Health outcomes of age at menarche in European women: a two-sample Mendelian randomization study

Abstract

Background

Observational studies have shown an association between age at menarche (AAM) and the risk of gynecological diseases. However, the causality cannot be determined due to residual confounding.

Methods

We conducted a Mendelian randomization (MR) study to evaluate the causal effect of AAM on several gynecological diseases, including endometriosis, female infertility, pre-eclampsia or eclampsia, uterine fibroids, breast cancer, ovarian cancer, and endometrial cancer. Single nucleotide polymorphisms were used as genetic instruments. The inverse variance weighted method was used as the primary approach and several other MR models were conducted for comparison. Cochran’s Q test, Egger’s intercept test, and leave-one-out analysis were conducted for sensitivity analysis. Radial MR analysis was conducted when detecting the existence of heterogeneity.

Results

After Bonferroni correction and thorough sensitivity analysis, we observed a robust causal effect of AAM on endometrial cancer (odds ratio: 0.80; 95% confidence interval: 0.72–0.89; P = 4.61 × 10⁻⁵) and breast cancer (odds ratio: 0.94; 95% confidence interval: 0.90–0.98; P = .003). Sensitivity analysis found little evidence of horizontal pleiotropy. The inverse variance weighted method also detected weak evidence of associations of AAM with endometriosis and pre-eclampsia or eclampsia.

Conclusions

This MR study demonstrated a causal effect of AAM on gynecological diseases, especially for breast cancer and endometrial cancer, which indicates AAM might be a promising index to use for disease screening and prevention in clinical practice.

Key messages

• What is already known on this topic – Observational studies have reported associations between age at menarche (AAM) and a variety of gynecological diseases but the causality has not been determined.

• What this study adds – This Mendelian randomization study demonstrated that AAM causally affects the risk of breast cancer and endometrial cancer.

• How this study might affect research, practice, or policy – The findings of our study imply that AAM could be a candidate marker for early screening of populations at higher risk of breast cancer and endometrial cancer.

Introduction

Menarche, a hallmark event of sexual development, is defined as the first day of menstrual bleeding for an adolescent girl [1]. Studies have shown that age at menarche (AAM) might be a reference for predicting future health outcomes [2, 3].

The associations between AAM and women’s health conditions have been well-documented. For instance, a large amount of literature has reported that early AAM was associated with an increased risk of breast cancer [4]. Studies have also found a negative association of AAM with endometrial cancer [5]. However, the existing literature has failed to determine whether the detected associations are attributable to confounding bias or causality [6]. Besides, the results were inconsistent in some gynecological diseases. For example, the association between AAM and pre-eclampsia or eclampsia remains equivocal as previous studies on this topic yielded inconsistent results [7–9]. Thereby, whether AAM is causally associated with women’s health outcomes remains to be determined. If AAM is identified as a promising index to predict some health consequences, then it would be a novel target for disease early screening and management in clinical practice.

Mendelian randomization (MR) is a technique for causal inference [10]. Epidemiologists and methodologists have reported that the evidence level of the MR method is merely secondary to randomized controlled trials (RCTs) [11]. Therefore, when RCTs on certain topics are restricted by ethics, manpower, or material resources, the MR method is definitely the optimum alternative approach for causal inference. Researchers have regarded the MR technique as a natural RCT design [12, 13]. In detail, the MR approach uses single nucleotide polymorphisms (SNPs) from genome-wide association studies (GWASs) as instrumental variables to proxy exposure phenotypes [14]. Considering that genotypes are formatted prior to phenotype formation and the genetic variants are allocated at conception, reverse causality and residual confounding are less likely [15]. Using MR, the impact of AAM on cardiovascular diseases, neurological diseases, and endocrine diseases has been widely investigated [16–18]. However, the existing MR studies about the impact of AAM on gynecological diseases were lacking or focused on only a limited set of them [19, 20]. Without broader exploration, additional potential health outcomes influenced by AAM might be overlooked or unrevealed, like breast cancer, endometriosis, uterine fibroids, and so on. Herein, we conducted a two-sample MR study to evaluate the causal effect of AAM on a range of female heath outcomes, including endometriosis, female infertility, pre-eclampsia or eclampsia, uterine fibroids, breast cancer, ovarian cancer, and endometrial cancer. Our findings would systematically clarify the diseases possibly attributable to AAM and shed light on the screening and prevention of gynecological diseases.

Materials and methods

Study design

We performed a two-sample MR study to investigate whether there is a causal effect of AAM on the risk of several health outcomes in European women. SNPs were utilized as instruments to proxy AAM. A reasonable MR study should at least meet the following three assumptions: (i) genetic instruments should be robustly associated with the exposure investigated; (ii) genetic instruments should not be associated with confounders; (iii) genetic instruments should exert effects on the outcome only through the exposure focused on [21]. No individual data were used in this MR study. Only publicly available GWAS data were used, and the ethics approvals were obtained in the original GWASs.

