Evaluation of inflammatory serum parameters as a diagnostic tool in patients with endometriosis: a case-control study

Patient data A total of 87 patients were included in the study, including 59 women with histomorphologically confirmed evidence of endometriosis (defined by the presence of at least two out of the three parameters: ectopic endometrioid glandular structures, CD10 + endometrial stromal cells, bleeding residuals with hemosiderin/ hematoidin and macrophage appearance) and 28 women with no evidence of endometriosis (control group). Cases within the control group underwent gynecologic therapy due to alternative diseases (infertility, fibroids, other benign lesions of the uterus or ovaries). All study patient cases underwent treatment and surgery at the Department of Gynecology and Obstetrics at Saarland University between 2013 and 2019. Subsequently, a sequential histomorphological investigation was conducted at the Institute of Pathology at Saarland University. As exclusion criteria we a priori defined: “patients with recurrence/postmenopausal endometriosis” , “patients with systematic inflammatory processes” , “hematological diseases (e.g., anemia)” , “diseases of iron metabolism” , or “neoplastic diseases” . See Supp. Table 1 for a tabular summary of our inclusion and exclusion criteria. In line with standard practice, endometriosis surgery is conducted preferably during follicular phase. By convention, a previous hormonal therapy should be discontinued 8 weeks prior to surgery. Within our non-endometriosis control group, patients in this clinical trial were enrolled throughout their menstrual cycle, reflecting a certain degree of variability that occurs in daily clinical practice; controls did not receive hormonal therapy. All data were handled according to the Declaration of Helsinki and the study protocol was approved by the regional ethics committee (Ethics Committee of the Saarland Medical Association; approval no. 46/21). Therefore, all study methods were performed in accordance with the relevant national regulations. Serum samples and analysis of inflammatory serum parameters All included cases’ serum samples were taken from our department’s endometriosis biobank, which comprises clinical patient data as well as serum samples and histopathological information from patients treated at our clinic. Prior to participation written informed consent was obtained from all individuals enrolled in this study; consecutive blood sample collection was performed at the day of surgery and stored in a deep freezer at -80 °C. For this study, serum levels of hepcidin (Human Hepcidin Quantikine ELISA kit; R&D Systems, USA), IL-6 (Human IL-6 Quantikine ELISA kit; R&D Systems, USA) and suPar (suPARnostic ® AUTO Flex ELISA kit; ViroGates A/S, Denmark) were tested in our laboratory employing Enzyme-linked Immunosorbent Assays (ELISA) kits according to the protocol established by the manufacturer. Additionally, serum levels of CRP , ferritin and soluble transferrin receptor (sTfR) were tested in the central laboratory of our hospital according to current clinical standards. Clinical parameters and laboratory data of the diagnostic routine (e.g., hemoglobin) were taken from the internal digital medical record. Statistical analysis and machine-learning A preliminary assessment of normality was performed using the Shapiro-Wilk test. In order to examine potential correlations within the dataset, a Spearman-Rho analysis was conducted. Additionally, the Mann-Whitney test was used to assess variable differences between the endometriosis group and the control group, and the Kruskal- Wallis Test was employed to evaluate differences between three or more groups (rASRM stage). Hereby, the cut-off for statistical significance was set at p < 0.05. For a subsequent machine-learning analysis, the data was initially visualized and explored using non-linear t-distributed Stochastic Neighbor Embedding (t-SNE). For our supervised learning approach, the data were split into a training cohort (77 patients) as well as an external validation cohort (10 patients: 6 patients with endometriosis and 4 patients as controls); the latter serves useful to detect potential tendencies of overfitting within a trained model. Within this study an approximate time-based splitting approach was established, integrating endometriosis patients as well as controls last treated in our clinic into the external validation cohort. Classifier training and selection as well as hyperparameter optimization of the selected model was performed using the MATLAB Classification Learner App within the MATLAB Toolbox (MathWork, Natick, MA) for Statistics and Machine-Learning™. Hereby, endometriosis/no endometriosis were set as response parameters and hepcidin/suPar/IL-6 serum levels served as predictor variables. For internal classifier validation, the method of 5-fold cross validation was chosen. The minimum Redundancy – Maximum Relevance (MRMR) algorithm allowed for identification of feature importance. Classifier performance was assessed using standard metrics such as sensitivity, specificity, positive predictive value, negative predictive value, as well as AUROC curve. While the evaluation of serum inflammatory parameters follows a pre-planned study strategy, our consecutively employed machine learning analysis was performed as an exploratory (post- hoc) approach.

