Impact of adenomyosis on pregnancy outcomes following frozen embryo transfer: development and validation of a single-center clinical predictive model

Adenomyosis is a complex, estrogen-dependent gynecological disorder defined by the presence of endometrial glands and stroma within the myometrium, accompanied by hyperplasia and hypertrophy of adjacent smooth muscle. These pathological changes contribute to myometrial thickening and an increased uterine volume ( 1 , 2 ). Clinically, adenomyosis commonly presents with pelvic pain, abnormal uterine bleeding, and impaired fertility, while approximately 30% of affected women remain asymptomatic ( 3 ). Studies suggest that isolated adenomyosis is detected in approximately 10% of women presenting for infertility evaluation ( 4 ), with higher detection rates reported among patients with recurrent miscarriage, recurrent implantation failure, and advanced maternal age seeking pregnancy ( 5 ), which collectively supporting a close association between adenomyosis and impaired fertility. The mechanisms by which adenomyosis compromises fertility are not yet fully elucidated, however, accumulating evidence suggests that it involves a combination of uterine structural and functional remodeling, impaired endometrial receptivity, and dysregulation of immune and molecular pathways. Structurally, adenomyosis is associated with thickening of the endometrial-myometrial junctional zone (JZ) and disruption of normal uterine peristalsis, which may impair sperm and embryo transport, alter intrauterine mechanical pressure, and ultimately hinder embryo implantation ( 6 ). At the molecular level, adenomyotic lesions and the surrounding eutopic endometrium exhibit a state of persistent chronic inflammation, characterized by increased proinflammatory cytokines. These changes are accompanied by aberrant expression of implantation-related factors, including HOXA10/HOXA11 and LIF, as well as progesterone resistance, leading to reduced endometrial receptivity and insufficient decidualization ( 7 , 8 ). In addition, adenomyosis is associated with disturbances in the endometrial immune microenvironment, such as altered natural killer cell and macrophage populations ( 9 ) and dysregulation of cellular homeostasis processes, including autophagy ( 10 ), which may further increase the risk of implantation failure and early pregnancy loss. These structural and molecular alterations will result in decreased natural pregnancy rates, as well as reduced implantation rates, clinical pregnancy rates, and live birth rates in assisted reproductive technologies (ART), alongside increased miscarriage rates. For patients with adenomyosis experiencing difficulties in achieving natural pregnancy or recurrent miscarriages, ART serves as an effective treatment option. However, embryo transfer failure not only signifies the loss of a single pregnancy opportunity but may also further exacerbate the negative impact of this condition on the patient’s physical and mental health. Multiple studies confirm that women with adenomyosis exhibit reduced clinical pregnancy and live birth rates in FET cycles, alongside significantly elevated spontaneous miscarriage rates ( 11 , 12 ), indicating poorer overall reproductive outcomes. Concurrently, recurrent implantation failure (RIF) is closely associated with diminished fertility-related quality of life and increased risks of anxiety and depression, imposing substantial psychological burdens and financial pressures. This may even compromise patients’ willingness and adherence to continued treatment ( 13 ). Therefore, systematically evaluating pregnancy outcomes in FET cycles among patients with adenomyosis and identifying early factors affecting key outcomes like live birth holds significant clinical importance for optimizing assisted reproductive strategies, improving prognosis, and alleviating patients’ physical and psychological burdens. This study aims to analyze factors influencing FET outcomes in adenomyosis patients, providing evidence for individualized clinical decision-making.

