Ultraviolet radiation and age at natural menopause in a nationwide, prospective US cohort.

A total of 105,631 women were eligible to be included in the analysis, and among them 63,801 reported natural menopause during 1,051,185 person-years. A total of 2,485 women reported menopause between ages 40-45. Table 1 summarizes characteristics of the participants by quartiles of cumulative average UV exposure. Overall, participants in the NHS II cohort were mostly white, parous, and never smokers. Throughout the study period, women with UV exposure in the highest quartile were more likely to be nulliparous, have an overall healthier diet, use perimenopausal hormones, and reside in neighborhoods with higher population density, family income, and home values; while women with exposure in the lowest quartile were more likely to be white and have ever worked in rotating shifts ( Table 1 ). No associations between UV exposure and menopause were detected in the basic and parsimonious models. However, after adjusting for all covariates, exposure to UV was associated with later menopause and this change of estimates mainly occurred after adjusting for perimenopausal hormone use. There was some evidence of a non-linear exposure-response relationship for cumulative average UV exposure (p for deviation from linearity < 0.01), but not for exposure in the previous year (p for deviation from linearity = 0.09). As shown in Table 2 , the HR was 0.96 (95% CI=0.94, 0.99) comparing women with cumulative average UV exposure in the highest quartile to those in the lowest quartile (p for trend <0.01). Similar inverse associations were observed for exposure in the previous year (p for trend < 0.01). Nevertheless, the observed associations were very modest. For example, the predicted median age at menopause was 1.5 months older for a woman with cumulative average UV exposure in the highest quartile than for a woman with exposure in the lowest quartile, conditional on all other factors. We found no associations with the timing of menopause for UV exposure at birth (HR comparing the highest to the lowest quartile=1.01, 95% CI=0.98, 1.04; p for trend =0.51) and at age 15 (HR comparing the highest to the lowest quartile=1.01, 95% CI=0.98, 1.03; p for trend =0.63), while exposure at age 30 was suggestively associated with delayed menopause (HR comparing the highest to the lowest quartile=0.97, 95% CI=0.95, 1.00; p for trend < 0.01) ( Table 3 ). Secondary analyses showed suggestive associations of UV and lower risk of menopause between age 40-45 for exposures in the previous year and at age 30 ( Table 4 ). For example, the HRs comparing the highest to the lowest quartile of UV was 0.91 (95% CI=0.81, 1.02, p for trend = 0.04) for exposure in the previous year and 0.89 (95% CI=0.78, 1.01, p for trend=0.03) at age 30 ( Table 4 ). Sensitivity analyses adjusting for sun exposure, sunscreen use, tanning bed use, outdoor physical activity, source-specific vitamin D intake, and particulate air pollution showed similar associations to those from the main model ( Table A.1 ). Restricting the analysis to women who never moved, who never used hormone therapy, or who never used oral contraceptives after age 40 also showed robust estimates ( Table A.2 ). No effect modification was observed by sun exposure, sensitivity, sunscreen use, vitamin D intake, vitamin D status, race/ethnicity, total and outdoor physical activity, BMI, menstrual cycle length, history of infertility, and smoking (p for interaction = 0.18-0.98) ( Table A.3 ).

