Female reproductive history in relation to chronic obstructive pulmonary disease and lung function in UK biobank: a prospective population-based cohort study

Objectives Sex differences in respiratory physiology and predilection for developing chronic obstructive pulmonary disease (COPD) have been documented, suggesting that female sex hormones may influence pathogenesis. We investigated whether aspects of female reproductive health might play a role in risk of COPD among women. Design Population-based prospective cohort study. Setting UK Biobank recruited across 22 centres in the UK between 2006 to 2010. Primary and secondary outcomes measures We examined a range of female reproductive health indicators in relation to risk of COPD-related hospitalisation/death (n=271 271) using Cox proportional hazards regression; and lung function (n=273 441) using linear regression. Results Parity >3 was associated with greater risk of COPD-related hospitalisation/death (adjusted HR 1.45; 95% CI: 1.16 to 1.82) and lower forced expiratory volume at 1 second/forced vital capacity ratio (FEV1/FVC) (adjusted mean difference −0.06; 95% CI: -0.07 to 0.04). Any oral contraception use was associated with lower risk of COPD-related hospitalisation/death (adjusted HR 0.85; 95% CI: 0.74 to 0.97) and greater FEV1/FVC (adjusted mean difference 0.01; 95% CI: 0.003 to 0.03). Late menarche (age >15) and early menopause (age <47) were also associated with greater risk of COPD-related hospitalisation/death (but not lung function), while endometriosis was associated with greater FEV1/FVC (not COPD-related hospitalisation/death). Early menarche (age <12 years) was associated with lower FEV1/FVC (but not COPD hospitalisation/death). Associations with polycystic ovary syndrome (PCOS) or ovarian cysts, any hormone replacement therapy (HRT) use, hysterectomy-alone and both hysterectomy and bilateral oophorectomy were in opposing directions for COPD-related hospitalisation/death (greater risk) and FEV1/FVC (positive association). Conclusions Multiple female reproductive health indicators across the life course are associated with COPD-related hospitalisation/death and lung function. Further studies are necessary to understand the opposing associations of PCOS/ovarian cysts, HRT and hysterectomy with COPD and objective measures of airway obstruction.

This is also reflected in the risk of obstructive lung disease. For example, boys have greater risk of asthma during early childhood compared to girls, but this sex difference reverses after puberty, when girls are found to have the greater risk of new-onset asthma. 2 Chronic obstructive pulmonary disease (COPD) has traditionally been seen as a disease predominantly affecting men. However, in more recent years, the overall burden of the disease has shifted more towards women. 2 3 Women with COPD also have, on average, fewer pack-years of smoking history than men with COPD; and nonsmokers with COPD are more often female. 1 This points towards a greater underlying susceptibility among women, and highlights the importance of understanding female-specific risk factors for COPD.
Female sex hormones are putative mediators of respiratory health, and regulate bronchodilation, cell proliferation, inflammation and metabolism of toxic cigarette smoke-related metabolites. 1 4-7 Notably, oestrogen and progesterone appear to have varying effects on lung function across the lifespan. 1 We know little, however, about female reproductive health indicators in relation to COPD risk. 1 These reproductive health indicators include markers of endogenous hormone exposure (age at menarche, age at menopause and length of reproductive lifespan), exogenous hormone exposure (oral contraceptive use and hormone replacement therapy), as well as pregnancy history. Of two existing studies, one study found no evidence of an association between hormone replacement therapy (HRT) and COPD risk, 8 while another reported a positive association between early age at first pregnancy (<20 years) and COPD-related mortality. 9 There is more evidence regarding the relationship between female reproductive health indicators and lung function. Early menarche, 10-12 nulliparity, 13 younger age at first pregnancy, 12 and postmenopausal status [14][15][16][17][18][19] have each been associated with reduced lung function. Evidence of the effects of HRT is contradictory. 14 15 20 21 In the present study, we therefore investigate associations of multiple female reproductive health indicators across the lifespan with incidence of COPD hospitalisation/death and lung function in the UK Biobank cohort.

Study population
Our study included female participants in UK Biobank (Figure 1), a population-based cohort study which recruited men and women age 40-69 years across 22 centres in England, Scotland and Wales between 2006-2010. 22 The response rate was 5.5%. 22 All participants provided written informed consent. Data on 273 441 women were available after withdrawals. At baseline, participants selfreported sociodemographic information, family medical history, early-life exposures, psychosocial history and medical history. Physical measurements included spirometry. Information on health outcomes after recruitment into the cohort was gathered prospectively via data linkages with hospital and death registers. Register data were available up to 28
We used linear regression in the analysis of standardised spirometry measurements. The multivariable linear regression models were adjusted for the same covariates as for hospitalisation/death from COPD, with the addition of asthma at baseline. We reported results as mean difference in the spirometry measurements with 95% CI.
We performed multiple imputation by chained equations to account for missing data. Exposures, covariates and outcomes were included in the imputation model. We imputed 20 datasets, which were then analysed using Rubin's rules. 24 Details of the imputation model are given in the online supplementary file (Table S1). All analyses were performed using Stata/MP 15 (StataCorp LLC, College Station, USA). Sensitivity analyses of the complete cases was performed for comparison with our imputed data.