This study is a single-center retrospective cohort study. Patients diagnosed with adenomyosis via ultrasound or MRI and undergoing FET at the Reproductive Center of Women and Children’s Hospital of Ningbo University between June 2018 and December 2024 were enrolled. Inclusion Criteria: Reproductive-age women with adenomyosis and normal ovarian reserve. Exclusion Criteria: Concurrent uterine fibroids>3cm; concurrent endometrial pathology; concurrent severe endocrine or systemic diseases; concurrent uterine malformations; concurrent stage III or higher endometriosis; concurrent chocolate cysts; concurrent hydrosalpinx or pyosalpinx. The flowchart is shown in Figure 1 . A total of 673 FET cycles were included. The study protocol was reviewed and approved by the hospital’s ethics committee (Approval No.: NBFE-2025-KY-210). For each woman, we attempted to retrieve information on whether the corresponding ovarian stimulation cycle followed a freeze-all strategy or included a fresh embryo transfer prior to subsequent frozen transfers. Detailed ovarian stimulation parameters (protocol, gonadotropin type/total dose, number of oocytes retrieved, and number of embryos obtained) were not consistently available in a linkable format in our database and therefore could not be analyzed. Flowchart. All patients underwent transvaginal three-dimensional ultrasound or pelvic MRI prior to transplantation, with images interpreted by experienced radiologists according to standardized criteria. Diagnosis of adenomyosis followed the revised Morphological Uterus Sonographic Assessment (MUSA) criteria ( 14 ), requiring the presence of ≥2 diagnostic features for inclusion. Key features included: Myometrial cysts; Hyperechogenic islands; Echogenic subendometrial lines and buds; Globular uterus; Asymmetrical myometrial thickening; Fan‐shaped shadowing; Translesional vascularity; Irregular junctional zone; Interrupted junctional zone. Adenomyosis types were classified as either focal (when normal myometrium surrounded >25% of the circumference of the lesion) or diffuse (if normal myometrium surrounded <25% or if it was difficult to differentiate focal from diffuse adenomyosis). Involvement of the junctional zone (JZ) is defined as lesion involvement or disruption of the endometrial-myometrial junctional zone. Uterine volume (V) is calculated using the ellipsoid formula by measuring longitudinal diameter (L), transverse diameter (W), and anteroposterior diameter (H): Pretreatment was categorized as no pretreatment, medical pretreatment, or surgical pretreatment prior to the index FET cycle. Medical pretreatment mainly consisted of depot gonadotropin-releasing hormone agonist (GnRH-a) injections (Leuprorelin Acetate Microspheres for Injection). The duration of GnRH-a pretreatment was individualized in routine practice, and the timing of embryo transfer was determined by experienced reproductive specialists based on overall clinical assessment and available indicators (e.g., symptom control), with FET initiated when the uterus was considered suitable for transfer. Surgical pretreatment referred to laparoscopic excision of adenomyosis lesions. To allow uterine recovery, embryo transfer was generally considered at least 6 months after surgery ( Supplementary Table 1 ). All frozen embryo transfers in this study involved day-3 cleavage-stage (D3) embryos. All included FET cycles were derived from a freeze-all strategy; no fresh embryo transfers were performed in the study population. This study defined two primary outcomes, with statistical units based on each transfer cycle. To avoid within-patient correlation, only the first eligible FET cycle per woman during the study period was included. Because longitudinal linkage of subsequent FET attempts beyond the index transfer was not feasible in our retrospective dataset, cumulative pregnancy and cumulative live birth rates across multiple frozen transfers could not be estimated. (i) Clinical Pregnancy Outcomes: Cycles labeled as “implantation failure” or “biochemical pregnancy” in “transfer outcomes” were defined as the non-clinical pregnancy group; live births and miscarriages were uniformly classified as the clinical pregnancy group. Biochemical pregnancy was defined