{"paper_id":"568ddd6d-e63b-4644-9594-522d847d48ef","body_text":"Clin. Exp. Obstet. Gynecol. 2026; 53(8): 49264\nhttps://doi.org/10.31083/CEOG49264\nCopyright: © 2026 The Author(s). Published by IMR Press.\nThis is an open access article under the CC BY 4.0 license .\nPublisher’s Note: IMR Press stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.\nOriginal Research\nCausal Effects of Circulating Sex Hormone Levels on the Risk of\nEndometriosis: A Two-Sample Mendelian Randomization Study\nJingnan Zhu1,2,†\n , Chong Zhu 1,†\n , Limiao Lu 1\n , Rubing Zheng 1\n , Y anwen Wang1\n ,\nY uchen Xue1\n , Mei Wang1\n , Qiang Y an2\n , Bin He 1,*\n1Department of Obstetrics and Gynecology, People’s Hospital of Pudong New Area, 200092 Shanghai, China\n2Institute of Maternal-Fetal Medicine and Gynecologic Oncology, Shanghai First Maternity and Infant Hospital, School of Medicine, Tongji University,\n200092 Shanghai, China\n*Correspondence: hebin@shpdph.com (Bin He)\n†These authors contributed equally.\nAcademic Editor: Osamu Hiraike\nSubmitted: 19 December 2025 Revised: 13 February 2026 Accepted: 9 March 2026 Published: 19 August 2026\nAbstract\nBackground: Endometriosis is a chronic, estrogen-dependent, immune-modulated disorder with unclear causal contributions from cir-\nculating sex hormones. Understanding whether hormone levels directly influence risk may guide prevention and targeted therapies. To\nevaluate the causal effects of circulating sex hormones—estradiol (E2), total testosterone (TT), free testosterone (FT), estrone (E1), and\nsex hormone–binding globulin (SHBG)—on the risk of endometriosis using a two-sample Mendelian randomization (MR) framework,\nwith clinical contextualization from a tertiary-care cohort. Methods: A two-sample MR study was performed using large, sex-stratified,\nEuropean-ancestry genome-wide association studies (GW AS) summary statistics for hormones and endometriosis harmonized via MR-\nBase/OpenGW AS. Independent genome-wide significant Single Nucleotide Polymorphisms (SNPs) served as instruments, with multiple\nsensitivity and pleiotropy-robust estimators (MR-Egger, weighted median/mode, MR-PRESSO, radial MR). Multivariable MR accounted\nfor SHBG and body mass index (BMI). A local cohort of 50 reproductive-age women with surgically or histologically confirmed en-\ndometriosis was analyzed descriptively (IBM SPSS Statistics 26.0) for phenotype benchmarking. Results: In primary analyses, higher\ngenetically proxied estradiol (E2) increased endometriosis risk (IVW OR 1.18), while total testosterone (TT) and free/bioavailable testos-\nterone (FT) were protective. SHBG showed a modest association that attenuated in multivariable models. Estrone (E1), evaluated as\nan exploratory exposure, was not statistically significant. E1 was not significant (OR 1.10, p = 0.100). Sensitivity analyses showed no\nmajor directional pleiotropy; >98% of SNPs satisfied Steiger directionality. Reverse MR found no evidence for endometriosis affecting\nhormone levels. The tertiary-care cohort reflected the typical clinical spectrum, reinforcing generalizability. Conclusions: This MR\nanalysis supports a causal role for higher estradiol in increasing, and higher testosterone in reducing, endometriosis risk, with SHBG\nacting indirectly via steroid modulation. Findings were robust across sensitivity checks and integrated with clinical and mechanistic\nevidence, suggesting endocrine-targeted strategies may have preventive or risk-stratification potential.\nKeywords: endometriosis; estradiol; testosterone; mendelian randomization; sex hormone–binding globulin\n1. Introduction\nEndometriosis is a chronic, estrogen-dependent in-\nflammatory condition characterized by the presence of\nendometrial-like tissue outside the uterine cavity, affecting\nan estimated 5–10% of women of reproductive age and a\nsubstantial proportion of those with pelvic pain or infer-\ntility [1]. Although once framed as a disorder confined to\nthe pelvis, contemporary clinical and translational perspec-\ntives increasingly recognize endometriosis as a systemic\ndisease with multi-organ involvement, complex symptoma-\ntology (pelvic pain, subfertility, fatigue, bowel and bladder\ndysfunction), and long diagnostic delays that erode qual-\nity of life and productivity [ 2]. Over the last decade, shifts\nin our understanding of its pathobiology—from retrograde\nmenstruation alone to integrated models involving immune\ndysregulation, neuroangiogenesis, and endocrine drivers—\nhave reshaped the research agenda and the therapeutic land-\nscape [ 3]. These advances are essential backdrops for\ncausal inquiry into circulating sex hormones: while obser-\nvational data suggest robust associations between hormone\nlevels and endometriosis risk or severity, such designs are\nsusceptible to confounding and reverse causation, under-\nscoring the need for genetic approaches that strengthen\ncausal inference [ 1,2]. Beyond classical pelvic manifesta-\ntions, endometriosis can also present in extra-pelvic loca-\ntions, including the thoracic cavity, abdominal wall, sur-\ngical scars, and rarely even distant organs. Thoracic en-\ndometriosis syndrome, for example, illustrates the capac-\nity of ectopic endometrial tissue to respond to cyclic hor-\nmonal signals outside the pelvis, reinforcing the systemic\nendocrine dependence of the disease. These atypical phe-\nnotypes further emphasize that circulating hormonal mi-\nlieus may influence not only pelvic lesion establishment\nbut also dissemination and persistence at distant anatomi-\ncal sites [3].\n\nA central tenet of endometriosis biology is estro-\ngen dependence coupled with altered progesterone signal-\ning within ectopic lesions and eutopic endometrium. Pro-\ngesterone resistance manifest as impaired decidualization,\naberrant receptor expression, and downstream transcrip-\ntional rewiring has emerged as a hallmark that may precede\nlesion establishment and perpetuate inflammation and pain\n[4]. Recent syntheses extend this concept to therapeutic in-\nnovation, proposing ways to bypass or reverse progesterone\ninsensitivity and aligning molecular readouts with clinical\nendpoints [5]. Complementing the progesterone narrative,\nestrogen receptor (ER) signaling through ERα and ERβ iso-\nforms modulates proliferation, inflammatory tone, and no-\nciception in lesion microenvironments; dysregulated ERβ,\nin particular, is implicated in apoptosis resistance and cy-\ntokine production [ 6]. These receptor-level insights mesh\nwith the demonstrable relevance of local estrogen biosyn-\nthesis, where aromatase expression and intracrine estrogen\nformation sustain lesion growth and symptom persistence,\nand where pharmacologic aromatase inhibition offers a bio-\nlogically plausible, if clinically nuanced, management strat-\negy [ 7]. Indeed, local estrogen formation via aromatase\nand other steroidogenic enzymes in ectopic tissue provides\na self-reinforcing loop that can operate even when systemic\nestrogen is nominal, highlighting why both circulating and\ntissue-level hormone dynamics warrant scrutiny in causal\nframeworks [8].\nThe endocrine-immune interface is increasingly rec-\nognized as a bidirectional driver of lesion establishment,\nneuroangiogenesis, and pain amplification. Cross-talk\namong steroid receptors, macrophage subsets, T cells,\nmast cells, and stromal fibroblasts reprograms cytokine mi-\nlieus and extracellular matrix, with endocrine cues shap-\ning immune cell recruitment and function while immune\nmediators in turn modulate steroidogenesis and receptor\nsignaling [ 9]. These mechanistic vistas dovetail with a\nbroader, integrative view of endometriosis that includes\ndiet, metabolism, and psychosocial stressors, each inter-\nsecting with hormonal axes to influence disease expression\nand quality of life. Comprehensive appraisals now empha-\nsize multimodal management that combines medical ther-\napy, surgery, lifestyle interventions, and attention to comor-\nbidities, while acknowledging persistent unmet needs and\nthe potential for precision approaches rooted in molecular\nphenotyping [ 10]. Within this context, disentangling the\ncausal role of circulating sex hormones in disease onset—as\ndistinct from correlates of established disease or treatment\neffects—has direct implications for prevention, risk strati-\nfication, and therapeutic development.\nGenomic discoveries have transformed endometriosis\nfrom an enigmatic clinical syndrome to a tractable complex\ntrait with reproducible risk loci and shared genetic under-\npinnings with pain and inflammatory conditions. Large-\nscale genome-wide association studies (GW AS) and meta-\nanalyses have delineated dozens of risk regions pointing\nto pathways in hormone signaling, inflammation, and tis-\nsue remodeling, and they reveal genetic correlations with\ncomorbid pain disorders, migraine, and autoimmune traits\n[11]. These insights invite causal questions: are circu-\nlating sex hormone levels upstream determinants of en-\ndometriosis risk, or do observed associations arise from\nconfounding by shared genetic architecture or lifestyle fac-\ntors? Moving beyond correlation requires analytic strate-\ngies that harness genetic variants as proxies for lifelong dif-\nferences in exposures—an approach well-suited to the high-\npolygenicity and modest effect sizes characteristic of hor-\nmone traits and endometriosis alike [ 11].