Utilizing Artificial Intelligence: Machine Learning Algorithms to Develop a Preoperative Endometriosis Prediction Model

Abstract

Objective: To evaluate the predictive value of clinical features in the diagnosis of endometriosis by utilizing machine learning algorithms (MLAs), aiming to develop an accurate, explainable prediction model.

Design: Retrospective case-control study from 2011 to 2022.

Setting: Tertiary referral center specializing in pelvic pain and minimally invasive gynecologic surgery.

Participants: All women aged 18 to 55 undergoing laparoscopic or robot-assisted excision of lesions of the ovary, pelvic viscera, or peritoneal surface by a single surgeon from 2011 to 2022. Exclusion criteria included women who required emergent surgery, as well as those lacking surgical specimens submitted for pathological analysis or a documented preoperative pelvic examination.

Interventions: A total of 209 clinical features, including demographics, presenting symptoms, gynecologic/obstetric history, and physical exam findings, were analyzed as predictors of endometriosis. The primary outcome was model performance in predicting endometriosis, evaluated using accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve. Feature importance was assessed using Shapley Additive Explanation values.

Results: Among 788 participants, 654 (83%) had pathology-confirmed endometriosis. The MLA, extreme gradient boosting, achieved an accuracy of 83%, sensitivity of 96%, and area under the receiver operating characteristic curve of 0.81. Shapley Additive Explanation analysis identified key predictors, including emesis (141 [21.56%] vs 10 [7.46%], p < .001), crampy pain (325 [49.69%] vs 38 [28.36%], p < .001), regular periods (429 [65.60%] vs 60 [44.78%], p < .001), severity of dysmenorrhea (0-3 Likert scale) (3 [1, 2] vs 3 [1, 2] p = .02), and retrocervical tenderness on rectovaginal exam (126 [19.27%] vs 7 [5.22%], p < .001).

Conclusion: This study demonstrates that MLAs have potential to predict endometriosis preoperatively utilizing clinical features. Identified predictors, such as retrocervical tenderness, crampy pain, and regular periods, can aid primary care providers in early recognition and referral. Further validation in diverse populations is necessary to develop a widely applicable clinical prediction tool.

Keywords: Artificial intelligence; Laparoscopy; Machine learning; Prediction model; Symptoms.