#Title: Delay in menarche is associated with reduced risks of seven cancers: a consortial analysis
BACKGROUND: The average age at menarche has fallen in European and U.S. populations over two centuries, a secular trend commonly attributed to improvements in nutrition and sanitation. Recent studies showing associations of early menarche with cardiovascular diseases and mortality risks have suggested that the timing of pubertal events may have broader implications for chronic disease risks in aging women than previously appreciated.
AIM: To determine the association of age at menarche with risks of most common cancers.
METHODS: We pooled data from nine prospective US and European cohorts with first enrollments in 1981-1998 and median follow-up times of 7-20 y. Cox regression was used to estimate hazard ratios and 95% confidence intervals for associations of age at menarche with the incidence of 19 site-specific primary invasive cancers. Models were adjusted for age, race, education, height, parity, history of oral contraceptive use, age at menopause, and baseline measures of smoking, alcohol, physical activity, and menopausal hormone therapy.
RESULTS: Among 536,450 women studied, 61,423 were diagnosed with 19 site-specific first primary cancers. Inverse linear associations were observed for 7 of 19 cancers studied. Each year of delay in menarche was associated with a small but statistically significant reduction in the risks of endometrial cancer (HR:0.91, 95% CI:0.89-0.94), liver cancer (HR:0.92, 95% CI:0.85-0.99), malignant melanoma (HR:0.95, 95% CI:0.93-0.98), and cancers of the bladder (HR:0.96, 95% CI:0.93-0.99), breast (HR: 0.96, 95% CI: 0.93-0.99), colon (HR:0.97, 95% CI:0.96-0.99) and lung (HR:0.98, 95% CI:0.96-0.99). Most estimates of association were attenuated but remained statistically significant when models additionally adjusted for body mass index (BMI). Associations of age at menarche with breast, endometrial, and melanoma skin cancers were more strongly inverse in never- compared to current-users of menopausal hormones (Pinteraction=0.02, 0.01, and 0.05, respectively). Risk associations were not modified by age, birth cohort or BMI.
CONCLUSIONS: Delayed menarche was associated with reduced risks of seven site-specific cancers. Modification of some risk associations by hormone therapy use is consistent with a shared causal pathway, perhaps related to estrogen exposure; mechanisms underlying the other associations should be investigated further.
Research Question: Which of nineteen site-specific cancers are associated with the age at menarche; do associations differ by body size, hormone use or smoking history?
Study Findings: In this analysis of pooled data from 12 US and European cohorts of middle-aged women, delay in menarche was associated with lower risks of seven cancers, including some not traditionally thought of as hormone-related. All but one of these associations remained evident even after accounting for measures of body size (height or BMI).
Meaning: It is likely that world-wide trends towards earlier menarche have contributed to rising rates of the most common cancers in the U.S. and Europe.
NOT SURE WHAT TO DO WITH THIS
Menarche, defined as the age of the first menstrual period, signals the initiation of monthly hormonal cycles and the beginning of the reproductive lifespan. It is also the closing event of female puberty, a phase of accelerated growth, weight gain, and secondary sexual development that begins in middle childhood and lasts several years (Emmanuel and Bokor, 2018).
Compiled data suggest that the age at menarche declined during the 19th and 20th centuries in Europe and the U.S. (Eveleth and Tanner 1990; Wyshak and Frisch, 1982; McDowell et al., 2007). While these changes have differed somewhat by place, social class, and ethnicity/race, secular trends towards lower age menarche have been observed in many countries, including Brazil (Silva and Padez, 2006), South Africa (Jones et al., 2009), Japan (Hosokawa et al., 2012), India (Pathak et al., 2014), and China (Song et al., 2015). The time-, place-, and population-specific estimates for average ages at menarche vary widely, from 12 years of age, observed in Hispanic and African American girls born in the U.S. during the 1980’s (McDowell et al., 2007) to 17 years of age, observed in girls from Northern European countries in the 19th century (Tanner, 1981).
The causes of secular trends in menarche are poorly understood, as are the causal relationships that explain associations of early menarche with chronic disease risks.