\nThe feasibility of such approaches pivots on the emer-\ngence of deeply phenotyped, genotyped population co-\nhorts that provide statistical power and harmonized pheno-\ntypes for both exposure and outcome. The UK Biobank,\nwith its breadth of biochemical assays, health records, and\ngenotypes on ~500,000 participants, has become a cor-\nnerstone for constructing genetic instruments for circulat-\ning hormones and for ascertaining endometriosis diagnoses\nwith linked hospital and primary-care data [ 12]. In par-\nallel, FinnGen, integrating national health registries with\nbiobank-scale genotyping in a founder-influenced popula-\ntion, offers complementary power and trait architecture, en-\nabling replication and generalization across Chinese ances-\ntries and providing dense case ascertainment for gyneco-\nlogic phenotypes [13]. Such large-scale biobanks have cat-\nalyzed a new wave of two-sample Mendelian randomiza-\ntion (MR), where exposure and outcome summary statistics\nfrom non-overlapping samples are combined to test causal\nhypotheses with fewer biases than traditional observational\ndesigns [12,13].\nMethodological and infrastructural advances have\nmade such analyses scalable and transparent. MR-Base has\nstandardized access to thousands of curated GW AS sum-\nmary datasets, coupled with instrument selection, harmo-\nnization, and sensitivity analyses that lower the barrier to\nrobust causal inference across the phenome [ 14]. Comple-\nmenting this, the MRC IEU OpenGW AS infrastructure ex-\nposes a growing corpus of harmonized GW AS through pro-\ngrammatic interfaces, facilitating reproducible two-sample\nMR pipelines that can integrate exposure GW AS for sex\nsteroids with outcome GW AS for endometriosis [ 15].\nThese platforms reduce analytic idiosyncrasy, foster sen-\nsitivity to pleiotropy and heterogeneity, and enable multi-\ndataset triangulation—features critical when interrogating\nhormones whose biosynthesis, transport, and receptor sig-\nnaling are enmeshed in complex physiological networks\n[14,15].\nRobust MR analysis depends on strong, specific ge-\nnetic instruments for the exposures of interest. Pertinent\nto sex hormones, several large GW AS have mapped com-\nmon variants that influence circulating testosterone, estra-\ndiol, estrone, and binding proteins, providing instruments\nwith sufficient F-statistics and biological interpretability.\nFor testosterone, sex-stratified analyses in population co-\nhorts have identified variants near SHBG, JMJD1C, and\n2\n\n\nSRD5A2, among others, and demonstrated sex-specific ar-\nchitecture that matters for instrument selection in female-\nfocused outcomes like endometriosis [ 16]. Estradiol, while\nchallenging to measure at scale due to assay sensitivity\nand cycle variability, has been interrogated via GW AS that\nnonetheless identify instruments and lend themselves to\ncausal analyses in bone and cardiometabolic traits—proof\nof concept that estradiol instruments can be informative\nfor disease outcomes [ 17]. Estrone, a key estrogen in\npostmenopausal physiology and a metabolite interlinked\nwith estradiol pathways, has recently been mapped geneti-\ncally, illuminating regulatory loci and offering new instru-\nments that may generalize, with caveats, to premenopausal\ncontexts through shared enzymatic pathways [ 18]. Be-\nyond single-hormone efforts, a large multi-hormone GW AS\nspanning over 200,000 individuals cataloged novel loci\nand sex-dependent effects for testosterone, sex hormone-\nbinding globulin (SHBG), and other sex steroids, expand-\ning the menu of instruments and enabling multivariable MR\nto parse correlated hormone effects [ 19].\nAccumulating MR evidence indicates that sex hor-\nmones exert causal influences on diverse diseases, setting\na precedent for interrogating endometriosis specifically.\nPhenome-wide MR leveraging UK Biobank instruments\nhas linked genetically proxied testosterone and SHBG to\ncardiometabolic outcomes, cancers, and reproductive traits,\nwhile emphasizing sex-specific causal patterns that cau-\ntion against naive pooling across sexes [ 20]. Focused two-\nsample MR has also begun to quantify the causal contri-\nbution of endogenous hormones to female cancer risks,\ndemonstrating both the feasibility of hormone MR at scale\nand the importance of dissecting hormone-dependent ma-\nlignancies with careful sensitivity analyses [ 21]. Multi-\nomics MR that integrates hormone instruments with adi-\nposity, glycemic traits, and reproductive endpoints further\nunderscores the intricate causal web connecting obesity,\nsex steroids, and reproductive health—relationships that are\ncrucial to consider when estimating direct hormone effects\non endometriosis risk and when guarding against confound-\ning by metabolic pathways [ 22].\nThe social and behavioral correlates of hormone lev-\nels are not merely epidemiologic curiosities but potential\nconfounders if left unaddressed. Genetic analyses linking\ntestosterone to socioeconomic position, educational attain-\nment, and health behaviors complicate causal interpreta-\ntions in conventional observational studies, where lifestyle,\nstress, and access to care are entangled with both hor-\nmone levels and endometriosis diagnosis [ 23]. MR’s use\nof germline proxies offers a route past many such con-\nfounders, but it also raises the bar for instrument valid-\nity and pleiotropy assessment—particularly for hormones\nwith broad systemic effects and for outcomes like en-\ndometriosis that have heterogeneous clinical presentations\nand care pathways [ 23]. These considerations motivate a\ndesign that pairs stringent instrument selection with sensi-\ntivity analyses capable of detecting and mitigating horizon-\ntal pleiotropy and residual confounding, while also contem-\nplating multivariable frameworks in which correlated hor-\nmones (e.g., estradiol, testosterone) and carriers (SHBG)\nare modeled jointly [ 19,20,22,23].\nFrom the vantage point of disease biology, genomic\nstudies have reinforced the plausibility that sex steroids\nare upstream drivers of endometriosis pathogenesis. Re-\nviews synthesizing GW AS loci, expression quantitative\ntrait loci (eQTLs), and functional annotations highlight con-\nvergence on hormone receptor signaling, steroidogenesis,\nand endometrial biology, providing mechanistic footholds\nfor interpreting any causal estimates that emerge from MR\n[24]. Equally salient is the recognition of endometriosis\nas an immunological disease, wherein macrophage acti-\nvation, complement pathways, and adaptive immune re-\nsponses are woven into lesion survival and pain; because\nsex steroids shape immune function—from antigen presen-\ntation to cytokine secretion—disentangling direct hormonal\neffects from immune-mediated pathways is both biologi-\ncally necessary and methodologically challenging [ 25]. In\npractice, this means that MR estimates for hormones should\nbe triangulated against immune-trait MR, genetic correla-\ntions, and pathway analyses to parse mediation versus direct\naction—an agenda that builds logically on the cited immune\nand endocrine interplay [ 24,25].\nTherapeutically, the endocrine background of en-\ndometriosis management remains central: combined oral\ncontraceptives, progestins, GnRH analogues and antago-\nnists, and aromatase inhibitors aim to suppress ovulation,\nreduce estrogen exposure, or overcome progesterone re-\nsistance to alleviate symptoms and reduce lesion activity\n[26]. While effective for many, these treatments are not\ncurative, and side-effect profiles, contraindications, and re-\ncurrence after discontinuation underscore the need to un-\nderstand whether lifelong differences in endogenous hor-\nmone levels causally alter risk—not only symptom tra-\njectories in established disease. If circulating hormone\nlevels are shown to causally influence incidence, preven-\ntive or risk-reducing strategies (pharmacologic or lifestyle)\nmight be identified for at-risk populations, and genetic risk\nprofiling could inform earlier evaluation for symptomatic\nindividuals [ 26]. Conversely, if MR suggests little or\nno causal effect of certain hormones on risk, that would\nredirect attention to lesion-autonomous steroidogenesis or\ndownstream immune-neural circuits as primary drivers—\nan insight equally valuable for rational drug development\n[24,25,26].\nPopulation-level burden estimates lend urgency to\nthese causal questions. Updated global burden analyses in-\ndicate that endometriosis remains common, with substan-\ntial years lived with disability attributable to pain, subfer-\ntility, and comorbid conditions; geographic variation re-\nflects differences in diagnosis, access to care, and perhaps\nenvironmental exposures that may interact with hormonal\npathways [ 27]. Temporal trends further suggest persistent\nunderdiagnosis and disparities across regions and health\n3\n\nsystems, magnifying the societal costs and underlining the\nneed for preventive frameworks grounded in causal biology\nrather than descriptive association [27]. These perspectives\nmake a compelling case for genetic epidemiology to com-\nplement clinical research: if we can infer how modifiable\nexposures and endogenous physiologic axes cause disease,\nwe can better target interventions, deploy resources, and re-\nfine diagnostic pathways.