This retrospective cohort study was conducted at a single tertiary referral center. Patient data were retrieved using the Severance Clinical Research Analysis Portal (SCRAP), an institutional electronic data warehouse that integrates clinical, laboratory, and administrative data from electronic medical records. We identified women who were newly diagnosed with ovarian endometrioma and presented to Severance Hospital between January 1, 2020 and December 31, 2023. In addition, a comparator cohort of women who were newly diagnosed with ovarian dermoid cyst and pathologically confirmed after surgery during the same period was identified from SCRAP.

Eligible patients were (1): women aged 20–45 years (2); newly diagnosed with ovarian endometrioma or ovarian dermoid cyst between January 1, 2020 and December 31, 2023 (3); not previously treated for endometriosis (surgery, medication, or sclerotherapy) or dermoid cyst (surgery); and (4) had serum AMH measured at the initial visit.

Dermoid cysts were selected as a pragmatic comparator because they share the clinical evaluation pathway of an adnexal mass but are not defined by the estrogen−dependent chronic inflammatory process that characterizes endometriosis (5), and have been used as a clinical comparator in prior studies assessing preoperative AMH in women with ovarian cysts (10). Because dermoid cysts were restricted to surgically treated and pathologically confirmed cases to minimize misclassification, the dermoid cohort should be interpreted as a surgically selected ovarian−cyst comparator rather than a population−based ‘healthy control’ group.

Patients were excluded if they had: missing AMH data or AMH ≤0.01 ng/mL; missing essential variables (height, weight, age, cyst size); prior treatment or recurrent disease; recent (<3 months) hormonal treatment, pregnancy, lactation, or ovarian stimulation; polycystic ovary syndrome, premature ovarian insufficiency, or early menopause; uncontrolled endocrine disorders; prior hysterectomy or adnexal surgery; history of malignancy, autoimmune disease, or severe hepatic/renal disease; or exposure to gonadotoxic drugs, chemotherapy, or pelvic radiotherapy. Because polycystic ovary syndrome and major endocrine/metabolic disorders can substantially affect AMH levels and are closely associated with BMI, they can introduce major confounding in studies of BMI–AMH associations. We therefore excluded these conditions a priori to improve interpretability of the estimated BMI effect. Patients with concomitant endometriosis were excluded from the comparator group. Likewise, patients in the endometriosis group were excluded if other types of ovarian cysts were present.

Demographic, anthropometric, and lifestyle data were obtained from EMRs. BMI (kg/m²) was calculated from measured height and weight, and analyzed both as a continuous variable and as a categorical variable according to the World Health Organization (WHO) Asian classification: underweight (<18.5), normal (18.5–22.9), overweight (23.0–24.9), and obese (≥25.0) Ovarian cyst size was defined as the sum of the maximum diameters of the right and left ovarian cysts. Smoking status (never vs ever), alcohol consumption (five categories by weekly frequency), and menstrual cycle characteristics (regular vs irregular; self-reported cycle length for regular cycles) were included as covariates.

The primary outcome was the association between BMI and serum AMH levels in women with ovarian endometrioma, with comparisons to a dermoid cyst control group (comparator). Secondary analyses assessed AMH differences across BMI and age subgroups, and evaluated the influence of potential covariates (cyst laterality, cyst size, menstrual cycle characteristics, parity, smoking, and alcohol use). Exploratory analyses tested for BMI×group interactions to determine whether the BMI–AMH relationship differed between endometrioma and dermoid cyst group.

