environmental and lifestyle risk factors (P = 0.004), but significance was lost when adjusting using all of the smaller group of familial variables (P = 0.138).

Supplementary Table S6 evaluates the potential for modification of the association between cumulative UVR and risk of BCC according to constitutional, lifestyle, and other risk factors. As shown, there is significant heterogeneity of absolute risk per unit exposure for Gaelic ancestry for all sites and among head and neck BCC (P < 0.001), with EAR per MJ/cm² among those with Gaelic ancestry of 30.05 (95% CI, 20.43–37.92) per 10,000 persons per year. There is also borderline significant (P = 0.035) heterogeneity by skin complexion for all sites, with risk particularly high for those with fair complexion, with overall EAR per MJ/cm² of 19.21 (95% CI, 8.50–29.93) per 10,000 persons per year (Supplementary Table S6). ERRs were significant or borderline significant by level of occupational radiation dose overall (P = 0.014), for head and neck BCC (P = 0.008), for upper extremity BCC (P = 0.050) and for upper+lower extremity and trunk BCC (P = 0.060), with EAR per unit cumulative radiant exposure tending to be higher among those receiving intermediate (0.05–0.1 Gy) doses (Supplementary Table S6). There were significant or borderline significant variations in EAR per unit UVR cumulative radiant exposure for upper+lower extremity and trunk BCC by smoking status (P = 0.011) or by exercise status (P = 0.065), with EAR tending to be higher for those exercising more, and not current smokers; the same pattern in relation to exercise status was also observed for upper extremity BCC (P = 0.024). A complication with most of these findings was non-convergence of the fitted models. Otherwise there was little variation of EAR either overall or for particular anatomic locations (Supplementary Table S6).

In general, increasing the lag period resulted in increasing ERR and EAR, so that with a latency of 2 years the log ERR per MJ/cm² was 1.21 (95% CI, 0.82–1.60; P < 0.001), which increased to 1.51 (95% CI, 0.99–2.02; P < 0.001) when the latency increased to 15 years, and likewise the EAR per MJ/cm² per 10,000 persons per year was 8.68 (95% CI, 6.19–10.63; P < 0.001), which increased to 14.70 (95% CI, 11.00–17.50; P < 0.001) when the latency increased to 15 years (Supplementary Table S7). Supplementary Table S8 shows that there was little evidence (P > 0.15) of variation of EAR per MJ/cm² cumulative radiant exposure per person year of follow-up with time after exposure, whether using the finer (four time window) or the broader (three time window) breakdown; a complication in interpreting these findings was non-convergence of the fitted models. However, there was evidence at borderline levels of statistical significance of variation of log ERR per MJ/cm² cumulative radiant exposure with time after exposure, whether using the finer (four time window) breakdown (P = 0.026) or the broader (three time window) (P = 0.043) breakdown; log ERR per MJ/cm² and the EAR per MJ/cm² appeared to be largest for the period 5 to 14 years after exposure, with smaller risks outside this time since exposure window (Supplementary Table S8).

When competing risk models were used there was little change in log ERR overall, which became 1.25 (95% CI, 0.84–1.66; P < 0.001) per MJ/cm² from 1.27 per MJ/cm² using the standard model, but there was considerable diminution of EAR overall, so that the EAR per MJ/cm² per 10,000 persons per year changed to 2.58 (95% CI, −2.42 to 5.97; P = 0.182) from 9.62 per MJ/cm² using the standard model (Table 3; Supplementary Tables S5 and S9).

In this U.S. study, one of the first large, nationwide, prospective cohort investigations to evaluate ambient cumulative UVR exposure and BCC risk by anatomic site, we observed that estimated excess relative and absolute risk of BCC at all anatomic sites combined rose with increasing estimated cumulative lifetime ambient UVR exposure. We used a high-quality lifetime estimated ambient UVR exposure assessment based on residence lived in longest during each of five age periods linked with AVGLO environmental measurements (31, 36). Relative risks varied little by anatomic site (Table 3), whereas there was more pronounced variation by site in absolute risks, with the highest excess absolute risks per unit of UVR cumulative radiant exposure observed for BCC on the head and neck while lower excess absolute risks per unit of UVR exposure were seen on the trunk. Excess absolute risks of BCC on the upper extremities rose modestly with cumulative UVR, but there was no excess absolute risk of BCC of the lower extremities with UVR (Supplementary Table S5). We have also demonstrated modification of the cumulative ambient UVR-BCC risks with lower log ERR, but higher EAR among those with Gaelic/Celtic ancestry, and higher EAR among those with fair complexion but in general no overall significant and convincing difference between the sexes or modification by other constitutional, lifestyle, reproductive or medical factors, although for particular anatomic sites there was some observed heterogeneity (Table 4; Supplementary Table S6). The fact that the excess relative risk for those with Gaelic ancestry is lower (Table 4) but the excess absolute risk is higher (Supplementary Table S6) is consistent with the fact that the underlying risk of BCC is higher in the group with Gaelic ancestry, implying a submultiplicative interaction between the effect of ancestry and UVR-associated risk. This is not infrequently observed between the effects of some carcinogen and the baseline (e.g., genetic) risk factors (37–39).