Genome-wide association study data for age at menarche

To use SNPs proxied for AAM, we obtained the GWAS data on AAM conducted by the ReproGen Consortium, comprising 182 416 women of European descent [22]. Women reporting AAM as <7 years old or >17 years old were excluded from the GWAS analysis. We extracted SNPs associated with AAM at P < 5 × 10⁻⁸ and then ensured the independence of the SNPs by setting the linkage disequilibrium (LD) r² at 0.001 within 10 000 kb. We also calculated F statistics to ensure that all the SNPs were sufficient in statistical strength. We firstly calculated the R² that reflects the proportion of variance in the phenotype explained by a given SNP using the following formula [23]:

In the formula, MAF is minor allele frequency, β is the genetic effect of a given SNP on AAM, SE is standard error, and N is the sample size of the AAM phenotype. We then calculated the F statistics with the following formula:

Specifically, the F statistics were recommended to be >10 to ensure the validity of a strong instrument [24]. Therefore, SNPs with F < 10 were excluded from the MR analysis.

Genome-wide association study data for outcomes

To evaluate the causal effect of AAM on health outcomes in European women, several diseases were included, such as endometriosis, female infertility, pre-eclampsia or eclampsia, uterine fibroids, breast cancer, ovarian cancer, and endometrial cancer. All the datasets were constrained to women of European ancestry from different consortia, including the Breast Cancer Association Consortium (BCAC) [25], Endometrial Cancer Association Consortium (ECAC) [26], Epidemiology of Endometrial Cancer Consortium (E2C2) [26], Ovarian Cancer Association Consortium (OCAC) [27], UK Biobank consortium (http://www.nealelab.is/uk-biobank/), and FinnGen consortium [28].

The genetic information on breast cancer came from the BCAC that combined OncoArray, iCOGS, and GWAS meta-analysis, comprising 122 977 breast cancer cases and 105 974 controls [25]. Genotype imputation and principal component analysis (PCA) were conducted using SHAPEIT2, IMPUTE, MACH, or Minimac. PCA was performed to adjust for potential population stratification.

Summary-level genetic data for endometrial cancer came from a published GWAS analysis that included 17 previously reported studies from the ECAC, the E2C2, and UK Biobank [26]. In total, 12 906 cases and up to 108 979 controls of European ancestry were included. Genotype imputation and PCA were conducted.

For ovarian cancer, we obtained data from the OCAC consisting of up to 25 509 epithelial ovarian cancer cases and 40 941 controls, which came from seven genotyping projects that passed quality control [27].

The phenotypes of endometriosis, female infertility, and pre-eclampsia or eclampsia were from the FinnGen consortium, a growing project integrating genotype data from Finnish biobanks with digital health record data from Finnish health registries, aiming at 500 000 individuals by the end of 2023 [28]. FinnGen participants were genotyped with Illumina and Affymetrix chip arrays. Quality control and PCA were conducted in the FinnGen GWAS analysis.

The genetic associations of SNPs with uterine fibroids were obtained from the UK Biobank, comprising 462 933 individuals of European ancestry, of which 7122 cases were diagnosed with uterine fibroids without other cancers. Specifically, the summary data were obtained from the Medical Research Council Integrative Epidemiology Unit (MRC-IEU) UK Biobank GWAS pipeline.

The full GWAS datasets were all obtained from the IEU Open GWAS Project (https://gwas.mrcieu.ac.uk/). More detailed information for the outcome phenotypes was also presented in Table 1.

Statistical analysis

To facilitate MR estimation of the effects of AAM on the health outcomes of European women, we extracted the SNPs and their genetic information from the outcome data, including the genetic effect (β), standard error, effect allele, other allele, effect allele frequency, and P values. When certain SNPs were missing in the outcome data, proxied SNPs in LD with the missing SNPs at r² > 0.8 were alternatively used. If no appropriate proxies were available, then the missing SNPs were eliminated from the MR analysis. We then integrated the genetic information of the exposure- and outcome-SNPs and made sure that the effect alleles were concordant between them. Palindromic SNPs with intermediate allele frequency > 0.3 were excluded, and SNPs with incompatible alleles were also eliminated. Therefore, we obtained a set of SNPs ultimately eligible for MR analysis.

