• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Sunday, November 22, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Health

Machine Learning Model Evaluates Endometriosis Risk Using Routine Laboratory Blood Markers

by
November 21, 2026
in Health
Reading Time: 6 mins read
0
Machine Learning Model Evaluates Endometriosis Risk Using Routine Laboratory Blood Markers

Machine Learning Model Evaluates Endometriosis Risk Using Routine Laboratory Blood Markers

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

A machine learning diagnostic model that integrates eight routinely available laboratory indicators demonstrated consistent discrimination in identifying existing endometriosis across internal and temporal validation cohorts, according to a retrospective diagnostic modeling study published in Frontiers in Medicine. Developed by researchers at the First Affiliated Hospital of Bengbu Medical University, the gradient boosting machine model achieved an area under the receiver operating characteristic curve of 0.8401 in an internal test cohort and 0.8565 in an independent temporal validation cohort. Despite this discriminative performance, the laboratory-based algorithm did not achieve a statistically significant improvement over an extended clinical model that combined established clinical symptoms, patient history, and transvaginal ultrasonography findings. Furthermore, at a fixed probability threshold of 0.50, the algorithm demonstrated relatively high specificity but limited sensitivity, missing a substantial proportion of confirmed cases and highlighting that routine laboratory indicators warrant investigation primarily as potential adjuncts rather than replacements for established clinical assessments.

Endometriosis is a common, chronic gynecological disorder defined by the presence of endometrial-like tissue outside the uterine cavity, affecting an estimated ten percent of women of reproductive age. The condition presents with heterogeneous clinical manifestations that frequently include dysmenorrhea, chronic pelvic pain, dyspareunia, and infertility, resulting in considerable impairments to physical health, overall quality of life, and economic productivity. Timely clinical diagnosis remains difficult because presenting symptoms often overlap substantially with various other gynecological and pelvic disorders, frequently leading to multi-year diagnostic delays across diverse healthcare delivery systems. Contemporary clinical guidelines emphasize integrated noninvasive evaluations incorporating comprehensive patient history, bimanual physical examinations, and dedicated pelvic imaging such as transvaginal ultrasonography and magnetic resonance imaging. Although modern protocols no longer mandate invasive laparoscopy with histological confirmation as an initial requirement for all patients, negative imaging findings cannot definitively exclude disease, particularly in superficial peritoneal lesions, maintaining ongoing scientific interest in accessible, circulating biomarkers.

Blood-based diagnostic markers represent an accessible modality capable of reflecting systemic biological processes, yet single circulating markers have proven insufficient for standalone clinical decision-making. Cancer antigen 125 represents the most extensively investigated circulating biomarker in endometriosis research, but while it demonstrates relatively high specificity, its clinical sensitivity is limited, particularly among individuals presenting with minimal or mild disease stages. Beyond single oncological markers, endometriosis involves complex systemic alterations including localized and peripheral immune activation, inflammatory signaling cascades, and shifting metabolic profiles. Innate and adaptive immune cell populations, including neutrophils, lymphocytes, and monocytes, participate in inflammatory processes and pelvic microenvironment remodeling, while altered circulating lipid concentrations have also been documented. Because individual hematological and metabolic parameters lack disease specificity when evaluated in isolation, multivariable machine learning algorithms offer an analytic framework to capture complex, nonlinear interactions among conventional laboratory values routinely measured during standard patient admissions.

To evaluate whether conventional laboratory measurements could inform noninvasive assessment, the investigators conducted a single-center retrospective investigation analyzing 23,606 patient records spanning six years of gynecological admissions between January 2020 and December 2025. Strict sequential exclusion criteria were applied to assemble a well-characterized cohort: patients were excluded if they were younger than 18 or older than 55 years of age, had final diagnoses outside the predefined gynecological scope, presented with concomitant malignant disease, had active systemic autoimmune disorders, or had received lipid-lowering agents within 90 days or antibiotic or immunomodulatory therapies before initial inpatient blood collection. These criteria yielded a final study population of 2,452 eligible women, comprising 839 individuals with surgically and histopathologically confirmed endometriosis and 1,613 disease controls admitted for alternative gynecological conditions, reflecting an overall disease prevalence of approximately 34.2 percent across the analyzed clinical cohorts.

