Female reproductive and hormonal factors and lung cancer mortality among never‐smokers: A prospective cohort study of 287 408 Chinese women

There is growing, but inconsistent evidence suggesting oestrogen may play a key role in lung cancer development, especially among never‐smoking women for whom lung cancer risk factors remain largely elusive. Using the China Kadoorie Biobank, a large‐scale prospective cohort with 302 510 women aged 30 to 79 years recruited from 10 regions in China during 2004 to 2008, we assessed the risk of lung cancer death among self‐reported never‐smoking women who were cancer‐free at baseline, in relation to age at menarche, age at menopause, time since menopause, prior use of oral contraceptives (OCP), number of livebirths, breastfeeding and age at first livebirth. Women were followed up to December 31, 2016 with linkage to mortality data. Hazard ratios (HR) and 95% confidence intervals (CI) were estimated using Cox regression, adjusting for key confounders including several socio‐demographic, environmental and lifestyle factors. Among 287 408 never‐smoking women, 814 died from lung cancer with a median follow‐up of 10.3 years. Women who had used OCP within 15 years prior to baseline had a significantly higher hazard of lung cancer death compared with never‐users: HR = 1.85 (95% CI: 1.14‐3.00) and risk increased by 6% with each additional year of use: HR = 1.06 (1.01‐1.10). Among parous women, the hazard of lung cancer death increased by 13% with each single livebirth: HR = 1.13 (1.05‐1.23); and among post‐menopausal women, the risk increased by 2% with each year since menopause: HR = 1.02 (1.01‐1.04). These results suggest that reproductive factors which were proxies for lower endogenous oestrogen level, for example, longer duration of OCP use, could play a role in lung cancer development.


What's new?
Increasing evidence suggests that oestrogen influences lung cancer development, especially among women who have never smoked. Lung cancer risk factors among never-smoking women, however, remain unclear. In this study, using data from the China Kadoorie Biobank, relationships between female hormonal factors and lung cancer among never-smoking women was investigated. Analyses reveal an elevated risk of lung cancer among women with a history of oral contraceptive use, among parous women, and among post-menopausal women. The findings indicate that reproductive factors that are associated with decreased endogenous oestrogen levels may play a role in lung cancer development in never-smoking women.

| INTRODUCTION
Lung cancer is the leading cause of cancer mortality globally and in China, accounting for 18% of total cancer deaths worldwide in 2020. 1,2 Although tobacco smoking is responsible for the great majority of lung cancer cases, about 25% of all lung cancer cases worldwide occur in lifelong never-smokers. 3 Lung cancer in neversmokers (LCINS) is more common among women than men, especially in Asian countries. 3,4 In China, 43.2% of all lung cancer cases in 2005 were never-smokers, and 60% of LCINS occurred in women. 5 In recent years, the incidence of lung cancer has been increasing among women globally as well as in China. 2 Despite the historically low smoking prevalence among Chinese women, which dropped from 3.1% to 2.7% during 2003 to 2013, 6 there was a 0.71% yearly increase in lung cancer incidence during the same period. 7 As LCINS is more common in women, it has been proposed that female sex-steroid hormones could play a key role in lung cancer development. Supporting evidence includes the elevated expression of oestrogen receptors (ERs) observed in both normal lung tissue and lung cancer cell lines. 8 However, there have been inconsistent findings from observational studies examining various female reproductive factors which have been used as a proxy measure of endogenous oestrogen activity. Some studies have reported reduced lung cancer risk associated with late menarche, 9 late age at first livebirth, 10,11 early menopause, 12 use of oral contraceptives (OCP), 13,14 or menopausal hormonal therapy (MHT), 13,14 while others have reported increased lung cancer risk associated with the same factors. 9,10,12,13,[15][16][17] The inconsistent results may be attributed to differences in study design, residual confounding from smoking, inadequate sample sizes, and/or racial differences among study populations.
The aim of the study was to assess the potential association of female reproductive factors with lung cancer development in the China Kadoorie Biobank (CKB) which contains one of the largest cohorts of never-smoking Chinese women, and near complete followup with linked death data. 18 We used self-reported information on female reproductive factors as a proxy measure of endogenous oestrogen activity based on the corresponding physiological mechanisms involved. [9][10][11]16,17 2 | MATERIALS AND METHODS