\nRecent assessments of global trends corroborate the\nscale of the challenge and hint at shifts in incidence and\ndetection that coincide with evolving diagnostic practices\nand awareness campaigns. Although differences in cod-\ning, imaging access, and surgical thresholds complicate\ncross-country comparisons, the enduring burden from 1990\nto 2021 and the ongoing need for high-quality surveil-\nlance emphasize why causal evidence that travels across\nsettings is vital [ 28]. MR, by leveraging germline vari-\nation fixed at conception, offers a tool comparatively in-\nsulated from many environmental and healthcare-system\nconfounders, enabling estimates that may generalize across\npopulations represented in the discovery GW AS. Nonethe-\nless, attention to ancestry, instrument transportability, and\nbiobank-specific ascertainment is essential when interpret-\ning and contextualizing findings, particularly for traits\ntied to reproductive physiology and care-seeking behavior\n[12,13,27,28].\nThe reproductive endocrine axis is broader than estra-\ndiol and testosterone alone. Anti-Müllerian hormone\n(AMH), a marker of ovarian reserve and folliculogenesis,\nhas robust heritability and now, with meta-analytic GW AS\nin premenopausal women, better-characterized genetic ar-\nchitecture [ 29]. While AMH is not the primary exposure\nin our focus, its genetic correlates and biological roles un-\nderscore the importance of considering ovarian reserve and\ncycle dynamics as potential mediators or confounders when\nevaluating the causal impact of sex steroids on endometrio-\nsis risk. Instruments for estradiol and testosterone may cap-\nture upstream regulators (e.g., hypothalamic-pituitary sig-\nnaling, steroidogenesis enzymes) that also modulate follicle\ndynamics; clarifying these relationships will improve the\ninterpretability of causal estimates and the biological sto-\nries we tell with them [ 17,18,19,29].\nFinally, polygenic risk approaches have begun to map\nthe phenomic footprint of endometriosis liability across the\nmedical record, revealing associations with gynecologic,\npain, and psychiatric phenotypes that invite mechanistic\nhypotheses and raise flags about potential collider bias in\nclinic-based samples [ 30]. Such PheW AS-scale observa-\ntions reinforce the need for careful design in causal analy-\nses: selection on clinical diagnosis can induce bias if ge-\nnetic liability to endometriosis co-varies with healthcare-\nseeking behavior or diagnostic intensity, and hormone lev-\nels themselves may influence contact with the healthcare\nsystem. Two-sample MR that draws exposure and outcome\nfrom distinct, large-scale GW AS—paired with harmoniza-\ntion, outlier detection, and pleiotropy-robust estimators—\noffers a principled way to mitigate these concerns, while\nsensitivity analyses and triangulation with related traits\n(e.g., SHBG, body mass index (BMI)) can probe the sta-\nbility of conclusions [ 14,15,19,20,22,30].\nTherefore, this study aims to utilize a two-sample\nMendelian randomization framework based on the aggre-\ngated data from genome-wide association studies (GW AS)\nof European ancestry to explore the effects of genetic\npredictors of circulating estradiol (E2), total testosterone\n(TT), free/bioavailable testosterone (FT), estrone (E1), and\nsex hormone-binding globulin (SHBG) on the risk of en-\ndometriosis. Secondary objectives include evaluating the\nindependence and robustness of these associations through\nmultivariate MR, reverse MR, and sensitivity analyses, and\nproviding contextual information on the genetic findings by\ncombining the descriptive data from the local tertiary med-\nical cohort.\n2. Materials and Methods\nWe conducted a two-sample Mendelian randomiza-\ntion (MR) study at a tertiary care, university-affiliated hos-\npital. The analytic framework was specified a priori to sat-\nisfy instrumental variable assumptions (relevance, indepen-\ndence, exclusion restriction) and to align with STROBE-\nMR reporting standards. Publicly available, de-identified\nGW AS summary statistics were used for the MR compo-\nnent (exposures: circulating sex hormones; outcome: en-\ndometriosis) (Fig. 1). A local hospital cohort (n = 50)\nwas used only for descriptive benchmarking and phenotype\nface-validity checks; it did not contribute to MR instru-\nment selection or effect estimation. Descriptive statistics\nfor the local cohort were generated using IBM SPSS Statis-\ntics 26.0 (IBM Corporation, Armonk, New Y ork, USA).\nContinuous variables were assessed for normality using the\nShapiro-Wilk test and summarized as mean ± SD or me-\ndian (IQR). Categorical variables were summarized as fre-\nquencies and percentages. No inferential comparisons or\nregression modeling were performed. The cohort analysis\nwas descriptive only.\n2.1 Data Sources\n2.1.1 Exposure GW AS (Sex Hormones)\nPrimary exposures were genetically proxied circu-\nlating sex hormone concentrations: estradiol (E2), total\ntestosterone (TT), free testosterone (FT; modeled via TT\nand SHBG in multivariable MR), estrone (E1), and sex\nhormone–binding globulin (SHBG). We used sex-stratified,\nEuropean-ancestry GW AS with rigorous assay quality con-\ntrol (QC) and adjustment for age, principal components\nof ancestry, genotyping batch, and cohort-specific covari-\nates, curated via OpenGW AS/MR-Base and recent consor-\ntia publications (Supplementary Tables 1,2).\nFinal MR analyses were restricted to European-\nancestry GW AS summary statistics for both the exposures\nand outcome to reduce bias arising from population strat-\n4\n\n\nFig. 1. Conceptual framework for mendelian randomiza-\ntion analysis. SNPs, Single Nucleotide Polymorphisms; GW AS,\nGenome-Wide Association Studies; MR, Mendelian Randomiza-\ntion.\nification. The UK Biobank, FinnGen, and consortium\ndatasets used or discussed in this study primarily com-\nprised participants of European ancestry. The estradiol\nGW AS was adjusted for age, genetic principal components\n(PC1–PC10), genotyping batch, and recruitment center.\nTestosterone and SHBG GW AS similarly adjusted for age,\nage2 (where specified), genetic principal components, assay\nbatch effects, and study-specific covariates.\n2.1.2 Outcome GW AS (Endometriosis)\nThe primary outcome was “any endometriosis”, de-\nfined in large biobank/consortium meta-analyses from\nICD-9 (617.x) and ICD-10 (N80.x) codes and/or surgi-\ncal/histologic confirmation. Where adequately powered,\nwe examined sub-phenotypes (ovarian endometrioma, deep\ninfiltrating endometriosis, superficial peritoneal disease).\nThe endometriosis GW AS was adjusted for age, genetic\nprincipal components, and recruitment center, with logis-\ntic regression performed under additive genetic models.\n2.1.3 Local Hospital Cohort (Contextualization Only; n =\n50)\nFrom the tertiary-care hospital Electronic Health\nRecord (EHR), we assembled a consecutive convenience\nsample of 50 reproductive-age patients with an endometrio-\nsis diagnosis. Inclusion criteria: (i) ICD-10 N80.x code or\nequivalent surgical/pathology confirmation; (ii) Reproduc-\ntive age was defined as 18–45 years; (iii) complete basic de-\nmographic and anthropometric data. Anthropometric vari-\nables included height (cm), weight (kg), BMI (kg/m 2), and\nparity status. Core variables included age, BMI, parity, di-\nagnosis confirmation, and treatment type. Cases missing\nany of these variables were excluded from descriptive sum-\nmaries. Exclusion criteria: (i) missing core variables af-\nter de-identification; (ii) malignant gynecologic disease at\nindex encounter. V ariables abstracted included age, BMI,\nparity, pain and infertility codes, imaging/surgical confir-\nmation, and key treatments. Patients with documented\nendocrine disorders (thyroid dysfunction, hyperprolactine-\nmia, Polycystic Ovary Syndrome.(PCOS), Cushing syn-\ndrome) were excluded from the descriptive cohort. These\ndata were analyzed descriptively in IBM SPSS Statistics\n26.0 and served solely to contextualize case mix at a ter-\ntiary center; no records from this cohort overlapped with\nexposure or outcome GW AS used for MR.\n2.1.4 Instrument Selection and Clumping\nFor each exposure, we selected independent Single\nNucleotide Polymorphisms (SNPs) associated at genome-\nwide significance ( p < 5 × 10 –8). Where instruments were\nsparse (e.g., E2), a prespecified sensitivity instrument set\nused p < 5 × 10 –6 with stricter strength criteria (mean F\n>20; per-SNP F >10). Independence was enforced via\nLD clumping (1000 Genomes EUR reference): r 2 < 0.001\nwithin a 10,000-kb window (primary) and r 2 < 0.01/5,000-\nkb (sensitivity). Missing exposure SNPs in the outcome set\nwere proxied with LD r 2 ≥ 0.80 within 500-kb when avail-\nable. Palindromic A/T or C/G SNPs with MAF 0.42–0.58\nwere excluded; others were harmonized using allele fre-\nquencies. Instrument strength (mean F-statistic) and expo-\nsure variance explained (R 2) were computed from reported\nbetas and SEs. Instrument strength was evaluated using the\nF-statistic, calculated as:\nF = (β2_exposure) / (SE2_exposure) For multi-SNP in-\nstruments, mean F-statistic was calculated across included\nvariants. An F-statistic >10 is conventionally consid-\nered indicative of strong instruments. We applied a more\nconservative threshold (mean F >20) to minimize weak-\ninstrument bias.