The study population included participants from the NHS II, an ongoing prospective cohort that recruited 116,429 female registered nurses who were 25-42 years old and resided in 14 states in 1989. During the follow-up, participants have moved and now reside in all 50 states and the District of Columbia since the mid-1990s. Follow-up questionnaires have been mailed to the residential address of the participants every 2 years, collecting information on major health risk factors and health conditions. In this analysis, we included women who were premenopausal, still responding to the follow-up questionnaires, and had at least one geocoded residential address in the contiguous US for exposure assessment at or after age 40. Women who reported a hysterectomy, oophorectomy (unilateral or bilateral), or cancer (except nonmelanoma skin cancer), or who died before age 40 were excluded. This study was approved by the Institutional Review Board of Brigham and Women‘s Hospital and the Human Subjects Committee of the Harvard T.H. Chan School of Public Health. Informed consent was implied through return of the questionnaires. Menopausal status, reasons of menopause (e.g. natural, surgical, and due to chemotherapy or radiation), and age at menopause were self-reported at baseline and in each follow-up questionnaire. Natural menopause was identified when a participant first indicated menopause due to natural causes. All self-reported menopausal status and reasons were verified by consistent reports in two adjacent questionnaires. The timing of menopause was determined by the self-reported age at menopause or by the age at return of the questionnaire if the age at menopause was missing. Women who reported menopause due to surgery, chemotherapy and/or radiation, or did not disclose reasons of menopause, who received a hysterectomy or oophorectomy, who were diagnosed with cancer (except for nonmelanoma skin cancer), or who died were censored at return of the questionnaire or at the time of event confirmation. Death was confirmed by reports from next-of-kin, postal authorities, or by searching the National Death Index. Cancer diagnosis was confirmed by medical record review and/or cancer registry. Women who failed to report their menopausal status on two consecutive questionnaires were considered as lost to follow-up. UV exposure was measured as erythemal UV, a measure of solar UV radiation indexing the capability of inducing erythema on white skin. Specifically, this measure covers the wavelength spectrum for both UV-A and UV-B and assigns greater weight to the latter, which is mainly involved in dermal synthesis of vitamin D. The average July noontime erythemal UV irradiation in the contiguous US for each year from 1980 to 2015 was obtained from a geospatial model. This model used a stratified area-to-point residual kriging approach to downscale National Aeronautics and Space Administration (NASA) satellite UV data, and incorporated geospatial information on predictors of UV to predict erythemal UV irradiation at a spatial resolution of 1 km 2 ( VoPham et al., 2016 ). The predicted July average UV levels were first linked to the geocoded residential address history for each participant throughout follow-up (1989-2015). We then calculated time-varying UV exposure as the cumulative average from age 40 onwards to represent long-term exposure in mid-adulthood and UV level in the previous year to represent recent exposure. We also considered early-life exposure in secondary analyses. As the residential address information before 1989 was not collected, we aggregated the earliest UV measure (1980) for each state and assigned these state-level estimates to the self-reported state of residence at birth and at age 15. For exposures at age 30, we used the state-level UV in 1980 for participants who were age 30 before 1980 (19% of participants), the state-level UV in the concurrent year for those who were age 30 between 1980-1989 (59% of participants), and the concurrent year UV at the residential address for those who were age 30 after 1989 (22% of participants). We considered time-varying and invariant covariates that have been suggested to be potential confounders between UV exposure and age at menopause or predictors of age at menopause. Information on time-varying covariates were collected from questionnaires every 2 years (every 4 years for diet and physical activity), including body mass index (BMI), smoking status and intensity, physical activity (metabolic equivalent task (MET) hours/week), alcohol consumption, overall diet quality measured by the 2010 Alternate Healthy Eating Index (AHEI, ( Chiuve et al., 2012 )), energy-adjusted vitamin D intakes (total, dietary, supplemental, and dairy, respectively), history of rotating shift work, marital status, history of reproductive health and behaviors (parity, age at first birth, breastfeeding, use of oral contraceptives and hormone replacement, and history of uterine fibroids and endometriosis), and neighborhood socioeconomic measures (Census tract population density, median family income, and median home values). Time-invariant covariates, including race/ethnicity and age at menarche were also included. We also obtained time-invariant information on skin reaction to sunlight and hair color as proxies