| Study design and population
Study design and methods of the CKB have been described elsewhere. 18 Briefly, a total of 512 715 adults (59% women) aged 30 to The present study was restricted to 290 328 female neversmokers (ie, women who smoked <100 cigarettes or equivalent in their lifetime) to eliminate the potential confounding effects from smoking in analyses. After excluding 1536 women with a selfreported diagnosis of any cancer at baseline, and 1384 who had missing, inconsistent, or biological implausible values for reproductive factors (see Section S1 for details), 287 408 women remained in the study population. ≥30 years); total duration of breastfeeding (in months) as a continuous and categorical variable (0, 0-12, 12-24, 24-36, 36-48, >48); OCP use (never-or ever-users); duration of OCP use as continuous and categorical variable (never-users, 0-5, 6-9, ≥10 years); and time since last use, calculated from the age when OCP was last stopped to the age at baseline, and categorised as 'never-users', 'ex-users ≥15 years', and 'ex-users <15 years or current users'. We also assessed the overall effect of duration of use and time since last use (never-users, ≤5 years of use and ex-users ≥ or <15 years, >5 years of use and ex-users ≥ or <15 years; see Table 3). To assess menopausal status, women with surgical menopause were excluded, and the remaining women were categorised into premenopausal, perimenopausal or postmenopausal, based on questions regarding whether they had reached menopause, were currently going through menopause, or already had natural menopause. Among postmenopausal women who had undergone natural menopause, age at menopause was analysed as a continuous and categorical variable (≤40; 41-45; 46-50; 51-55; ≥56 years). Time since menopause, defined as the number of years between the age at menopause and age at baseline, was analysed as a continuous and categorical variable (≤10; >10-20; >20 years). Reproductive years was defined as the duration between age at menarche and age at menopause, and was analysed as a continuous and categorical variable (≤25; >25-30; >30-35; >35 years).

| Ascertainment of exposures and confounders
Confounders were selected a priori based on the literature, and included marital status, highest education, region of residence, height, physical activity, body mass index, exposure to second-hand smoke, exposure to household air pollution, cooking oil use, chronic obstructive pulmonary disease, diabetes, alcohol consumption, and dietary intake of fruits, meat, dairy products, and fish/seafood, and mutual adjustment with other reproductive factors. Full details on exposure and confounder ascertainment are described in Tables S1 and S2, respectively.

| Ascertainment of lung cancer death
The vital status of all participants was monitored regularly through local residential records and death certificates reported to the regional CDC of China's Disease Surveillance Points (DSP) system. 18 Fatal events were coded using the International Classification of Diseases, 10th Revision (ICD-10) by trained DSP staff members blinded to baseline information. 19 The 10 regions for CKB were selected because they were covered by the DSPs system to ensure the mortality follow-up for the participants are as complete as possible. For deaths not captured through the routine procedures, standardised verbal autopsy and/or medical record reviews were conducted by trained health workers to ascertain the most probable cause of death. 18 With these standardised approaches and stringent quality control for data collection, the ascertainment of death records has high completeness and quality.
The main outcome of interest in this study was death from lung cancer (ICD-10: C34). All participants were followed up from baseline to the date of death, or "censored" upon death due to other causes, loss to follow-up (1.0%) or end of the study period (December 31, 2016), whichever occurred first. This study only used CKB mortality data as lung histopathology data and incidence data were not available for analysis.

| Statistical analyses
Baseline characteristics of the study population are described as means and SDs (SDs) or percentages by categories. The reproducibility of some reported exposure data (ever use of OCP and number of livebirths) was assessed using kappa statistics (κ-value) 20 in the subsample with complete exposure information at baseline and resurvey.
For the analyses investigating the association between lung cancer death and number of livebirths, age at first livebirth, and total duration of breastfeeding, nulliparous women (n = 3689) were excluded. For the analysis of menopausal status, those with surgical menopause (n = 13 473) were excluded, while analyses for menopausal age, time from menopause and reproductive years were confined to postmenopausal women who had undergone natural menopause. For the remaining reproductive factors, analyses were conducted among all women in the study population (see Table S3). Each exposure of interest is analysed in a separate Cox model, adjusted for the confounders and the other relevant exposures of interest. Multivariable Cox proportional hazards regression, with age as the underlying time scale, was used to estimate adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for lung cancer death associated with each exposure variable, adjusting for the full set of confounders in all analyses. Subgroup analyses were also conducted by area of residence (urban; rural), secondhand smoking exposure (ever; never) and body weight (elevated BMI; normal or below normal BMI). Additionally, two sensitivity analyses were conducted: first, excluding data in the first 2 years of follow-up (ie, to start the follow-up analysis from the third year onwards and exclude any women who died in the first 2 years) to minimise the potential risk of bias from women who might have lived with earlystage or subclinical lung cancer at baseline; second, excluding current users of OCP when assessing time since last use, due to their small number of women (672) and lung cancer deaths (2).
The proportional hazards assumption in the Cox regression models was examined using the method by Lin et al. 21 Statistical analyses were performed in SAS version 9.4 and significance was defined as P < .05.