\n2.1.5 Harmonization\nExposure and outcome summary statistics were har-\nmonized to the exposure-increasing allele with removal\nof ambiguous/discordant variants. We minimized sam-\nple overlap by choosing distinct biobanks/consortia; where\noverlap could not be definitively excluded, we relied on\npleiotropy-robust estimators in sensitivity analyses. Expo-\nsure and outcome GW AS summary statistics were derived\nfrom distinct European-ancestry consortia datasets. Based\non consortium documentation and recruitment sources, di-\n5\n\nrect participant-level overlap is unlikely. However, because\ncomplete individual-level cross-referencing was not avail-\nable, residual overlap cannot be entirely excluded.\nIn the presence of sample overlap, weak instruments\nmay bias estimates toward the confounded observational\nassociation. The relatively strong instrument strength ob-\nserved (mean F >27 for all exposures) reduces this con-\ncern, and concordance across pleiotropy-robust estimators\nfurther mitigates overlap-related bias.\n2.1.6 Primary MR Analysis\nThe primary estimator was inverse-variance weighted\n(IVW) MR with multiplicative random effects, report-\ning odds ratios (OR) for endometriosis per 1-SD geneti-\ncally predicted increase in hormone level with 95% CIs.\nCochran’s Q assessed heterogeneity.\n2.2 Sensitivity and Robustness Analyses\nTo probe horizontal pleiotropy and robustness, we per-\nformed:\n● MR-Egger regression (intercept test and slope es-\ntimate);\n● Weighted median and weighted mode estimators;\n● MR-PRESSO global test with outlier removal and\noutlier-corrected IVW;\n● Radial MR (radial IVW/Egger) to detect high-\ninfluence points (|standardized residual| >3);\n● Leave-one-out analyses;\n● Steiger directionality to confirm variance ex-\nplained is greater for exposure than outcome.\nWe screened instruments in phenotype association re-\nsources (e.g., PhenoScanner-like catalogs) to flag associ-\nations ( p < 1 × 10 –5) with potential confounders (BMI,\nsmoking, age at menarche, PCOS). Biologically implausi-\nble or evidently pleiotropic variants were prespecified for\nexclusion in sensitivity runs, with side-by-side reporting. A\nscreening threshold of p < 1 × 10 –5 was selected in pheno-\ntype association databases to identify potential pleiotropic\nassociations while avoiding excessive exclusion of valid\ninstruments. A stricter genome-wide threshold ( p < 5 ×\n10–8) may fail to detect moderate but biologically plausi-\nble pleiotropic effects.\n2.3 Multivariable Mendelian Randomization (MVMR)\nBecause SHBG modulates bioavailability and adipos-\nity can confound hormone–endometriosis relationships, we\nran:\n● Model A: E2 + SHBG\n● Model B: TT + SHBG (proxying FT)\n● Model C (extended): E2 + TT + SHBG + BMI\nJoint instrument sets were built from the union of\nexposure-specific instruments; conditional F-statistics were\ninspected to confirm adequate instrument strength in the\nmultivariable setting. Estimates were interpreted as direct\neffects conditional on the other exposures.\n2.4 Bidirectional MR\nWe tested reverse causation using genome-wide sig-\nnificant endometriosis instruments as the exposure and hor-\nmone GW AS as outcomes, applying IVW and MR-Egger\nwith Steiger tests for directionality.\n2.5 Subtype and Stratified Analyses\nWhere available and adequately powered, we repeated\nMR for ovarian endometrioma, deep infiltrating, and super-\nficial disease. Female-specific hormone GW AS were pri-\noritized; combined-sex instruments were used only in sen-\nsitivity analyses (extracting female-specific effects where\navailable) because of sex-heterogeneous genetic architec-\nture.\n2.6 Ancestry\nThe primary analyses were restricted to GW AS sum-\nmary statistics derived from populations of European an-\ncestry to reduce bias from population stratification and dif-\nferences in linkage disequilibrium structure. No cross-\nancestry meta-analysis was performed. Accordingly, the\ngeneralizability of the findings to non-European popula-\ntions requires further investigation.\n2.7 Multiple Testing\nWe controlled family-wise error for the five primary\nexposures (E2, TT, FT via TT+SHBG, E1, SHBG) using\nBonferroni correction (α = 0.05/number of primary tests).\nSecondary/subtype analyses were additionally evaluated\nwith Benjamini–Hochberg False Discovery Rate (FDR), la-\nbeled exploratory.\n2.8 Power\nWe estimated power using standard non-centrality\napproximations from exposure R 2, outcome case/control\ncounts, and α as above. For transparency, we report the\nminimum detectable OR at 80% power per 1-SD increase\nfor each exposure under the primary instrument set. For\ntraits with modest R 2 (e.g., E2), interpretations emphasize\neffect-size precision and triangulation across estimators.\n2.9 Quality Control\nWe enforced MAF ≥0.01 and imputation Information\nMetric (INFO) ≥0.8 (when provided), verified allele align-\nment, removed strand-ambiguous variants with intermedi-\nate MAF, and conducted outlier/influence diagnostics as\nspecified. Data pulls were version-controlled with recorded\nGW AS builds, releases, and sample sizes.\n2.10 Data Analysis\nDescriptive analyses were carried out in IBM SPSS\nStatistics 26.0. Normality was assessed with Shapiro-Wilk.\nContinuous variables are presented as mean (SD) or me-\ndian (IQR) and compared with t-test or Mann-Whitney U\nas appropriate; categorical variables are counts (percent)\n6\n\n\nTable 1. Baseline characteristics of the local tertiary-care endometriosis cohort (n = 50) .\nV ariable V alue\nAge (years), mean ± SD 31.80 ± 5.40\nBMI (kg/m2), mean ± SD 24.90 ± 3.80\nParity status\nNulliparous — n (%) 32 (64.00%)\nParous — n (%) 18 (36.00%)\nSubtype of endometriosis\nOvarian endometrioma — n (%) 22 (44.00%)\nDeep infiltrating — n (%) 14 (28.00%)\nSuperficial peritoneal — n (%) 9 (18.00%)\nMixed/unspecified — n (%) 5 (10.00%)\nClinical features\nPelvic pain documented — n (%) 40 (80.00%)\nInfertility code present — n (%) 18 (36.00%)\nDiagnostic confirmation\nLaparoscopic — n (%) 38 (76.00%)\nHistologic — n (%) 29 (58.00%)\nInitial management\nCombined oral contraceptives — n (%) 20 (40.00%)\nProgestin-only therapy — n (%) 15 (30.00%)\nGnRH analogues/antagonists — n (%) 6 (12.00%)\nExpectant/analgesics — n (%) 9 (18.00%)\nBMI, body mass index; SD, Standard Deviation; GnRH, Gonadotropin-Releasing Hormone.\ncompared with χ2 tests. These summaries contextualize the\ntertiary-care case mix and do not influence MR estimation.\nMR analyses were performed in R (TwoSampleMR, ieug-\nwasr/OpenGW AS, MRPRESSO, RadialMR, Mendelian-\nRandomization; plus data.table and ggplot2). Descriptive\nstatistics for the hospital cohort used SPSS 26.0. Random\nseeds were set for reproducibility, and all scripts were kept\nunder version control.\nMR used only public, de-identified GW AS summary\ndata and was deemed non-human subjects research by the\nInstitutional Ethics Committee. The local EHR descriptive\ncomponent (n = 50) used de-identified records. Written in-\nformed consent was obtained from all participants included\nin the local EHR cohort. All procedures adhered to the Dec-\nlaration of Helsinki and institutional policies.\nThe local tertiary-care cohort was included exclu-\nsively for descriptive contextualization of case mix and dis-\nease phenotype at a referral center. These data were not\nused in instrument derivation, causal estimation, validation,\nor triangulation analyses, and no inferential comparisons\nwere performed between the cohort and genetic findings.\n3. Results\n3.1 Baseline Characteristics\nThe local tertiary-care cohort (n = 50) reflected a typi-\ncal reproductive-age endometriosis population (mean age\n31.8 ± 5.4 years; BMI 24.9 ± 3.8 kg/m 2). Ovarian en-\ndometrioma was the most common subtype (44%), fol-\nlowed by deep infiltrating disease (28%). Pelvic pain was\npresent in 80% and infertility codes in 36%. Most diagnoses\nwere laparoscopically confirmed (76%). These characteris-\ntics align with patterns described in tertiary referral centers\n(Table 1).\nWith respect to disease presentation, ovarian en-\ndometrioma was the most frequently observed subtype\n(44.00%), followed by deep infiltrating endometriosis\n(28.00%), superficial peritoneal disease (18.00%), and\nmixed/unspecified forms (10.00%). The predominance of\novarian endometrioma mirrors patterns seen in surgical se-\nries and imaging-based studies. Symptomatically, pelvic\npain was highly prevalent, reported in 80.00% (n = 40)\nof participants, while 36.00% (n = 18) carried an infertil-\nity diagnosis code (Table 1). These descriptive findings\nare not intended to provide inferential evidence regard-\ning hormone–endometriosis relationships. Diagnosis was\npredominantly achieved through direct visualization at la-\nparoscopy (76.00%), and histologic confirmation was avail-\nable in over half of the cases (58.00%), reflecting adher-\nence to gold-standard diagnostic pathways in the majority.\nRegarding initial management, 40.00% received combined\noral contraceptives, 30.00% were managed with progestin-\nonly regimens, 12.00% were started on GnRH analogues\nor antagonists, and 18.00% were managed expectantly or\nwith analgesics alone. This distribution of therapies demon-\nstrates a spectrum of approaches influenced by symptom\nseverity, fertility goals, and patient preference.\n7\n\nTable 2. Genetic instrument characteristics for circulating sex hormone exposures (primary sets) .