Serum AMH concentrations were measured using the cobas e801 modules (Roche Diagnostics International AG; Rotkreuz, Switzerland) according to the manufacturer’s instructions. The intra- and inter-assay coefficients of variation were <5%. For adnexal mass evaluation, all patients underwent transvaginal ultrasonography; if the diagnosis was inconclusive, additional MRI or contrast-enhanced abdominopelvic CT was performed. Ovarian cyst size was defined as the sum of the maximum diameters of right and left ovarian cysts, measured on transvaginal or transrectal ultrasonography by obstetrician−gynecologists or dedicated sonographers in the Department of Obstetrics and Gynecology, and on MRI/CT (when performed) based on radiology reports interpreted by board−certified radiologists. Because imaging was obtained as part of routine clinical care, interpreters were not formally blinded to clinical information; serum AMH results were available in the electronic medical record and could be accessed if needed.

The primary analysis focused on the association between BMI and log−AMH within the endometrioma group. A priori, we powered the study to detect a small incremental association of BMI with log−AMH, operationalized as a partial R² =0.01–0.02 (two−sided α=0.05), consistent with published effect sizes of roughly 8–12% lower AMH per 5 kg/m² higher BMI (11). Accordingly, we estimated that 700 endometrioma participants would provide 80% power (12). Dermoid participants were enrolled as a comparator to enable secondary/exploratory assessment of effect modification (BMI×diagnosis); subgroup/interaction analyses were not formally powered.

Baseline characteristics were compared using t−tests or Mann–Whitney U tests, and χ² or Fisher’s exact tests for categorical variables; standardized mean differences (SMDs) were reported. For AMH between groups, analysis of covariance (ANCOVA) was performed on log−AMH with age as a covariate and results back−transformed to percent differences with 95% confidence intervals (CIs).

To characterize age–AMH trajectories, we used segmented (piecewise) regression by group to estimate data−driven breakpoints and compared pre−/post−break slopes.

We regressed log−transformed AMH on BMI (continuous) using multivariable linear models, adjusting for age, diagnosis, ovarian cyst maximum diameter, bilaterality, parity, smoking, alcohol use, cycle regularity, and cycle length. Age, cycle length, and cyst size were specified with restricted cubic splines (RCS, df=4). Cyst size and laterality were included a priori to partially account for potential cyst−related effects on AMH across cyst phenotypes (10). BMI non−linearity was screened in age−adjusted models via nested F−tests comparing linear vs spline terms; BMI was kept linear in the primary specification. Spline-based curves were used for model diagnostics and visualization, and any inference about the shape of the BMI–AMH relationship (e.g., attenuation across BMI ranges) was treated as exploratory. A group × BMI interaction tested differential effects by diagnosis. Coefficients on the log scale were exponentiated to express percent change in AMH per 1 kg/m² increase in BMI.

Pre−specified secondary analyses included (i) age−stratified BMI–AMH associations using breakpoints from segmented models with formal BMI × age−group tests; (ii) categorical BMI analyses (WHO categories) with age−adjusted geometric means and group × category interactions (Dunnett adjustment); and (iii) BMI−range−restricted sensitivity analyses (≤30, and ≤35 kg/m²) to assess robustness and potential range restriction; interpretation of differences across BMI ranges was considered exploratory. Model assumptions were checked with standard diagnostics.

All tests were two−sided with α = 0.05. Analyses and figure generation were performed using R version 4.5.1 (R Foundation for Statistical Computing, Vienna, Austria).

This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Institutional Review Board of Severance Hospital (IRB No. 4-2025-1027). The requirement for informed consent was waived owing to the retrospective nature of the study.

A total of 951 women were included: 717 with endometriomas and 234 with dermoid cysts (Supplementary Figure S1; see Additional file). Baseline characteristics differed between groups (Table 1; see end of document). Women with endometriomas were older (31.9 ± 6.1 vs 29.9 ± 5.9 years, p < 0.001) and had lower BMI (21.1 ± 3.0 vs 22.4 ± 4.1 kg/m², p < 0.001). Menstrual cycles were more often regular (92.1% vs 78.9%, p < 0.001) and shorter (28.8 vs 30.6 days, p < 0.001) in the endometrioma group. Median serum AMH was 2.52 ng/mL in the endometrioma group and 2.70 ng/mL in the dermoid group (p = 0.024 by rank-sum), but age-adjusted geometric means did not differ (2.18 vs 2.34 ng/mL; p = 0.245; Supplementary Table S1; see Additional file).

