Long-term exposure to particulate matter and roadway proximity with age at natural menopause in the Nurses' Health Study II Cohort.

Characteristics of all participants throughout follow-up are shown in Table 1 . Among the 105,996 NHS II participants who were still premenopausal at age 40, 66% were never smokers, 83% had at least 1 full-term pregnancy, and 74% never used hormone therapy during follow-up. The mean [± standard deviation (SD)] levels of cumulative average PM 10 , PM 2.5-10 , and PM 2.5 were 22.6 (± 6.5), 9.0 (± 4.7), and 13.6 (± 3.1) μg/m 3 , respectively. Within each time window, PM 10 was highly correlated with PM 2.5-10 (Spearman r=0.85-0.86) and moderately with PM 2.5 (Spearman r=0.69-0.71), while the correlation between PM 2.5-10 and PM 2.5 was low (Spearman r=0.26-0.29). We also observed very high correlations for PM 10 , PM 2.5-10 , and PM 2.5 , respectively, between exposure windows (Spearman r=0.90-0.99). A total of 64,340 reports of natural menopause were identified from 105,996 eligible NHS II participants and 1,059,229 person-years of follow-up. Exclusions before age 40 and censoring during the follow-up are showed in Table S1 . The median age at natural menopause was 51 years old. No statistically significant non-linear exposure-response relationships were observed (p for deviations from linearity=0.33-0.97). As shown in Table 2 , after adjusting for all covariates, we found residential PM 10 , PM 2.5-10 , and PM 2.5 were associated with a slightly earlier onset of menopause. For example, the HRs and 95% CIs of natural menopause for a 10 μg/m 3 increase of the cumulative averages of PM 10 , PM 2.5-10 , and PM 2.5 was 1.02 (95% CI=1.00, 1.04), 1.03 (95% CI=1.00, 1.05), and 1.03 (95% CI=1.00, 1.06), respectively. Similar associations were also found for exposures to all three PM fractions in age 40-45. Hazard ratios from models with both PM 2.5-10 and PM 2.5 were similar to the single pollutant models. The predicted median age at natural menopause among women with cumulative average PM 10 , PM 2.5-10 , and PM 2.5 at the 75 th percentile (PM 10 : 25.4 μg/m 3 , PM 2.5-10 : 11.0 μg/m 3 , and PM 2.5 : 15.6 μg/m 3 ) was 0.6, 0.6, and 0.4 months earlier compared to those with exposure levels at the 25 th percentile (PM 10 : 18.4 μg/m 3 , PM 2.5-10 : 5.8 μg/m 3 , and PM 2.5 : 11.5 μg/m 3 ), respectively. There were 53,526 reports of natural menopause over 892,262 person-years of follow-up in 91,343 NHS II participants for the analysis of roadway proximity. Our fully adjusted model suggested women who lived within 50 m from A1-A3 roads at age 40 and in the previous year had earlier onset of menopause compared to those who lived 500 m or more from the roads (age 40: HR=1.03, 95%CI=1.00, 1.06; previous year: HR=1.02, 95%CI=0.99, 1.05). Similar HRs were also observed for distance to A1-A2 roads, although they were not statistically significant ( Table 3 ). The predicted median age at menopause was 1 month earlier among women who lived within 50 m to all major roads at age 40 compared to those who lived more than 500 m away from such roads. Both PM and road proximity showed similar associations with natural menopause in sensitivity analyses when the follow-up was ended in 2009, and among women who never used oral contraceptives after age 40. Besides, the HRs were comparable to the main results when restricting the analysis to women who never used hormone therapy and who never changed their addresses after age 40, although the 95% CIs included the null except for PM among those who never moved. Additional adjustment for residential NDVI and for exclusion before age 40 showed similar HRs and 95%CIs with natural menopause. ( Table S2 and S3 ). We did not observe evidence of effect modification by BMI, physical activity, menstrual cycle length at age 18-22, and Census tract population density for any PM fraction in any time windows (p values for interaction=0.07-0.96). We observed associations of PM 2.5 and PM 10 with earlier menopause only among never smokers (except for the exposure in the previous year) ( Figure 1 ). There was suggestive evidence of effect modification by region as women who lived in the West had greater and statistically significant HRs than women in other regions ( Figure 1 ). No effect modification was observed for analyses of roadway proximity (p values for interaction=0.08-0.85).

The NHS II cohort is an ongoing prospective cohort including 116,429 female registered nurses who completed the baseline questionnaire in 1989. At baseline, the participants were between 25 to 42 years of age and from 14 states in the US (California, Connecticut, Indiana, Iowa, Kentucky, Massachusetts, Michigan, Missouri, New York, North Carolina, Ohio, Pennsylvania, South Carolina, and Texas). However, they have resided in all 50 states and the District of Columbia since the mid-1990’s. A follow-up questionnaire has been mailed to the participant’s residential address every 2 years collecting information on major health risk factors and health outcomes. For this analysis, we included women who were premenopausal, still responding to the questionnaires, and had at least 1 geocoded residential address in the continental US after age 40. Women who reported a hysterectomy, oophorectomy, or cancer (except basal and squamous cell carcinoma), or had 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, and informed consent was implied by return of the questionnaires. Menopausal status, reasons for menopause (natural, surgical, chemotherapy or radiation), and age at menopause were self-reported by the NHS II participants in every questionnaire. For this analysis, we considered menopause due to natural causes (natural menopause) as the outcome of interest. To avoid outcome misclassification, all self-reported menopausal status and reasons were verified by their consistency on two consecutive questionnaires. If a participant did not indicate age at menopause in the questionnaire, we used her age at the return of that questionnaire as a surrogate. Participants who reported menopause due to surgery, chemotherapy or radiation, or did not report the reason of menopause were censored at the time of report. We also censored participants who failed to report menopausal status in two subsequent questionnaires after the last valid report of being premenopausal, who reported hysterectomy or oophorectomy (both bilateral and unilateral), women who died, and women who were diagnosed with cancer (except basal and squamous cell carcinoma). Death was confirmed by next-of-kin and postal authorities or by searching the National Death Index. Cancer diagnosis was identified from self-report and confirmed via medical records or from cancer registries. The residential address of each participant were geocoded to obtain latitude and longitude. We assumed that each woman who moved did so at the beginning of the questionnaire cycle. Monthly levels of particulate matter (PM) with an aerodynamic diameter less than or equal to 10 microns (PM 10 ), between 2.5 to 10 microns (PM 2.5-10 ), and less than or equal to 2.5 microns (PM 2.5 ) at residential addresses in the continental US were obtained from spatio-temporal prediction models ( Yanosky et al., 2014 ). These models used monthly monitoring data from several monitoring networks across the US and geospatial predictors, including road proximity, elevation, urban land use, population density, point sources of PM, and meteorological variables such as monthly average wind speed, temperature, and total precipitation, to predict monthly PM 10 and PM 2.5 between January 1989 to December 2007. Monthly PM 2.5-10 was calculated by subtracting the predicted PM 2.5 from PM 10 at each location. Cross-validation results showed good predictive performance with strong agreement between the predictions and observed values. We used these monthly predictions to calculate the annual means for PM 10 , PM 2.5-10 , and PM 2.5 , respectively, at each address. Predictions for 2007 were carried forward for all subsequent years. We created 3 time-varying long-term exposure metrics for each PM fraction: the cumulative average from age 40 with a 1 year lag to represent exposures during mid-adulthood; the average from age 40 to 45 to represent exposures in early mid-adulthood (cumulative average from age 40 with a 1 year lag when participants were still under 45); and the annual average in the previous year to represent recent exposures. Roadway proximity was used to represent traffic-related exposures and was measured by distance to major roadways for each residential address using the ESRI StreetMap Pro 2007 road segment data. We defined major roadways using the US Census Feature Class Codes: A1 (primary roads, typically interstate highways with limited access, division between the opposing directions of traffic, and defined exits), A2 (non-interstate highways and major roads without limited access), and A3 (secondary roads usually with more than 2 lanes). Distance to the largest (A1), the two largest (A1-A2), and all (A1-A3) road segments were calculated at age 40 and in the previous year to represent early mid-adulthood and recent exposure, and was categorized by 0-49 m, 50-199 m, 200-499 m, and 500 m or above, based on the distribution of exposure in the cohort and exposure studies ( Karner et al., 2010 ). Time-varying and invariant information on potential confounders were collected from