Early menarche extends the female reproductive life span, but has also been associated with unfavorable outcomes, including shorter stature, greater risk of adverse pregnancy outcomes, and greater mid-life risks of diabetes, cardiovascular diseases, and premature mortality. Thus, secular trends towards earlier menarche may have implications for population health.
We have long known that the age at menarche is inversely associated with the risk of breast cancer.16-19 Similarly inverse associations have been seen for endometrial17,20 and colon cancers.17,21
The implications of trends in menarche for risks of other cancers are not firmly established.
More recently, it has been noted that early menarche predicts greater risks of all-cause mortality,11-13 cardiovascular mortality,14,15 and cancer-associated mortality among women.14
We have therefore tested for associations of site-specific cancer risks with the age at menarche using a standardized analytic approach, in a consortial data set that includes 536,450 women drawn from nine prospective cohorts in the U.S. and Europe. Next we assessed the extent to which these associations are mediated by adult BMI, and then tested whether the observed associations differ by birth cohort, height, adiposity, parity, history of smoking, or use of postmenopausal hormones.
Study Design and Population
This analysis was carried out using a data resource developed originally through the National Cancer Institute Cohort Consortium for a study of obesity and mortality,36 and then further developed for a study of leisure-time physical activity and cancer risk.37 Participating cohorts provided data on menarche, anthropometry, physical activity, smoking status, postmenopausal hormone use and first primary incident cancers; data was compiled and harmonized in a collaborative effort that involved NCI investigators and cohort representatives.
Of twelve cohorts that participated in the study of physical activity, two were not included because they had male participants only. Of those remaining, nine cohorts agreed to participate, including six from the U.S. and three from Europe (Table 1). Thus the study included women (>18 years of age) from nine cohorts.
We considered as potential outcomes the following cancers for which we projected adequate statistical power: cancers of the bladder, brain, breast, colon, endometrium, gastric cancer (cardia only), head and neck, kidney, liver, and lung; lymphocytic leukemia, myeloma, and myeloid leukemia, non-Hodgkin lymphoma, and cancers of the ovary, pancreas, rectum, skin (melanoma only), small intestine, and thyroid.
Cancer ascertainment: Incident first primary cancers were identified by cohort staff through follow-up questionnaires with subsequent review of medical records (Zhang et al., 2006; Rajaraman et al., 2015), cancer registry linkage; (Larsson, 2006; Moore, 2009; Roswall, 2017), or both (Leitzman, 2008; Lacey et al., 2009; Hildebrand, 2013). Cancer types were defined using the Surveillance Epidemiology and End Results site recode and the International Classification of Diseases for Oncology, Third Edition (see Supplementary Table 1) (World Health Organization, 2000). Participants were followed from baseline to the earliest date of cancer diagnosis, death, or end of follow-up. Site-specific cancers were selected for this analysis if there were at least 300 cases across the nine participating cohorts. For each cancer, only cohorts with at least 15 cases were included for analysis.
Statistical Methods: For two cohorts in which menarche was queried using categories, mid-points for each category were used to impute continuous measures of the age at menarche. The baseline body mass index (BMI) was calculated for each participant as her weight in kilograms divided by the square of the height in meters. When modeling the risks of ovarian and endometrial cancers, women who reported histories of oophorectomy and hysterectomy at baseline, respectively, were excluded from analysis. Statistical analyses were done in SAS 9.4.
Cox proportional hazards models were used to estimate hazard ratios (HR) and 95% confidence intervals (CIs) for each association of age at menarche with a site-specific cancer risk. Hazard ratios (HR) were estimated per additional year in attaining menarche, with study time as the underlying time-metric. First, we used restricted cubic splines and likelihood ratio tests to evaluate whether associations of menarche and cancer risks were consistent with linearity. Because all associations but one were best described as linear (Figure 1), we modeled all risk associations using age at menarche as a continuous linear basis for all subsequent analyses. Separate models were produced for each cohort and outcome, and random-effects meta-analysis was used to generate summary risk estimates (DerSimonian and Laird 1986). Statistical heterogeneity was evaluated using Cochran’s Q (Cochran 1954).