In both groups, AMH declined significantly with increasing age (Supplementary Figure S2; see Additional file). Segmented regression identified breakpoints at 35.7 years (endometrioma group) and 40.4 years (dermoid group) (Supplementary Table S2; see Additional file). The post-break decline was steeper in the dermoid group (β = −0.51 ± 0.15) than in the endometrioma group (β = −0.21 ± 0.02), with a between-group difference reaching significance (p = 0.047), whereas pre-break slopes did not differ (p = 0.790).

After age adjustment using restricted cubic splines (RCS), higher BMI was associated with lower AMH (Figure 1). The RCS model fit the data better than a linear specification (F = 4.08, p = 0.007; Supplementary Table S3; see Additional file), indicating modest nonlinearity, but the overall trend remained inverse. In multivariable models (Table 2; see end of document), age was the strongest determinant (per 9 years: −38.9%, p < 0.001 overall), and women with endometriomas had lower AMH than those with dermoid cysts (−15.8%, 95% CI −24.9% to −5.7%; p = 0.002). For BMI, each 1−kg/m² increase was associated with a 2.3% decrease in AMH overall (p = 0.003). For interpretability, this corresponds to an estimated 11% lower AMH per 5 kg/m² higher BMI, conditional on covariates. In group-specific models, the inverse BMI–AMH association was significant in the dermoid group (β = −0.03; −2.8% per kg/m²; p = 0.009) but not in the endometrioma group (β = −0.02; −1.9% per kg/m²; p = 0.060). However, the BMI × group interaction was not significant (Model 5 interaction p = 0.538; Supplementary Table S4; see Additional file), indicating no evidence of a differential BMI effect between groups. Overall model fit was modest (adjusted R² = 0.22). BMI provided minimal incremental explanatory value after covariate adjustment (partial R² = 0.01), accounting for 1% of the remaining (residual) variance in log−AMH (Supplementary Table S5; Additional file).

Across stepwise models (Models 1–5), the inverse BMI effect persisted in direction in both groups (Supplementary Table S4; Figure 2; see Additional file). In Model 1 (age-adjusted), the association per 1 kg/m² was −1.9% (p = 0.047) in the endometrioma group and −2.8% (p = 0.020) in the dermoid group; in Model 5 (fully adjusted), estimates were −1.5% (p = 0.135) and −2.5% (p = 0.048), respectively, while interactions remained non−significant (p = 0.468–0.573). Model fit and incremental variance explained are summarized in Supplementary Table S5 (Additional file).

When stratifying by age breakpoints, BMI remained inversely related to AMH within strata, and BMI × age group interactions were not significant (endometrioma p = 0.453; dermoid p = 0.063; Supplementary Table S6; see Additional file). Using WHO BMI categories, age-adjusted geometric mean AMH showed no significant group × BMI-category interaction (p = 0.211), indicating similar categorical BMI patterns across groups (Supplementary Table S7; Supplementary Figure S3; see Additional file).

Restricting to BMI ≤35 kg/m², the inverse BMI–AMH association persisted in both groups (dermoid −3.17% per kg/m², p=0.020; endometrioma −2.05%, p=0.044). In contrast, when restricting to BMI ≤30 kg/m², estimates attenuated and were not significant (dermoid −3.11%, p=0.084; endometrioma −1.16%, p=0.303), and group × BMI interactions remained non−significant across ranges (Supplementary Table S8; Supplementary Figure S4; see Additional file). This pattern is consistent with range restriction and a modestly non−linear BMI effect suggested by spline tests (Supplementary Table S3; see Additional file). Model diagnostics did not reveal major violations of assumptions (Supplementary Figure S5; see Additional file).