Clinical characteristics of women with and without endometriosis Women with endometriosis (refer to Supp. Table  2 for detailed listing of localization) as well as without endometriosis were matched in age [median (range) (years): 30 (19–40) vs. 29 (22–41) respectively] and with Scientific Reports | (2025) 15:20172 2| https://doi.org/10.1038/s41598-025-05719-1 www.nature.com/scientificreports/ respect to body mass index (BMI) [median (Range) (kg/m 2): 22.4 (17.6–37) vs. 22.1 (17.7–35.8) respectively]. Table 1 presents an overview of additional clinical variables, such as a menstrual cycle, catamenial pain, dysmenorrhea, dyspareunia, and hormonal contraception. Association of inflammatory biomarkers in endometriosis patients Initial univariate analysis determined different levels of suPar (Mann–Whitney test; p = 0.024) and IL-6 (Mann– Whitney test; p < 0.001) between the endometriosis group and our control group; levels of hepcidin, CRP , sTfR and ferritin did not vary significantly between groups, refer to Table 2 for an optical display. With regard to the clinical patient characteristics age and BMI, a consecutive Spearman correlation analysis determined a distinct association of IL-6 (endometriosis group: r = 0.426, p < 0.001; control group: r = 0.724, p < 0.001) and patient’s BMI in both groups but, interestingly, a significant association of suPar concentration with both age (r = 0.462, p = 0.013) and BMI (r = 0.601, p < 0.001) in solely our non-endometriosis control group whereas not in patients with endometriosis (BMI: r = 0.165, p = 0.231; age: r = 0.032, p = 0.811), see also (Table 3). While hepcidin levels initially did not differ between our cohorts, a subsequent analysis of our groups’ percentile rankings (Q25, Q25-50, Q75) indicates differences in relative positions within the respective data sets collected [Q75 ranks: median (Range) (pg/ml): 26031 (17947–34973) in the endometriosis group ( n = 15) vs. 14869 (12914–22755) in the control group ( n = 7; Mann–Whitney test, p = 0.002)]. Therefore, all three inflammatory parameters (suPar, hepcidin, IL-6) collected in the studies were included in a further machine learning classification for the diagnosis of endometriosis. Evaluating our inflammatory markers of interest with regard to the stage of endometriosis (rASRM), we did not determine any association between the degree of inflammation and the Control group (n = 28) Endometriosis group (n = 59) p-value Hepcidin (pg/ml) 8538 (153-22755) 9534 (127-34973) NS suPar (ng/ml) 1.9 (1.3–3.6) 2.3 (0.8–4.2) 0.024 IL-6 (pg/ml) 1.3 (0.4-6.0) 2.9 (0.3–50.3) < 0.001 CRP (mg/ml) 1.2 (0.5–13.7) 1.1 (0.5–26.6) NS sTfR (mg/ml) 2.5 (1.6–4.9) 2.6 (1.9–7.2) NS Ferritin (ng/ml) 40.5 (8-118) 55.0 (6-194) NS Table 2. Results of serum level of inflammatory biomarker put to test, presented as median (range). p-value was determined using Mann-Whitney-test. Control group (n = 28) Endometriosis group (N = 59) Age (years) [median (range)] 30 (19–40) 29 (22–41) BMI (kg/m2) [median (range)] 22.4 (17.6–37) 22.1 (17.7–35.8) Regular menstrual cycle  Y es 21 (75%) 27 (46%)  No 7 (25%) 14 (24%)  Not known 0 (0%) 18 (30%) Pain  Y es 0 (0%) 36 (61%)  No 28 (100%) 19 (32%)  Not known 0 (0%) 4 (7%) Dysmenorrhea  Y es 0 (0%) 31 (52%)  No 28 (100%) 23 (39%)  Not known 0 (0%) 5 (9%) Dyspareunia  Y es 0 (0%) 12 (20%)  No 28 (100%) 42 (71%)  Not known 0 (0%) 5 (9%) Hormonal contraception  Y es 11 (39%) 19 (32%)  No 17 (61%) 37 (63%)  Not known 0 (0%) 3 (5%) Table 1. Clinical characteristics of our endometriosis group as well as our control group. Scientific Reports | (2025) 15:20172 3| https://doi.org/10.1038/s41598-025-05719-1 www.nature.com/scientificreports/ clinical affection of endometriosis (Kruskal-Wallis Test; Hepcidin and rASRM I-IV: p = 0.492, suPar and rASRM I-IV: p = 0.100, IL-6 and rASRM I-IV: p = 0.626), see (Supp. Fig. 1). Machine learning assisted diagnosis of endometriosis based on inflammatory biomarkers A preliminary data display (t-SNE) did not reveal any relevant similarities between our groups. To establish a profound algorithm performance, we split our data into a training cohort for classifier training and an external validation cohort for following testing of the