The study population was organized into three distinct methodological partitions to separate model derivation from subsequent independent verification. Patients evaluated during the initial four-year development period between January 2020 and December 2023 constituted the development cohort of 1,470 participants, which was randomly split using stratified sampling into a model-training dataset of 1,029 women (352 endometriosis cases and 677 controls) and a held-out internal test dataset of 441 women (151 endometriosis cases and 290 controls). To rigorously test the transportability of the resulting algorithms over time, a separate temporal validation cohort was established comprising 982 consecutive eligible patients treated during the subsequent two-year period between January 2024 and December 2025 (336 endometriosis cases and 646 controls). Neither the internal test partition nor the temporal validation cohort was accessed during candidate predictor selection, algorithm comparison, or model hyperparameter tuning.

All cases of endometriosis in the study were established through surgical visualization and histopathological verification. The disease-control comparator group comprised patients admitted for other defined gynecological pathologies, including adenomyosis, benign ovarian cysts, pelvic inflammatory disease, uterine leiomyomas, functional dysmenorrhea, chronic pelvic pain, and unexplained infertility. Across the training, internal test, and temporal validation partitions, 77.1 percent, 75.5 percent, and 78.2 percent of disease controls underwent surgical procedures for their primary conditions, while the remainder were diagnosed through comprehensive clinical evaluations and ancillary testing. Because surgical verification was not universally performed across every disease control, the authors noted that occult peritoneal endometriosis could not be completely excluded in every non-surgical control, appropriately defining the comparator cohort as a clinical disease-control population rather than a definitively verified healthy cohort.

Twelve candidate predictors were prespecified based on their routine availability in hospital laboratories: age, body mass index, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, lymphocyte count, neutrophil count, monocyte count, white blood cell count, platelet count, and cancer antigen 125. All laboratory measurements were obtained from fasting blood samples collected outside of active menstruation. Missing values were handled during model development using multiple imputation by chained equations across five imputed datasets. Feature selection combined least absolute shrinkage and selection operator regression using the one-standard-error criterion with the random forest-based Boruta algorithm. White blood cell count was excluded prior to final selection due to a Pearson correlation exceeding 0.90 with neutrophil count. Majority voting across the imputed datasets identified eight final predictors: cancer antigen 125, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, lymphocyte count, neutrophil count, monocyte count, and platelet count.

Using the training dataset of 1,029 participants, the researchers compared six candidate machine learning algorithms using five repeats of nested ten-fold cross-validation, with internal five-fold cross-validation dedicated to hyperparameter optimization. The evaluated algorithms included logistic regression, support vector machines, neural networks, adaptive boosting, extreme gradient boosting, and gradient boosting machines. Gradient boosting achieved the highest mean cross-validation area under the receiver operating characteristic curve at 0.8483, closely followed by extreme gradient boosting at 0.8470, adaptive boosting at 0.8397, neural networks at 0.8351, support vector machines at 0.8275, and logistic regression at 0.8113. Based on the prespecified selection criterion, the gradient boosting machine was selected for final model construction, configured with an ensemble of 150 trees, an interaction depth of three, a shrinkage rate of 0.05, a minimum terminal node size of ten, and a bag fraction of 0.8.

When evaluated in the independent validation partitions, the final locked gradient boosting machine model achieved an area under the curve of 0.8401 (95 percent confidence interval, 0.8006 to 0.8797) in the internal test cohort and 0.8565 (95 percent confidence interval, 0.8320 to 0.8810) in the temporal validation cohort. The model demonstrated superior discrimination compared to cancer antigen 125 alone, which recorded areas under the curve of 0.7353 and 0.7583 in the respective cohorts, and outperformed a basic clinical model combining the biomarker with age, body mass index, and pregnancy history, which yielded 0.7684 and 0.7683. However, when benchmarked against an extended clinical model incorporating symptoms of dysmenorrhea, pelvic pain, infertility, and transvaginal ultrasonography findings, the performance differences narrowed dramatically. The extended clinical model achieved areas under the curve of 0.8326 in internal testing and 0.8410 in temporal validation; paired DeLong tests confirmed that the incremental discrimination offered by the machine learning algorithm was not statistically significant in either cohort (P = 0.758 and P = 0.294, respectively).

The operational classification metrics further characterized the clinical trade-offs associated with the model. Applying a fixed classification probability threshold of 0.50 without post-hoc re-optimization yielded a specificity of 91.38 percent in the internal test cohort and 88.85 percent in the temporal validation cohort. Conversely, sensitivity reached only 55.63 percent and 59.52 percent in the respective cohorts, indicating that more than forty percent of patients with surgically confirmed endometriosis received false-negative model classifications at this threshold. Model calibration showed Brier scores of 0.1471 internally and 0.1421 temporally, with calibration intercepts of 0.0754 and 0.0137 and calibration slopes of 1.0889 and 1.1295, reflecting modest departures from ideal calibration where predicted probabilities were slightly less dispersed than observed outcome proportions. Exploratory decision curve analysis indicated positive cohort-specific net benefit across intermediate probability thresholds, but these curves were unadjusted for external population prevalence.