RESULTS
Of the 273 441 women included in this study, 2170 had prevalent COPD at baseline (Figure 1). These prevalent cases were excluded from the analysis of incident COPD-related hospitalisation/death. By the end of follow-up (median 6 years), there were 1138 incident cases of hospitalisation/death from COPD, yielding an incidence rate of 0.19 per 100 000 person years.
In UK Biobank, 28.4% of women did not have spirometry data, 16.1% were uncertain of or did not report menopause status and 14.3% did not report smoking history. The proportion of missing information on the other covariates were modest (Table 1). Overall, women with spirometry data were younger, more often White, had higher educational attainment, higher annual household income, less deprivation (lower Townsend index), greater smoking history and were more likely to have asthma than women who did not undergo spirometry assessment (online supplementary file Table S2). The distribution of participants' characteristics across the imputed datasets is shown in Table 1.
Risk of hospitalisation/death from COPD did not differ between pre-and postmenopausal women (    (Table 3). These associations did not change after further adjustment for HRT ever use (results not shown).
Hospitalisation/death from COPD did not differ between parous and nulliparous women, but parity >3 was associated with increased risk (HR 1.45; 95% CI: 1.16, 1.82) compared to nulliparity. Women who were nulliparous had lower FEV 1 and FVC but greater FEV 1 /FVC compared to parous women, while women with parity >3 had greater FEV 1 and FVC but lower FEV 1 /FVC compared to nulliparous women (Table 3).
A history of PCOS or ovarian cysts was associated with increased risk of hospitalisation/death from COPD (HR 1.61; 95% CI: 1.12, 2.32) and greater FEV 1 /FVC (mean difference 0.05; 95% CI: 0.02, 0.09), while there was no evidence of an association with FEV 1 or FVC. A history of endometriosis was not associated with COPD hospitalisation/death, FEV 1 or FVC (Tables 2 and 3), though we observed a modest positive association with FEV 1 /FVC (mean difference 0.07; 95% CI: 0.03, 0.10).
We observed several associations between exogenous hormone use and COPD hospitalisation/death. Women who had ever used OC had lower risk of hospitalisation/death from COPD compared to never users (HR 0.85; 95% CI: 0.74, 0.97) ( Table 2). This appeared to be driven by OC use of >9 years (Table 2). In line with the observed findings for COPD, OC ever use and duration of use showed positive associations with FEV 1 , FVC and FEV 1 /FVC (Table 3).
Ever using HRT was associated with a modest increased risk of hospitalisation/death from COPD (HR 1.15; 95% CI: 1.01, 1.30) ( Table 2). We observed no associations of HRT ever use with FEV 1 or FVC, but a positive association with FEV 1 /FVC (Table 3). When we examined the duration of HRT use, we only observed increased hospitalisation/death from COPD (HR 1.33; 95% CI: 1.12, 1.57) and greater FEV 1 /FVC (mean difference 0.02; 95% CI: 0.01, 0.04) with 1-2 years of use (Tables 2 and 3). There was no evidence of a dose-response relationship. in FEV 1 . 26 Spirometry assessment at one time point may therefore not be as accurate at identifying clinically-relevant COPD compared to register data on hospitalisation/death.
A study of 424 797 postmenopausal women in the United States reported no difference in COPD mortality by parity. 9 In contrast, our present study reports higher risk of COPD hospitalisation/death among women with more than three births compared to nulliparous women. However, we investigated both COPD-related hospitalisation and death, and we were therefore able to capture less severe COPD cases. A study of 70 965 women in the Nurses' Health Study reported no evidence of an association between HRT ever use and COPD incidence, although the investigators included both probable (self-reported physician diagnosis along with a record of spirometry or imaging at the time of diagnosis or COPD documented on a death register) and definite (self-reported physician diagnosis and FEV 1 <80% predicted) cases of COPD. 8 Here we report a positive association between HRT use and COPD hospitalisation/death. This is the first study to assess COPD risk with regards to age at menarche, menopause status, hysterectomy, gynaecological surgery, endometriosis or PCOS.
Two previous studies reported reduced FEV 1 in adult women who experienced menarche before age 10 10 12 but neither found evidence of an association between age at menarche and FEV 1 /FVC. A Mendelian randomisation study of adolescent and adult women across multiple European cohorts, including a subset from UK Biobank, estimated a 24.8 mL increase in FVC per year increase in age at menarche among adult women and a -56.6 mL decrease in FVC per year increase in age at menarche among adolescent girls. 11 This study reported no association between age at menarche and FEV 1 /FVC. 11 Notably, this Mendelian randomisation study did not explore non-linear associations between age at menarche and measures of lung function. We observed lower FEV 1 and FVC in adult women with late menarche, and lower FVC and FEV 1 /FVC in adult women with early menarche.
A recent systematic review, which included studies from UK Biobank, 16 reported that postmenopausal status was associated with reduced FEV 1 and FVC but not FEV 1 /FVC. 18 Current HRT use has been associated with both higher 14 and lower 15   association between HRT use and greater FEV 1 /FVC. 14 Others have identified differences in lung function according to continuous/cyclic and combined/oestrogen-only HRT. 20 21 Based on the information from participants in UK Biobank, we could not stratify by HRT subtypes.
There are sex differences in lung anatomy and physiology throughout the life course. 1 2 4 During gestation, female foetuses have fewer bronchi, smaller lungs and produce surfactant earlier than male foetuses. 1 Female neonates and children have greater lung function and faster lung growth than their male counterparts, and boys are more likely to be diagnosed with asthma during childhood than girls, while girls are twice as likely as boys to be diagnosed with asthma after puberty. 1 These findings suggest that female sex steroids can accelerate lung maturation and termination of lung growth, which may partially explain why early menarche associates with lower FEV 1 /FVC in adulthood. During the reproductive years, parity and PCOS relate to an increase in circulating oestrogen, which has been implicated in the pathogenesis of inflammatory lung diseases including COPD. 6 27-29 Although OC use appears to be protective against COPD, some researchers have suggested that OC use attenuates the fluctuations in sex steroids, which correlates with fluctuations in endothelial nitric oxide, a potent bronchodilator. 1 Additionally, exogenous hormones may not be as influential during reproductive years but may become more influential in after menopause, when endogenous hormone production is low. 6 This is supported by our findings of opposing associations with COPD risk for OC use (lower risk) during reproductive period and HRT use (higher risk) after menopause.
The effect of female sex hormones on COPD and lung health may vary depending on the net effects on airway smooth muscle, inflammation and cigarette smoke metabolism. Oestrogens can regulate the synthesis of nitric oxide, a potent bronchodilator, within human bronchial epithelium, 1 30 as well as regulate airway bronchoconstriction in response to acetylcholine or histamine. 31 32 There is also evidence that progesterone may affect airway smooth muscle constriction, although these reports are limited to animal studies. 33 In addition to regulating bronchodilation and bronchoconstriction, in vitro studies report that oestrogen can also modulate the proliferation of airway smooth muscle cells, 34 35 which is a key feature of COPD. 36 Oestrogen's pro-and anti-inflammatory effects have been documented in other female-predominant diseases 37 and may also play a role in COPD, as chronic inflammation is another hallmark of the disease. Finally, oestrogen is an important regulator of toxic cigarette smoke metabolites through cytochrome P450 enzymes. 1 In a mouse model of COPD, Glassberg et al. reported that oestradiol administration after ovariectomy protected against cigarette smoke-induced alveolar septal destruction, macrophage infiltration and decreased other markers of inflammation compared to ovariectomy without oestradiol administration. 38 In the same mouse model, however, non-ovariectomised female mice also had increased small airway remodelling following cigarette smoke exposure when compared to both male mice and ovariectomised female mice. 39 These timing-, dose-and cell-dependent signalling pathways may explain why exposure to female sex steroids, and oestrogen in particular, have potentially contradictory effects on respiratory health.
We cannot exclude a potential influence of selection bias, due to the low participation rate in UK Biobank. Women in the cohort are less likely to have ever smoked, have greater educational attainment, but a similar number of offspring, as compared to the general UK population (https://www.ons.gov.uk/). There is also a lower prevalence of self-reported COPD in UK Biobank (0.8%) compared to physician-diagnosed COPD in the general female UK population (1.68%). 40 Overall, women participating in UK Biobank reflect a higher-educated, more health-conscious group, with more modest underlying risk for COPD compared to the general population. We can only speculate as to how these selection factors might have influenced our associations of interest, but we propose that the most likely direction of any selection bias is towards the null (resulting in an underestimation). Strengths of this study include the large, prospective design. However, we relied on participants to self-report reproductive history which might have led to measurement error.
Years of OC or HRT use were estimated as the difference between ages at first and last use. This likely resulted in overestimation of the duration of use, which might have attenuated our findings. In addition, we were unable to distinguish between reversible and non-reversible airway obstruction at baseline, as a bronchodilator was not administered before spirometry was conducted. Although we adjusted for self-reported asthma at baseline, this may have led to overestimation of COPD-related airway obstruction among participants with undiagnosed asthma.
Associations with PCOS or ovarian cysts; any HRT use; hysterectomy-alone; and both hysterectomy and bilateral oophorectomy were in opposing directions for COPD-related hospitalisation/death (greater risk) and FEV 1 /FVC (positive association).   We relied on participants to self-report reproductive history and we cannot exclude a potential influence of selection bias, due to the low participation rate in UK Biobank.

