The effects of sunlight exposure on mortality: a systematic review of epidemiological studies

Patient and public involvement

We formed a public advisory group comprising four individuals with different skin types (three with Afro-Caribbean ancestry and one with very pale skin). We obtained input into the protocol from this group, although their ability to influence it was limited since the project was commissioned. They explained that people with darker skin can be very concerned about the harms of sunlight exposure as well as the possibility of vitamin D deficiency, reinforcing the need to examine data by subgroups defined by skin colour and type. They also made it clear that more refined guidance would be useful for people with all types of skin. They strongly supported the project and advised us that the findings would be of great interest. They further advised us on how we might disseminate the findings to those responsible for national guidelines on sun exposure; they felt that their health providers (e.g. general practitioners) were not sufficiently informed to provide targeted advice. Members of the public helped us write a lay summary of our plans for our website in advance of the project start (https://bristol-esg.org/projects-2/sunlight-exposure/) and a member of the public advisory group helped us write a plain language summary to disseminate the results.

Eligibility criteria and search methods

Criteria for inclusion of articles in the review were: (i) reporting a primary epidemiological study with a cohort design (including randomized trials of interventions to alter relevant behavioural exposures), a case-control design or an ecological design; (ii) in a general population; (iii) using any measure(s) of long-term sunlight exposure; and (iv) reporting outcome data on all-cause mortality, cancer-specific mortality or cardiovascular disease (CVD) mortality. To perform a comprehensive assessment, we used the exposure term ‘sunlight’ broadly to include measures of radiation (such as sunlight, UVR and UVB); proxies for radiation measures (such as latitude and geographical location); and behaviours associated with sun exposure (such as episodes of sunburn and recreational sun exposure). We excluded: (i) studies of artificial sources of UVR exposure, such as sunbed use and prescribed phototherapy and (ii) studies restricted to people with pre-existing disease (e.g. reporting survival rates in people diagnosed with melanoma).

We searched MEDLINE, Embase, Web of Science and the Cochrane Central Register of Controlled Trials to November 2023 using relevant controlled vocabularies, text‐words and search syntax appropriate to each resource. See Appendix S1 in the extended data²¹ for the full search strategy. Additionally, we performed forwards and backwards citation searches of included articles identified in the search and scanned the reference lists of relevant systematic reviews. We also checked for any relevant retraction statements or errata of included studies. Two reviewers independently assessed all reports for eligibility.

Data extraction and analysis

Data were extracted using standardized data extraction forms developed in Microsoft Excel, which had first been piloted on a small sample of articles and adapted as necessary. We collected the following data: study design (nested case-control, case-control, cohort, ecological, trial, quasi-experimental), funding sources (public, industry, mixed), study location, sex, age, ethnicity/race, skin type/colour, occupation, inclusion criteria, method/definition of sunlight exposure, period of exposure (childhood, adolescence, adulthood), length of exposure (specific, lifetime) and target condition (all-cause, cancer, CVD). We extracted summary data relating to the association between sun exposure and mortality overall and for CVD and cancer-related causes. This was reported differently across the studies included in this review, and included odds ratios (OR), risk ratios (RR), hazard ratios (HR), regression coefficients, correlation coefficients and narrative reporting.

Our main outcomes were death from any cause (all-cause mortality), death due to any cancer (all-cancer mortality) and death due to any cardiovascular cause²² (all-CVD mortality). We also examined death due to skin cancer (melanoma and NMSC) and the five most common causes of cancer-related mortality in the UK according to Cancer Research UK²³: breast, prostate, lung, bowel and pancreatic cancer; as well as specific causes of CVD.

We assessed risk of bias in results using the Risk Of Bias In Non-randomized Studies of Exposure (ROBINS-E) tool for observational studies²⁴. Risk-of-bias assessments were not carried out on articles that reported results as correlations, mortality rates or narratively because the ROBINS-E tool aims to assess risk of bias in estimates of causal effects. Two reviewers independently assessed all reports considered potentially relevant for inclusion, extracted descriptive data and results of included articles, and assessed the risk of bias in results. Any disagreements were resolved by consensus or discussion with a third reviewer.