Subgroup analyses revealed that model discrimination varied descriptively according to anatomical presentation and disease severity. In patients with revised American Society for Reproductive Medicine stage III to IV disease, the model achieved areas under the curve of 0.8625 in internal testing and 0.8849 in temporal validation, whereas discrimination among patients with stage I to II disease was lower at 0.8141 and 0.8286. Similarly, the algorithm performed better in the presence of ovarian endometriomas (areas under the curve of 0.8822 and 0.8963) than in their absence (0.8147 and 0.8430). When restricted to the 987 participants across both validation cohorts who possessed complete numerical measurements for all eight indicators, the frozen prediction pipeline reproduced stored probabilities to machine precision, confirming computational reproducibility. Nevertheless, the authors emphasized key limitations, including the retrospective single-center design, lack of uniform surgical verification among disease controls, absence of menstrual cycle standardization, and the inability of the standalone software pipeline to process incomplete clinical records without external imputation.

Subject of Research: Machine learning prediction of endometriosis using routine laboratory indicators

Article Title: Development and temporal validation of a gradient boosting machine model for endometriosis using routine laboratory indicators

Article References: Hu, Q., Li, H., Ma, S., Zhao, Z., Rong, T., Wang, L., Wei, J., & Ye, G. (2026). Development and temporal validation of a gradient boosting machine model for endometriosis using routine laboratory indicators. Frontiers in Medicine, 13, Article 1934454. https://doi.org/10.3389/fmed.2026.1934454

Image Credits: AI Generated

DOI: 10.3389/fmed.2026.1934454

Keywords: endometriosis, machine learning, gradient boosting machine, routine laboratory indicators, temporal validation, CA125, lipid profile, hematological indicators, diagnostic modeling, clinical prediction models, receiver operating characteristic, discrimination, calibration, decision curve analysis, gynecological assessment, transvaginal ultrasonography, internal test cohort

News Source: Teresa Odom. (November 21, 2026). Machine Learning Model Evaluates Endometriosis Risk Using Routine Laboratory Blood Markers. Scienmag.

Tags: CA125calibrationclinical prediction modelsdecision curve analysisDiagnostic ModelingdiscriminationEndometriosisGradient Boosting Machinegynecological assessmentHematological Indicatorsinternal test cohortlipid profileMachine Learningreceiver operating characteristicroutine laboratory indicatorstemporal validationtransvaginal ultrasonography
Share12Tweet7Share2ShareShareShare1

Related Posts

Loneliness Strongly Associated With Depression Among Older Adults in Golestan Iran

Loneliness Strongly Associated With Depression Among Older Adults in Golestan Iran

November 22, 2026
Self-Powered Neuroprosthesis Harvests Walking Energy to Improve Mobility After Stroke

Self-Powered Neuroprosthesis Harvests Walking Energy to Improve Mobility After Stroke

November 22, 2026

Trifluoroacetate Counterion Reduces Atherosclerosis and Plasma Lipids in Mice via PPAR-α

November 22, 2026

RAGE, Not TLR4, Drives Inflammatory Pain in Human Skin

November 22, 2026

POPULAR NEWS

  • RAGE, Not TLR4, Drives Inflammatory Pain in Human Skin

    RAGE, Not TLR4, Drives Inflammatory Pain in Human Skin

    29 shares
    Share 12 Tweet 7
  • Phloem Girdling Drives Rhizosphere Microbial Adaptation to Belowground Carbon Limitation in Tobacco

    29 shares
    Share 12 Tweet 7
  • Direct Integration of GIS and AI in Tuberculosis Studies Remains Rare

    29 shares
    Share 12 Tweet 7
  • Cortical Folding Motifs Reveal Distinct Brain Shape Variations and Asymmetry From Infancy

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Loneliness Strongly Associated With Depression Among Older Adults in Golestan Iran

Self-Powered Neuroprosthesis Harvests Walking Energy to Improve Mobility After Stroke

Comparative Genomics Unveils Evolutionary Lineages and Stress Genes Across Four Vigna Legumes

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 89 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.