Conclusions
 Data on environmental tobacco smoke were not available and we cannot exclude residual bias from this exposure.

INTRODUCTION
There are clear sex differences in lung anatomy and physiology throughout the life course. 1 This is also reflected in the risk of obstructive lung disease. For example, boys have greater risk of asthma during early childhood compared to girls, but this sex difference reverses after puberty, when girls are found to have the greater risk of new-onset asthma. 2 Chronic obstructive pulmonary disease (COPD) has traditionally been seen as a disease predominantly affecting men. However, in more recent years, the overall burden of the disease has shifted more towards women. 2 3 Women with COPD also have, on average, fewer pack-years of smoking history than men with COPD; and nonsmokers with COPD are more often female. 1 This points towards a greater underlying susceptibility among women, and highlights the importance of understanding female-specific risk factors for COPD.
Female sex hormones are putative mediators of respiratory health, and regulate bronchodilation, cell proliferation, inflammation and metabolism of toxic cigarette smoke-related metabolites. 1 4-7 Notably, oestrogen and progesterone appear to have varying effects on lung function across the lifespan. 1 We know little, however, about female reproductive health indicators in relation to COPD risk. 1 These reproductive health indicators include markers of endogenous hormone exposure (age at menarche, age at menopause and length of reproductive lifespan), exogenous hormone exposure (oral contraceptive use and hormone replacement therapy), as well as pregnancy history. Of two existing studies, one study found no evidence of an association between hormone replacement therapy (HRT) and COPD risk, 8 while another reported a positive association between early age at first pregnancy (<20 years) and COPD-related mortality. 9 There is more evidence regarding the relationship between female reproductive health indicators and lung function. Early menarche, [10][11][12] nulliparity, 13 younger age at first pregnancy, 12 and postmenopausal status [14][15][16][17][18][19] have each been associated with reduced lung function. Evidence of the effects of HRT is contradictory. 14 15 20 21 In the present study, we therefore investigate associations of multiple female reproductive health indicators across the lifespan with incidence of COPD hospitalisation/death and lung function in the UK Biobank cohort.

Study population
Our study included female participants in UK Biobank (Figure 1), a population-based cohort study which recruited men and women age 40-69 years across 22 centres in England, Scotland and Wales between 2006-2010. 22 The response rate was 5.5%. 22
Amaral et al. previously reported decreased lung function in postmenopausal women in UK Biobank. 16 We repeat these analyses for completeness alongside analyses of other female reproductive health indicators.

Respiratory outcomes
Prevalent COPD at baseline was identified using self-report and registrations before baseline in the hospital registers. Incident hospitalisations/deaths from COPD after baseline were identified using hospital and death registers. The following International Classification of Diseases (ICD) codes were used to capture COPD: ICD-9 codes 490-492, 494 and 496; or ICD-10 codes J40-J44.
Forced expiratory volume at one-second (FEV 1 ) and forced vital capacity (FVC) were measured using Vitalograph Pneumotrac 6800 (Vitalograph, UK) at baseline. Spirometry was not performed if participants had experienced a chest infection in the last month; had a lifetime history of detached retina or collapsed lung; had experienced a heart attack, eye surgery or surgery to the chest or abdomen in the last 3 months; were currently pregnant in the first or third trimester; or were currently using medication for tuberculosis. Two measurements were conducted if the difference between these measurements was ≤5%. A third measurement was conducted if the difference was >5%. We used the single greatest measurements of FEV 1 and FVC per participant. Postbronchodilator spirometry was not available, although drug treatment was not withheld. Spirometry measurements were converted to internally standardised z-scores by age and height based on the Global Lung Initiative 2012 recommendations 23 to reduce bias related to age. These z-scores were used for all analyses.