Few previous studies have evaluated UVR and BCC by anatomic site, and none have involved large prospective cohort investigations examining dose–response employing cumulative estimates of exposure based on UVR measurements among residents from a large range of latitudes with broad assessment of potential confounders and effect modifiers. In a clinical series of 202 patients identified from skin cancer clinics in Queensland, Australia, Khalesi and colleagues (16) found that risk of BCC on solar-shielded skin (compared with solar-exposed skin) increased with increasing time spent outdoors and with numbers of actinic keratoses, but the correlation disappeared after adjusting for confounders (16). However, the study did not estimate excess absolute risks, and did not report potential effect modification by constitutional or lifestyle risk factors. Among 254 patients with histologically confirmed BCC ascertained from dermatology practices in Puerto Rico nonmelanoma skin cancers in sun-exposed areas were more common in participants who received high and very high UVR doses (17). The study design limitations of these two clinical series, both conducted in geographic regions with very high solar exposure precludes comparison with our large prospective cohort study at a broad range of latitudes. A population-based case–control study of 192 BCC and 700 controls (all identified in a population-based survey) from Western Australia found that risk of BCC on the head and neck and on the limbs declined whereas risk of BCC on the trunk rose with increasing total (cumulative occupational and recreational) sun exposure based on lifetime residential history and time outdoors on workdays and non-work days during warmer and cooler months (18). A study of 528 hospital-based BCC cases and 512 hospital controls with acute conditions from 16 centers in northern and central Italy found associations of recreational sun exposure and risk of both head/neck and trunk BCC and of occupational sun exposure with head/neck but not truncal BCC, although long duration of exposure was not linked with higher risk than short duration for any of these associations (40). Comparisons of the results of the two case–control studies with our prospective study are difficult due to study design differences, the differences in measurements (questionnaires about time spent outdoors vs. ground-based measurements), the emphasis on personal (case–control studies) versus ambient exposures (our cohort study), and metrics used (relative risk only vs. relative and excess absolute risk per unit UVR) and make comparison of results difficult. A small skin cancer cohort study in the town of Nambour, Queensland, Australia (22) assessed detailed anatomic site of 177 first BCC (598 total BCC) but had only a relatively crude 3-level occupational coding of likely UVR exposure (mainly outdoors, partly indoors and outdoors, mostly indoors).

The notably higher percentage of BCCs occurring on the face and head is well documented (41), but no studies have analyzed cumulative ambient UVR and risk of BCC by anatomic location. Our findings revealed that UVR-associated BCCs are also linked with increasing estimated cumulative ambient UVR on the trunk and, to a modest extent on the upper extremities. The trunk and upper extremities likely receive less exposure than the head and neck, possibly due to behavioral factors and clothing, but data on these factors were not collected. The findings may also be the result of some sort of abscopal effect, that is to say an effect in a part of the body distant to that exposed, something that has been suggested in a Ptch1^+/− mouse model in relation to ionizing radiation exposure (42), and as also suggested by the UVA- and UVB-associated mutagenic bystander effects in vitro (43, 44). There are known involvements of UVR exposure on systemic immune function (45, 46), which could conceivably modify BCC risk on solar-unexposed parts of the body. These should be priority areas for future work in this area.

The metric used here, cumulative radiant UVR exposure, although not commonly used in epidemiologic studies, incorporates two important determinants of UVR, namely, latitude and age. As outlined previously, there are strong biophysical reasons why it should be employed (31, 36). Using data taken from that employed in the article of Little and colleagues (31), very similar to that used here, it can be estimated that at ages 65 and 80 an increase of 1 degree of latitude (mean taken over the ages 0–65, 0–80, respectively) corresponds to a reduction of cumulative radiant exposure of about 0.0240 and 0.0298 MJ cm^–2, respectively. These are aggregate figures over the whole dataset; the precise degree of reduction for an individual will depend on their latitude and other factors.

