Introduction

There is ongoing debate about whether avoidance of sunlight or vitamin D deficiency is a major risk factor for health. The findings of two recent reviews on the impact of vitamin D were completely different, with one showing that no firm conclusions could be drawn 1 and the other demonstrating a population attributable risk of death in the same range as smoking, inactivity or obesity 2. Studies regarding sun exposure are rare, but recently, we reported that the mortality rate was doubled in women in the Melanoma in Southern Sweden (MISS) cohort who avoided active sun exposure, compared to those with the highest sun exposure 3. In addition, we found no differences in all-cause or cutaneous malignant melanoma (MM) mortality between those who expose themselves to and those who avoid the sun.

Most studies have analysed the relationship between the upper extreme of sun exposure and skin cancer and have showed an increased incidence. Therefore, it is difficult to investigate sun exposure without taking skin cancer into consideration. Skin cancer is usually divided into three types according to increasing severity: basal cell carcinoma (BCC), squamous cell carcinoma (SCC) and cutaneous MM. The two former are often grouped as nonmelanoma skin cancer (NMSC) due to their similarity and generally nonfatal prognosis. SCC is mostly related to cumulative exposure to UV light, whilst UV light mainly increases the risk of MM through episodic sunburn and excessive exposure including frequent use of tanning beds 4. The incidence of MM in Sweden has doubled during the last 15 years, whilst the mortality rate has been constant since 1980s 5.

What causes the excess mortality amongst women in the small subgroup (5.8%) who avoid sun exposure is currently unknown. In this study, we have classified mortality into three main categories, death due to cardiovascular disease (CVD), cancer and noncancer/non-CVD, and analysed all-cause death in a competing risk scenario. The aim of this study was to determine how sun exposure is related to these main causes of death.

Materials and methods

The study was approved by the Ethics Committee of Lund University (LU 632-03). The MISS study, initiated in 1990, included approximately 1000 Sweden-born women of each age from 25 to 64 years (n = 39 973) who had no history of malignancy. Subjects were selected from the general population registry of the South Swedish Health Care Region by random computerized selection and represented 20% of the female population of South Sweden at each age.

Women were invited to complete a standardized written questionnaire concerning risk factors for MM. The initial questionnaire was administered from 1990 to 1992 and resulted in 29 518 women participating in the study (response rate 74%). The questionnaire was a detailed inquiry into several factors of potential interest for mortality, such as sun exposure habits, marital status, educational level, smoking, alcohol consumption and the number of births. A total of 184 women emigrated during the study period and were censored after emigration. We collected information on mean personalized family income between 1990 and 1993 from official income and taxation records at Statistics Sweden (further details available at http://www.scb.se/en_/). Four predetermined questions were posed regarding sun exposure: (i) How often do you sunbathe during the summertime? (never, 1−14 times, 15−30 times, >30 times); (ii) Do you sunbathe during the winter, such as on vacation to the mountains? (no, 1−3 days, 4−10 days, >10 days); (iii) Do you use tanning beds? (never, 1−3 times per year, 4−10 times per year, >10 times per year); and (iv) Do you go abroad on vacation to swim and sunbathe? (never, once every 1–2 years, once a year, two or more times per year). The four questions were dichotomized into yes/no in the analysis (i.e. sometimes versus no or never). We created a four-score variable as a measure of sun exposure depending on the number of ‘yes’ responses to the above questions on a scale from 0 (avoid sun exposure: reference) to 4 (greatest sun exposure). Sun exposure habits were categorized into three groups: zero ‘yes’ responses (avoidance of sun exposure; the main study group); ‘yes’ responses to one or two questions (moderate exposure); and ‘yes’ responses to three or four questions (greatest exposure). Vital statistics and cancer data were determined from the National Population Register up to 1 January 2011. The presence of skin cancer was recorded in the following hierarchical order: MM, NMSC or no skin cancer. Thus, a woman with NMSC was reclassified to MM upon MM diagnosis.

With regard to smoking habits, women were recorded as either smokers or nonsmokers at baseline. As a measure of comorbid illness at the start of the study, we created a dummy variable termed ‘comorbidity’ to identify women who had been treated with antidiabetic [Anatomical Therapeutic Chemical (ATC) classification system A : 10] or anticoagulant (ATC B : 01) drugs or medication for CVD (ATC C : 01–C : 10) for more than 1 month.

Age was categorized into 10-year intervals. For comparison of ages, approximately 50 and 60 years of age referred to women in the age groups 45–54 and 55–64 years at the start of the study. Data regarding BMI and physical exercise were recorded at the second questionnaire in the year 2000.

Statistical analysis

Descriptive statistical analysis was performed using cross-tabulation with 95% confidence interval (CI). Cox regression was performed to assess all-cause mortality, as the dependent variable, with sun exposure, age, smoking, education, marital status, disposable income and comorbidity as independent variables. Subdistribution Cox regression analysis was performed to determine whether avoidance of sun exposure is a risk factor for CVD, cancer and noncancer/non-CVD. When a specific death was used as the dependent variable, the other two causes of death were censored, and sun exposure and other confounders were introduced as independent variables. In the final cause-specific regression, we included comorbidity, smoking, sun exposure, age, education, marital status and disposable income. The subdistribution hazard ratio (sHR) and 95% CI were used to formally assess whether the resulting cumulative incidence functions differed significantly by level of sun exposure. Fine and Gray regression models were used to estimate cause-specific cumulative incidence functions for death due to cancer, CVD and noncancer/non-CVD in the presence of competing risks 6. The model-based cause-specific cumulative incidence functions were used to quantify the absolute as well as the relative contribution of each cause of death to all deaths (Fig. 1). The competing risks regression models were adjusted for the same potential confounding factors as the cause-specific Cox regression models.

As a complement to the competing risk models, we also quantified the loss in average life expectancy over a 20-year observation period by estimating the differences in restricted mean survival (RMS), that is the area under the survival curve between two time-points. This provides a measure of average survival between exposure groups. We predicted the RMS based on a flexible parametric survival model that uses restricted cubic splines to model the baseline hazard function 7. Specifically, we calculated the difference in RMS between the three different sun exposure groups over a 20-year follow-up period, adjusted for age at study inclusion, comorbidities, disposable income and smoking status. The results are presented for smokers and nonsmokers of different ages who had a previous record of comorbid conditions and a low disposable income.

Both the Cox regression and Fine and Gray competing risks analyses used time from inception as the timescale. Time from inception was calculated from inclusion to cause-specific death (cancer, CVD and other causes), emigration or 1 January 2011, whichever occurred first.

IBM SPSS 21 (Statistical Package for the Social Sciences, SPSS Inc., Chicago, IL, USA) software was used for descriptive analysis, and Stata 12 (Statacorp, College Station, TX, USA) was used for the regression modelling. P-values <0.05 were considered statistically significant.