Multivariable-adjusted models included baseline age, smoking, alcohol, race/ethnicity, education, and hormone replacement therapy, oral contraceptive use, age at menopause, parity and height. In order to evaluate whether anthropometric measures confound and/or mediate observed associations, we ran models that additionally adjusted for body mass index (Model 2). Multiple imputation procedures were used to accommodate missing data within each cohort; across all included participants and variables we observed <3% missingness.26 We evaluated multiplicative effect modification by birth year, height, and adult BMI (<25 kg/m2; > 25 kg/m2), smoking status (current; former; never smokers), postmenopausal hormone therapy (ever-; former-; or never-users), and follow-up time (<5 years of follow up; > 5 years of follow up) using the Wald test for homogeneity. Interactions were declared if p-values were less than 0.05. The proportional hazards assumption was tested by creating an interaction term between categorical age at menarche and follow-up time, and the Wald test was used to assess the statistical significance of the interaction.
In a retrospective analysis we first considered determinants of the age at menarche and observed that it varied significantly by cohort and birth year. Average ages at menarche declined monotonically across birth decades from 1900 to 1960 in six of the nine cohorts, and in all cohorts across the decades of 1920-1940. In a pooled sample including all cohort participants, we observed that women in three cohorts (SMC, WLH, and PLCO) had significantly later menarche, while women in AARP had earlier menarche, than would be predicted by birth year alone. Cohort-adjusted mean ages at menarche declined from 13.4 to 12.6 between 1905 and 1941, occurring 8 days earlier for each successive birth year while the proportion of women who experienced menarche before age 12 y grew from 10% to 25% during the same period (Figure 1). The age at menarche was directly associated with attained height (p<0.001), and inversely associated with baseline BMI (p<0.001).
We modeled associations of site-specific cancer risks with the continuous age at menarche using splines (Figure 1). Results of likelihood ratio tests suggest that the association of menarche with melanoma risk was best described as non-linear, with a marked change in slope at the median age of 12 y; the spline suggests no change in risk observed across menarcheal ages before the median age of 12 y and a strong inverse trend at ages >12 y (P for non-linearity=0.01). Since this association did show reasonable fit with a linear trend, and all other site-specific cancer risks were best modeled as linear functions, we decided to use linear models to obtain comparable estimates.
Using linear models we observed statistically significant inverse associations of the age at menarche with the risks of 7 out of 19 cancers. As shown in table 1, each year of delay in menarche was associated with a small but statistically significant reduction in the risks of endometrial cancer (HR:0.91, 95% CI:0.89-0.94), liver cancer (HR:0.92, 95% CI:0.85-0.99), malignant melanoma (HR:0.95, 95% CI:0.93-0.98), and cancers of the bladder (HR:0.96, 95% CI:0.93-0.99), breast (HR: 0.96, 95% CI: 0.93-0.99), colon (HR:0.97, 95% CI:0.96-0.99) and lung (HR:0.98, 95% CI:0.96-0.99) (Table 1). When these models were additionally adjusted for BMI, the summary HR were altered by <5% and remained statistically significant for all cancer sites except liver (Table 2)
Because of the strong trend in the age at menarche by birth year, we assessed whether tested associations were modified or confounded by birth cohort (Table 3). Year of birth was seen to modify the association of menarcheal age with melanoma risk; in strata defined by the median year of birth seen in the pooled sample, an inverse association with melanoma risk was seen in women born before 1934, but not in those born afterwards (HRbefore=0.92, 95% CI=0.87-0.97 vs. HRafter=0.99, 95% CI=0.96-1.02, Pinteraction=0.02).
Effect modification was not observed by baseline age, BMI, or parity, however statistically significant interactions were observed by height for ovarian cancer (HRtall=0.92, 95% CI=0.87-0.98 vs. HRshort=1.05, 95% CI=0.98-1.12, Pinteraction=0.01), and by smoking status for cancers of the lung (HRnever=0.90, 95% CI=0.85-0.95 vs. HRever=1.09, 95% CI=1.00-1.19, Pinteraction=0.02) and thyroid (HRnever=0.90, 95% CI=0.85-0.95 vs. HRever=1.09, 95% CI=1.00-1.19, Pinteraction<0.001). Associations of age at menarche with breast, endometrial, and melanoma skin cancer risks were more strongly inverse in never- compared to current-users of hormone therapy (Pinteraction=0.02, 0.01, and 0.05, respectively). None of these interactions remained statistically significant after controlling for the false discovery rate (Benjamini and Hochberg, 1995).