Due to the diversity in the types and units of measurement of the sunlight exposure measures, we did not consider it appropriate to conduct meta-analyses. Instead, we produced a narrative synthesis of the findings for each mortality outcome and examined any comparisons across people of different skin type/colour or ethnicity.

Articles reporting results as a relative risk (e.g., RR, HR; or data sufficient to derive these) were grouped by type of exposure and presented in forest plots, with estimates inverted as appropriate to illustrate findings using consistent direction of effect. Where an article reported only separate risk estimates for males and females, we combined these using fixed-effect meta-analysis to provide a result that additionally controlled for sex/gender.

Where results were presented in forms other than relative risks for two or more studies, we listed these in tables. Where such results were reported only separately for males and females these were treated as one analysis result (to be consistent with the combined relative risk results), but were reported as separate results in the tables.

We address certainty in the evidence through consideration of each of the five GRADE domains in the discussion.

Search results and included studies

The database searches identified 8501 records, and 5181 records were identified using other methods. After examination of potentially relevant full text articles, 73 articles^25–97 met the criteria for inclusion in the review (Figure 1). For a comprehensive summary of the characteristics of these 73 articles, see Table S1 in the extended data²¹.

Figure 1. Identification of relevant studies.

Where multiple articles had used the same cohort, same exposure type and reported the same outcome, we classed these as ‘overlapping’. In these cases, we chose a main article as the source of data. The selection of data was based on an algorithm that included the type of analysis performed (ratio data preferred to linear regression or correlation coefficients), adjustment within analysis (most relevant confounders controlled for preferred), population (whole population preferred to specific subgroups), follow-up time (longer preferred), number of units of observation (e.g. counties preferred to states) and number of participants. As the majority of the included studies were ecological, we applied the same process to national datasets. We considered articles using national mortality data for the same country and date range to be using the same cohort, and used the same selection algorithm to select a main article in such cases. Table S2 in the extended data²¹ provides a review of all overlapping studies.

After the selection of main articles, this resulted in 55 articles being included in the narrative synthesis. All subsequent analysis is conducted on these 55 articles. Eight of the articles reported data on all-cause mortality, eight on all-CVD mortality, and 17 on all-cancer mortality. Eleven described cohort studies, three case-control studies and 41 ecological studies.

Twenty-four (44%) of the articles were conducted in the USA, five were conducted worldwide (including between 34 and 172 countries), four in China, and three each in Japan, Spain and Sweden. Other articles were conducted in the USA and Canada, the UK, and Switzerland (all n = 2); as well as Australia, Chile, Europe-wide, France, Italy, New Zealand and Turkey (all n = 1).

We provide descriptions of the exposures in Table S3 in the extended data²¹. The majority of the articles (n = 42, 76%) measured a single exposure; ten (18%) measured two exposures, two (4%) measured three exposures and one measured four exposures. Thus, we reported details of 72 exposures in total. Of those, 40 were measures of radiation, 17 were proxy measures of radiation and 15 were measures of behaviours associated with sunlight exposure. Radiation measures included solar radiation, UVR, UVA and UVB, DNA- and erythemal-weighted UVB and UVR, UV Index, solar incidence, insolation, irradiation, sunlight hours and sunshine. Among the proxy measures for radiation, 16 were latitude (e.g. residential at county or state level) and one was geographic location of deployment during war (tropical vs European). Measures of sunlight exposure-related behaviour included four of occupational exposure, four of recreational exposure and one of both occupational and recreational exposure combined; three used skin damage, two used NMSC mortality rates and one used melanoma mortality rates.

Of the 55 included main articles, 34 (62%) reported information on participants’ skin type/colour or ethnicity. Most were conducted in only White participants, whilst two only contributed data for Black participants. In six articles, the authors reported that skin type/colour or ethnicity was mixed. Of these, four included Black and White participants (although, where proportions were reported, the populations were predominately White). One included Black, White and Hispanic participants, and one included Black, White, Hispanic, Asian and Native American participants.