the follow-up questionnaires every 2 years (every 4 years for diet and physical activity). For this analysis, we considered covariates that have been suggested to be associated with age at natural menopause and/or with exposure to PM or roadway proximity: body mass index (BMI, kg/m 2 ), smoking status (never, past, current) and intensity (< 25 or ≥ 25 cigarettes/day), physical activity (metabolic equivalent task (MET) hours/week), marital status, alcohol consumption (0 g/day, 0.1-4.9 g/day, 5.0-14.9 g/day, and ≥ 15.0 g/day), US Census Bureau regions of residence (Northeast, Midwest, West, and South), history of rotating shift work (never, past, current), oral contraceptive use (never, past, current), pre- or peri-menopausal use of hormone therapy (never, past, current), parity (nulliparous, 1-2, 3 or more full-term pregnancies), history of breastfeeding (less than 1 month, 1-12 months, 13-24 months, and more than 24 months), age at first birth (under 20, 20-25, 26-30, and ≥ 31 years old). Overall diet quality was measured via the 2010 Alternate Healthy Eating Index (AHEI) ( Chiuve et al., 2012 ). We also controlled for diagnosis of uterine fibroids and endometriosis, which are major indicators for receiving hysterectomy and oophorectomy, to reduce bias by censoring on the report of both procedures ( Whiteman et al., 2008 ; Wright et al., 2013 ). Neighborhood socioeconomic status was measured by Census tract-level population density, median income, and median home values. Race/ethnicity and age at menarche were also included as time invariant covariates. Information on menstrual cycle length at age 18-22 was provided at baseline. All missing values were modeled using missing indicators. We further assessed exposures to residential greenness measured by the Normalized Difference Vegetation Index (NDVI) in July of each year from the Advanced Very High Resolution Radiometer satellite at 1 km resolution. We calculated person-years for all participants from age 40 until self-reported natural menopause, censoring, or the return of the 2015 questionnaire, whichever came first. We further excluded participants without addresses geocoded to the street level for the analysis of roadway proximity. Then we computed the hazard ratios (HRs) and 95% confidence intervals (CIs) of natural menopause for a 10 μg/m 3 increase in each PM fraction and in each exposure time window, separately, using a time-varying Cox proportional hazard model with age as the time scale. All models were stratified by participants’ age in 1989 to control for possible time trends. After assessing the correlations between size fractions, we fitted a two-pollutant model including PM 2.5-10 and PM 2.5 for each exposure time window. Associations of roadway proximity with age at natural menopause were also computed as HRs and 95% CIs for both continuous and categorical variables. For each exposure measure, we fitted a basic model with age and calendar years only, a parsimonious model additionally adjusting for BMI, smoking, race/ethnicity, region, and socioeconomic status, and a full model adjusting for all covariates. Non-linear exposure-response relationships were examined using cubic splines for all continuous exposure measures. Hazard ratios greater than 1 indicate that the exposure is associated with earlier menopause, and HRs less than 1 indicate an association with later menopause. We also predicted the median age at menopause from the models and compared the predicted menopausal age between PM levels at the 25 th to 75 th percentile, and between the nearest category (0-49 m) of distance to major roads to the furthest category (≥ 500 m). To examine the robustness of our findings, we conducted sensitivity analyses by ending the follow-up period in 2009 to avoid carryforward of exposure predictions; restricting to women who never changed their residential addresses; and restricting to women who never reported using hormone therapy and who never used oral contraceptives after age 40. As air pollution and greenspace exposure were negatively correlated and one previous study has linked residential greenness with later menopause, we further adjusted analyses for residential greenness ( Triebner et al., 2019 ). To control for possible selection bias from exclusion before age 40, we calculated stabilized inverse probability weights of being included at age 40 using exposures and covariates measured at the baseline year (1989), and incorporated these personal weights in the Cox model, respectively. We also explored effect modification by region, smoking status, BMI (< 25, 25-29.9, and ≥ 30 kg/m 2 ), physical activity ( 27 MET hours/week), menstrual cycle length at age 18-22 (≤ 25 days, 26-31 days, 32-39 days, and ≥ 40 days or irregular), and Census tract level population density (< 1,000 and ≥ 1,000 people per km 2 ) in separate models, and statistical significance was examined by adding multiplicative interaction terms to the model. All statistical tests were 2-sided with an α level of 0.05. All analyses were conducted in SAS 9.4.