algorithm’s performance accuracy on new data points as mentioned above. Using a training set of 77 patients (53 of the endometriosis group, 24 of the control group) and an internal 5-fold cross validation we employed a decision tree classifier which resulted in an overall accuracy (right classifications/all classifications) of 81.8% with corresponding AUROC value = 0.7948. Figure 1 shows the ROC curve, here for both classes defined (default option when using the MATLAB Classification Learner App). Hereby, the TPR (true positive rate) for endometriosis detection was 92.5% and the TPR for our control group was 58.3%, the resulting PPV (positive predictive value) for diagnosing endometriosis based on our established machine learning algorithm and the aforementioned inflammatory serum biomarkers put to test was 83.1%, the PPV for excluding endometriosis in a patient not affected by the disease was 77.8%; see (Fig. 2). As suggested by our preceding analysis, determination of feature importance proved IL-6 (MRMR, importance score: 0.0897) and suPar (MRMR, importance score: 0.0246) as most useful for diagnostic purposes. Figure 3 additionally depicts internal cut-off points of our decision tree classifier that can manually be followed from the top node to leaf nodes following the branches in dependence of the predictor’s value. When testing our established classifier on the holdout external validation set (6 of the endometriosis group, 4 of the control group), we confirmed our reported metrics with an overall accuracy of 80.0%, suggesting solid classification capabilities without any suspicious signs of potential overfitting. Fig. 1. displays the ROC curve of our supervised learning approach (decision tree) for the classification of endometriosis based on serum levels of suPar, hepcidin, IL-6. The dotted line represents a random-chance classification. Control group Endometriosis group Age BMI Age BMI suPar ng/ml r = 0.462 p = 0.013 r = 0.601 p < 0.001 r = 0.032 p = 0.811 r = 0.165 p = 0.213 IL-6 pg/ml r = 0.042 p = 0.883 r = 0.724 p < 0.001 r = 0.124 p = 0.351 r = 0.426 p < 0.001 Table 3. Correlation analysis of SuPar and IL-6 with clinical parameters according to each a priori defined subgroup (endometriosis/control). Scientific Reports | (2025) 15:20172 4| https://doi.org/10.1038/s41598-025-05719-1 www.nature.com/scientificreports/

Reciprocal relationships between existing endometriosis and increased inflammatory parameters have been confirmed in several studies. From a pathophysiological point of view interactions between ectopic endometrial tissue and pathways of the innate and adaptive immune system as well as regional inflammatory processes have been described so far and are reflected in altered immune cell function but also impaired peritoneal immunosurveillance20,21. However to date, it is not been fully understood whether such phenomena are causative conditions in the etiopathogenesis of endometriosis or vice versa represent reactive conditions22. Fig. 3. Individual cut-off points of our decision tree classifier displayed from the top node including all leaf nodes and branches. suPar soluble urokinase-type plasminogen activator receptor, IL-6 interleukin-6. Fig. 2. Confusion matrices depicting both PPV (positive predictive value)/FDR (false discovery rate) as well as TPR (true positive rate)/FNR (false negative rate) of our decision tree classification and our training data set based on the internal 5-fold cross validation. Scientific Reports | (2025) 15:20172 5| https://doi.org/10.1038/s41598-025-05719-1 www.nature.com/scientificreports/ These findings, which are originally often based on translational research results 23–25are also reflected by numerous diagnostic studies focusing for example on the pro- inflammatory cytokine IL-6 in patients with endometriosis; IL-6 is produced by macrophages, which are among the most prevalent cell types within endometriomas and are frequently observed in close proximity to peritoneal ectopic endometrial lesions. In that regard the team of Kokot et al. determined elevated IL-6 serum levels in patients with endometriosis in comparison to a healthy control group and furthermore described a stage dependent increase