Diagnosis of Endometriosis Based on Comorbidities: A Machine Learning Approach

Area of Science:

• Reproductive Medicine
• Medical Informatics
• Computational Biology

Background:

• Endometriosis, characterized by endometrial-like tissue outside the uterus, remains difficult to diagnose and treat.
• Its association with other diseases is a common clinical observation, suggesting complex underlying mechanisms.
• Current diagnostic methods for endometriosis often involve invasive procedures and can lead to delays in treatment.

Purpose of the Study:

• To develop and evaluate machine learning models for improved endometriosis diagnosis.
• To identify key clinical features predictive of endometriosis using a large dataset.
• To assess the feasibility of artificial intelligence in supporting clinical decision-making for endometriosis.

Main Methods:

• Utilized a large dataset of 627,566 cases, including endometriosis patients and controls.
• Developed a machine learning platform with algorithms: logistic regression, decision tree, random forest, AdaBoost, and XGBoost.
• Employed Shapley Additive Explanation (SHAP) values to determine feature importance.

Main Results:

• The XGBoost model demonstrated superior performance with an Area Under the Curve (AUC) of 0.725.
• Achieved a sensitivity of 68.6% and specificity of 62.9% on the test set.
• Top predictive features included age, infertility, uterine fibroids, anxiety, and allergic rhinitis.

Conclusions:

• Machine learning shows promise as an auxiliary tool for endometriosis diagnosis, offering valuable insights.
• The predictive model achieved a high negative predictive value (99.58%), useful for ruling out the condition.
• Further research with more informative features is needed to enhance model performance and clinical utility.