as serum β-hCG≥7 mIU/mL without ultrasound confirmation. Clinical pregnancy was defined as ultrasound detection of an intrauterine gestational sac, embryo, or fetal heartbeat. (ii) Live Birth Outcomes: Among cycles achieving clinical pregnancy, delivery at ≥24 weeks gestation with viable fetal was defined as live birth. Pregnancy loss occurring before 24 weeks after clinical pregnancy was defined as miscarriage. Analysis of clinical pregnancy outcomes included all cycles meeting inclusion criteria; analysis of live birth outcomes included only cycles achieving clinical pregnancy. All statistical analyses were performed using R software version 4.3.1. Continuous variables were first tested for normality using the Shapiro-Wilk test. Variables approximating normal distribution were expressed as mean ± standard deviation (SD), and intergroup comparisons were performed using the independent samples t-test. For non-normally distributed variables, intergroup comparisons were conducted using the Mann-Whitney U test. Categorical variables were expressed as n (%), and intergroup comparisons were performed using the chi-square (χ²) test or Fisher’s exact test. Univariate logistic regression analyses were conducted with “clinical pregnancy outcome” (non-clinical pregnancy = 0, clinical pregnancy = 1) and “live birth outcome” (miscarriage = 0, live birth = 1) as dependent variables. Odds ratios (ORs) with 95% confidence intervals (CIs) and two-tailed P values were reported. Candidate predictors were pre-specified based on clinical relevance and prior evidence, including age, BMI, duration of infertility, AMH, PBAC, HB, CA125, uterine volume, dysmenorrhea severity, adenomyosis type, JZ involvement, and pretreatment modality, and were entered simultaneously into multivariable logistic regression models. To construct parsimonious nomogram-based prediction models, we retained predictors that remained statistically significant after multivariable adjustment. Because the primary aim was prognostic model development rather than multiple hypothesis testing, no formal multiplicity correction (e.g., Bonferroni) was applied. Instead, P values were interpreted cautiously alongside model performance and internal validation. Separate multivariable models were established for clinical pregnancy and live birth, and adjusted ORs with 95% CIs were reported. For improved clinical interpretability, continuous predictors were rescaled to meaningful increments. CA125 was expressed per 10 U/mL increase, and uterine volume (recorded in mm 3 ) was expressed per 10,000 mm 3 (i.e., 10 cm 3 ) increase. Corresponding ORs/aORs and 95% CIs were transformed accordingly. Based on the multivariate models, the rms package was used to construct nomograms estimating individual outcome probabilities. Model discriminative power was evaluated via ROC curves and AUC. Bootstrap resampling (B = 500) generated calibration curves to assess consistency between predicted probabilities and actual outcomes. Decision curve analysis (DCA) compared the net benefit of “model-based decision”, “all interventions”, and “no intervention” at different threshold probabilities to evaluate the models’ potential clinical utility. Patients were stratified into three probability groups based on model-predicted probabilities: low, moderate, and high. Actual clinical pregnancy rates and live birth rates were compared across groups to evaluate the model’s prognostic stratification capability. All tests were two-sided, with P<0.05 indicating statistically significant differences.

A total of 673 FET cycles were included, comprising 457 non-clinical pregnancies and 216 clinical pregnancies (32.10%). Among the 216 cycles achieving clinical pregnancy, 160 resulted in live births (23.77%) and 56 in miscarriages (25.93%). All transfers involved day-3 (D3) embryos. All included cycles were frozen embryo transfers derived from a freeze-all strategy in the index stimulation cycle. No fresh embryo transfer was performed prior to the frozen transfers included in this analysis. Detailed distributions of adenomyosis phenotype/severity and pretreatment modality are provided in Supplementary Table 2 . Within the clinical pregnancy analysis cohort, no statistically significant differences were observed between the non-clinical pregnancy group and the clinical pregnancy group in terms of age, BMI, duration of infertility, or AMH (all P>0.05, Table 1 ), indicating overall comparability of baseline characteristics between the two groups. Further stratification within the clinically pregnant cohort revealed no significant differences in age, BMI, duration of infertility, or AMH levels between the miscarriage and live birth groups (all P>0.05, Table 2 ). Comparison of baseline characteristics between non-clinical and clinical pregnancy groups. 