In this large, prospective, nationwide female cohort, we observed an modest association of long-term exposure to ambient particulate air pollution (PM 10 , PM 2.5-10 , and PM 2.5 ) after age 40 with earlier onset of menopause. Similarly, we also found women who lived close to all major roadways (< 50 m) at age 40 and in the previous year had slightly earlier menopause compared to women living further away (≥ 500 m). Our results were robust in sensitivity analyses. In addition, slightly stronger associations of PM and earlier menopause were found among never smokers and among women who lived in the West; however, there was no evidence of effect modification by body weight, physical activity, and population density. No evidence of effect modification was observed for roadway proximity. To our knowledge, this is the first study examining the associations between ambient air pollution and roadway proximity with age at natural menopause. An earlier age at menopause has been suggested to be associated with higher risk of cardiovascular diseases, osteoporosis, depression, and shorter life expectancy ( Atsma et al., 2006 ; Hu et al., 1999 ; Jung et al., 2015 ; Kuh et al., 2016 ; Muka et al., 2016 ; Savonitto et al., 2018 ; Shuster et al., 2010 ; Sullivan et al., 2017 ; Xu et al., 2020 ). It is believed that natural menopause occurs when the number of follicles reduces to a critical threshold of around 1,000, while other evidence suggests that the timing of natural menopause is also determined by the rate of follicle atresia ( Thomford et al., 1987 ). One study has examined the associations of residential PM 2.5 exposures with antral follicle counts among women (mean age of 35) from a fertility clinic. This study found associations of a 2 μg/m 3 increase of 3-month average PM 2.5 with a 7.2% reduction of antral follicle counts, and this effect was comparable with the effect of a 2-year increase of age ( Gaskins et al., 2019 ). Several studies in animals also observed PM or diesel exhaust from traffic were associated with significant reductions in ovarian reserve ( Gai et al., 2017 ; Liao et al., 2020 ; Ogliari et al., 2013 ; Veras et al., 2009 ). A recent cross-sectional study showed inverse associations of PM exposure and road proximity with ovarian reserve hormone levels among 67 women ( Abareshi et al., 2020 ). Ovarian follicle atresia is mainly driven by the cyclic change of hormone milieu in the ovary, and can also be triggered by inflammation and oxidative stress ( Agarwal et al., 2012 ; Boots and Jungheim, 2015 ; Kaipia et al., 1996 ; Lu et al., 2018 ). One study among 820 NHS II participants showed associations of blood inflammatory markers and early menopause (menopause before age 45) ( Bertone-Johnson et al., 2019 ). Abundant evidence suggests that PM and traffic related exhaust can promote systemic inflammation and oxidative stress ( Brook et al., 2010 ; Mazzoli-Rocha et al., 2010 ; Rao et al., 2018 ). Other studies also demonstrated chemical components in particulate air pollution and diesel exhaust (e.g. polycyclic aromatic hydrocarbon) may induce oocyte toxicity and thus promote follicular atresia and arrest ( Borman et al., 2000 ; Meek, 1998 ; Misaki et al., 2008 ; Mori et al., 2002 ; Oh et al., 2008 ; Takeda et al., 2004 ). However, more mechanistic evidence is needed, especially evidence focusing on potential mechanistic pathways (e.g. DNA damage, inflammation, and oxidative stress) linking PM exposure and oocyte survival, and also on exposure in midlife. Besides, as evidence has shown that endocrine disrupting chemicals can adversely affect follicle development and ovulation, it is important to understand whether PM and its endocrine disrupting constituents can induce similar effects and which stages are the most susceptible ( Gore et al., 2015 ). We observed consistent associations of slightly earlier menopause with exposure to PM 10 , PM 2.5-10 , and PM 2.5 exposures at age 40-45, and with roadway proximity in age 40, a time period that is considered to be the late reproductive and early perimenopausal stage in women’s reproductive lifespan ( Harlow et al., 2012 ). It has also been suggested that the rate of follicle atresia accelerates after age 37 in most women possibly as a consequence of accumulation of cellular damage and lower damage repair capacity with age ( Djahanbakhch et al., 2007 ; Faddy, 2000 ; Faddy and Gosden, 1996 ; Wallace and Kelsey, 2010 ). Therefore, our results may indicate PM and traffic-related pollution could accelerate the already expedited follicle loss in early midlife. Meanwhile, when considering exposure in the previous year, only PM10 and road proximity were found to be associated with menopause. This may further indicate PM and traffic-related exposure could lead to excess atresia in addition to the accumulated aging-related damage in oocytes. Although we observed statistically significant associations of PM and roadway proximity with menopause, these associations were quite modest when compared to the HRs of other potential predictors of age at natural menopause in the model (e.g. smoking) ( Table S3 ). The predicted median age at menopause in our analysis was less than 1 month younger for women with PM exposure at the 75 th percentile to the 25 th percentile, while previous literature has suggested women who smoke had 1-2 years younger menopausal age ( Gold, 2011 ) . On the other hand, we observed stronger and statistically significant associations of PM with earlier menopause in never smokers but not in past or current smokers. It is possible that the stronger association of smoking with earlier menopause masked the modest impact of PM. Besides, PM 2.5 was shown to have slightly stronger associations with earlier menopause among women who lived in the West. This effect heterogeneity of PM 2.5 was consistent with a cohort study in the US, which showed PM 2.5 was associated with higher breast cancer risk only among women in the west and suggested this heterogeneity may be due to the differences in chemical compositions of PM 2.5 by region ( White et al., 2019 ). Our study had a few notable limitations. First, we had to rely on self-reported menopause queried as whether the participants’ periods had ceased permanently in the questionnaire. This defining criterion of menopause was less accurate than the definition commonly used in other epidemiological studies as the cessation of menstruation for at least 12 consecutive months ( World Health Organization, 1996 ). Besides, as this information was collected every two years, we had very limited ability to identify the exact timing of menopause, which would lead to increased measurement error, weakening our ability to detect effects. However, one previous validation study in the Nurses’ Health Study suggested the prospective assessment of menopausal age is quite robust, with a 98.8% consistency in self-reported menopause status and types of menopause and a mean difference of 0.06 years in reported age at natural menopause over a two-year period ( Colditz et al., 1987 ). Second, exposure measurement error was possible. Although the predicted PM from our geospatial models revealed satisfactory agreement with real time monitoring data, there may be misclassification due to lack of information on the amount of time spent at home and the use of household air filtration by participants ( Kioumourtzoglou et al., 2014 ). Additionally, because all exposure measurements were only available until 2007, it is likely that there is more measurement error in the subsequent years after as we carried the predictions forward. However, our sensitivity analysis ending the follow-up in 2009 showed very similar HRs. Third, the very high correlations between different exposure time windows for all PM fractions, and the high correlations of PM 10 with both PM 2.5-10 and PM 2.5 may limit our ability to fully distinguish the actual critical effect time window for air pollution and age at menopause, especially between the cumulative average and exposure in age 40-45, and which PM fraction contributed most to the observed associations. In addition, as the exposure was calculated after age 40, associations of air pollution exposure in earlier life periods such as childhood and adolescent, were not examined. Early exposure may also contribute to left censoring in our data. Although our sensitivity analysis correcting for baseline exclusion supported robustness of the main findings, our earliest exposure measure may not fully capture early exposure, especially in potentially important developmental windows. We were also unable to examine the association of exposure to other major gaseous air pollutants with the outcome due to data availability, although