\nExposure SNPs (n) R 2 (%) Mean F-statistic F >10 (%) Steiger-consistent SNPs (%) Palindromic removed (n) LD proxies used (n)\nEstradiol (E2) 7 0.32 27.90 100.00 100.00 1 2\nTotal Testosterone (TT) 145 3.20 48.50 100.00 98.62 6 4\nFree/Bioavailable Testosterone (FT)* 168 4.10 45.20 100.00 98.21 8 5\nEstrone (E1) 18 0.80 32.40 100.00 100.00 1 1\nSHBG 171 6.50 60.70 100.00 99.42 7 3\n*FT modeled via TT and SHBG in multivariable MR. SHBG, sex hormone–binding globulin.\nTable 3. Primary MR results for any endometriosis (IVW as primary estimator) .\nExposure SNPs (n) IVW OR (95% CI) p (IVW) Q (df) p (Q) I 2 (%) MR-Egger OR (95% CI) p (Egger slope) Egger intercept p (intercept) Weighted median OR ( p) Weighted mode OR ( p)\nE2 7 1.18 (1.05–1.33) 0.006 12.50 (6) 0.051 52.00 1.12 (0.95–1.32) 0.180 0.003 0.410 1.16 (0.012) 1.15 (0.037)\nTT 145 0.92 (0.87–0.97) 0.003 198.20 (144) 0.002 27.30 0.95 (0.90–1.01) 0.110 −0.001 0.090 0.93 (0.004) 0.94 (0.021)\nFT 168 0.89 (0.82–0.96) 0.002 226.40 (167) 0.001 26.20 0.92 (0.83–1.02) 0.118 −0.002 0.132 0.90 (0.006) 0.91 (0.028)\nE1 18 1.10 (0.98–1.24) 0.100 20.80 (17) 0.232 18.30 1.07 (0.90–1.27) 0.430 0.002 0.482 1.09 (0.142) 1.08 (0.210)\nSHBG 171 1.07 (1.02–1.12) 0.005 245.90 (170) 0.001 30.90 1.05 (0.99–1.11) 0.094 0.001 0.210 1.06 (0.012) 1.05 (0.048)\nTable 4. Multivariable MR (MVMR) estimates — direct effects adjusting for correlated traits .\nModel Exposure Conditional F Direct OR (95% CI) p-value\nA: E2 + SHBG Estradiol (E2) 18.70 1.14 (1.02–1.28) 0.022\nSHBG 31.60 1.04 (0.99–1.09) 0.110\nB: TT + SHBG Total Testosterone (TT) 29.80 0.91 (0.86–0.97) 0.003\nSHBG 34.90 1.05 (1.00–1.10) 0.048\nC: E2 + TT + SHBG + BMI\nEstradiol (E2) 16.40 1.12 (1.00–1.26) 0.049\nTotal Testosterone (TT) 28.90 0.92 (0.86–0.98) 0.008\nSHBG 35.70 1.03 (0.98–1.09) 0.210\nBMI (per 1-SD) 45.00 1.06 (1.02–1.10) 0.003\nTable 5. Sensitivity, pleiotropy, and reverse MR checks .\nExposure MR-PRESSO ( p) Outliers removed (n) IVW after outlier OR ( p) Radial outliers (n) Max leave-one-out change (%) Egger intercept ( p) Steiger-consistent SNPs (%) Reverse MR β (SE) p (reverse)\nE2 0.071 1 1.17 (0.008) 1 3.40 0.410 100.00 0.004 (0.006) 0.480\nTT 0.012 5 0.91 (0.002) 6 2.10 0.090 98.62 −0.003 (0.003) 0.340\nFT 0.015 7 0.88 (0.001) 7 2.85 0.132 98.21 −0.004 (0.004) 0.300\nE1 0.220 0 1.10 (0.100) 0 1.90 0.482 100.00 0.002 (0.005) 0.700\nSHBG 0.009 8 1.06 (0.004) 9 2.50 0.210 99.42\n8\n\n\nFig. 2. Genetic instrument characteristics for circulating sex hormone exposures (primary sets) .\n3.2 Genetic Instrument Characteristics\nThe strength and validity of the genetic instruments\nunderpin the credibility of the MR findings. For E2,\n7 genome-wide significant, independent SNPs explained\n0.32% of the variance, with a mean F-statistic of 27.90,\nwell above the conventional threshold for strong instru-\nments. Importantly, 100.00% of these SNPs were Steiger-\nconsistent, indicating correct causal directionality at the\ninstrument level. TT and FT had the largest instrument\nsets—145 and 168 SNPs respectively—explaining 3.20%\nand 4.10% of the variance, with robust mean F-statistics\n(48.50 for TT, 45.20 for FT). The very high proportion of F\n>10 SNPs (100.00%) and Steiger consistency (>98%) sup-\nports the reliability of these instruments (Table 2 and Fig.\n2). E1 instruments consisted of 18 SNPs explaining 0.80%\nof the variance, with a mean F-statistic of 32.40, meeting\naccepted thresholds for instrument strength. SHBG had the\nlargest variance explained (6.50%) and the highest mean F-\nstatistic (60.70) among all exposures, reflecting its highly\nheritable nature and the extensive GW AS data available.\nAcross all exposures, palindromic SNPs were minimal and\ncarefully handled, and LD proxies were only used when\nnecessary.\n3.3 Primary MR Results (Pre-Specified Exposures)\nThe primary IVW analyses provided evidence for\ncausal roles of specific hormones in endometriosis risk. Ge-\nnetically predicted higher estradiol levels were associated\nwith a significantly increased risk of endometriosis (OR:\n1.18, 95% CI: 1.05–1.33, p = 0.006). This association per-\nsisted despite moderate heterogeneity (I 2 = 52.00%) and\nwas not driven by directional pleiotropy (Egger intercept\np = 0.410) (Table 3).\nConversely, higher genetically predicted TT and FT\nwere associated with a protective effect. TT had an OR of\n0.92 (p = 0.003) and FT an OR of 0.89 (p = 0.002), both with\nmodest heterogeneity (I 2 around 26–27%) and no signifi-\ncant pleiotropy evidence. Estrone showed a nonsignificant\ntrend toward increased risk (OR 1.10, p = 0.100), suggest-\ning any causal role may be smaller or context-dependent.\nSHBG displayed a small but statistically significant posi-\ntive association (OR 1.07, p = 0.005), which was consistent\nacross weighted median and weighted mode estimators (Ta-\nble 3 and Fig. 3).\nBonferroni correction (α = 0.01) confirmed statisti-\ncal robustness for E2, TT, FT, and SHBG. Weighted me-\ndian and mode results aligned with IVW for these traits,\nstrengthening confidence in the findings.\n3.4 Multivariable MR\nMultivariable MR allowed dissection of independent\nhormone effects after accounting for correlations, particu-\nlarly with SHBG and BMI. In the E2 + SHBG model, E2\nmaintained a direct, statistically significant association with\nendometriosis risk (OR: 1.14, p = 0.022), while SHBG’s ef-\nfect became nonsignificant (p = 0.110), suggesting E2 is the\nprincipal driver when both are considered together. In the\nTT + SHBG model, TT remained significantly protective\n9\n\nFig. 3. Primary MR results for any endometriosis (IVW as primary estimator) .\n(OR: 0.91, p = 0.003), and SHBG displayed a borderline\npositive effect (OR: 1.05, p = 0.048). This indicates that\nthe SHBG signal in univariable MR may partly reflect its\ninfluence on TT bioavailability (Table 4 and Fig. 4). The\nfull E2 + TT + SHBG + BMI model demonstrated that both\nE2 (OR: 1.12, p = 0.049) and TT (OR: 0.92, p = 0.008)\nretained independent effects, while SHBG was nonsignifi-\ncant. BMI emerged as an independent risk factor (OR: 1.06,\np = 0.003), consistent with known hormonal and inflamma-\ntory pathways linking adiposity to endometriosis risk. Con-\nditional F-statistics above 10 for all exposures confirmed\ninstrument strength in the multivariable setting.\n3.5 Sensitivity, Pleiotropy, and Reverse MR\nMR-PRESSO identified a varying number of potential\noutlier SNPs across hormone traits: one outlier for E2, no\noutliers for E1, and 5–8 outliers for TT, FT, and SHBG. Re-\nmoval of these outliers did not materially alter the direction,\nmagnitude, or statistical significance of the corresponding\nIVW estimates. Radial MR identified no influential SNPs\nfor E1, one influential SNP for E2, and multiple radial out-\nliers for TT, FT, and SHBG (6, 7, and 9 SNPs, respectively),\nas shown in Table 5.\nE1 was pre-specified but considered exploratory due\nto lower instrument strength and biological context. Egger\nintercept tests were nonsignificant for all exposures, ruling\nout major directional pleiotropy. Steiger tests confirmed\nthat >98% of SNPs explained more variance in the expo-\nsure than in the outcome, supporting correct causal direc-\ntion. Subtype and stratified analyses were considered ex-\nploratory and interpreted cautiously.\nAlthough MR-PRESSO detected potential horizontal\npleiotropy for TT, FT, and SHBG, exclusion of identified\noutliers did not materially alter the IVW effect estimates.\nThe direction, magnitude, and statistical significance of the\nassociations remained stable after outlier removal. Con-\ncordant findings across IVW, weighted median, weighted\nmode, and MR-Egger analyses further support the robust-\nness of the causal inference. Reverse MR analyses showed\nno evidence that liability to endometriosis causally alters\ncirculating E2, TT, FT, or E1 levels ( p = 0.480, 0.340,\n0.300, and 0.700, respectively). Thus, all available reverse-\nMR findings were non-significant ( p ≥ 0.300). The corre-\nsponding reverse-MR result for SHBG should be reported\nseparately once the β estimate, standard error, and p value\nare available (Table 5 and Fig. 5).\n10\n\n\nFig. 4. Multivariable Mendelian randomization estimates of the direct effects of circulating sex hormones and body mass index\non endometriosis risk . (A) Model including estradiol (E2) and sex hormone–binding globulin (SHBG). (B) Model including total\ntestosterone (TT) and SHBG. (C) Model including E2, TT, SHBG, and body mass index (BMI). Effect estimates are presented as odds\nratios (ORs) with 95% confidence intervals (CIs) per genetically predicted 1-standard-deviation increase in each exposure. The vertical\nreference line at OR = 1.00 indicates no association.\nFig. 5. Sensitivity and pleiotropy analyses for the associations between circulating hormone traits and endometriosis .