Statistical analyses
We used Cox proportional hazards regression to examine the association of each female reproductive health indicator in relation to incidence of COPD hospitalisation/death. Participants without COPD at baseline were followed from enrolment until hospitalisation/death from COPD, death from other causes, or the end of follow-up if the participants were still alive and disease-free.
The time axis for the Cox regression was calendar time. We examined deviations from the proportional hazard assumption using Schoenfeld residuals. 24 Age-and confounder-adjusted estimates are reported as hazard ratios (HR) with 95% confidence intervals (95% CI). The following potential confounders were included in multivariable analyses: age, height, log-transformed body  We used linear regression in the analysis of standardised spirometry measurements. The multivariable linear regression models were adjusted for the same covariates as for hospitalisation/death from COPD, with the addition of asthma at baseline. We reported results as mean difference in the spirometry measurements z-scores with 95% CI. These mean differences in zscores can be converted to raw values by multiplication with the standard deviation in spirometry measurement for the given age and height group. Sensitivity analyses for standardised spirometry measurements were also conducted with: 1) menopause status excluding participants age <45 years or >60 years at baseline; 2) smoking history adjusted as two covariates (duration in years and average number of cigarettes/day); 3) further adjustment for baseline smoking status; and 4) further adjustment for comorbidities baseline (cardiovascular disease or diabetes).
We performed multiple imputation by chained equations to account for missing data. Exposures, covariates and outcomes were included in the imputation model. We imputed 20 datasets, which were then analysed using Rubin's rules. 26 Details of the imputation model are given in the online supplementary file (Table S1). All analyses were performed using Stata/MP 15 (StataCorp LLC, College Station, USA). Sensitivity analyses of the complete cases was performed for comparison with our imputed data.

Patient and public involvement
The present study did not involve patients and the public.

RESULTS
Of the 273 441 women included in this study, 2170 had prevalent COPD at baseline ( Figure 1). These prevalent cases were excluded from the analysis of incident COPD-related hospitalisation/death. By the end of follow-up (median 6 years), there were 1138 incident cases of hospitalisation/death from COPD, yielding an incidence rate of 0.19 per 100 000 person years. In UK Biobank, 28.4% of women did not have spirometry data, 16.1% were uncertain of or did not report menopause status and 14.3% did not report smoking history. The proportion of missing information on the other covariates were modest (Table 1). Overall, women with spirometry data were younger, more often White, had higher educational attainment, higher annual household income, less deprivation (lower Townsend index), greater smoking history and were more likely to have asthma than women who did not undergo spirometry assessment (online supplementary file Table S2). The distribution of participants' characteristics across the imputed datasets is shown in Table 1.
Risk of hospitalisation/death from COPD did not differ between pre-and postmenopausal women ( Postmenopausal status at baseline and early natural menopause were each associated with lower FEV 1 and FVC but not FEV 1 /FVC (Table 3). These associations did not change after further adjustment for HRT ever use (results not shown). After restricting the sample to participants who were age 45-60 years (inclusive) at the time of reporting menopause status to reduce residual bias due to age, we found that menopause status was positively associated with COPD-related hospitalisation/death (HR  (Tables 2 and 3), though we observed a modest positive association with FEV 1 /FVC (mean difference 0.07; 95% CI: 0.03, 0.10).
We observed several associations between exogenous hormone use and COPD hospitalisation/death. Women who had ever used OC had lower risk of hospitalisation/death from COPD compared to never users (HR 0.85; 95% CI: 0.74, 0.97) ( Table 2). This appeared to be driven by OC use of >9 years (Table 2). In line with the observed findings for COPD, OC ever use and duration of use showed positive associations with FEV 1 , FVC and FEV 1 /FVC (Table 3).
Ever using HRT was associated with a modest increased risk of hospitalisation/death from COPD (HR 1.15; 95% CI: 1.01, 1.30) ( Table 2). We observed no associations of HRT ever use with FEV 1 or FVC, but a positive association with FEV 1 /FVC (Table 3). When we examined the duration of HRT use, we only observed increased hospitalisation/death from COPD (HR 1.33; 95% CI: 1.12, 1.57) and greater FEV 1 /FVC (mean difference 0.02; 95% CI: 0.01, 0.04) with 1-2 years of use (Tables 2 and 3). There was no evidence of a dose-response relationship. A history of hysterectomy (with or without bilateral oophorectomy), but not bilateral oophorectomy-alone, were associated with greater risk of COPD hospitalisation/death (Table 2). At the same time, hysterectomy was associated with higher, rather than lower, FEV 1 /FVC (Tables 3). Hysterectomy with bilateral oophorectomy was also associated with lower FVC ( Table 3).
The directions of association between individual female reproductive health indicators with COPD risk and spirometry measures are summarised in Table 4. 55 929 women had reported either a respiratory illness (asthma, COPD, tuberculosis and/or pneumonia) and/or FEV 1 /FVC<0.7 at baseline.
However, our findings after excluding these women from our analyses of COPD-related hospitalisation/death were similar to our main findings (online supplementary file Table S3).
We report the results of our sensitivity analyses, after 1) further adjusting for baseline smoking status; 2) adjusting for smoking history as two covariates including duration in years of smoking and average number of cigarettes per day; 3) excluding women who have ever smoked; or 4) further adjusting for cardiovascular disease or diabetes reported at baseline, in the online supplementary file (Tables S4-S8). The direction and magnitude of associations remained similar for risk of COPD hospitalisation/death (online supplementary file Tables S4 and S6) and change in spirometry   measures (online supplementary file Tables S7 and S8) after further adjustment for the above covariates. In our analyses of never smokers (n=161 626), however, associations of COPD-related hospitalisation/death with early menopause, parity>3, PCOS/ovarian cysts, ever using oral contraception or hormone replacement therapy became null (online supplementary file Table S5).
Early age at menarche (<12 years) was also positively associated with COPD-related hospitalisation/death in this restricted sample of never smokers (HR 1.39; 95% CI: 1.02, 1.90), whilst the remaining associations remained unchanged from the main analyses.
We also investigated associations of reproductive health indicators with risk of COPD-related mortality, and not hospitalisation, among women of UK Biobank. The associations from these analyses of 273 441 women (including those with COPD reported at baseline) were null for all  Table S10) and lung function (online supplementary file Table S11) also produced similar results to those from our imputed dataset.