Below we provide our findings for our primary outcomes: all-cause, all-CVD and all-cancer mortality, both overall and by type of exposure. Subsequently, we provide a brief summary of the findings for our secondary, cause-specific mortality outcomes. We report the full risk-of-bias assessments, including justifications for judgments, in Table S4 in the extended data²¹.

All-cause mortality

Overview. In total, eight articles investigated the effect of sunlight on all-cause mortality. Six used a cohort design and two had an ecological design. One article reported findings for two exposures (nine analyses reported across the eight articles). The findings were mixed (see Figure 2 for reported relative risks). Five analysis results^52,77,89,97 were in the direction of a beneficial association between sunlight and all-cause mortality. Four analysis results^47,67,76,83 were in the direction of a harmful effect of sunlight.

Figure 2. Forest plot of studies showing the effect of exposure on all-cause mortality.

*Fixed-effects meta-analysis performed to combine gender subgroups. †Result inverted to reflect an increase in exposure. NR: ethnicity of population not reported.

Radiation. Three articles looked at the effect of radiation on all-cause mortality (Figure 2). The results of all three articles were considered to be at high risk of bias. Goggins et al.⁵² conducted an ecological study of the Hong Kong population showing that a 10 W/m² increase in solar radiation was associated with a 10% decrease in all-cause mortality (RR = 0.90, 95% confidence interval (CI) 0.85 to 0.94). Lin et al.⁷⁶ measured the residential address of cohort participants across six states and two metropolitan areas in the USA to determine average July erythemal UVR, defined as biological damage per square meter (BD/m²). This measure was subsequently split into quartiles. When comparing the second quartile with the first, there was no evidence of a difference in all-cause mortality (RR = 1.00, 95% CI 0.98 to 1.03; Figure 3). However, when comparing the third and fourth quartiles with the first there were 8% and 6% increases in all-cause mortality, respectively.

Figure 3. Forest plot of studies showing the effect of exposure on all-CVD mortality.

†Result inverted in order to reflect an increase in exposure. NR: ethnicity of population not reported.

In addition to the articles included in Figure 2, Fu and Wang⁴⁷ examined the relationship between the average hours of daily sunshine in China and all-cause mortality using regression. They found that an increase of 0.1 to 0.2 in the average daily sunshine duration rate (equivalent to 2.9 hours of additional daily sunshine) was associated with an increase in all-cause mortality rate the following year (β = 11.509, 95% CI 1.87 to 21.15; high risk of bias).

Proxy for radiation. Two articles looked at the effect of proxy radiation exposures on all-cause mortality (Figure 2). The results of both articles were considered to be at high risk of bias. Stevenson et al.⁸⁹ examined the effect of latitude in the UK. They found that more southerly latitudes were associated with lower all-cause mortality, when compared with residences 300km further north (HR = 0.94, 95% CI 0.92 to 0.96). Page et al.⁸³ examined a cohort of WWII veterans who were deployed to either Pacific or European battlefronts, arguing that those who were deployed to the Pacific area would have experienced higher levels of sun exposure than those in Europe. They observed a small increase in risk associated with being deployed in the Pacific compared with Europe, however the wide confidence intervals were compatible with both a benefit and harm (odds ratio = 1.03, 95% CI 0.89 to 1.19).

Behavioural. Three articles examined the effect of sunlight exposure behaviour (Figure 2). He et al.⁶⁷ looked at the effect of physician-assessed actinic skin damage. The findings suggested that greater sun exposure, as indicated by skin damage, is associated with an increase in the risk of mortality. When comparing those with severe skin damage with those who were assessed to have no damage, there was a 45% increase in risk (HR = 1.45, 95% CI 1.22 to 1.72; high risk of bias). The association does not appear to be driven by smoking (an important common cause of skin damage and CVD), which was controlled for in the analysis.