\n4. Discussion\nThe small tertiary-care cohort included for contex-\ntual purposes demonstrates a phenotype broadly consistent\nwith published referral-center series. However, it was not\ndesigned for analytic inference and should be interpreted\nstrictly as descriptive background. The predominance of\novarian endometrioma (44.00%) with substantial deep in-\nfiltrating disease (28.00%) mirrors patterns in surgical and\n11\n\nimaging series and aligns with the well-recognized hetero-\ngeneity of clinical presentation emphasized by Zondervan\net al. (2016) [31] and the mechanistic spectrum summarized\nby Saunders (2022) [24]. The largely normal-to-overweight\nBMI distribution (mean 24.90 kg/m 2) is consistent with\nthe mixed literature on adiposity and endometriosis, while\nreminding us that inflammatory and hormonal milieus—\nrather than BMI alone—likely dominate pathogenesis, an\ninterpretation that dovetails with immune-endocrine trian-\ngulation in recent MR work on immune traits (Pan et al.\n(2024) [ 32]; Peng et al. (2024) [ 33]) and on metabolic\nfactors such as blood lipids (Wang et al. (2024) [ 34]).\nThe high proportion of laparoscopic (76.00%) and histo-\nlogic (58.00%) confirmation underscores gold-standard di-\nagnosis in most cases and provides a credible local back-\ndrop for interpreting the genetic (MR) results. Finally,\nthe spread of initial management—combined Oral Con-\ntraceptive Pills (OCPs) (40.00%), progestin-only regimens\n(30.00%), GnRH analogues/antagonists (12.00%), and ex-\npectant/analgesics (18.00%)—echoes contemporary prac-\ntice variation and the endocrine focus of therapy highlighted\nin translational overviews (Saunders (2022) [24]), reinforc-\ning the biological plausibility of our hormone-centric causal\nanalysis.\nGenetic instrument characteristics (foundation for\ncausal inference). Instrument quality was strong across ex-\nposures: all SNP sets met conventional strength thresholds\n(mean F >27.90) and showed near-universal Steiger con-\nsistency (>98%), minimizing concern about reverse causa-\ntion at the instrument level. These metrics are precisely the\nconditions recommended by methodological benchmarks\nthat advocate robust instruments and comprehensive sensi-\ntivity analyses to guard against horizontal pleiotropy (V er-\nbanck et al. (2018) [ 35]; Zhao et al. (2020) [ 36]; Bow-\nden et al. (2018) [ 37]; Morrison et al. (2020) [ 38]). The\nbreadth and depth of our testosterone and SHBG instru-\nments (145–171 SNPs; R 2 up to 6.50%) reflect advances\nin large-scale, sex-stratified hormone GW AS (Leinonen et\nal. (2023) [ 19]), while the availability and harmoniza-\ntion of summary statistics via MR-ready infrastructures en-\nsured transparent and reproducible data flows (Hemani et\nal. (2018) [ 14]; Elsworth et al. (2020) [ 15]). E1 and E2\ninstruments explained less variance (R 2 0.80% and 0.32%\nrespectively)—a known limitation of estrogen assays and\nsample sizes in GW AS—but still attained mean F-statistics\ncompatible with unbiased MR estimation. Screening of in-\nstruments against genotype–phenotype catalogs is a best-\npractice step for pleiotropy risk management, consistent\nwith tools like PhenoScanner (Kamat et al. (2019) [ 39]).\nAltogether, the instrument profile justifies confidence that\nthe Table 3 estimates arise from adequately strong and di-\nrectionally valid genetic proxies.\nAlthough estradiol instruments met conventional\nstrength thresholds (mean F >10), the variance explained\n(R2 = 0.32%) remains modest. Therefore, while the direc-\ntion of effect appears consistent and biologically plausible,\neffect size precision should be interpreted cautiously. Mea-\nsurement heterogeneity in estradiol GW AS—particularly\ndue to menstrual cycle variability and assay sensitivity—\nmay contribute to residual noise in genetic instruments.\nPrimary MR results (E2 risk-increasing; androgens\nprotective; SHBG modestly risk-increasing) Three con-\nverging patterns emerge. First, genetically proxied higher\nestradiol increased endometriosis risk (IVW OR: 1.18, p\n= 0.006). This is biologically coherent with estrogen-\ndependence of lesion growth and progesterone resistance\nin the endometrium and ectopic tissue emphasized in ge-\nnetic and functional syntheses (Saunders (2022) [ 24]; Zon-\ndervan et al. (2016) [ 31]). The heterogeneity we observed\n(I2 = 52.00%) is unsurprising given cycle variability and\nassay heterogeneity in estrogen GW AS; nonetheless, the\nnull Egger intercept argues against directional pleiotropy\nbeing the driver—an interpretation in line with pleiotropy-\naware MR frameworks (V erbanck et al. (2018) [ 35]; Zhao\net al. (2020) [ 36]). Second, higher genetically proxied to-\ntal and free/bioavailable testosterone were associated with\nlower endometriosis risk (TT OR: 0.92, p = 0.003; FT OR:\n0.89, p = 0.002). These inverse associations are direction-\nally consistent with an androgen-protective signal recently\nreported in a dedicated MR of androgens and endometrio-\nsis by Gjorgoska et al. (2024) [ 40], who also concluded\nthat androgens may mitigate risk. While effect magnitudes\nacross studies depend on instrument sets, transformations\n(per-SD vs. per-unit), and outcome definitions, the conver-\ngence toward small-to-moderate protective ORs strength-\nens the inference that androgens are not mere correlates but\npotential causal modulators of risk. Third, SHBG showed\na modest positive association (OR: 1.07, p = 0.005). Be-\ncause SHBG determines the bioavailable fraction of sex\nsteroids, this pattern may reflect its role as an upstream\nregulator that reduces androgen bioavailability or alters es-\ntrogen dynamics. The nuanced SHBG finding echoes the\ncomplex, shared genetic architecture between SHBG and\ntestosterone loci noted in sex-stratified GW AS (Leinonen et\nal. (2023) [ 19]). Notably, E1 showed only a nonsignificant\ntrend (OR: 1.10, p = 0.100), which could indicate smaller\nor context-dependent effects, or simply limited power due\nto lower R 2—both anticipated challenges for estrogen phe-\nnotypes in MR. Methodologically, our reliance on mul-\ntiple estimators (weighted median/mode) and heterogene-\nity/pleiotropy checks follows the best-practice toolkit pro-\nmulgated by Bowden et al. (2018) [ 37], Zhao et al. (2020)\n[36], and V erbanck et al. (2018) [ 35], thereby increasing\nthe credibility of the primary inferences.\nMultivariable MR (disentangling direct effects of E2\nand TT from SHBG and BMI) Multivariable modeling clar-\nifies that the univariable SHBG signal is at least partly ex-\nplained by its correlation with sex steroids. When condi-\ntioning on SHBG, estradiol retained a direct risk-increasing\neffect (OR: 1.14, p = 0.022); conversely, when condition-\ning on SHBG, total testosterone remained directly protec-\ntive (OR: 0.91, p = 0.003). In the full model (E2 + TT\n12\n\n\n+ SHBG + BMI), E2 (OR: 1.12, p = 0.049) and TT (OR:\n0.92, p = 0.008) continued to show independent, opposing\neffects, while SHBG was attenuated and nonsignificant—\nexactly the kind of disentanglement that multivariable MR\nwas designed to achieve (Sanderson et al. (2019) [ 41]).\nThe emergence of BMI as an additional risk factor (OR:\n1.06, p = 0.003) situates our hormone findings within a\nbroader metabolic context that resonates with MR evidence\nlinking endometriosis with circulating lipids (Wang et al.\n(2024) [34]) and with coagulation biology (Li et al. (2023)\n[42]), as well as with immune cell influences (Pan et al.\n(2024) [ 32]; Peng et al. (2024) [ 33]). These triangula-\ntions suggest that endocrine, metabolic, and immune axes\nare not isolated but interdependent in shaping endometriosis\nrisk—an integrated view long argued for in translational re-\nviews (Zondervan et al. (2016) [31]; Saunders (2022) [24]).\nFrom a translational angle, the persistence of an E2-risk\nand TT-protective profile after conditioning implies that (i)\nestrogen-lowering or estrogen-modulating strategies have a\ncausal rationale beyond symptomatic control, and (ii) selec-\ntively enhancing androgenic signaling—or preventing its\nsuppression via high SHBG—could have preventive rele-\nvance, though any such approach must be balanced against\nsystemic effects and patient-specific contraindications. The\nindependence from SHBG in multivariable models cautions\nagainst interpreting SHBG as a direct driver once underly-\ning steroid levels are accounted for, consistent with genetic\narchitecture laid out by Leinonen et al. (2023) [19]. It is im-\nportant to emphasize that MR estimates reflect genetically\ninfluenced lifetime exposure gradients and should not be\nequated with pharmacologic hormone manipulation effects\nin adulthood.\nSensitivity, pleiotropy, and reverse MR (robustness of\ninferences) Comprehensive sensitivity work supports the\nstability of our findings. MR-PRESSO detected horizon-\ntal pleiotropy for TT, FT, and SHBG, but outlier removal\n(5–8 SNPs) preserved effect directions and significance.\nThis is the practical scenario MR-PRESSO was built for—\nidentifying and mitigating distortions without discarding\nthe entire causal signal—consistent with broader pleiotropy\ndetection principles advocated by V erbanck et al. (2018)\n[35] and the robust-adjusted profile score approach by Zhao\net al. (2020) [ 36]. Radial MR flagged few high-influence\nvariants, and leave-one-out analyses showed minimal sen-\nsitivity to single instruments (maximum change 3.40%),\naligning with visualization and influence diagnostics pro-\nposed by Bowden et al. (2018) [ 37]. Directionality was\nalso well supported: steiger tests favored the exposure-\nto-outcome direction for ≥98.21% of SNPs; Egger inter-\ncepts were uniformly nonsignificant; and reverse MR found\nno evidence that genetic liability to endometriosis causally\nshifts circulating hormone levels (all p > 0.30). Collec-\ntively, this suite of checks addresses core MR assumptions\nand typical threats (correlated and uncorrelated pleiotropy),\nechoing the logic of model classes like CAUSE (Morrison\net al. (2020) [ 38]) and the deployment of multivariable MR\nwhen correlation structures are suspected (Sanderson et al.