In this pooled analysis of 536,470 middle-aged and elderly women drawn from nine cohorts in the U.S. and Europe, we observed that each year of delay in menarche is associated with modest but statistically significant reductions in the risks of seven cancers, including endometrial and liver cancers, malignant melanoma, and cancers of the bladder, colon, lung and breast. These associations show no evidence of heterogeneity by cohort, and most are not modified by birth cohort, height, body mass index, smoking or parity.
The inverse association of the age at menarche with endometrial cancer risk is consistent with previous literature (summarized in Gong 2015). In our data, the association is linear and no inflection point is observed. The association of EC with the age at menarche was substantially attenuated with additional adjustment for BMI, but it remained statistically significant. Several previous studies showed that genetic risk scores predicting early menarche are also associated with EC risk (Day 2017, Wang 2015). Findings of several studies suggest that these associations are mediated, in part, by effects on lifetime menstrual cycles (Yang 2016, Day 2017), and in part, by BMI (Wang 2015). Estimates of the associations of AAM with EC risk have not shown statistically significant heterogeneity by subtype (Setiawan 2013, Yang 2013).
Our finding adds to the limited data available on the role of menarche in liver cancer, summarized in a meta-analysis which identified three prospective studies on AAM and primary liver cancer and reported a summary estimate of RR=0.50 (95% CI 0.32, 0.79) for late vs. early age at menarche with the risk of primary liver cancer. Liver cancer has rarely been studied in association with reproductive factors and the number of cases available for analysis are limited since it is a rare disease. One previous study, which included many of the same cohorts in the present study but considered HCC rather than PLC, observed a non-statistically significant trend towards reduced risk in associated with delayed menarche (OR=0.64, 95% CI: 0.40, 1.03). In another prospective study, which included 218 cases of incident primary liver cancer, Yu observed that early menarche (LT12 vs GT 16) increased risk of HCC in carriers of hepatitis B antigens (multivariate-adjusted OR, 6.96; 95% CI, 2.52-19.18) but that no increased risk was apparent in non-carriers (Pinteraction=0.005). In this study, similar to most previous studies, data on viral infections was not available for our use.
Few studies have reported on associations of melanoma risk with reproductive factors, and most have been underpowered or retrospective. Our finding of an inverse association of melanoma risk with the age at menarche is consistent with the only published study that explicitly tested for an association with the age at menarche, a prospective study by Kvaskoff and colleagues in which significantly reduced risk of cutaneous melanoma was observed in women with late vs. the median age at menarche (Kvaskoff et al. 2011).
Several studies have tested for associations of bladder cancer risk with the age at menarche and other reproductive factors. In the large AARP cohort, with 651 incident cases of incident bladder cancer observed among women, Daugherty et al. found reduced risk in association with late menarche (age > 15 y vs. <10 y, RR= , Ptrend=0.02). Other studies are more poorly powered and offer inconsistent findings (Cantwell et al., 2006; Huang et al., 2009; McGrath et al., 2006; Prizment et al., 2006).
Few studies of reproductive factors in the etiology of lung cancer have been published and those in the literature show mixed results that could reflect heterogeneity by tumor subtypes. In the largest prospective study done to date, Brinton and colleagues observed reduced risk of lung cancer in women with late vs. early menarche.
Our finding of a significant inverse association of breast cancer risk with the age at menarche is consistent with a large body of available epidemiologic data including a pooled analysis that drew data from 117 observational studies (Collaborative Group on Hormonal Factors in Breast Cancer 2012). There is some evidence that distinct breast cancer subtypes have different associations with menarche. In the Million Women’s Study Reeves and colleagues observed statistically significant heterogeneity in the risk associations observed for distinct histologic subtypes, with a strongly inverse association of menarche with the risk of lobular breast cancers and weaker inverse associations with ductal, lobular, tubular, and mixed ductal/lobular subtypes; in contrast, non-statistically significant direct associations were seen with the risks of mucinous and medullary subtypes of breast cancer (Reeves et al. 2009).
The causes underlying secular trends in menarche are poorly understood, as are the causal relationships that explain associations of early menarche with chronic disease risks.