1 Mean ± SD. Comparison of baseline characteristics between miscarriage and live birth groups among clinically pregnant cycles. 1 Mean ± SD. Univariate logistic regression analysis with “clinical pregnancy” as the outcome showed a mild negative correlation between PBAC score and clinical pregnancy (OR = 0.99, 95%CI 0.98-1.00, P = 0.049). For each 10 U/mL increase in CA125, the odds of clinical pregnancy decreased (OR = 0.73, 95% CI 0.63-0.84, P<0.001). Similarly, for each 10,000 mm 3 (10 cm 3 ) increase in uterine volume, the odds of clinical pregnancy decreased (OR = 0.64, 95% CI 0.58-0.72, P<0.001). Hemoglobin levels showed no significant association with clinical pregnancy. Compared to no dysmenorrhea, moderate-to-severe dysmenorrhea significantly reduced the likelihood of clinical pregnancy (OR = 0.44, 95%CI 0.29-0.68, P<0.001), while mild dysmenorrhea had no significant effect. Focal adenomyosis was associated with a higher likelihood of clinical pregnancy than diffuse adenomyosis (OR = 2.84, 95%CI 2.00-4.04, P<0.001). The highest clinical pregnancy rate was observed in those without JZ involvement, while the odds ratio for clinical pregnancy significantly decreased when JZ was involved (OR = 0.34, 95%CI 0.24-0.48, P<0.001). Pretreatment modalities also demonstrated a gradient effect: compared with no pretreatment, surgical intervention showed no significant change in odds ratio, whereas medical therapy significantly increased the clinical pregnancy rate (OR = 6.95, 95%CI 4.54-10.64, P<0.001) ( Table 3 ). Univariate logistic regression analysis of factors associated with clinical pregnancy in adenomyosis patients undergoing FET. OR, odds ratio; CI, confidence interval. After incorporating age, BMI, duration of infertility, AMH, PBAC, HB, CA125, uterine volume, dysmenorrhea severity, adenomyosis type, JZ involvement, and pretreatment modality into a multivariate logistic regression model, the effects of some indicators stabilized. CA125 remained an independent risk factor after adjustment. Per 10 U/mL increase, the adjusted OR was 0.84 (95% CI 0.72-0.99, P = 0.037). Increased uterine volume also remained independently associated with reduced clinical pregnancy. Per 10,000 mm 3 (10 cm 3 ) increase, the adjusted OR was 0.70 (95% CI 0.61-0.79, P<0.001). Regarding symptoms and structural factors, moderate-to-severe dysmenorrhea remained associated with lower clinical pregnancy rates (adjusted OR = 0.55, 95%CI 0.32-0.94, P = 0.029); Compared with diffuse adenomyosis, focal adenomyosis showed a higher odds ratio for clinical pregnancy (adjusted OR = 1.88, 95%CI 1.20-2.96, P = 0.006); involvement of JZ significantly reduced the probability of clinical pregnancy (adjusted OR = 0.41, 95%CI 0.27-0.61, P<0.001). Among pretreatment modalities, only medical therapy maintained significant benefit in the multivariable model (adjusted OR = 9.45, 95%CI 5.65-16.25, P<0.001), suggesting that medical intervention was independently associated with higher clinical pregnancy rates after controlling for other factors, whereas surgical therapy alone did not reach statistical significance ( Table 4 ). Multivariable logistic regression analysis of factors associated with clinical pregnancy in adenomyosis patients undergoing FET. OR: odds ratio; CI: confidence interval. Adjusted ORs with 95% CIs are shown. For categorical variables, the P value on the variable header row is the likelihood ratio test (overall effect); level-specific P values are Wald tests. P<0.001 is displayed as <0.001. Based on the significant predictors identified in the multifactorial analysis (CA125, uterine volume, dysmenorrhea severity, adenomyosis type, JZ involvement, and pretreatment modality), a multifactorial logistic regression model was constructed with “clinical pregnancy” as the outcome. A nomogram was subsequently developed based on this model ( Figure 2A ). The nomogram assigns corresponding points to each variable, converting the cumulative total score into the probability of clinical pregnancy, thereby enabling visual risk estimation at the individual level. Clinical pregnancy prediction model in adenomyosis patients undergoing FET. (A) Nomogram for predicting the probability of clinical pregnancy based on CA125, uterine volume, dysmenorrhea severity, adenomyosis type, junctional zone (JZ) involvement, and pretreatment modality. (B) Receiver operating characteristic (ROC) curve of the model (AUC=0.830). (C) Calibration curve showing good agreement between predicted and observed clinical pregnancy rates. (D) Decision curve analysis demonstrating favorable net benefit across a range of clinically relevant threshold probabilities. The ROC curve showed that the model had an AUC of 0.830, indicating strong discriminatory ability between “non-clinical pregnancy” and “clinical pregnancy” ( Figure 2B ). In the calibration curve, the model’s predicted probabilities generally approached the ideal 45° line relative to observed clinical pregnancy rates, indicating good calibration ( Figure 2C ). Decision curve analysis further demonstrated that within commonly used clinical threshold probability ranges, this predictive model yields higher net benefit compared to strategies of “treating all as clinical pregnancies” or “treating all as non-clinical pregnancies”, suggesting its potential clinical utility ( Figure 2D ). To evaluate the model’s performance in practical stratified management, the 673 cycles were divided into three groups based on the model-predicted clinical pregnancy probability: low (n=225), medium (n=224), and high (n=224). The median predicted probabilities (Q1-Q3) for the three groups were 0.083 (0.037-0.125), 0.231 (0.194-0.288), and 0.605 (0.484-0.818), respectively. The corresponding actual clinical pregnancy rates were 11.1%, 16.5%, and 68.8%, with statistically significant differences between groups (P<0.001) ( Table 5 ). As the predicted probability stratification increased, the actual clinical pregnancy rate showed a marked gradient increase, indicating that this model not only possesses good statistical performance but also has clear clinical predictive probability stratification capability. Clinical pregnancy rates according to model-predicted probability strata. Patients were stratified into low, medium, and high probability groups according to model-predicted clinical pregnancy probabilities. Among 216 cycles achieving clinical pregnancy, univariate logistic regression with “live birth” as the outcome revealed significant associations between CA125, uterine volume, adenomyosis type, JZ involvement, and pretreatment modality with live birth outcomes. In contrast, age, BMI, duration of infertility, AMH, PBAC, and HB showed no significant relationship with live birth rates. For each 10 U/mL increase in CA125, the odds of live birth decreased (OR = 0.28, 95% CI 0.18-0.45, P<0.001). Increased uterine volume was also associated with reduced live birth. Per 10,000 mm³ (10 cm³) increase, the OR was 0.49 (95% CI 0.39-0.63, P<0.001). Regarding structural and therapeutic factors, patients with focal adenomyosis had a higher live birth rate than those with diffuse adenomyosis (OR = 2.40, 95%CI 1.25-4.59, P = 0.009); involvement of JZ significantly reduced live birth rates (OR = 0.09, 95%CI 0.04-0.19, P<0.001). Compared with no pretreatment, both surgical and medical interventions were associated with higher live birth rates, with medical treatment showing the greatest advantage (OR = 5.40, 95%CI 2.50-11.64, P<0.001) ( Table 6 ). Univariate logistic regression analysis of factors associated with live birth in adenomyosis patients undergoing FET. OR, odds ratio; CI, confidence interval. In the multivariable model, after incorporating the same covariates as the clinical pregnancy model, several indicators still maintained independent predictive value for live birth outcomes. CA125 levels maintained a negative association after