DISCUSSION
In this large, prospective study, several female reproductive health indicators were associated with COPD hospitalisation/death and/or lung function. Parity greater than three was associated with decreased baseline FEV 1 /FVC, a marker of airway obstruction, and increased risk of hospitalisation/death from COPD during follow-up. Ever using OC was associated with greater FEV 1 /FVC and lower risk of COPD hospitalisation/death. Associations of hysterectomy, HRT use, and a history of PCOS or ovarian cysts showed inconsistent findings for FEV 1 /FVC and COPD hospitalisation/death. This inconsistency may be explained by the fact that spirometry was measured in participants at baseline (age 40-69) and FEV 1 and FEV 1 /FVC may have declined during follow-up. Conversely, many patients diagnosed with stage 1 COPD do not go on to develop COPD symptoms or COPD resulting in hospitalisation/death. 27 There is also high intra-individual variability in FEV 1 . 28 Spirometry assessment at one time point may therefore not be as accurate at identifying clinically-relevant COPD compared to register data on hospitalisation/death.
A study of 424 797 postmenopausal women in the United States reported no difference in COPD mortality by parity. 9 In contrast, our present study reports higher risk of COPD hospitalisation/death among women with more than three births compared to nulliparous women. However, we investigated both COPD-related hospitalisation and death, and we were therefore able to capture less severe COPD cases. A study of 70 965 women in the Nurses' Health Study reported no evidence of an association between HRT ever use and COPD incidence, although the investigators included both probable (self-reported physician diagnosis along with a record of spirometry or imaging at the time of diagnosis or COPD documented on a death register) and definite (self-reported physician diagnosis and FEV 1 <80% predicted) cases of COPD. 8 Here we report a positive association between HRT use and COPD hospitalisation/death. This is the first study to assess COPD risk with regards to age at menarche, menopause status, hysterectomy, gynaecological surgery, endometriosis or PCOS.
Two previous studies reported reduced FEV 1 in adult women who experienced menarche before age 10 10 12 but neither found evidence of an association between age at menarche and FEV 1  A recent systematic review, which included studies from UK Biobank, 16 reported that postmenopausal status was associated with reduced FEV 1 and FVC but not FEV 1 /FVC. 18 Current HRT use has been associated with both higher 14 and lower 15 lung function. One study reported an association between HRT use and greater FEV 1 /FVC. 14 Others have identified differences in lung function according to continuous/cyclic and combined/oestrogen-only HRT. 20 21 Based on the information from participants in UK Biobank, we could not stratify by HRT subtypes.
There are sex differences in lung anatomy and physiology throughout the life course. 1  puberty. 1 These findings suggest that female sex steroids can accelerate lung maturation and termination of lung growth, which may partially explain why early menarche associates with lower FEV 1 /FVC in adulthood. During the reproductive years, parity and PCOS relate to an increase in circulating oestrogen, which has been implicated in the pathogenesis of inflammatory lung diseases including COPD. 6 29-31 Although OC use appears to be protective against COPD, some researchers have suggested that OC use attenuates the fluctuations in sex steroids, which correlates with fluctuations in endothelial nitric oxide, a potent bronchodilator. 1 Additionally, exogenous hormones may not be as influential during reproductive years but may become more influential in after menopause, when endogenous hormone production is low. 6 This is supported by our findings of opposing associations with COPD risk for OC use (lower risk) during reproductive period and HRT use (higher risk) after menopause.
The effect of female sex hormones on COPD and lung health may vary depending on the net effects on airway smooth muscle, inflammation and cigarette smoke metabolism. Oestrogens can regulate the synthesis of nitric oxide, a potent bronchodilator, within human bronchial epithelium, 1 32 as well as regulate airway bronchoconstriction in response to acetylcholine or histamine. 33 34 There is also evidence that progesterone may affect airway smooth muscle constriction, although these reports are limited to animal studies. 35 In addition to regulating bronchodilation and bronchoconstriction, in vitro studies report that oestrogen can also modulate the proliferation of airway smooth muscle cells, 36 37 which is a key feature of COPD. 38 Oestrogen's pro-and anti-inflammatory effects have been documented in other female-predominant diseases 39 and may also play a role in COPD, as chronic inflammation is another hallmark of the disease. Finally, oestrogen is an important regulator of toxic cigarette smoke metabolites through cytochrome P450 enzymes. 1 In a mouse model of COPD, Glassberg et al. reported that oestradiol administration after ovariectomy protected against cigarette smoke-induced alveolar septal destruction, macrophage infiltration and decreased other markers of inflammation compared to ovariectomy without oestradiol administration. 40 In the same mouse model, however, non-ovariectomised female mice also had increased small airway Overall, women participating in UK Biobank reflect a higher-educated, more health-conscious group, with more modest underlying risk for COPD compared to the general population. We can only speculate as to how these selection factors might have influenced our associations of interest, but we propose that the most likely direction of any selection bias is towards the null (resulting in an underestimation). Strengths of this study include the large, prospective design. However, we relied on participants to self-report reproductive history which might have led to measurement error.
Years of OC or HRT use were estimated as the difference between ages at first and last use. This likely resulted in overestimation of the duration of use, which might have attenuated our findings. In addition, we were unable to distinguish between reversible and non-reversible airway obstruction at baseline, as a bronchodilator was not administered before spirometry was conducted. Although we adjusted for self-reported asthma at baseline, this may have led to overestimation of COPD-related airway obstruction among participants with undiagnosed asthma. Lastly, we were unable to adjust for environmental tobacco smoke exposure as these data were not collected in UK Biobank. We therefore cannot exclude the possibility of residual bias from environmental tobacco smoke.