Lindqvist et al.⁷⁷ measured self-reported sun exposure in a cohort of Swedish women. The evidence suggested that there may be a decrease in mortality with increased sun exposure. Those who reported the highest level of sun seeking behaviour had a 38% decreased risk of dying from any cause compared with those who reported the lowest level (HR = 0.62, 95% CI 0.50 to 0.80; high risk of bias). This finding is supported by Yang et al.⁹⁷ in another cohort of Swedish women, though their results were considered to be at very high risk of bias. Those who reported spending one week or more annually on sunbathing vacations had a 30% decreased risk of mortality compared with those who never went on sunbathing vacations (HR = 0.70, 95% CI 0.60 to 0.90). They also reported an association between higher number of sunburns experienced and lower all-cause mortality, however the wide confidence intervals were compatible with both a benefit and harm (HR = 0.90, 95% CI 0.70 to 1.20).

All-CVD mortality

Overview. In total, eight articles investigated the effect of sunlight on overall CVD mortality, with one reporting findings for two exposure types (thirteen analyses reported across the eight articles). Six of the articles had a cohort design and two were ecological. There were mixed findings (Figure 3). Six analysis results^{26,52,78,89,97} suggested that higher levels of sunlight are associated with lower risk of CVD mortality. In contrast, five analyses^26,67 suggested that higher levels of sunlight are associated with a higher risk of CVD mortality. Two analyses^40,76 produced mixed findings.

Radiation. Three articles looked at the effect of radiation, all were considered to be at high risk of bias (Figure 3). Al-Hamdan et al.²⁶ performed a national ecological study in the USA. They found that a 100 kJ/m² increase in solar radiation was associated with a 1% higher risk of dying from CVD for White participants, a 3% higher risk for Black, Hispanic and Asian participants, but a 3% lower risk for Native American participants. Goggins et al.’s⁵² Hong Kong-based study found that an increase in solar radiation of 10 W/m² was associated with a lower risk of dying from CVD (RR = 0.87, 95% CI 0.78 to 0.97). In Lin et al.’s⁷⁶ dose response analysis of average July residential erythemal UVR exposure (BD/m²), there were mixed findings. Risk of CVD mortality increased with dose. However, across all levels of comparison, the confidence intervals were compatible with both a benefit and harm, or with a null effect.

Proxy for radiation. Stevenson et al.⁸⁹ found evidence to suggest a beneficial effect of sunlight exposure, measured via latitude in the UK. Compared with northerly latitudes, residences 300 km further south were associated with a 9% lower risk of CVD mortality (HR = 0.91, 95% CI 0.86 to 0.95; high risk of bias).

Behavioural. Four articles examined the effect of sunlight exposure behaviour (Figure 3). Lindqvist’s et al.⁷⁸ Swedish cohort study found that those who had self-reported high recreational sun exposure were less at risk of dying from CVD than those who reported no exposure (HR = 0.45, 95% CI 0.31 to 0.67; high risk of bias).

Yang et al.⁹⁷ found that people who reported spending one or more weeks a year on sunbathing vacations were half as likely to die from CVD than those who never spent time on sunbathing vacations (HR = 0.50, 95% CI 0.30 to 0.80). Additionally, there was an association between higher frequency of sunburning and lower CVD mortality, however the wide confidence intervals indicate uncertainty in this finding (HR = 0.30, 95% CI 0.10 to 1.10). Furthermore, the results reported in this article were considered to be at very high risk of bias.

In contrast, the findings in He et al.⁶⁷ suggested a harmful effect of sunlight. They found that greater physician-assessed actinic skin damage was associated with a higher risk of CVD mortality. Those whose skin damage was considered severe had a 64% increased risk of CVD mortality compared with those with no actinic skin damage (HR = 1.64, 95% CI 1.29 to 2.10; high risk of bias). As shown for all-cause mortality, the association does not appear to be driven by smoking, which was controlled for in the analysis.