\n(2019) [41]). Finally, external triangulation increases con-\nfidence: the androgen-protective signal we observe aligns\nwith the independent MR by Gjorgoska et al. (2024) [ 40];\nimmune and metabolic MR analyses (Pan et al. (2024) [32];\nPeng et al. (2024) [ 33]; Wang et al. (2024) [ 34]; Li et\nal. (2023) [ 42]) frame plausible intermediary pathways;\nand shared genetic architecture across gynecologic and pain\ntraits (Adewuyi et al. (2020) [ 43]) and even ovarian cancer\nsubtypes (Wang et al. (2023) [ 44]) underscores the need\nfor careful interpretation when phenotypes overlap a theme\nlong advocated in the endometriosis genetics field (Zonder-\nvan et al. (2016) [ 31]).\nA local control group was not included, as the de-\nscriptive cohort was not intended for comparative anal-\nysis. Causal inference relied exclusively on large-scale\nGW AS case-control datasets comprising thousands of cases\nand controls. Future institutional studies may benefit from\nincluding matched control groups for detailed phenotypic\ncomparison.\nIntegrated interpretation and implications Putting the\npieces together, your data show: (1) a clinical profile typ-\nical for tertiary care; (2) strong, directionally valid ge-\nnetic instruments (Table 2); (3) a risk-increasing effect of\nestradiol and protective effects of total and bioavailable\ntestosterone, with a modest SHBG signal (Table 3); (4) in-\ndependence of E2-risk and TT-protection after condition-\ning on SHBG and BMI (Table 4); and (5) extensive ro-\nbustness to pleiotropy and directionality violations (Table\n5). These findings are mechanistically plausible within\nan estrogen-dependent, immune-modulated disease model\n(Saunders (2022) [ 24]), converge with independent andro-\ngen MR evidence (Gjorgoska et al. (2024) [ 40]), and sit\nalongside causal signals reported for immune cells, coagu-\nlation factors, and lipids (Pan et al. (2024) [ 32]; Peng et al.\n(2024) [33]; Li et al. (2023) [ 42]; Wang et al. (2024) [ 34]).\nTwo caveats merit emphasis. First, estrogen instruments\nexplain modest variance—an assay and sample-size limita-\ntion acknowledged across estrogen GW AS—so precision is\nlower for E2 and E1 than for TT/SHBG. Second, MR cap-\ntures lifelong, genetically influenced exposure differences;\ntranslating these into interventional strategies requires care-\nful consideration of timing, dose, and off-target effects.\nNonetheless, the concordance between your E2-risk and\nTT-protective estimates and external MR signals, the atten-\nuation of SHBG in multivariable models, and the robust-\nness across sensitivity frameworks together build a coherent\ncausal narrative that can inform both mechanistic hypothe-\nses (e.g., endocrine-immune crosstalk) and strategic pre-\nvention or risk-stratification approaches in future work—\nprecisely the translational arc envisioned by comprehensive\ngenetics roadmaps (Zondervan et al. (2016) [ 31]) and con-\ntemporary genomic syntheses (Saunders (2022) [ 24]).\n13\n\nLimitations\nThis study is limited by the relatively small local co-\nhort size (n = 50), which constrains the power of clinical\ncorrelations despite robust genetic analyses. Estradiol and\nestrone instruments explained modest variance, reducing\nprecision for estrogen estimates compared to testosterone\nand SHBG. Mendelian randomization reflects lifelong ge-\nnetically influenced hormone levels, which may not fully\ncapture short-term or treatment-induced hormonal changes.\n5. Conclusions\nOur two-sample MR analysis supports a causal role for\nhigher estradiol in increasing, and higher total and bioavail-\nable testosterone in reducing, the risk of endometriosis,\nwith SHBG showing a modest positive association atten-\nuated in multivariable models. These results are robust\nto pleiotropy and reverse causation checks, aligning with\nindependent genetic evidence and biological plausibility.\nIntegration with clinical and mechanistic data highlights\nthe interconnected roles of endocrine, metabolic, and im-\nmune pathways in disease pathogenesis. In primary anal-\nyses, genetically proxied estradiol demonstrated a risk-\nincreasing effect, whereas testosterone showed protective\nassociations. Exploratory findings for estrone were null.\nThese results were robust across sensitivity frameworks and\nmultivariable adjustment. While MR reflects lifelong expo-\nsure gradients rather than pharmacologic interventions, the\nfindings strengthen causal inference regarding endocrine\ncontributions to endometriosis susceptibility. These find-\nings strengthen the biological evidence for endocrine in-\nvolvement in disease susceptibility. However, Mendelian\nrandomization reflects lifelong genetically proxied expo-\nsure differences rather than short-term modifiable hormonal\ninterventions. Translation into preventive or therapeutic\nstrategies will require careful clinical investigation.\nAvailability of Data and Materials\nThe data supporting the findings of this study have\nbeen included in the study as well as in the attached sup-\nplementary data.\nAuthor Contributions\nJZ and CZ designed and planned the study, analyzed\nthe data, and interpreted the results. LL and RZ collected\nthe data and clinical materials and supervised the project.\nLL and RZ edited and revised the manuscript with a focus\non important intellectual content. YW and YX collected,\nassessed, and interpreted the data. MW contributed to data\ninterpretation and manuscript preparation. QY and BH con-\nducted the laboratory investigations and statistical analy-\nses and provided substantial intellectual input during the\ndrafting and revision of the manuscript. All authors con-\ntributed to critical revision of the manuscript for important\nintellectual content. All authors read and approved the fi-\nnal manuscript. All authors have participated sufficiently\nin the work and agreed to be accountable for all aspects of\nthe work.\nEthics Approval and Consent to Participate\nThis study was approved by the Ethics Committee of\nPeople’s Hospital of Pudong New Area, (Approval NO.\n2024-LW-11). Informed consent was obtained from all par-\nticipants, and the study was conducted in accordance with\nthe Declaration of Helsinki.\nAcknowledgment\nNot applicable.\nFunding\nThis project was supported by the Medical Disci-\npline Construction Program of Shanghai Pudong New\nArea Health Commission (No. PWZxk2022-28)and\nShanghai Pudong New Area People’s Hospital Project\n(PRYQH202504).\nConflicts of Interest\nThe authors declare no conflicts of interest.\nDeclaration of AI and AI-Assisted\nTechnologies in the Writing Process\nDuring the preparation of this work, the authors used\nChatGPT 4.0 for language editing, grammar refinement,\nand structural clarity. After using this tool, the authors care-\nfully reviewed and edited the content as needed and take full\nresponsibility for the content of the publication. No AI tool\nwas used for data generation, statistical analysis, or scien-\ntific interpretation.\nSupplementary Material\nSupplementary material associated with this article\ncan be found, in the online version, at https://doi.org/10.\n31083/CEOG49264.\nReferences\n[1] Zondervan KT, Becker CM, Koga K, Missmer SA, Taylor RN,\nViganò P . Endometriosis. Nature Reviews. Disease Primers.\n2018; 4: 9. https://doi.org/10.1038/s41572-018-0008-5\n[2] Taylor HS, Kotlyar AM, Flores V A. Endometriosis is a chronic\nsystemic disease: clinical challenges and novel innovations.\nThe Lancet. 2021; 397: 839–852. https://doi.org/10.1016/\nS0140-6736(21)00389-5\n[3] Chapron C, Marcellin L, Borghese B, Santulli P . Rethinking\nmechanisms, diagnosis and management of endometriosis. Na-\nture Reviews. Endocrinology. 2019; 15: 666–682. https://doi.or\ng/10.1038/s41574-019-0245-z\n[4] Zhang P , Wang G. Progesterone Resistance in Endometriosis:\nCurrent Evidence and Putative Mechanisms. International Jour-\nnal of Molecular Sciences. 2023; 24: 6992. https://doi.org/10.\n3390/ijms24086992\n[5] Tang HC, Lin TC, Wu MH, Tsai SJ. Progesterone resistance\nin endometriosis: A pathophysiological perspective and poten-\n14\n\n\ntial treatment alternatives. Reproductive Medicine and Biology.\n2024; 23: e12588. https://doi.org/10.1002/rmb2.12588\n[6] Chantalat E, V alera MC, V aysse C, Noirrit E, Rusidze M, Weyl\nA, et al. Estrogen receptors and endometriosis. International\nJournal of Molecular Sciences. 2020; 21: 2815. https://doi.or\ng/10.3390/ijms21082815\n[7] Ferrero S, Gillott DJ, V enturini PI, Remorgida V . Use of aro-\nmatase inhibitors to treat endometriosis-related pain symptoms:\na systematic review. Reproductive Biology and Endocrinology.\n2011; 9: 89. https://doi.org/10.1186/1477-7827-9-89\n[8] Mori T, Ito F, Koshiba A, Kataoka H, Takaoka O, Okimura H, et\nal. Local estrogen formation and its regulation in endometriosis.\nReproductive Medicine and Biology. 2019; 18: 305–311. https:\n//doi.org/10.1002/rmb2.12285\n[9] Zhang H, Sheng S, Pan Z, Zhao L, Y ang C, Li C, et al. Immune\nand endocrine regulation in endometriosis: what we know. Jour-\nnal of Endometriosis and Uterine Disorders. 