Funding 22
Give the source of funding and the role of the funders for the present study and, if applicable, for the original study on which the present article is based 2 *Give information separately for cases and controls in case-control studies and, if applicable, for exposed and unexposed groups in cohort and cross-sectional studies. also associated with greater risk of COPD-related hospitalisation/death (but not lung function), while endometriosis was associated with greater FEV 1 /FVC (not COPD-related hospitalisation/death). Early menarche (age <12 years) was associated with lower FEV 1 /FVC (but not COPD hospitalisation/death).
Associations with PCOS or ovarian cysts; any HRT use; hysterectomy-alone; and both hysterectomy and bilateral oophorectomy were in opposing directions for COPD-related hospitalisation/death (greater risk) and FEV 1 /FVC (positive association).   We relied on participants to self-report reproductive history and we cannot exclude a potential influence of selection bias, due to the low participation rate in UK Biobank.

Conclusions
 Data on environmental tobacco smoke were not available and we cannot exclude residual bias from this exposure. (COPD) has traditionally been seen as a disease predominantly affecting men. However, in more recent years, the overall burden of the disease has shifted more towards women. 2 3 Women with COPD also have, on average, fewer pack-years of smoking history than men with COPD; and nonsmokers with COPD are more often female. 1 This points towards a greater underlying susceptibility among women, and highlights the importance of understanding female-specific risk factors for COPD.
Female sex hormones are putative mediators of respiratory health, and regulate bronchodilation, cell proliferation, inflammation and metabolism of toxic cigarette smoke-related metabolites. 1 4-7 Notably, oestrogen and progesterone appear to have varying effects on lung function across the lifespan. 1 We know little, however, about female reproductive health indicators in relation to COPD risk. 1 These reproductive health indicators include markers of endogenous hormone exposure (age at menarche, age at menopause and length of reproductive lifespan), exogenous hormone exposure (oral contraceptive use and hormone replacement therapy), as well as pregnancy history. Of two existing studies, one study found no evidence of an association between hormone replacement therapy (HRT) and COPD risk, 8 while another reported a positive association between early age at first pregnancy (<20 years) and COPD-related mortality. 9 There is more evidence regarding the relationship between female reproductive health indicators and lung function. Early menarche, [10][11][12] nulliparity, 13 younger age at first pregnancy, 12 and postmenopausal status [14][15][16][17][18][19] have each been associated with reduced lung function. Evidence of the effects of HRT is contradictory. 14 15 20 21 In the present study, we therefore investigate associations of multiple female reproductive health

Study population
Our study included female participants in UK Biobank (Figure 1), a population-based cohort study which recruited men and women age 40-69 years across 22 centres in England, Wales and Scotland between 2006-2010. 22 The response rate was 5.5%. 22  Applications to access the resource are approved by the UK Biobank Access Sub-Committee and Ethics & Governance Council.
Amaral et al. previously reported decreased lung function in postmenopausal women in UK Biobank. 16 We repeat these analyses for completeness alongside analyses of other female reproductive health indicators.

Respiratory outcomes
Prevalent COPD at baseline was identified using self-report and registrations before baseline in the hospital registers. Incident hospitalisations/deaths from COPD after baseline were identified using hospital and death registers. The following International Classification of Diseases (ICD) codes were used to capture COPD: ICD-9 codes 490-492, 494 and 496; or ICD-10 codes J40-J44.
Forced expiratory volume at one-second (FEV 1 ) and forced vital capacity (FVC) were measured using Vitalograph Pneumotrac 6800 (Vitalograph, UK) at baseline. Spirometry was not performed if participants had experienced a chest infection in the last month; had a lifetime history of detached retina or collapsed lung; had experienced a heart attack, eye surgery or surgery to the chest or abdomen in the last 3 months; were currently pregnant in the first or third trimester; or were currently using medication for tuberculosis. Two measurements were conducted if the difference between these measurements was ≤5%. A third measurement was conducted if the difference was >5%. We used the single greatest measurements of FEV 1 and FVC per participant. Postbronchodilator spirometry was not available, although drug treatment was not withheld. Spirometry measurements were converted to internally standardised z-scores by age and height based on the Global Lung Initiative 2012 recommendations 23 to reduce bias related to age. These z-scores were used for all analyses.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59   We used linear regression in the analysis of standardised spirometry measurements. The multivariable linear regression models were adjusted for the same covariates as for hospitalisation/death from COPD, with the addition of asthma at baseline. We reported results as mean difference in the spirometry measurements z-scores with 95% CI. These mean differences in zscores can be converted to raw values by multiplication with the standard deviation in spirometry measurement for the given age and height group. Sensitivity analyses for standardised spirometry measurements were also conducted with: 1) menopause status excluding participants age <45 years or >60 years at baseline; 2) smoking history adjusted as two covariates (duration in years and average number of cigarettes/day); 3) further adjustment for baseline smoking status; and 4) further adjustment for comorbidities baseline (cardiovascular disease or diabetes).
We performed multiple imputation by chained equations to account for missing data. Exposures, covariates and outcomes were included in the imputation model. We imputed 20 datasets, which were then analysed using Rubin's rules. 26 Details of the imputation model are given in the online supplementary file (Table S1). All analyses were performed using Stata/MP 15 (StataCorp LLC, College Station, USA). Sensitivity analyses of the complete cases was performed for comparison with our imputed data.

Patient and public involvement
The present study did not involve patients and the public.