Donneyong et al.⁴⁰ looked at the self-reported frequency of outdoor recreational activity. The results were mixed, with wide confidence intervals that were compatible with both a benefit and harm, and considered to be at high risk of bias.

All-cancer mortality

Overview. In total, 17 articles looked at the effect of sunlight exposure on all-cancer mortality, with some reporting multiple exposure types, outcomes and/or date ranges (22 analyses reported across the 17 articles). Twelve of the articles had an ecological design and five had a cohort design. The majority of analysis results (20 analyses) were in the direction of a beneficial association between sunlight and all-cancer mortality. However, two analyses^67,76 suggested there may be a harmful effect of sunlight. See Figure 4 for reported relative risks and Table 1 for results reported in other formats.

Figure 4. Forest plot of studies showing the effect of exposure on all-cancer mortality.

*Fixed-effects meta-analysis performed to combine gender subgroups. †Result inverted in order to reflect an increase in exposure. ‡Ratio calculated via exponentiating logistic regression beta coefficient. NR: ethnicity of population not reported.

Radiation. Ten articles looked at the effect of radiation on all-cancer mortality. The majority (n = 9) found a potentially beneficial effect of sunlight, with higher levels of radiation associated with lower levels of cancer mortality (Figure 4 and Table 1). In Altug and Kilçiksiz²⁷, a national ecological study in Turkey, the results indicated that a one (unspecified) unit increase in sunlight duration was associated with a 56% reduction in the risk of cancer mortality (RR = 0.44, 95% CI 0.32 to 0.60; high risk of bias).

Chen et al.³⁵ performed a national ecological study in China, looking at the effect of UVB exposure on cancer mortality. The findings suggested that a 10 mW/m² increase in average daily UVB irradiance was associated with a 4% lower risk of mortality (rate ratio = 0.96, 95% CI 0.95 to 0.97; high risk of bias). Similarly, Fukuda et al.’s⁴⁸ ecological study in Japan found that, per MJ/m² of solar radiation, there was a 1.3% reduction in cancer mortality risk (RR = 0.99, 95% CI 0.98 to 0.99; some concerns over risk of bias). The findings in Grant and Garland⁵⁶ and Grant⁵⁷ suggest a similar association. Higher levels of UVB were found to be associated with lower cancer mortality rates across both Black and White males and females between 1970 and 1994 (some concerns over risk of bias). A relationship between higher levels of radiation and lower cancer mortality was also observed in Apperly²⁸, Camara and Brandao³⁴, Ezzati et al.⁴² (some concerns over risk of bias) and Grant and Garland⁵⁶ (1950–1969; some concerns over risk of bias).

Goggins et al.⁵² found an association between higher solar radiation and lower cancer mortality, however the confidence intervals were compatible with a null effect (RR = 0.93, 95% CI 0.85 to 1; high risk of bias). In contrast, Lin et al.⁷⁶ found that increasing levels of residential erythemal UVR (BD/m²) were associated with an increased risk of cancer mortality. Comparing the fourth with the first quartile showed a 6% increase in the risk of mortality (HR = 1.06, 95% CI 1.02 to 1.11; some concerns over risk of bias).

Proxy for radiation. Four articles looked at the effect of proxy radiation exposures on cancer mortality, with all finding a potentially beneficial effect of sunlight (Figure 4 and Table 1). The findings of all four of the articles were assessed to be at high risk of bias.

Stevenson et al.⁸⁹ reported that, compared with UK northerly latitudes, residences 300 km further south had a 7% reduction in mortality risk (HR = 0.93, 95% CI 0.90 to 0.96). Grant⁵⁹ conducted a national ecological study in China. They found an association between higher latitudes (i.e., lower levels of sunlight) and higher cancer mortality (β = 0.75, p < 0.001). A similar relationship was found by Grant in France⁶² and California⁶⁴.