adjusting for confounders. Per 10 U/mL increase, the adjusted OR was 0.45 (95% CI 0.25-0.84, P = 0.012). The relationship between enlarged uterine volume and reduced live birth also remained significant. Per 10,000 mm 3 (10 cm 3 ) increase, the adjusted OR was 0.35 (95% CI 0.21-0.56, P<0.001). JZ involvement reduced the adjusted odds ratio for live birth to 0.11 (95%CI 0.04-0.31, P<0.001), indicating a more pronounced negative impact on pregnancy outcomes. Compared with no pretreatment, medication showed a significant positive association with live birth (adjusted OR = 22.13, 95%CI 6.12-105.49, P<0.001), whereas surgical intervention lost statistical significance after multivariable adjustment. Other variables did not demonstrate independent effects in the multivariable model ( Table 7 ). Multivariable logistic regression analysis of factors associated with live birth in adenomyosis patients undergoing FET. OR, odds ratio; CI, confidence interval. Adjusted ORs with 95% CIs are shown. For categorical variables, the P value on the variable header row is the likelihood ratio test (overall effect); level-specific P values are Wald tests. P<0.001 is displayed as <0.001. A live birth prediction model for clinical pregnancies was constructed based on four independent predictors: CA125, uterine volume, JZ involvement, and pretreatment modality. This model therefore predicts the probability of live birth conditional on having achieved clinical pregnancy, and is intended for early-pregnancy prognostic stratification rather than pre-transfer estimation of overall live birth probability. A corresponding nomogram was generated ( Figure 3A ). Cumulative scores enable intuitive estimation of the probability of achieving a live birth for each clinical pregnancy cycle, providing quantitative evidence for early pregnancy prognosis assessment and follow-up strategies. The ROC curve revealed an AUC of 0.921 ( Figure 3B ), indicating strong discriminatory ability between “miscarriage” and “live birth”. To examine potential overfitting, we further evaluated model optimism using bootstrap internal validation; the optimism-corrected performance metrics are reported in Supplementary Table 3 . Calibration curves demonstrated overall good fit between predicted probabilities and actual live birth rates, with only minor deviations observed in the high-prediction probability range ( Figure 3C ). Decision curve analysis suggests that within a broad range of threshold probabilities, this live birth prediction model yields greater net clinical benefit than either “all intervention” or “no intervention” strategies, supporting its potential value as a clinical decision-making tool ( Figure 3D ). Live birth prediction model among clinical pregnancy cycles. (A) Nomogram for predicting live birth probability based on CA125, uterine volume, JZ involvement, and pretreatment modality. (B) Receiver operating characteristic (ROC) curve of the model (AUC=0.921). (C) Calibration curve showing good concordance between predicted and observed live birth outcomes. (D) Decision curve analysis indicating superior net clinical benefit over “all intervention” and “no intervention” strategies. Similarly, based on model predictions of live birth probability, 216 clinical pregnancy cycles were categorized into low, medium, and high probability groups (72 cases each). The median predicted probabilities (Q1-Q3) for the low, medium, and high probability groups were 0.335 (0.109-0.570), 0.903 (0.845-0.933), and 0.987 (0.975-0.994), respectively. The corresponding actual live birth rates were 34.7%, 88.9%, and 98.6%, with significant differences between groups (P<0.001) ( Table 8 ). While the observed live birth rate increased markedly across the three predicted-probability strata, the very high predicted probabilities in the highest stratum should be interpreted with caution given potential optimism in internally developed models, underscoring the need for external validation. Live birth rates according to model-predicted probability strata. Analysis was restricted to cycles achieving clinical pregnancy (n = 216). Patients were stratified into low, medium, and high probability groups according to model-predicted live birth probabilities.