in concentration (rASRM Stage III vs. Stage IV; p = 0.0409)26. Taking a different approach, the team of Ghodsi et al. analyzed IL-6 concentrations within follicular fluids from endometriosis patients and determined a higher concentration in comparison to healthy controls27and Incognito et al. examined in a systematic approach the relevance of IL-6 as predictive marker for endometriosis-associated infertility 13. However, an ultimate step of transformation of these findings from benchmark to clinical routine has not yet been fulfilled. It is precisely this gap that some research teams have been addressing recently: by using machine-learning algorithms, they are proposing an alternative way of analyzing data to promote the development of biomarker-based diagnostics towards clinical utility. Vodolazkaia et al. employed inter alia a least squares support vector machines based multivariate analysis of selected plasma biomarkers (VEGF , annexin V , CA-125, glycodelin) in order to detect endometriosis even in patients without prior sonographical evidence (training data set: accuracy 81%; test data set: accuracy 74%)1. In contrast to this, Knific et al. analyzed 40 cytokine plasma biomarkers in 116 endometriosis patients and a healthy control group and did not determine acceptable accuracies when using different machine-learning algorithms (e.g., random forest analysis, decision tree) for classification4. Within our study, we report the machine-learning based classification of endometriosis based on inflammatory serum biomarkers with an internal overall accuracy of 81.8%; our proposed model performs sufficiently comparable even when classifying new data (external validation data set). As revealed already by MRMR algorithm analysis, a closer look at the decision-making basis of our decision tree proves the importance of IL-6 (> 1.98 pg/ml) and suPar (> 2.1 ng/ml) within our supervised learning approach – both markers have been shown to be significantly altered within the endometriosis group even in our preceding univariate analysis. Therefore, our findings are in line with current literature proposing a significant role of suPar within the inflammatory process in patients with endometriosis28. From a critical perspective, the overall sample size of 87 patients hampered the additional uptake of further predictor variables - such as clinical symptoms – which would potentially allow for an improved classification ability of the established algorithm; however, we refrained from doing so in order to ensure a proper relationship between sample size and number of predictor variables (one in ten rule)29. That said, it is worth mentioning that even in the small field of endometriosis research different types of predictor parameters are put to test, aiming for classification/diagnosis relying on clinical symptoms or imaging variables 9. Prospectively, studies with a sufficient number of participants could take into account a larger combination of most promising predictor variables. Additionally, future multi-center study approaches would allow for algorithm validation with further diversified test data sets. Within the scope of this study, solely a selected panel of parameters of interest could be analyzed, leaving out traditional markers in the field of gynecological research such as CA125 (MUC-16, mucin-16), estrogen, or anti-müllerian hormone (AMH). While a study of Somigliana et al. in 2004 did not postulate any significant diagnostic advantages of testing both IL-6 and CA125, future study approaches may not only determine the association of suPar as well as Hepcidin with e.g., CA125, but prospectively also their combined prognostic potential 30. Last but not least, potential confounders possibly arising from individual constitutional differences, inter-individual hormonal discrepancies, or non-controlled environmental influences (e.g., smoking) could always hamper the presented study results. Albeit we thoroughly controlled for known potential confounders, which may impact individual systematic inflammatory processes (refer to Supp. Table 1), a potential risk of (yet) unknown bias always needs to be considered when cautiously interpreting data from a clinical study.