Behavioural. Five articles studied the effect of sunlight exposure behaviour on cancer mortality (Figure 4 and Table 1). He et al.⁶⁷ found that those with greater physician-assessed actinic skin damage had a greater risk of cancer mortality. Participants whose skin damage was considered severe had a 78% higher risk of mortality compared with those with no actinic skin damage (HR = 1.78, 95% CI 1.20 to 2.63; high risk of bias).

In contrast, Grant’s⁶⁴ ecological study in California, using population-level NMSC mortality rates as a proxy for sunlight exposure, found a relationship between higher NMSC mortality rates and lower overall cancer mortality (β = −0.69, p < 0.001; high risk of bias). Apperly²⁸ looked at the proportion of the population engaged in agricultural work and reported a relationship between higher proportions of agricultural workers and lower cancer mortality. However, no margins of error were reported with this estimate.

Two articles, Lindqvist et al.⁷⁸ and Yang et al.⁹⁷, found that recreational sunbathing behaviour was associated with lower cancer mortality. However, in both cases there were wide confidence intervals which were compatible with both a beneficial and harmful effect of sunlight exposure (Figure 4). Furthermore, they were considered to be at high and very high risk of bias, respectively.

Specific cancers and CVD causes

Skin cancers. In total, 23 articles investigated the effect of sunlight on skin cancer mortality with some reporting multiple exposure types, outcomes or date ranges (46 analyses reported across the 23 articles; 28 analyses measuring effect on melanoma, 16 measuring effect on NMSC, and two measuring effect on both combined). Seventeen of the articles were ecological, three had a cohort design and three used a case-control design. There were 25 analyses looking at the effect of radiation, 16 on the effect of proxy radiation exposures and five looked at the effect of sunlight exposure behaviour.

Most analysis results (n=34) suggested that higher levels of sunlight were associated with a higher risk of both melanoma and NMSC mortality (Figure 5 and Table 2). However, the results of six were in the direction of a beneficial effect of sunlight: two found that higher latitude (i.e. lower sunlight) was associated with lower melanoma³⁹ and NMSC mortality²⁵; two suggested an association between higher levels of UVB and lower melanoma mortality^51,90; one reported a weak association between higher annual insolation and lower melanoma mortality²⁵ and one reported an association between higher NMSC mortality rates and lower melanoma mortality⁵⁸. The remaining six analyses either produced mixed results^45,58,96, had no direction reported⁴⁴, showed little to no effect⁷² or had very wide confidence intervals that were compatible with both a benefit and harm⁸³.

Figure 5. Forest plot of studies showing the effect of exposure on skin cancer mortality.

*Fixed-effects meta-analysis performed to combine gender subgroups. †Result inverted in order to reflect an increase in exposure. ‡Ratio calculated via exponentiating logistic regression beta coefficient. NR: ethnicity of population not reported.

Breast cancer mortality. In total, 11 articles looked at the effect of sunlight on breast cancer mortality, with some reporting multiple exposure types and/or date ranges (17 analyses reported, across the 11 articles). Nine articles had an ecological design and one each used cohort and case-control designs. There were ten analysis results examining the effect of radiation, four looked at proxy measures of radiation and three looked at behavioural measures of exposure. Overall, the evidence suggested that sunlight may provide a protective effect (Figure 6 and Table 3). Thirteen analysis results from Boscoe and Schymura³⁰, Chen et al.³⁵, Freedman et al.⁴⁶, Grant and Garland⁵⁶, and Grant^53,57,58,62 indicated that higher levels of sunlight were associated with reduced breast cancer mortality. Three analyses from Lin et al.⁷⁶, Fukuda et al.⁴⁸, and Grant⁵⁸ suggested that there may be a harmful association between sunlight and breast cancer mortality, and one from Fleischer and Fleischer⁴⁴ was reported without a direction of effect.

Figure 6. Forest plot of studies showing the effect of exposure on breast cancer mortality.

*Fixed-effects meta-analysis performed to combine gender subgroups. †Result inverted in order to reflect an increase in exposure. ‡Ratio calculated via exponentiating logistic regression beta coefficient. NR: ethnicity of population not reported.