This study, based on a large single-center cohort of patients with adenomyosis undergoing FET cycles, defined “implantation failure + biochemical pregnancy” as non-clinical pregnancy group and “miscarriage + live birth” as clinical pregnancy group. Among cycles achieving clinical pregnancy, miscarriages were categorized as miscarriage group and live births as live birth group. Multivariate prediction models were constructed separately for clinical pregnancy and live birth. Results showed that uterine volume, involvement of the endometrial-myometrial junctional zone (JZ), and pretreatment modality were stable independent predictors in both models. Dysmenorrhea severity and adenomyosis type were primarily associated with achieving clinical pregnancy, while serum CA125 demonstrated stronger association with live birth outcomes. The decision curves constructed based on these variables demonstrated good discrimination (AUC of approximately 0.830 for the clinical pregnancy model and approximately 0.921 for the live birth model) and calibration. Decision curves and probability stratification further suggested that the models possess certain potential for clinical application. In this adenomyosis-only cohort, the clinical pregnancy rate after FET was 32.10%. Because the present study did not include a non-adenomyosis control group, we did not perform between-group comparisons, and the observed pregnancy outcomes should be interpreted as descriptive estimates within women with adenomyosis. Comparisons with other infertility populations should rely on external literature rather than inference from our dataset. Uterine structural remodeling and involvement of the uterine consistently correlated closely with both clinical pregnancy outcomes and live birth outcomes in this study, suggesting these may represent key pathological mechanisms through which adenomyosis impacts reproductive prognosis. On one hand, increased uterine volume was significantly associated with reduced clinical pregnancy and live birth rates in both models, while JZ involvement remained a strong adverse factor (OR approximately 0.3-0.4) in multivariate analysis. Previous studies have indicated that adenomyosis causes thickening, disruption, and abnormal rhythmic contractions of the JZ, impairing sperm and embryo transport as well as uterine pressure, thereby reducing endometrial receptivity during the implantation window ( 6 ). More critically, JZ dysfunction and impaired local blood perfusion may persist during early pregnancy, increasing the risk of miscarriage and placental complications ( 15 ). Although we did not longitudinally assess JZ changes during early pregnancy, prior work suggests that impaired remodeling of the inner myometrium/junctional zone before conception may predispose to defective deep placentation and adverse pregnancy outcomes ( 16 – 18 ). Recent reviews further support the role of junctional zone structure/function in adenomyosis-related infertility and obstetric complications ( 6 ). Our findings directly link these structural and functional abnormalities to two key clinical outcomes “achieving clinical pregnancy” and “ultimately delivering a live birth.” This indicates that uterine volume and JZ involvement should be incorporated as critical considerations in risk assessment and intervention decisions for assisted reproductive management in adenomyosis patients, rather than merely serving as diagnostic imaging information. Clinical and laboratory indicators reflecting disease activity and burden play distinct roles across different outcome stages. Moderate-to-severe dysmenorrhea significantly reduced the probability of clinical pregnancy in both univariate and multivariate models using clinical pregnancy outcomes as endpoints, but its independent impact on live birth outcomes was weaker. This suggests pain primarily reflects lesion activity and abnormal contractions directly interfering with the implantation phase. Once pregnancy is established, its maintenance through to live birth relies more heavily on the local and systemic inflammatory, vascular, and immune environments within the uterus ( 19 , 20 ). Serum CA125 levels showed significant negative correlations with both clinical pregnancy and live birth in this study, particularly emerging as an independent risk factor in the live birth model. This aligns with CA125’s role as a marker of adenomyosis lesion extent and inflammatory burden ( 21 , 22 ). Although some studies reported no significant association between CA125 levels prior to hormone replacement cycle FET and outcomes in adenomyosis patients ( 23 ), these studies often failed to distinguish between different outcome stages and did not adequately account for important confounding factors such as JZ involvement and pretreatment modality. This study suggests that in adenomyosis patients, the relationship between CA125 and uterine volume, JZ involvement, and symptom severity should be interpreted holistically. CA125 should be viewed as a component within the “disease burden-uterine environment-pregnancy outcome” chain rather than a standalone linear predictor. Pretreatment modality was strongly associated with both clinical pregnancy and live birth in this study, however, these findings must be interpreted in the context of treatment indications. In routine practice, treatment selection (particularly surgery) is typically driven by baseline disease