To sum up: The results presented in our study highlight the diagnostic potential not only of IL-6 but recognizable also suPar as a pro-inflammatory serum biomarker in endometriosis patients. By establishing a supervised machine learning algorithm that provides robust accuracy even on an external validation dataset, we introduce a straightforward computational method for integrating our findings into a concrete clinical tool, opening up future prospects for non-surgical diagnostic methods. Data availability Interested parties are warmly invited to contact the corresponding author for individual options. Received: 24 September 2024; Accepted: 4 June 2025

1. Vodolazkaia, A. et al. Evaluation of a panel of 28 biomarkers for the non-invasive diagnosis of endometriosis. Hum. Reprod. 27, 2698–2711 (2012). 2. Domínguez, F . et al. Lipidomic profiling of endometrial fluid in women with ovarian endometriosis†. Biol. Reprod. 96, 772–779 (2017). 3. Cosar, E. et al. Serum MicroRNAs as diagnostic markers of endometriosis: a comprehensive array-based analysis. Fertil. Steril. 106, 402–409 (2016). 4. Knific, T. et al. Multiplex analysis of 40 cytokines do not allow separation between endometriosis patients and controls. Sci. Rep. 9, 16738 (2019). Scientific Reports | (2025) 15:20172 6| https://doi.org/10.1038/s41598-025-05719-1 www.nature.com/scientificreports/ 5. Wang, L., Zheng, W ., Mu, L. & Zhang, S. Identifying biomarkers of endometriosis using serum protein fingerprinting and artificial neural networks. Int. J. Gynecol. Obstet. 101, 253–258 (2008). 6. Parlatan, U. et al. Raman spectroscopy as a non-invasive diagnostic technique for endometriosis. Sci. Rep. 9, 19795 (2019). 7. Perrotta, A. R. et al. The vaginal Microbiome as a tool to predict rASRM stage of disease in endometriosis: a pilot study. Reproductive Sci. 27, 1064–1073 (2020). 8. Akter, S. et al. Machine learning classifiers for endometriosis using transcriptomics and methylomics data. Front. Genet. 10, (2019). 9. Sivajohan, B. et al. Clinical use of artificial intelligence in endometriosis: a scoping review. NPJ Digit. Med. 5, 109 (2022). 10. Manolov, V ., Marinov, B. & Vasilev, V . Serum Hepcidin levels in endometriosis. Akush Ginekol. (Mosk) 54 (1), 32–38 (2015). 11. Izumi, Y . et al. Hepcidin as a key regulator of iron homeostasis triggers inflammatory features in the normal endometrium. Free Radic Biol. Med. 209, 191–201 (2023). 12. Sillem, M. Soluble urokinase-type plasminogen activator receptor is over-expressed in uterine endometrium from women with endometriosis. Mol. Hum. Reprod. 3, 1101–1105 (1997). 13. Incognito, G. G. et al. Interleukin-6 as A useful predictor of Endometriosis-Associated infertility: A systematic review. Int. J. Fertil. Steril. 17, 226–230 (2023). 14. Li, S. et al. Role of Interleukin-6 and its receptor in endometriosis. Med. Sci. Monit. 23, 3801–3807 (2017). 15. Krygere, L., Jukna, P ., Jariene, K. & Drejeriene, E. Diagnostic potential of cytokine biomarkers in endometriosis: challenges and insights. Biomedicines 12, 2867 (2024). 16. Prima, A. T., Hendry, D. & Bachtiar, H. Correlation between levels of Interleukin-6 in peritoneal fluid with degree of pain, adhesion, and endometriosis fertility index score in endometriosis. Andalas Obstet. Gynecol. J. 5, 193–205 (2021). 17. Backes, Y . et al. Usefulness of SuPAR as a biological marker in patients with systemic inflammation or infection: a systematic review. Intensive Care Med. 38, 1418–1428 (2012). 18. Nemeth, E. et al. Hepcidin, a putative mediator of anemia of inflammation, is a type II acute-phase protein. Blood 101, 2461–2463 (2003). 19. Guan, Y ., Carlberg, M., Bruse, C., Carlström, K. & Bergqvist, A. Effects of hormones on uPA, PAI-1 and SuPAR from cultured endometrial and ovarian endometriotic stromal cells. Acta Obstet. Gynecol. Scand. 