| Baseline characteristics
Among 287 408 never-smoking women, the median age was 50.5 years (Table 1). Mean (SD) age at menarche, at birth of first child and at menopause were 15.4 (1.9), 23.4 (3.2) and 48.6 (4.0) years, respectively. Only 10% of women ever used OCP, and the mean duration of use was 3.6 years among the ever-users. Almost all women (98.8%) were parous, with an average of 2.2 live births, and the majority reported breastfeeding (97.2%) with the average total breastfeeding time being 29.7 months. Compared with rural women, urban T A B L E 1 Distribution of baseline characteristics of never-smoking women by area of residences and outcome (lung cancer death) in the (Continues) women (45%) were older on average at baseline, more educated, less physically active, had a higher income, and fewer were exposed to passive smoking or smoky coal at home. A higher proportion of urban women were overweight/obese, nulliparous, and those who were parous had fewer children, compared with rural women. Overall, compared with other women, those who died of lung cancer were older at baseline (mean 60.1 vs 50.0 years old), less educated, less physically active and had more children. In the re-survey of 11 314 never-smoking women, the κ-value for ever use of OCP and number of livebirths was 0.67 and 0.92 respectively (see Table S4).  additional year of use (HR = 1.06; 95% CI 1.01-1.10), and the highest point estimate was observed for those who reported OCP use of ≥10 years (HR = 1.97; 95% CI 1.24-3.13). In addition, the overall hazard for those who used OCP for >5 years was higher than those who used for ≤5 years regardless of when OCP use ceased (Table 3). In sub-group analyses, a similar pattern of associations between OCP use and lung cancer death was observed among urban dwelling women (with additional association for ever vs never OCP use) but not among rural women (see Tables 2 and 3), among women ever exposed to secondhand smoke but not those never exposed, and among women with elevated BMI but not those with normal or under normal BMI (see Table S5). In sub-group analyses, association between number of livebirths and lung cancer death was observed among rural parous women but not urban women (Table 2), among parous women with normal or below normal BMI but not those with elevated BMI, and among parous women regardless of their secondhand smoke exposure (Table S5).

| Female reproductive factors and lung cancer death
Among post-menopausal women, each additional year since menopause was associated with a 2% increased risk of lung cancer death (HR = 1.02; 95% CI 1.01-1.04; P = .0087). Compared with those who were < 10 years postmenopausal, the HR of lung cancer death was 52% higher for women who were more than 20 years postmenopausal (HR = 1.52; 95% CI 1.11-2.07). In sub-group analyses, a similar pattern of results was observed for post-menopausal urban women, but not rural women (Table 2), for women ever exposed to secondhand smoke but not those never exposed, and for women with elevated BMI but not those with normal or below normal BMI (see Table S5).
In the sensitivity analysis which excluded data in the first 2 years of follow-up, the association of lung cancer death with duration of OCP use, number of livebirths, and time since menopause did not change appreciably. Although the association between lung cancer death and time since last OCP use among all women was not statistically significant, it remained significant among urban women. When current users of OCP were excluded from the analysis of time since last OCP use, similar results were observed as that in the main analysis (see Table S6). There was no evidence of non-proportionality in the Cox regressions.