of sensitivity to sun exposure. Behavioral factors related to UV exposure, including time spent under direct sunlight and tanning bed use at age 25-35, were collected in 2005; sunscreen use while a teenager and in the past summer were obtained in 1993. Menstrual cycle length at age 18-22 was obtained at baseline and history of infertility was collected throughout follow-up. Time-varying exposure to ambient air pollution including particulate matter with aerodynamic diameters of 2.5-10 (PM 2.5-10 ) and ≤ 2.5 microns (PM 2.5 ) at each residential address were predicted from geospatial models ( Yanosky et al., 2014 ). We calculated person-years from age 40 until self-reported natural menopause, censoring events, or return of the 2015 questionnaire, whichever came first. Hazard ratios (HRs) and 95% confidence intervals (95% CIs) were computed using the time-varying Cox proportional hazards regression with age as the time scale and stratified by calendar years to control for time trends. Potential non-linear exposure-response relationships were examined using cubic splines for all exposure measures. UV exposure in different time windows was introduced into the model by quartiles and as a continuous variable, respectively. Tests for trend were conducted using the median values of UV exposure of each quartile in the models. We fitted a basic model with age and calendar year only, a parsimonious model additionally adjusting for race/ethnicity, smoking, BMI, and neighborhood socioeconomic status, and a full model additionally adjusting for physical activity, diet, rotating shift work, marital status, and history of reproductive health and behaviors. Age at menarche and reproductive history (e.g. parity, breastfeeding, age at first birth, and history of endometriosis and uterine fibroids) were not adjusted for UV exposure at birth and at age 15 as they were potential mediators. HRs less than 1 suggest that the exposure is associated with a later onset of menopause and HRs above 1 suggest an association with earlier menopause. We conducted secondary analyses of UV exposure and menopause between age 40 and 45. We also conducted sensitivity analyses examining for potential confounding by sun exposure, tanning bed use, and air pollution (calculated as time-varying cumulative averages). To examine for potential residual confounding by different sources of vitamin D, we separately adjusted for vitamin D from diet, supplements, and dairy products in place of total vitamin D intake ( Purdue-Smithe et al., 2017 ). To further control for time spent outdoors, we replaced the total physical activity levels with METs calculated from exercises that are more likely to be outdoors, including walking, running, jogging, swimming, cycling, and tennis, in the covariates. Finally, we restricted the analyses to women who never changed residential address, who never used oral contraceptives after age 40, and who never used perimenopausal hormone therapy. Effect heterogeneity by sun exposure, sensitivity (skin reactions to sun exposure and hair color), sunscreen use (as a teenager and in the past summer of 1993), vitamin D intake (total and supplemental, respectively), vitamin D status, physical activity (total and from outdoor exercises. respectively), BMI, race\ethnicity (white and non-white), menstrual cycle length at age 18-22, history of infertility, and smoking were examined by adding multiplicative interaction terms to the model; stratum-specific estimates were obtained from separate models. Vitamin D status was computed from a prediction model established previously using circulating 25(OH)D measured in a subset of NHSII participants and known predictors (i.e., age, BMI, race/ethnicity, physical activity, dietary and supplementary vitamin D intake, UV-B flux in state of resident, alcohol intake, season, and menopausal status and postmenopausal hormone use) ( Bertrand et al., 2012 ), and were classified as deficient (predicted 25(OH)D < 20 ng/mL), insufficient (predicted 25(OH)D ≥ 20 and < 30 ng/mL), and sufficient (predicted 25(OH)D ≥ 30 ng/mL) in stratification analysis ( Holick et al., 2011 ). All statistical analyses were conducted in SAS 9.4 (SAS Inc, NC USA). All tests were two-sided with an α level of 0.05.

We found very modest associations between midlife UV exposure and later onset of menopause in this nationwide, prospective, female cohort. Of the early-life exposures, we examined, only those at age 30 were suggestively associated. Women who had higher UV exposure after age 30 had slightly lower risk of menopause between age 40-45, although the confidence intervals were wider and included the null. There were no evident effect modification by sun exposure, sensitivity, sunscreen use, vitamin D intakes, race/ethnicity, physical activity, body mass, menstrual cycle length, history of infertility, and smoking. Although menopause is a natural consequence of ovarian aging, an earlier onset of menopause has been associated with higher risks of cardiovascular diseases, osteoporosis, and shorter life expectancy ( Atsma et al., 2006 ; Gold, 2011 ; Hu et al., 1999 ; Kuh et al., 2016 ; Muka et al., 2016 ; Sullivan et al., 2017 ). One important determinant of the timing of menopause is the follicle atresia throughout lifetime ( Leidy, 1994 ; Thomford