81, 389–397 (2002). 20. da Riccio, L. Immunology of endometriosis. Best Pract. Res. Clin. Obstet. Gynaecol. 50, 39–49 (2018). 21. Patel, B. G. et al. Pathogenesis of endometriosis: interaction between endocrine and inflammatory pathways. Best Pract. Res. Clin. Obstet. Gynaecol. 50, 50–60 (2018). 22. Paul Dmowski, W . & Braun, D. P . Immunology of endometriosis. Best Pract. Res. Clin. Obstet. Gynaecol. 18, 245–263 (2004). 23. Vigano, P . Expression of intercellular adhesion molecule (ICAM)-1 mRNA and protein is enhanced in endometriosis versus endometrial stromal cells in culture. Mol. Hum. Reprod. 4, 1150–1156 (1998). 24. Bilotas, M. et al. Effect of GnRH analogues on apoptosis and expression of Bcl-2, bax, Fas and FasL proteins in endometrial epithelial cell cultures from patients with endometriosis and controls. Hum. Reprod. 22, 644–653 (2007). 25. Selam, B., Kayisli, U. A., Garcia-Velasco, J. A. & Arici, A. Extracellular Matrix-Dependent regulation of Fas ligand expression in human endometrial stromal cells. Biol. Reprod. 66, 1–5 (2002). 26. Kokot, I., Piwowar, A., Jędryka, M., Sołkiewicz, K. & Kratz, E. M. Diagnostic significance of selected serum inflammatory markers in women with advanced endometriosis. Int. J. Mol. Sci. 22, 2295 (2021). 27. Ghodsi, M., Hojati, V ., Attaranzadeh, A. & Saifi, B. Evaluation of IL-3, IL-5, and IL-6 concentration in the follicular fluid of women with endometriosis: A cross-sectional study. Int. J. Reprod. Biomed. (IJRM) 213–220. https://doi.org/10.18502/ijrm.v20i3.10713 (2022). 28. Grande, G. et al. Cervical mucus proteome in endometriosis. Clin. Proteom. 14, 7 (2017). 29. Harrell, F . E., Lee, K. L. & Mark, D. B. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat. Med. 15, 361–387 (1996). 30. Somigliana, E. Use of the concomitant serum dosage of CA 125, CA 19 – 9 and interleukin-6 to detect the presence of endometriosis.

from a series of reproductive age women undergoing laparoscopic surgery for benign gynaecological conditions. Hum. Reprod. 19, 1871–1876 (2004).

Figures 2 and 3 were created with Biorender.com. All data was handled in alignment with Ethics Committee of Saarland (approval no. 46/21). Written patient consent was obtained from all individuals enrolled in this study. During the revision process, certain sentences and phrases in the manuscript were proofread by ChatGPT 3.5 (OpenAI; USA) to improve both language proficiency and readability. After using these tools, the authors re - viewed and edited the content as needed and take full responsibility for the publication’s content. Author contributions Study design: GGK, MK, PS. Acquisition of samples and data: MK, IM, BHH, EFS, MPN, PS. Statistical analysis: MK, GGK. Data analyzation/interpretation: GGK, MK, IM, MPN. Coordination and supervision: IM, BHH, EFS, GGK. GGK and MK wrote the first draft. All authors contributed to the final draft. Funding Open Access funding enabled and organized by Projekt DEAL. Declarations Competing interests The authors declare no competing interests. Additional information Supplementary Information The online version contains supplementary material available at h t t p s : / / d o i . o r g / 1 0 . 1 0 3 8 / s 4 1 5 9 8 - 0 2 5 - 0 5 7 1 9 - 1 . Correspondence and requests for materials should be addressed to G.G.K. Reprints and permissions information is available at www.nature.com/reprints. Scientific Reports | (2025) 15:20172 7| https://doi.org/10.1038/s41598-025-05719-1 www.nature.com/scientificreports/ Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. © The Author(s) 2025 Scientific Reports | (2025) 15:20172 8| https://doi.org/10.1038/s41598-025-05719-1 www.nature.com/scientificreports/