RESULTS
Of the 273 441 women included in this study, 2170 had prevalent COPD at baseline (Figure 1). These prevalent cases were excluded from the analysis of incident COPD-related hospitalisation/death. By the end of follow-up (median 6 years), there were 1138 incident cases of hospitalisation/death from COPD, yielding an incidence rate of 0.19 per 100 000 person years.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  In UK Biobank, 28.4% of women did not have spirometry data, 16.1% were uncertain of or did not report menopause status and 14.3% did not report smoking history. The proportion of missing information on the other covariates were modest (Table 1). Overall, women with spirometry data were younger, more often White, had higher educational attainment, higher annual household income, less deprivation (lower Townsend index), greater smoking history and were more likely to have asthma than women who did not undergo spirometry assessment (online supplementary file Table S2). The distribution of participants' characteristics across the imputed datasets is shown in Table 1.
Risk of hospitalisation/death from COPD did not differ between pre-and postmenopausal women (Table 2), and this finding was similar after further adjusting for HRT use (HR 1.02; 95% CI: 0.78, 1.35). In examining age at menopause and risk hospitalisation/death from COPD among women who experienced natural menopause (n=159 571), the risk was higher in women who experienced menopause age <47 years (HR 1.44; 95% CI: 1.19, 1.75) when compared to age 50-52 (Table 2). This association remained after further adjustment for HRT use (HR 1.43; 95% CI: 1.18, 1.74).
Postmenopausal status at baseline and early natural menopause were each associated with lower FEV 1 and FVC but not FEV 1 /FVC (Table 3). These associations did not change after further adjustment for HRT ever use (results not shown). After restricting the sample to participants who were age 45-60 years (inclusive) at the time of reporting menopause status to reduce residual bias due to age, we found that menopause status was positively associated with COPD-related hospitalisation/death (HR  while there was no evidence of an association with FEV 1 or FVC. A history of endometriosis was not associated with COPD hospitalisation/death, FEV 1 or FVC (Tables 2 and 3), though we observed a modest positive association with FEV 1 /FVC (mean difference 0.07; 95% CI: 0.03, 0.10).
We observed several associations between exogenous hormone use and COPD hospitalisation/death. Women who had ever used OC had lower risk of hospitalisation/death from COPD compared to never users (HR 0.85; 95% CI: 0.74, 0.97) ( Table 2). This appeared to be driven by OC use of >9 years (Table 2). In line with the observed findings for COPD, OC ever use and duration of use showed positive associations with FEV 1 , FVC and FEV 1 /FVC (Table 3).
Ever using HRT was associated with a modest increased risk of hospitalisation/death from COPD (HR 1.15; 95% CI: 1.01, 1.30) ( Table 2). We observed no associations of HRT ever use with FEV 1 or FVC, but a positive association with FEV 1 /FVC (Table 3). When we examined the duration of HRT use, we only observed increased hospitalisation/death from COPD (HR 1.33; 95% CI: 1.12, 1.57) and greater FEV 1 /FVC (mean difference 0.02; 95% CI: 0.01, 0.04) with 1-2 years of use (Tables 2 and 3). There was no evidence of a dose-response relationship. A history of hysterectomy (with or without bilateral oophorectomy), but not bilateral oophorectomy-alone, were associated with greater risk of COPD hospitalisation/death (Table 2). At the same time, hysterectomy was associated with higher, rather than lower, FEV 1 /FVC (Tables 3). Hysterectomy with bilateral oophorectomy was also associated with lower FVC ( Table 3).
The directions of association between individual female reproductive health indicators with COPD risk and spirometry measures are summarised in Table 4. 55 929 women had reported either a respiratory illness (asthma, COPD, tuberculosis and/or pneumonia) and/or FEV 1 /FVC<0.7 at baseline.
However, our findings after excluding these women from our analyses of COPD-related hospitalisation/death were similar to our main findings (online supplementary file Table S3).
We report the results of our sensitivity analyses, after 1) further adjusting for baseline smoking status; 2) adjusting for smoking history as two covariates including duration in years of smoking and average number of cigarettes per day; 3) excluding women who have ever smoked; or 4) further adjusting for cardiovascular disease or diabetes reported at baseline, in the online supplementary file (Tables S4-S8). The direction and magnitude of associations remained similar for risk of COPD hospitalisation/death (online supplementary file Tables S4 and S6) and change in spirometry   measures (online supplementary file Tables S7 and S8) after further adjustment for the above covariates. In our analyses of never smokers (n=161 626), however, associations of COPD-related hospitalisation/death with early menopause, parity>3, PCOS/ovarian cysts, ever using oral contraception or hormone replacement therapy became null (online supplementary file Table S5).
Early age at menarche (<12 years) was also positively associated with COPD-related hospitalisation/death in this restricted sample of never smokers (HR 1.39; 95% CI: 1.02, 1.90), whilst the remaining associations remained unchanged from the main analyses.
We also investigated associations of reproductive health indicators with risk of COPD-related mortality, and not hospitalisation, among women of UK Biobank. The associations from these analyses of 273 441 women (including those with COPD reported at baseline) were null for all  (online supplementary file Table S9), except age at menarche <12 years compared to 12-15 years (HR 1.52; 95% CI: 1.08, 2.13). However, there were only 179 COPDrelated deaths in UK Biobank and many categories of our reproductive health indicators experienced fewer than ten events. Complete-case analyses of the risk of COPD hospitalisation/death (online supplementary file Table S10) and lung function (online supplementary file Table S11) also produced similar results to those from our imputed dataset.