road proximity may partly reflect exposures to nitric oxides and traffic noise. Finally, compared to the general female population in the US, our participants were mostly white, worked as health professionals, and had relatively high socioeconomic status. Therefore, our ability to examine potential disparities by race was limited and our results may only apply to white women with similar characteristics rather than the general female population in the US or in other countries with higher levels of air pollution. Despite the limitations, this large nation-wide cohort provided us with sufficient power to detect a modest association of PM and road proximity with age at natural menopause. With more than 90% of the NHS II participants were pre-menopausal in 1989 and around 1% remained premenopausal in 2015, we were able to capture the entire menopausal transition for the majority of the participants over 20 years of follow-up. We had detailed information on the participants’ lifestyles, reproduction history, and health status updated biennially, which allowed adjustment of time-varying confounding in the analysis. Finally, with the high spatio-temporal resolution PM models and residential history updated continuously, we were able to examine time-varying exposures in different life stages in the analysis.

Exposure to higher levels of ambient particulate air pollution and living near to major roadways after age 40 were modestly associated with earlier onset of natural menopause in this large, prospective, and nationwide cohort of female nurses in US. Exposure from age 40-45, which corresponds to the late reproductive and early premenopausal stage, showed more consistent associations with earlier menopause compared to later exposure. We also observed some effect heterogeneity by smoking and by region. A previous study suggested a 3% increase of cardiovascular disease risk for a 1-year younger menopausal age, therefore, the observed associations of PM with age at menopause are unlikely be of important clinical significance ( Hu et al., 1999 ). Nevertheless, our results were consistent with previous observations of air pollution and traffic with accelerated aging. As this is the first study of air pollution and traffic with age at natural menopause, more evidence is needed for confirmation.

Natural menopause is clinically defined as the cessation of menstruation for at least 12 consecutive months ( World Health Organization, 1996 ). It is a natural consequence of ovarian aging and age related oocyte depletion. The current median age at natural menopause in the US has been reported to be approximately 51-52 years and it varies by geographic region and by race/ethnicity ( Gold et al., 2013 ; McKnight et al., 2011 ; Morabia and Costanza, 1998 ; Thomas et al., 2001 ). So far, it has been suggested that genetics and a number of modifiable factors, including smoking, body weight, diet, physical activity, reproductive health behaviors, and socioeconomic status, may explain part of this variation ( Gold, 2011 ; Gold et al., 2013 ; He et al., 2009 ; He and Murabito, 2014 ; Lujan-Barroso et al., 2018 ; Nagel et al., 2005 ; Pyun et al., 2014 ; Schoenaker et al., 2014 ; Zhu et al., 2018 ). However, evidence on the possible influence of ambient environment exposure with age at natural menopause remains scarce and only one recent study in Europe investigated the association between an environmental exposure, surrounding greenness, and age at menopause ( Triebner et al., 2019 ). Previous studies have shown that air pollution and traffic-related exposures were associated with shorter telomere length, a biomarker of aging ( Martens and Nawrot, 2018 , 2016 ; Miri et al., 2019 ). Although the mechanisms are still not clear, two studies have linked shorter telomere length with earlier age at menopause in women ( Gray et al., 2014 ; Shenassa and Rossen, 2015 ). However, to the best of our knowledge, no study has examined the association of air pollution and traffic exposure with age at natural menopause. Our primary objective was to investigate these associations in a large US prospective cohort study – the Nurses’ Health Study II (NHS II). We compared the associations in different exposure time windows and explored whether the associations were modified by lifestyle factors that have been previously associated with age at menopause.