Prostate cancer mortality. There were 15 articles looking at the effect of sunlight on prostate cancer mortality, with some reporting multiple exposure types and/or date ranges (25 analyses reported across the 15 articles). There were 12 articles with an ecological design, two used a cohort and one used a case-control design. The findings were mixed (Figure 7 and Table 4). Fourteen analysis results looked at the effect of radiation, five looked at proxy radiation measures and six examined the effect of sunlight exposure behaviour. Most results suggested there may be a beneficial effect of sunlight on prostate cancer mortality (14 analyses). However, six results from Lin et al.⁷⁶, Grant and Garland⁵⁶, Freedman et al.⁴⁶, John et al.⁶⁹ and Grant⁵⁸, indicated that sunlight may have a harmful effect on prostate cancer mortality. Three results from Grant⁵⁸, Grant and Garland⁵⁶ and Mizoue⁸⁰, found little evidence of an effect; one result from Colli and Grant³⁷ produced very wide confidence intervals compatible with both a benefit and harm and one article⁴⁴ did not report the direction of effect.

Figure 7. Forest plot of studies showing the effect of exposure on prostate cancer mortality.

†Result inverted in order to reflect an increase in exposure. NR: ethnicity of population not reported.

Lung cancer mortality. There were 10 articles investigating the effect of sunlight on lung cancer mortality, with some reporting multiple exposure types (12 analyses reported across the 10 articles). Nine articles had an ecological design and one had a cohort design. There were six analyses looking at the effect of radiation, three looked at proxy radiation measures and three looked at sunlight exposure behaviour. The findings were mixed (Figure 8 and Table 5). The majority of analyses (n = 7), reported in Chen et al.³⁵, Fukuda et al.⁴⁸, Fleischer and Fleischer⁴⁴, Grant^57,59,64, and Garland et al.⁵⁰ found that higher levels of sunlight were associated with a decreased risk of lung cancer mortality. However, three analyses in Lin et al.⁷⁶ and Grant⁵⁸ suggested that there may be a harmful effect of sunlight. The findings of two analyses from two articles^58,62 were mixed, with different findings reported for males and females.

Figure 8. Forest plot of studies showing the effect of exposure on lung cancer mortality.

*Fixed-effects meta-analysis performed to combine gender subgroups. ‡Ratio calculated via exponentiating logistic regression beta coefficient. NR: ethnicity of population not reported.

Bowel cancer mortality. There were 14 articles measuring the effect of sunlight on bowel cancer mortality, with some reporting multiple exposure types, outcomes and/or date ranges (31 analyses reported across the 14 articles). There were 11 articles with an ecological design, two used a cohort and one had case-control design. Sixteen analysis results examined radiation measures, eight looked at proxy radiation measures and seven looked at the effect of exposure behaviour measures. Overall, the majority of analysis results (n = 22) suggested that higher levels of sunlight were associated with a decreased risk of bowel cancer mortality (Figure 9 and Table 6). However, four analyses from Grant⁵⁸ and Veach et al.⁹⁴ indicated that there may be a harmful association between sunlight and mortality. Three analyses from Grant⁵⁸, Lin et al.⁷⁶ and Freedman et al.⁴⁶ produced mixed findings, with conflicting results found either across dose levels or between gender. The result in Page et al.⁸³ produced very wide confidence intervals that were compatible with both a benefit and a harm, whilst in Fleischer and Fleischer⁴⁴ the direction of effect was not reported.

Figure 9. Forest plot of studies showing the effect of exposure on bowel cancer mortality.

*Fixed-effects meta-analysis performed to combine gender subgroups. †Result inverted in order to reflect an increase in exposure. ‡Ratio calculated via exponentiating logistic regression beta coefficient. NR: ethnicity of population not reported.