2023; 4: 100049.\nhttps://doi.org/10.1016/j.jeud.2023.100049\n[10] Cano-Herrera G, Salmun Nehmad S, Ruiz de Chávez Gascón\nJ, Méndez Vionet A, van Tienhoven XA, Osorio Martínez MF,\net al. Endometriosis: A Comprehensive Analysis of the Patho-\nphysiology, Treatment, and Nutritional Aspects, and Its Reper-\ncussions on the Quality of Life of Patients. Biomedicines. 2024;\n12: 1476. https://doi.org/10.3390/biomedicines12071476\n[11] Rahmioglu N, Mortlock S, Ghiasi M, Møller PL, Stefansdot-\ntir L, Galarneau G, et al. The genetic basis of endometriosis\nand comorbidity with other pain and inflammatory conditions.\nNature genetics. 2023; 55: 423–436. https://doi.org/10.1038/\ns41588-023-01323-z\n[12] Bycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp\nK, et al. The UK Biobank resource with deep phenotyping and\ngenomic data. Nature. 2018; 562: 203–209. https://doi.org/10.\n1038/s41586-018-0579-z\n[13] Kurki MI, Karjalainen J, Palta P , Sipilä TP , Kristiansson K, Don-\nner KM, et al. FinnGen provides genetic insights from a well-\nphenotyped isolated population. Nature. 2023; 613: 508–518.\nhttps://doi.org/10.1038/s41586-022-05473-8\n[14] Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V , Baird\nD, et al. The MR-Base platform supports systematic causal in-\nference across the human phenome. elife. 2018; 7: e34408.\nhttps://doi.org/10.7554/eLife.34408\n[15] Elsworth B, Lyon M, Alexander T, Liu Y , Matthews P , Hallett j,\net al. The MRC IEU OpenGW AS data infrastructure. BioRxiv.\n2020. (preprint) https://doi.org/10.1101/2020.08.10.244293\n[16] Ruth KS, Day FR, Tyrrell J, Thompson DJ, Wood AR, Mahajan\nA, et al. Using human genetics to understand the disease impacts\nof testosterone in men and women. Nature Medicine. 2020; 26:\n252–258. https://doi.org/10.1038/s41591-020-0751-5\n[17] Schmitz D, Ek WE, Berggren E, Höglund J, Karlsson T, Jo-\nhansson Å. Genome-wide association study of estradiol levels\nand the causal effect of estradiol on bone mineral density. The\nJournal of Clinical Endocrinology and Metabolism. 2021; 106:\ne4471–e4486. https://doi.org/10.1210/clinem/dgab507\n[18] Y u C, Bakshi A, Bell RJ, Islam RM, Handelsman DJ, McNeil JJ,\net al. Genome-wide association study identifies genetic regula-\ntion of oestrone concentrations and association with endometrial\ncancer risk in postmenopausal women. eBioMedicine. 2024;\n101: 104997. https://doi.org/10.1016/j.ebiom.2024.104997\n[19] Leinonen JT, Mars N, Lehtonen, LE, Ahola-Olli A, Ruot-\nsalainen S, Lehtimäki T, et al. Genetic analyses implicate com-\nplex links between adult testosterone levels and health and dis-\nease. Communications Medicine. 2023; 3: 4. https://doi.org/10.\n1038/s43856-022-00226-0\n[20] Y uan S, Wang L, Sun J, Y u L, Zhou X, Y ang J, et al. Ge-\nnetically predicted sex hormone levels and health outcomes:\nphenome-wide Mendelian randomization investigation. Interna-\ntional Journal of Epidemiology. 2022; 51: 1931–1942. https:\n//doi.org/10.1093/ije/dyac036\n[21] Denos M, Sun YQ, Brumpton BM, Li Y , Albanes D, Burnett-\nHartman A, et al. Sex hormones and risk of lung and col-\norectal cancers in women: a Mendelian randomization study.\nScientific Reports. 2024; 14: 23891. https://doi.org/10.1038/\ns41598-024-75305-4\n[22] V enkatesh SS, Ferreira T, Benonisdottir S, Rahmioglu N,\nBecker CM, Granne I, et al. Obesity and risk of female repro-\nductive conditions: A Mendelian randomisation study. PLoS\nMedicine. 2022; 19: e1003679. https://doi.org/10.1371/journa\nl.pmed.1003679\n[23] Harrison S, Davies NM, Howe LD, Hughes A, et al. Testos-\nterone and socioeconomic position: Mendelian randomization\nin 306,248 men and women in UK Biobank. Science Advances.\n2021; 7: eabf8257. https://doi.org/10.1126/sciadv.abf8257\n[24] Saunders PTK. Insights from genomic studies on the role of sex\nsteroids in the aetiology of endometriosis. Reproduction & Fer-\ntility. 2022; 3: R51–R65. https://doi.org/10.1530/RAF-21-0078\n[25] Bao C, Wang H, Fang H, et al. Genomic Evidence Supports\nthe Recognition of Endometriosis as an Inflammatory Systemic\nDisease and Reveals Disease-Specific Therapeutic Potentials of\nTargeting Neutrophil Degranulation. Frontiers in Immunology.\n2022; 13: 758440. https://doi.org/10.3389/fimmu.2022.758440\n[26] V annuccini S, Clemenza S, Sorbi M, Petraglia F. Hormonal\ntreatments for endometriosis: the endocrine background. Re-\nviews in Endocrine & Metabolic Disorders. 2021; 23: 333–355.\nhttps://doi.org/10.1007/s11154-021-09666-w\n[27] Y an H, Li X, Dai Y , Shi J, Wu Y , Gu Z, et al. Global, regional, and\nnational burdens of endometriosis from 1990 to 2021: a trend\nanalysis. Frontiers in Medicine. 2025; 12: 1562196. https://doi.\norg/10.3389/fmed.2025.1562196\n[28] Y u X, Shi L, Deng X, Zhang Y , Wang H. Global burden of en-\ndometriosis from 1990 to 2021 and projections to 2050: a com-\nprehensive analysis based on the global burden of disease study\n2021. Frontiers in Global Women’s Health. 2025; 6: 1613468.\nhttps://doi.org/10.3389/fgwh.2025.1613468\n[29] Pujol-Gualdo N, Karjalainen MK, Võsa U, Arffman RK, Mägi\nR, Ronkainen J, et al. Circulating anti-Müllerian hormone lev-\nels in pre-menopausal women: novel genetic insights from a\ngenome-wide association meta-analysis. Human Reproduction.\n2024; 39: 1564–1572. https://doi.org/10.1093/humrep/deae117\n[30] McGrath IM, International Endometriosis Genetics Consortium,\nMontgomery GW, Mortlock S. Polygenic risk score phenome-\nwide association study reveals an association between en-\ndometriosis and testosterone. BMC Medicine. 2023; 21: 482.\nhttps://doi.org/10.1186/s12916-023-03184-z\n[31] Zondervan KT, Rahmioglu N, Morris AP , Nyholt DR, Mont-\ngomery GW, Becker CM, et al. Beyond Endometriosis Genome-\nWide Association Study: From Genomics to Phenomics to the\nPatient. Seminars in Reproductive Medicine. 2016; 34: 242–\n254. https://doi.org/10.1055/s-0036-1585408\n[32] Pan L, Chen Y , Zhou Z, Ma S, Cao Y , Ma Y . The correlation\nbetween immune cells and endometriosis: a bidirectional two-\nsample mendelian randomization study. BMC Women’s Health.\n2024; 24: 641. https://doi.org/10.1186/s12905-024-03493-2\n[33] Peng Y , Li Y , Wang L, Lin S, Xu H. Causality of immune\ncells and endometriosis: a bidirectional mendelian randomiza-\ntion study. BMC Women’s Health. 2024; 24: 574. https://doi.or\ng/10.1186/s12905-024-03417-0\n[34] Wang Z, Zhan C, Liao L, Luo Y , Lin S, Y an S. Bidirec-\ntional causality between the levels of blood lipids and en-\ndometriosis: a two-sample mendelian randomization study.\nBMC Women’s Health. 2024; 24: 387. https://doi.org/10.1186/\ns12905-024-03213-w\n[35] V erbanck M, Chen CY , Neale B, Do R. Detection of widespread\nhorizontal pleiotropy in causal relationships inferred from\nMendelian randomization between complex traits and diseases.\n15\n\nNature Genetics. 2018; 50: 693–698. https://doi.org/10.1038/\ns41588-018-0099-7\n[36] Zhao Q, Wang J, Hemani G, Bowden J, Small DS. Statistical\ninference in two-sample summary-data Mendelian randomiza-\ntion using robust adjusted profile score. The Annals of Statistics.\n2020; 48: 1742–1769. https://doi.org/10.1214/19-AOS1866\n[37] Bowden J, Spiller W, Del Greco M F, Sheehan N, Thomp-\nson J, Minelli C, et al. Improving the visualization, interpreta-\ntion and analysis of two-sample summary data Mendelian ran-\ndomization via the Radial plot and Radial regression. Interna-\ntional Journal of Epidemiology. 2018; 47: 1264–1278. https:\n//doi.org/10.1093/ije/dyy101\n[38] Morrison J, Knoblauch N, Marcus JH, Stephens M, He X.\nMendelian randomization accounting for correlated and uncor-\nrelated pleiotropic effects using genome-wide summary statis-\ntics. Nature Genetics. 2020; 52: 740–747. https://doi.org/10.\n1038/s41588-020-0631-4\n[39] Kamat MA, Blackshaw JA, Y oung R, Surendran P , Burgess S,\nDanesh J, et al. PhenoScanner V2: an expanded tool for search-\ning human genotype-phenotype associations. Bioinformatics.\n2019; 35: 4851–4853. https://doi.org/10.1093/bioinformatics/b\ntz469\n[40] Gjorgoska M, Rizner TL. The effect of androgens on the risk\nof endometriosis sub-phenotypes and ovarian neoplasms: A\nMendelian randomization study. The Journal of Steroid Bio-\nchemistry and Molecular Biology. 2024; 239: 106482. https:\n//doi.org/10.1016/j.jsbmb.2024.106482\n[41] Sanderson E, Davey Smith G, Windmeijer F, Bowden J. An\nexamination of multivariable Mendelian randomization in the\nsingle-sample and two-sample summary data settings. Interna-\ntional Journal of Epidemiology. 2019; 48: 713–727. https://doi.\norg/10.1093/ije/dyy262\n[42] Li Y , Liu H, Y e S, Zhang B, Li X, Y uan J, et al. The effects of\ncoagulation factors on the risk of endometriosis: a Mendelian\nrandomization study. BMC Medicine. 2023; 21: 195. https://do\ni.org/10.1186/s12916-023-02881-z\n[43] Adewuyi EO, Sapkota Y , International Endogene Consortium\nIec, andMe Research Team, International Headache Genet-\nics Consortium Ihgc, Auta A, et al. Shared Molecular Ge-\nnetic Mechanisms Underlie Endometriosis and Migraine Co-\nmorbidity. Genes. 2020; 11: 268. https://doi.org/10.3390/gene\ns11030268\n[44] Wang L, Li X, Wang Y , Li G, Dai S, Cao M, et al. Endometrio-\nsis and epithelial ovarian cancer: a two-sample Mendelian ran-\ndomization analysis. Scientific Reports. 2023; 13: 21992. https:\n//doi.org/10.1038/s41598-023-49276-x\n16","source_license":"CC0","license_restricted":false}