The current study has several strengths including a large nationwide U.S. cohort of both sexes (80% female), residing in a wide range of latitudes, with long term follow-up and lifetime residential history. Use of ground-based AVGLO measurements of ambient UVR is a significant strength, combining the advantages of fine (1 km² grid) geographical resolution, full-day average of solar exposure taking account of cloud cover, non-erythemally weighted measured irradiances, and ongoing data collection from 1960 to the present (36). In addition, our study provides the anatomic site of nearly two-thirds of the self-reported BCC, a large proportion of the BCCs were validated against medical records, and quantitative measurements for ambient UVR. The USRT cohort also had detailed information about constitutional, lifestyle, reproductive, and medical factors, as well as occupational information on ionizing radiation exposure that enabled us to evaluate potential effect modification across a wide variety of anatomic locations, and also assess latency and variations of EAR and ERR with time since exposure (Supplementary Tables S7 and S8), somewhat paralleling previous analysis (36).

Limitations of our study include self-report of anatomic location, possible mis-report of BCC not validated by medical records, restriction of the study population to White, indoor workers, and ambient UVR being based on self-reported residential history locations for five age periods, with no assessment of intermittent UVR exposures (e.g., during vacations) or other residential location not recorded in these five age periods. We also did not collect baseline information about number of hours spent outdoors or the clothing worn when subjects were outdoors, both of which are likely to correlate with latitude, or the use of tanning sunbeds. We do not have data on certain medical conditions and treatments, particularly disorders and medications associated with immune dysfunction or immune suppression (organ transplant, immunosuppressive drugs) and genetic factors (other than Gaelic ancestry or cancer in siblings at young age) that have been associated with increased risk of BCC (3). However, there is no reason to suppose that the prevalence of these factors and conditions would correlate with UVR exposure and so would not be expected to appreciably confound any trends of BCC with UVR. As noted in the Results, there are problems with lack of convergence, which particularly affect the absolute risk models (Supplementary Tables S5, S6, and S8). Only persons who survived and answer one of the first two questionnaires and the third questionnaire are included in this analysis. A degree of selection is thereby introduced. This would not be expected to bias log ERR, as risks are estimated in relation to a nominal baseline group, which would be equally affected by the selection, although downward bias would be expected for the measure of EAR, as excess cases would be removed by the selection.

We have demonstrated that BCC varies with increasing estimated lifetime ambient UVR exposure strongly for the head and neck, and to a somewhat lesser extent for the trunk, and modestly for the upper extremities. Our findings, of higher UVR-associated absolute excess risk associated with exposure of solar-exposed parts of the body such as the head and neck, and also of sites that might be presumed less solar exposed such as the trunk (Supplementary Table S5) parallel those suggested in other studies, albeit ones without the more comprehensive lifetime ambient UVR measures that we have (22). Our findings underscore the importance of skin cancer prevention guidelines to focus on the development of long-term sun protection habits, especially for the head and neck.

No disclosures were reported.

M.P. Little: Conceptualization, data curation, software, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing. T. Lee: Formal analysis, methodology, writing–review and editing. M.G. Kimlin: Conceptualization, visualization, methodology, writing–review and editing. C.M. Kitahara: Investigation, methodology, project administration, writing–review and editing. R. Zhang: Formal analysis, validation. B.H. Alexander: Resources, investigation, methodology, writing–review and editing. M.S. Linet: Resources, supervision, funding acquisition, methodology, project administration, writing–review and editing. E.K. Cahoon: Conceptualization, formal analysis, supervision, validation, visualization, methodology, writing–review and editing.

The authors would like to thank the two referees and the associate editor for their detailed and helpful remarks. The authors thank Elizabeth Platz and Paul Strickland for their input in the design and review of early study results. We would like to thank the radiologic technologists who participated in the USRT Study, Jerry Reid of the ARRT for continued support, and Diane Kampa and Allison Iwan of the University of Minnesota for study management and data collection.

This work was supported by the Intramural Research Program of the NIH, NCI, Division of Cancer Epidemiology and Genetics.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.