| DISCUSSION
This study evaluates the relationship between female reproductive factors and lung cancer death using a prospective cohort of 287 408 never-smoking women in China, which is one of the largest studies to date. After adjusting for key confounders including second-hand smoke exposure and household air pollution, we observed significantly increased hazards of lung cancer death with recent use of OCP and longer duration of use, as well as increased hazards with a higher number of livebirths and longer time since menopause. Lung cancer death was not significantly associated with other reproductive factors such as age at menarche and breastfeeding. Subgroup analyses indicated that the increased risk of lung cancer death observed among women who used OCP and those with longer time since menopause was restricted to urban dwelling women, women ever exposed to secondhand smoke, and women with elevated BMI; while the positive association with higher number of livebirths was observed among rural women, and among women with normal or below normal BMI.
All HRs in the subgroup analyses were largely consistent with those in the main analysis. Also, the age-standardised lung cancer mortality rate for this cohort (20/100000) was comparable to that in a recent publication for Chinese females (19/100000) 23 (see Table S7).
Since OCP became available in the early 1960s, it has been studied for its potential link to several female cancers, and was classified by the IARC as Group 1 carcinogen for breast and cervical cancer. 24 Research on the relationship between OCP and lung cancer has increased in the last two decades due to growing interests in the roles of female hormonal factors in lung cancer, 25 however the weight of evidence to date suggests a null or inverse association. 11,13,14,16,26 Nevertheless, two prior studies observed a positive association between lung cancer risk and OCP use: the Nurses' Health Study (NHS) 15 and a case-control study of Chinese women in Fujian. 27  Our results also showed that post-menopausal women were at a higher risk of lung cancer death with longer time since menopause, with a 2% increased risk for each year since menopause, and a 52% increase for more than 20 years compared with <10 years since menopause. This is comparable to the WHI Studies in which time in years since menopause was positively associated with lung cancer risk, with more than 20 years post-menopause associated with a 50% increase in risk compared with <5 years since menopause. 16 Menopause is a T A B L E 3 Adjusted hazard ratios (HR) and 95% confidence intervals (CI) for lung cancer death in relation to oral contraceptive (OCP) use among never-smoking women in the milestone of hormonal changes due to ageing characterised by marked reduction of endogenous oestrogen synthesis associated with ovarian dysfunction. Resulting low plasma oestrogen levels in postmenopausal women gradually decline over time, and the variable 'time since menopause' corresponds to this decline which could be a more sensitive variable than 'menopausal age'. According to our analysis, with increasing time since menopause, postmenopausal women may have a higher risk of lung cancer death, which may partly explain the higher incidence of lung cancer among never-smoking women than never-smoking men. 35 Several mechanisms have been hypothesised to explain the relationship between reproductive factors and the risk of lung cancer.
Collectively, these factors act as proxies for the hormonal effects of endogenous oestrogen-either causative or protective-on lung cancer risk. We note that factors associated with lung cancer diagnosis and death may differ, as could the direction of effect-for example, a risk factor that increases risk of developing lung cancer may confer a survival advantage due to better treatment response. Although our study was based on mortality data as we did not have data on lung cancer incidence, due to its high fatality and high proportion diagnosed at late stage of disease, 36  Interpretations of our results must be made with the consideration of several factors. Data on lung cancer incidence and tumour characteristics were not available for analysis, which meant that our results were limited to mortality outcomes. While incidence of highly fatal diseases such as lung cancer may be reasonably approximated using mortality-based data 37 the (likely small) proportion of participants diagnosed with curable disease were not included in estimates of disease risk. Risk estimates may also be biased by access to treatment with curative intent. In addition, genetic information related to lung cancer development and progression was not available for analyses. A more comprehensive view may thus be offered by followup studies including genetic information, for example, genes involved in the absorption, distribution, metabolism, and excretion of drugs (socalled ADME genes) that have previously been shown to play a role in metabolising steroid hormones and be associated with patient survival. 44 Risk estimates reported in this study relate to a population with particularly high rates of lung cancer among never smoking women.
The higher lung cancer risk in Asian-born never smoking women could be due to either genetic or environmental factors or both, which can be difficult to disentangle. 45 Reproductive behaviour and characteristics among women in mainland China have generally differed from those in Western countries, with later age at menarche, earlier menopausal age and universal breastfeeding among parous women. 43 However, over the past few decades there have been marked changes in these patterns, mainly due to periodic political and social turmoils including wars and famines, the "One-child Policy", and rapid urbanisation. The "One-child Policy" not only restricted the number of children in a family but also encouraged late marriage and childbirth, and applied more strictly to urban than rural residents. 46

DATA AVAILABILITY STATEMENT
The datasets generated and/or analysed during the current study are accessible on reasonable request (https://protect-au.mimecast.com/ s/O4QPCyojxQTNop6qnfMKhRN?domain=ckbiobank.org). Further information is available from the corresponding author upon request.

ETHICS STATEMENT
This study has been conducted using the China Kadoorie Biobank, the ethical approval of which was obtained from the Oxford Tropical