Pancreatic cancer mortality. There were seven articles measuring the effect of sunlight on pancreatic cancer mortality, with some reporting multiple exposure types or date ranges (11 analyses reported across the seven articles). Six articles had an ecological design and one used a cohort. There were seven analyses examining the effect of radiation, two articles looked at proxy for radiation measures and two looked at sunlight exposure behaviours. The majority of analyses (n = 9) suggested that higher levels of sunlight are associated with a decreased risk of pancreatic cancer mortality, as reported in Boscoe and Schymura³⁰, Fukuda et al.⁴⁸, Lin et al.⁷⁶, Neale et al.⁸¹, Grant and Garland⁵⁶, and Grant⁵⁸. However, one analysis in Grant⁵⁸ indicated there may be a harmful association between melanoma mortality rates and pancreatic cancer mortality. One analysis in Fleischer and Fleischer⁴⁴ was reported narratively, without a direction of effect (Figure 10 and Table 7).

Figure 10. Forest plot of studies showing the effect of exposure on pancreatic cancer mortality.

*Fixed-effects meta-analysis performed to combine gender subgroups. †Result inverted in order to reflect an increase in exposure. ‡Ratio calculated via exponentiating logistic regression beta coefficient. NR: ethnicity of population not reported.

Cause-specific CVD mortality. Six articles looked at cause-specific CVD mortality (7 analyses across the 6 articles); four with an ecological design and two using a cohort design. All seven analyses examined radiation measures. Four results looked at the effect on heart disease mortality and three looked at the effect on stroke mortality. The findings were mixed (Figure 11 and Table 8). Two analyses suggested a beneficial effect of radiation on heart disease mortality^42,87. However, three analyses reported a harmful effect on stroke mortality^42,52,76. One produced mixed results⁴³, and one did not report a direction of effect³⁴ (Figure 11 and Table 8).

Figure 11. Forest plot of studies showing the effect of exposure on cause-specific CVD mortality.

†Result inverted in order to reflect an increase in exposure. NR: ethnicity of population not reported.

Subgrouping by skin type/colour or ethnicity

We sought to investigate differences in effect between people with different skin types/colours or ethnicity. However, information to allow this was limited. Most articles reported findings that examined the whole population (62%), and several limited their population to White people only (24%).

One study in the USA examined all-CVD mortality by ethnicity subgroup²⁶, reporting slightly higher mortality risk associated with sunlight exposure for White, Black, Hispanic and Asian people, but a slightly lower risk among Native Americans. Also in the USA, Boscoe and Schymura³⁰ found that residing along the southern border (erythemally-weighted UVB exposure of roughly 1540 kJ/m²/year) was associated with a decreased risk of breast cancer mortality compared with residing along the northern border (roughly 650 kJ/m²/year) among both White women (RR = 0.87, 95% CI 0.85 to 0.88) and Black women (RR = 0.90, 95% CI 0.86 to 0.94). Pennello et al.⁸⁴ found a harmful relationship between UVB and both melanoma and NMSC mortality for both Black and White people. Finally, Veach et al.⁹⁴ found that higher solar radiation (Weber/m² by state) was associated with higher risk of bowel cancer mortality in Black (β = 0.003, p < 0.001) and Hispanic men (β = 0.001, p = 0.033), but lower risk in White men (β = −0.001, p = 0.001).

It was sometimes possible to make indirect comparisons across people of different skin types/colours or ethnicity. Grant and Garland⁵⁶ and Grant⁵⁷ reported data for the White population and the Black population, respectively, in the USA between 1970 and 1994. The results were suggestive of a beneficial effect of sunlight on all-cancer, breast cancer and bowel cancer mortality for both Black and White people. Boscoe and Schymura³⁰ reported that higher levels of sunlight exposure were associated with a decreased risk of prostate cancer in White men in the USA between 1993 and 2002 (RR = 0.85, 95% CI 0.84 to 0.87); whilst conversely, Colli and Grant³⁷ observed a small positive association between higher winter UV Index and prostate cancer mortality among Black men in the USA between 1992 and 2001, though the confidence intervals were compatible with both benefit and harm (β = 0.20; 95% CI −2.4 to 2.8).