The inverse variance weighted (IVW) method [29] was undertaken as the primary method for causality inference as it has the strongest statistical power. The IVW estimation is elicited by combining the Wald ratios of the instrument included for analysis, and this method has been broadly applied in conventional MR analysis. In the present MR investigation, we used the random-effect model rather than fixed-effect model for IVW analysis because the random-effect IVW remains a conservative estimation with the existence of heterogeneity. As several outcomes were investigated, Bonferroni correction was conducted to calculate the multiple-testing-corrected P value. In detail, a significant MR estimate was defined as P < .007. Besides, MR estimates with .007 < P < .05 were identified as nominally significant results.

Association patterns derived from other MR models were also implemented to evaluate whether the MR estimations were robust. Specifically, we performed the weighted median method and MR-Egger regression method to assess the causal associations [30, 31]. The weighted median method is slightly less powerful than the random-effect IVW models, but is more conservative in inferring causalities as it assumes that <50% of the instruments are invalid. MR-Egger regression is the model with the least power as it assumes that all the SNPs are invalid. This method is typically applied for direction judgement. In addition, as the slope term of MR-Egger regression is not constrained to the origin, the derived intercept term could be used to assess horizontal pleiotropy [32]. Typically, consistent directions among distinct MR models could enhance the reliability of the MR results.

Sensitivity analysis was then conducted to evaluate whether the MR assumptions were violated. We conducted Cochran’s Q test to determine whether heterogeneity existed. A Cochran’s Q–derived P value < .05 suggests that heterogeneity is detected [33]. It should be noted that heterogeneity does not detract from the MR estimations as we used a random-effect model for the IVW method. Despite this, we still conducted radial MR to identify potential outliers that contribute to heterogeneity [29]. Once the outliers were detected, we further conducted MR analysis to evaluate the outliers-corrected estimates. We also investigated pleiotropy by evaluating the intercept term of the aforementioned MR-Egger regression method. More specifically, an intercept-derived P value < .05 indicated pleiotropy was detected, which could bias the MR results. For the significant MR results, we also conducted a leave-one-out analysis to investigate the stability of the pooled IVW estimates. Specifically, SNPs were dropped one by one and the IVW analyses were replicated after dropping the corresponding SNPs.

All the analyses were performed using the “TwoSampleMR” package (version 0.5.6) and the “RadialMR” package (version 1.0) in the R program (version 4.1.3).

Results

We finally obtained a set of SNPs for MR analyses. All the SNPs were strong enough with F > 10. As the outcomes were all binary phenotypes, MR estimations were reported as odds ratio (OR) with a 95% confidence interval (CI).

The IVW method detected a statistically significant effect of AAM on endometrial cancer (OR: 0.80; 95% CI: 0.72–0.89; P = 4.61 × 10⁻⁵) (Fig. 1). The weighted median and MR-Egger regression obtained consistent results (Table 2). Cochran’s Q test found little evidence of heterogeneity and Egger’s intercept test found no evidence of horizontal pleiotropy (Table 3). The leave-one-out analysis showed no SNPs to be exerting a strong effect on the overall IVW estimation (Fig. S1). Moreover, IVW also indicated a causal effect of AAM on pre-eclampsia or eclampsia (OR: 0.84; 95% CI: 0.72–0.99; P = .03) and no heterogeneity or pleiotropy was detected (Table 3). However, the weighted median and MR-Egger estimates were not significant (Table 2), and the leave-one-out analysis showed that the IVW estimate was not robust (Fig. S1).

For breast cancer, IVW did not find any evidence of a causal effect of AAM on it (OR: 0.98; 95% CI: 0.92–1.05; P = .58). However, the weighted median found that AAM played a protective role (OR: 0.93; 95% CI: 0.87–0.99; P = .02). Cochran’s Q test detected considerable heterogeneity, which might bias the MR results. Therefore, further radial MR was conducted to identify and exclude outliers to replicate the MR analyses (Fig. 2). The outliers-corrected MR estimates showed a significant causal effect of AAM on the risk of breast cancer (Table 4). Leave-one-out analysis indicated the overall IVW estimate was robust.

Likewise, MR estimates found no evidence of causal associations between AAM and endometriosis. Given that heterogeneity was detected (Table 3), radial MR was conducted and the outliers-corrected IVW showed a significant effect. However, the leave-one-out analysis indicated the causality was unstable (Fig. S1).

For female infertility, ovarian cancer, and uterine fibroids, no evidence of causality with AAM was observed in the primary MR analysis or the outliers-corrected MR analysis.

Discussion

Previous studies have focused much on the relationship between AAM and women’s health outcomes. However, owing to the flaws of conventional observational studies like unmeasurable confounders, the causality of AAM on women’s health outcomes remains to be elucidated. The present study found a protective role of later AAM on two women’s cancers, including breast cancer and endometrial cancer, using a two-sample MR design. We also found some evidence of an association between AAM and pre-eclampsia or eclampsia as well as endometriosis.