severity, symptom burden, and prior treatment failures, making confounding by indication likely. To evaluate the impact of this potential bias, we conducted sensitivity analyses excluding pretreatment modality from the models; discrimination remained acceptable, supporting the robustness of the non-pretreatment predictors ( Supplementary Table 3 ). Consistent with clinical practice, fertility-sparing surgery for adenomyosis is often considered a “last resort,” usually reserved for patients with markedly enlarged uterine volume, severe dysmenorrhea, multiple failed embryo transfers, or symptoms significantly impairing quality of life ( 24 , 25 ). Therefore, the surgical group in our cohort likely represented patients with a greater underlying disease burden and a more complex treatment history. Accordingly, the lower clinical pregnancy and live birth rates observed in the surgical group compared with the medical group are more plausibly attributable to baseline severity and prior treatment failures rather than indicating that surgery itself is “ineffective” or “harmful.” Taken together, the associations between pretreatment modality and pregnancy outcomes in our cohort should be interpreted as observational correlations rather than causal effects. Moreover, GnRH-a pretreatment was individualized and the decision to proceed with embryo transfer was made once the uterus was considered suitable for transfer, which may introduce additional heterogeneity and selection effects. Future prospective studies with standardized pretreatment protocols and explicit criteria for treatment selection are warranted. Finally, the two predictive models developed in this study, together with their nomograms, ROC curves, calibration plots, decision curves, and probability stratification, provide intuitive tools for clinical risk assessment. In the clinical pregnancy model, the observed clinical pregnancy rate increased from approximately 11% to nearly 70% across the three predicted-probability strata (lowest to highest). In the live birth model, the corresponding live birth rate rose from about 35% to nearly 100%, demonstrating a clear outcome gradient that supports the use of these models for prognostic stratification. In practice, patients classified as low probability may benefit from closer evaluation and shared decision-making regarding pre-transfer optimization (e.g., timing of transfer and broader optimization of the uterine environment), however, whether any specific intensified intervention improves outcomes cannot be inferred from this retrospective analysis and requires prospective confirmation. Conversely, for patients with higher predicted probabilities, unnecessary escalation of interventions may be avoided, potentially reducing treatment burden. The strengths of this study include a uniformly diagnosed adenomyosis cohort based on imaging, inclusion of FET cycles only, integration of symptom, laboratory, imaging, and pretreatment variables, and comprehensive model evaluation using discrimination, calibration, decision curves, and risk stratification. Nevertheless, several limitations should be acknowledged. First, the single-center retrospective design is susceptible to selection and information biases. Second, the absence of a non-adenomyosis control group precludes direct comparative conclusions regarding whether pregnancy rates are lower than those of other infertility populations, and residual within-patient correlation may remain, potentially leading to underestimated standard errors. In addition, assessments of JZ involvement and adenomyosis type lacked standardized quantitative criteria. Key determinants of FET outcomes—such as blastocyst status, endometrial thickness at transfer, and endometrial preparation protocols—were not consistently recorded and therefore could not be incorporated, which may introduce residual confounding. Furthermore, our retrospective database did not allow reliable longitudinal linkage of subsequent FET attempts beyond the index transfer, therefore, cumulative clinical pregnancy and cumulative live birth rates across multiple frozen transfers could not be estimated. Detailed ovarian stimulation-cycle characteristics (e.g., stimulation protocol, number of oocytes retrieved, and number of embryos obtained) were also not consistently recorded in a linkable format. Although we reported whether a freeze-all strategy or a preceding fresh transfer occurred, the absence of comprehensive stimulation parameters may limit generalizability and introduces potential residual confounding. Notably, all transfers in this cohort used cleavage-stage (D3) embryos, partially reducing heterogeneity related to embryo stage. Moreover, the live birth model was developed among cycles achieving clinical pregnancy and thus is intended for prognostic stratification conditional on clinical pregnancy rather than comprehensive pre-transfer estimation of overall live birth probability. Treatment regimens were heterogeneous, and the models have undergone internal validation only; external validation in multicenter prospective cohorts (ideally using a unified imaging grading system and incorporating embryo-related variables and pregnancy complications) is required before broader implementation. Future studies evaluating whether risk-stratified management improves real-world outcomes will further clarify the clinical utility of these models in adenomyosis patients.