et al., 1987 ). Therefore, factors related to follicle survival and atresia may affect women’s age at menopause. Solar UV-B exposure is essential for cutaneous vitamin D synthesis, which can account for 90% of the total vitamin D produced in the human body ( Holick, 2007 ). Evidence has suggested vitamin D can affect the follicles by stimulating ovarian steroidogenesis ( Irani and Merhi, 2014 ). One study found lower expression and activities in aromatase, the enzyme responsible for estrogen conversion, in the ovary of vitamin D receptor null mutant mice. These mice also showed impaired folliculogenesis and hypergonadotropic hypogonadism, indicating vitamin D can facilitate ovarian estrogen biosynthesis ( Kinuta et al., 2000 ). Experiments in human prostate cancer cells and ovarian cells (e.g. granulosa, stromal, and theca cells) also showed vitamin D can regulate the production of estradiol and estrone as well as the expression of anti-Mullerian hormone, which are important hormonal factors for follicle survival and development ( Malloy et al., 2009 ; Parikh et al., 2010 ). Vitamin D has also been suggested to reduce cellular oxidative stress and inhibit apoptosis by inducing antioxidant expression and maintaining the normal mitochondrial function, which may also be beneficial to follicle survival and therefore delay menopause ( Wimalawansa, 2019 ). We found very modest associations of long-term residential UV exposure and later onset of menopause in this nationwide US cohort. A similar weak association was observed when restricting the analysis to women who reported menopause between age 40-45. To date, one cross-sectional study has examined lifelong sun exposure and age at natural menopause among 156 women in Turkey, and the results showed a strong association between low sun exposure throughout the lifetime and earlier onset of menopause (HR=6.38 comparing women with low to those with high lifetime sun exposure) ( Aydin et al., 2005 ). Previous analyses in the NHS and NHS II have found weak or modest associations between vitamin D intake and the timing of menopause. For example, one study in NHS found weak and not statistically significant associations of total vitamin D intake from food and supplements with delayed menopause (vitamin D in the 5 th vs the 1 st quintile, HR=0.95 and 0.96 for menopause before and after age 51, respectively) ( Carwile et al., 2013 ). In NHS II, dietary and dairy vitamin D intake were modestly associated with lower risk of early menopause (HR=0.83 and 0.85 comparing the 5 th to the 1 st quintile of dietary and dairy vitamin D, respectively), while no associations were observed for total and supplemental vitamin D ( Purdue-Smithe et al., 2017 ). Another analysis in a subsample of NHS II showed suggestive yet not statistically significant inverse associations for plasma total and free 25-hydroxyvitamin D levels and early menopause risk ( Purdue-Smithe et al., 2018 ). Other studies examining circulating vitamin D levels and markers of ovarian reserve reported mixed results ( Alavi et al., 2020 ; Dennis et al., 2012 ; Drakopoulos et al., 2017 ; Jukic et al., 2018a , 2015b ; Merhi et al., 2012 ; Moridi et al., 2020 ; Purdue-Smithe et al., 2018 ). However, most of these studies used a single measure of vitamin D, which could be affected by the very recent UV exposure and/or diets. Results from these studies suggested that the associations between vitamin D and ovarian reserve may differ by characteristics such as fertility, menstrual pattern, and race/ethnicity; however, no effect modification was found for these factors in our stratified models ( Moridi et al., 2020 ; Moslehi et al., 2017 ). Long-term UV exposure in this study was measured using predictions from a high-resolution geospatial model, which has been used in previous analyses ( Lam et al., 2020 ; VoPham et al., 2019 ). Advantages of this exposure measure include that it is an objective measure of environmental UV-B exposure and is minimally affected by recall errors compared to the self-reported behavioral proxies of sun exposure (e.g. time spent outdoors, sunscreen use, and clothing). However, this ambient UV measure does not consider individual behaviors, which likely contribute to the between-individual variations in personal UV exposure ( Dadvand et al., 2011 ). Nevertheless, our sensitivity analyses showed similar estimates and no effect modification was observed by these factors. We observed the largest change in the HRs between the crude, parsimonious, and fully adjusted model after adjusting for perimenopausal hormone use, which appeared to be a major confounder. Although the use of hormone replacement does not directly lead to menopause, this behavior may indicate that a woman is experiencing symptomatic estrogen withdrawal and may be approaching menopause (i.e., hormone users were more likely to report menopause). On the other hand, using exogenous hormone can lead to uterine bleeding, likely resulting in a delayed report of menopause. A previous survey in the US showed women in the South were more likely to use hormones, which was also observed in our cohort ( Gass et al., 2015 ). Therefore, adjusting for hormone use as a covariate in the model may remove some residual confounding by geographic region, which was not controlled