Age at menarche and breast cancer

Previously, observational studies have found a relationship between AAM and breast cancer. Bui et al. [34] found that earlier AAM was associated with an increased risk of breast cancer in a retrospective case–control study conducted in Northern Vietnam. Similarly, the Collaborative Group on Hormonal Factors in Breast Cancer [35] found that the risk of breast cancer increased in those younger at menarche in a meta-analysis of 117 epidemiological studies comprising 118 964 women with breast cancer. In line with previous studies, the present MR study confirmed a causal effect of AAM on the risk of breast cancer. Specifically, later AAM was associated with a decreased risk of breast cancer. Early onset of menarche indicates an earlier and greater cumulative exposure to estrogen, which can increase the risk of breast cancer. More mechanisms underlying the effect of AAM on breast cancer risk are warranted for further investigation.

Age at menarche and endometrial cancer

The association between AAM and endometrial cancer has long been noted. Gong et al. [5] found that each 2-year delay of AAM was associated with a 4% decreased risk of endometrial cancer. Fuhrman et al. [36] also found an inverse association between AAM and the risk of endometrial cancer in a large-scale longitudinal study containing 536 450 women from nine cohorts. Though the relationship between AAM and endometrial cancer has been widely investigated, the causality could not be determined based on the existing evidence as observational studies are susceptible to confounding. To this end, we conducted an MR study and demonstrated a protective role of later AAM in endometrial cancer risk, which is less vulnerable to confounding or reverse causality. Estrogen receptors are expressed in endometrial tumors and thus a higher circulating estrogen level is associated with an increased risk of endometrial cancer. Studies have revealed that exposure to estrogen could cause malignant transformations through somatic mutations of endometrial cells through increasing mitotic activity, DNA replication, and somatic mutations in endometrial cells [37–39]. In this regard, later AAM indicates a delayed exposure to estrogen, leading to a decreased risk of endometrial cancer.

Age at menarche and other women’s health outcomes

The present MR investigation also detected weak evidence of associations of AAM with endometriosis and pre-eclampsia or eclampsia. However, sensitivity analyses suggested the observed causalities might not be robust, thus making conclusive interpretation unavailable. Previous studies on this topic also yielded discrepant results. Marcellin and colleagues found no evidence of the association between AAM and endometriosis in a cross-sectional study comprising 789 women [40]. Nnoaham et al. [41], however, in a meta-analysis of case–control analyses, observed a detrimental effect of early AAM on the risk of endometriosis. Likewise, several studies observed a negative association between AAM and pre-eclampsia or eclampsia and some studies did not [7–9]. Based on the equivocal results derived from previous studies and the relatively weak evidence observed in our MR analysis, the evidence about the causality of AAM on endometriosis and pre-eclampsia or eclampsia could not be established.

Our MR study also showed that female infertility, ovarian cancer, and uterine fibroids might not be attributed to AAM. Some studies but not all studies have observed an association of AAM with them. The inconsistent results might be due to confounders commonly existing in observational studies. Our MR study found no evidence of causality between them and sensitivity analysis suggested the results were robust. Therefore, AAM might not be a candidate index to screen specific populations with higher risks of female infertility, ovarian cancer, or uterine fibroids.

Strengths and limitations

Using an MR framework, our study has several strengths. First, benefiting from the availability of large-scale GWAS data, we comprehensively investigated a range of women’s health outcomes that might correlate with AAM. Most studies focused merely on a limited range of women’s health outcomes before. Investigating the distinct impacts of AAM on different gynecological diseases would help understand the heterogeneity of biological mechanisms among them. Second, by combining the MR estimations and sensitivity analysis, the present MR study is less vulnerable to reverse causality and confounders which are typically inevitable in observational studies, thus making the causality inference more reliable. Third, all the participants were constrained to European women in the present MR investigation, and thereby bias owing to population stratification was less likely.

There are some limitations in this MR study. First, although constricting participants to European ancestry could avoid population heterogeneity, whether the current findings can be generalized to other populations remains unclear. Therefore, future studies are warranted to confirm the generalization of our results in other populations. Second, given that only summary-level data were publicly available, we were not allowed to perform stratified analysis. Third, it should be noted that although we found no evidence of associations between AAM and the risk of certain gynecological diseases, it does not mean that AAM had no impact on the disease prognosis as factors that influence the risk and prognosis of a disease could be distinct.

Conclusion

Our MR study demonstrates a causal effect of AAM on the risk of breast cancer and endometrial cancer, which implies that AAM could be a promising marker for screening populations with a higher risk of breast or endometrial cancer.

Acknowledgements

We are grateful for all the GWASs making the summary-level data publicly available.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Conflict of interest statement: The authors declare no conflict of interest.

References