DISCUSSION
In this large, prospective study, several female reproductive health indicators were associated with COPD hospitalisation/death and/or lung function. Parity greater than three was associated with decreased baseline FEV 1 /FVC, a marker of airway obstruction, and increased risk of hospitalisation/death from COPD during follow-up. Ever using OC was associated with greater FEV 1 /FVC and lower risk of COPD hospitalisation/death. Associations of hysterectomy, HRT use, and a history of PCOS or ovarian cysts showed inconsistent findings for FEV 1 /FVC and COPD hospitalisation/death. This inconsistency may be explained by the fact that spirometry was measured in participants at baseline (age 40-69) and FEV 1 and FEV 1 /FVC may have declined during follow-up. Conversely, many patients diagnosed with stage 1 COPD do not go on to develop COPD symptoms or COPD resulting in hospitalisation/death. 27 There is also high intra-individual variability in FEV 1 . 28 Spirometry assessment at one time point may therefore not be as accurate at identifying clinically-relevant COPD compared to register data on hospitalisation/death.
A study of 424 797 postmenopausal women in the United States reported no difference in COPD mortality by parity. 9 In contrast, our present study reports higher risk of COPD hospitalisation/death among women with more than three births compared to nulliparous women. However, we investigated both COPD-related hospitalisation and death, and we were therefore able to capture less severe COPD cases. A study of 70 965 women in the Nurses' Health Study reported no evidence of an association between HRT ever use and COPD incidence, although the investigators included both probable (self-reported physician diagnosis along with a record of spirometry or imaging at the time of diagnosis or COPD documented on a death register) and definite (self-reported physician diagnosis and FEV 1 <80% predicted) cases of COPD. 8 Here we report a positive association between HRT use and COPD hospitalisation/death. This is the first study to assess COPD risk with regards to age at menarche, menopause status, hysterectomy, gynaecological surgery, endometriosis or PCOS.
Two previous studies reported reduced FEV 1 in adult women who experienced menarche before age 10 10 12 but neither found evidence of an association between age at menarche and FEV 1 /FVC. A Mendelian randomisation study of adolescent and adult women across multiple European cohorts, including a subset from UK Biobank, estimated a 24.8 mL increase in FVC per year increase in age at menarche among adult women and a -56.6 mL decrease in FVC per year increase in age at menarche among adolescent girls. 11 This study reported no association between age at menarche and FEV 1 /FVC. 11 Notably, this Mendelian randomisation study did not explore non-linear associations between age at menarche and measures of lung function. We observed lower FEV 1 and FVC in adult women with late menarche, and lower FVC and FEV 1 /FVC in adult women with early menarche.
A recent systematic review, which included studies from UK Biobank, 16 reported that postmenopausal status was associated with reduced FEV 1 and FVC but not FEV 1 /FVC. 18 Current HRT use has been associated with both higher 14 and lower 15 lung function. One study reported an association between HRT use and greater FEV 1 /FVC. 14 Others have identified differences in lung function according to continuous/cyclic and combined/oestrogen-only HRT. 20 21 Based on the information from participants in UK Biobank, we could not stratify by HRT subtypes.
There are sex differences in lung anatomy and physiology throughout the life course. 1 2 4 During gestation, female foetuses have fewer bronchi, smaller lungs and produce surfactant earlier than male foetuses. 1 Female neonates and children have greater lung function and faster lung growth than their male counterparts, and boys are more likely to be diagnosed with asthma during childhood than girls, while girls are twice as likely as boys to be diagnosed with asthma after puberty. 1 These findings suggest that female sex steroids can accelerate lung maturation and termination of lung growth, which may partially explain why early menarche associates with lower FEV 1 /FVC in adulthood. During the reproductive years, parity and PCOS relate to an increase in circulating oestrogen, which has been implicated in the pathogenesis of inflammatory lung diseases including COPD. 6 29-31 Although OC use appears to be protective against COPD, some researchers have suggested that OC use attenuates the fluctuations in sex steroids, which correlates with fluctuations in endothelial nitric oxide, a potent bronchodilator. 1 Additionally, exogenous hormones may not be as influential during reproductive years but may become more influential in after menopause, when endogenous hormone production is low. 6 This is supported by our findings of opposing associations with COPD risk for OC use (lower risk) during reproductive period and HRT use (higher risk) after menopause.
The effect of female sex hormones on COPD and lung health may vary depending on the net effects on airway smooth muscle, inflammation and cigarette smoke metabolism. Oestrogens can regulate the synthesis of nitric oxide, a potent bronchodilator, within human bronchial epithelium, 1 32 as well as regulate airway bronchoconstriction in response to acetylcholine or histamine. 33 34 There is also evidence that progesterone may affect airway smooth muscle constriction, although these reports are limited to animal studies. 35 In addition to regulating bronchodilation and bronchoconstriction, in vitro studies report that oestrogen can also modulate the proliferation of airway smooth muscle cells, 36 37 which is a key feature of COPD. 38 Oestrogen's pro-and anti-inflammatory effects have been documented in other female-predominant diseases 39 and may also play a role in COPD, as chronic inflammation is another hallmark of the disease. Finally, oestrogen is an important regulator of toxic cigarette smoke metabolites through cytochrome P450 enzymes. 1 In a mouse model of COPD, Glassberg et al. reported that oestradiol administration after ovariectomy protected against cigarette smoke-induced alveolar septal destruction, macrophage infiltration and decreased other markers of inflammation compared to ovariectomy without oestradiol administration. 40 In the same mouse model, however, non-ovariectomised female mice also had increased small airway We cannot exclude a potential influence of selection bias, due to the low participation rate in UK Biobank. Women in the cohort are less likely to have ever smoked, have greater educational attainment, but a similar number of offspring, as compared to the general UK population (https://www.ons.gov.uk/). There is also a lower prevalence of self-reported COPD in UK Biobank (0.8%) compared to physician-diagnosed COPD in the general female UK population (1.68%). 42 Overall, women participating in UK Biobank reflect a higher-educated, more health-conscious group, with more modest underlying risk for COPD compared to the general population. We can only speculate as to how these selection factors might have influenced our associations of interest, but we propose that the most likely direction of any selection bias is towards the null (resulting in an underestimation). Strengths of this study include the large, prospective design. However, we relied on participants to self-report reproductive history which might have led to measurement error.
Years of OC or HRT use were estimated as the difference between ages at first and last use. This likely resulted in overestimation of the duration of use, which might have attenuated our findings. In addition, we were unable to distinguish between reversible and non-reversible airway obstruction at baseline, as a bronchodilator was not administered before spirometry was conducted. Although we adjusted for self-reported asthma at baseline, this may have led to overestimation of COPD-related airway obstruction among participants with undiagnosed asthma. Lastly, we were unable to adjust for environmental tobacco smoke exposure as these data were not collected in UK Biobank. We therefore cannot exclude the possibility of residual bias from environmental tobacco smoke.