for to avoid over-adjustment. Previous studies on timing of menopause considered hormone use as either a covariate in the model ( Gold et al., 2013 ; Luoto et al., 1994 ; Schoenaker et al., 2014 ) or as a censoring event ( Ding et al., 2020 ; Triebner et al., 2019 ), and so far it is not yet clear which is a better option. Nevertheless, our sensitivity analysis by restricting to women who never used hormone showed similar trends, although the HRs were slightly attenuated. Limitations of this study include that the UV exposure was measured at the residential address, while time spent in the other locations (e.g. workplace or recreational location) or outdoors was not available. This exposure was estimated in July of each year and thus does not capture seasonal variation of UV. For early-life exposure, we used the state average measured in 1980 to represent historical UV exposure at birth and at age 15, which is likely associated with exposure measurement error and reduced spatial and temporal variation. Although we considered several behavioral factors related to sun exposure in sensitivity analyses, these variables were collected retrospectively and/or were measured only once during the follow-up, which limits our ability to fully adjust for changes of these behaviors over time. As for the outcome, a previous validation study has demonstrated high reproducibility and reliability in the self-reported age at menopause in NHS ( Colditz et al., 1987 ). However, this annual scale measure was relatively crude and may limit our ability to detect stronger associations. Finally, NHS II participants were predominantly white, which may limit our ability to examine for possible effect heterogeneity by race/ethnicity as skin pigmentation (i.e., melanin) has been suggested to absorb UV radiation and thus reduce the amount of UV that penetrates the dermis for vitamin D production ( Neer, 1975 ; Yuen and Jablonski, 2010 ). Despite the limitations, we used a high spatial resolution model to objectively measure the time-varying UV exposure of our participants throughout the follow-up. We examined time-varying information on the covariates from over 20 years of follow-up in this cohort, thus allowing for better control of confounding. We were able to capture menopausal transition (either due to natural or non-natural causes) for most of the participants during follow-up, and therefore had sufficient statistical power to detect even a very modest association in the analysis.

We examined associations of UV exposure, a major source of human bioactive vitamin D, with age at natural menopause in this large, prospective, female cohort. Our results showed a very modest association between higher ambient UV exposure in adulthood and later onset of menopause, while no associations were observed for UV exposures in early life (i.e. at birth and age 15). Our results were consistent with previous findings on vitamin D with reproductive aging in the same population. Although vitamin D may have an important physiological role in the ovary, our findings may have very limited clinical significance.

Solar ultraviolet (UV) radiation within the wavelength spectrum of 280-315 nm (UV-B) is the most predominant source of vitamin D in humans as it can penetrate the skin and convert 7-dehydrocholesterol to previtamin D and vitamin D ( Holick, 2007 ; Wacker and Holick, 2013 ). The health benefits of vitamin D have been well-documented in the literature, and evidence has suggested protective roles of vitamin D in musculoskeletal health, autoimmune disease, cancers, and cardiovascular disease ( Holick, 2004 ). Experimental s have found vitamin D receptors in female reproductive tissue, suggesting vitamin D may regulate reproductive activity among women ( Dokoh et al., 1983 ; Johnson et al., 1996 ; Kinuta et al., 2000 ). However, current population-based studies on vitamin D and female reproductive health outcomes such as fertility and menstrual regularity showed inconsistent results. For example, evidence from observational and intervention studies reported positive or null associations between vitamin D status and ovarian reserve ( Moridi et al., 2020 ; Moslehi et al., 2017 ), while other studies have linked low vitamin D levels with long menstrual cycle length or irregular cycles ( Jukic et al., 2015a , 2016 , 2018b ; Łagowska, 2018 ; Moridi et al., 2020 ; Moslehi et al., 2017 ). Previous analyses in the Nurses’ Health Study (NHS) and the Nurses’ Health Study II (NHS II) found modest or null associations for vitamin D intake or circulating 25-hydroxy-vitamin D with the timing of menopause ( Carwile et al., 2013 ; Purdue-Smithe et al., 2017 , 2018 ). However, whether solar UV exposure, as a critical factor of dermal synthesis of vitamin D in human, can affect female reproductive aging remains unclear. In this study, we sought to use time-varying UV exposure estimated from a high resolution spatiotemporal prediction model to examine the associations of ambient UV exposure in mid-adulthood with age at natural menopause in a large, prospective cohort. We also considered associations with early-life UV exposure at birth, at age 15, and at age 30. Possible effect modifications by sun exposure, sensitivity to sunlight, sunscreen use, and vitamin D intake were also examined.