In the setting of breast cancer, the underlying causal mechanisms have typically been framed in terms of roles for timing of reproductive and somatotropic hormones in cancer pathogenesis.32
In contrast, data suggests that the association of early menarche with cardiovascular risk is mediated through its correlations with short stature, elevated body mass index, and metabolic syndrome.
The timing of menarche is associated with several inter-related early life factors including birthweight, the timing of other pubertal events, and attained height; it will be important to sort out which specific developmental events are most strongly associated with chronic disease risks. Menarche is a late event in an integrated series of biological events that make up the pubertal transition. Female puberty typically begins with adrenarche, marked by a rise in adrenal androgens, which may be followed by thelarche, the initiation of breast development, or by pubarche, the development of pubic hair. Evidence suggests that early puberty may be triggered by the catch-up growth that occurs in children experiencing catch-up growth as a consequence of intrauterine growth retriction or negative energy balance in the postnatal period; insulin resistance and accelerated aging are common sequelae of this scenario and it is therefore possible that early menarche is associated with related health outcomes as consequences of a common cause. Prospective studies of development could be used to sort out which developmental events or phenotypes are most strongly associated with chronic disease risks in mid-life.
Recent studies have suggested that the association of early menarche with cardiovascular risks are mediated, in part, through associations with shorter stature, greater adiposity,29 and with components of metabolic syndrome, including central adiposity, dyslipidemia, hypertension14 and hyperglycemia.30,31 Many cancers are thought to be associated with metabolic syndrome, including breast, colon and endometrial cancers. Dyslipidemia and hyperglycemia have each been shown to be independently associated with breast and endometrial cancers.
Our data could be interpreted as supporting this for some cancers. In this study we observed that risk associations for endometrial, breast, and melanoma skin cancers were more strongly inverse in never users of hormone therapy compared to current users, suggesting that menarche and menopausal hormones may influence cancer risks through a common pathway. Endometrial, breast, and melanoma skin tumors in particular can be promoted by exposure to endogenous or exogenous estrogens. Similarly, WHI investigators observed greater risk of lung cancer in users of HRT. By contrast, research suggests that estrogens may play a preventive role for liver cancer. Liver cancer (HCC) has a striking gender difference in incidence; women are at lower risk than men; recent investigations have suggested that estrogens are protective and act through ER- mediated pathways. Similarly, the incidence of bladder cancer shows sexual dimorphism, with higher incidence in men compared to women, which persist even after adjusting for differences in smoking histories.38
Strengths of the study include large sample size, prospectively collected data, and a uniform analytic approach. All studied cancers had at least 300 cases. Even so, the number of outcomes for rarer cancers may have limited our ability to detect true associations. It is increasingly recognized that cancer subtypes may have different etiologic causes but we did not have data on histologic or intrinsic subtypes. In all studies, menarche was retrospectively reported. Because the exposure was reported prior to any cancer diagnosis, errors introduced by difficulties in recall would likely be non-differential, and therefore could result in a reduced ability to detect true associations. The data used in this study comes from cancer cohorts in developed nations only, and participants were predominantly white women; this could limit the generalizability of our findings.
These findings suggest that long-standing secular trends towards earlier menarche have likely contributed to the incidence of many of the most common cancers affecting women in the developed world. If the observed associations are causal, cases attributable to the one-year shift in the age at menarche observed retrospectively across birth years would account for 9.9% of endometrial cancer cases in this population (or n=448), 8.7% of liver cancer cases (n=29), 5.3% of melanoma cases (n=184), 4.2% of bladder cancer cases (n=63), 3.1% of colon cancer cases (n=166), 2.0% of lung cancer cases (n=143), and 2.0% of breast cancer cases (n=512). This assumes, of course, that the effects of the timing of menarche on cancer risks are independent of changes in the prevalence of other risk factors.
Several ecologic analyses have demonstrated strong inverse correlations (r2>40%) between group-level measures of female life expectancy and average ages at menarche, but the underlying causal relationships remain poorly understood (Thomas et al., 2001). The timing and tempo of pubertal events vary markedly both within and across human populations, and are predictive of adult phenotypes and health risks. This broad variation in menarche seen both within- and across populations, suggests an adaptive plasticity operating within the constraints of human biology and the impacts on reproductive fitness and survival. The timing and pace of pubertal development are thought to reflect reflect strategic choices about the allocation of resources toward growth, maintenance and reproduction.
In conclusion, declining age at menarche is associated with higher risk of more cancers than previously thought, including some cancers in which estrogens are thought to increase risk, and other cancers in which estrogens are thought to reduce risk. These findings highlight the need to understand the causes of secular trends in menarche and the causal pathways linking menarche with health risks in later life. The range of cancer outcomes associated with the age at menarche suggests that consideration of common underlying causal pathways could suggest broadly effective strategies for cancer prevention.
Menarche is the most dramatic milestone in the secondary sexual development of girls. Menarche marks the end of puberty, signaling the initiation of monthly hormonal cycles and a deceleration in vertical growth.1
An earlier event, adrenarche, is more subtle and can be detected biochemically as an increase in serum DHEAS, or clinically as development of adult body odor, greasy hair, and growth of axillary and pubic hair.
The timing of milestones in secondary sexual development (adrenarche, menarche, thelarche) is less correlated than it was in previous generations. (reference?)
It is clear that both genetic traits and environmental factors influence the age at menarche, but the broad variations in this phenotype, observed both within- and across populations, and its stereotypical alteration in response to modern life suggest that environmental cues play an important role (Gillette and Folinsbee, 2012). Because population-level declines in the average age at menarche in Europe occurred synchronously with historical trends towards greater attained heights, improvements in nutritional status have been thought a key factor driving both trends. More recent evidence suggests that improvements in sanitation and hygiene could also influence growth and development since delayed or altered exposures to microbes in early life can have profound and life-long effects on energy balance (Beard and Blaser, 2002; Cho and Blaser, 2012).
The association of menarche with breast cancer risks and with intermediate measures such as mammographic density tend to be modeled as linear relationships. Categorizations of the age at menarche that lump the earliest and median categories for age at menarche may mask variations in risk that are non-linear. It is possible that subtypes have distinct associations with menarche which would become more clear in subtype specific analyses or case-case comparisons.
While the age at menarche is often considered as a linear predictor of cancer risks, the shape of the associations with metabolic syndrome and cardiovascular risks are not linear but rather U shaped, with the greatest contrast observed when comparing women with early menarche to those who had menarche at the median age. It is therefore worth noting, that although most dose-response associations were adequately described using linear models, splines suggest changes in the slope of many associations at the median age of 12 y. This may reflect pleiotropic effects of menarche on risk whose effects may take effect at distinct parts of the range for menarcheal ages.
Early menarche occurs before age 11 y, and may be heralded by premature adrenarche, or an increase in adrenal production of androgens before age 8 y.
Adipocytes have estrogen receptors and androgen receptors.
The effects of estrogens and androgens on the distribution of body fat are influenced at least
in part through differences by sex and by depot-specific responses to sex-steroid hormones.
Mesenchymal stem cells give rise to adipocytes, chondrocytes,
The onset of puberty
Peak height velocity
Over the course of the 20th century the age at menarche fell declined by 3 months per decade.
Adrenarche is detected biochemically as an increase in serum DHEAS, or clinically as development of adult body odor, greasy hair, and growth of axillary and pubic hair.
It is clear that both genetic traits and environmental factors influence the age at menarche, but the broad variations in this phenotype, observed both within- and across populations, and its stereotypical alteration in response to modern life suggest that environmental cues play an important role (Gillette and Folinsbee, 2012). Several lines of evidence suggest that trajectories of pubertal growth and development are set in infancy or in the womb (D’Aloisio et al., 2013; Mishra et al., 2009). In utero growth restriction, low birthweight, and accelerated weight gain in infancy are each associated with lower ages at menarche . Because population-level declines in the average age at menarche in Europe occurred synchronously with historical trends towards greater attained heights, improvements in nutritional status have been thought a key factor driving both trends. More recent evidence suggests that improvements in sanitation and hygiene could also influence growth and development since delayed or altered exposures to microbes in early life can have profound and life-long effects on energy balance (Beard and Blaser, 2002; Cho and Blaser, 2012).