Uncovering Symptom-Lesion Associations Through Machine Learning.

OBJECTIVE

to evaluate the association between symptoms and the site of endometriosis lesions using machine learning analysis.

DESIGN

retrospective study.

SETTING

Two tertiary hospitals.

PARTICIPANTS

A total of 726 patients undergoing laparoscopic treatment for endometriosis with histological confirmation between 2009 and 2019 and at least one endometriosis-related symptom.

INTERVENTIONS: Clinical data was retrieved from pre-operative consultations and after laparoscopic surgery for endometriosis in an online database. Data was analyzed using machine learning, specifically multiple correspondence analysis (MCA), to evaluate associations between symptoms and lesion location.

RESULTS

MCA revealed three major dimensions explaining 56.9% of the variance. The arrangement of categories in the factorial plane (Dim.1 × Dim.2 × Dim. 3) indicated the presence of three distinct profiles. The first group profile predominantly presents severe pain symptoms, including dysmenorrhea, dyspareunia, and acyclic pelvic pain associated with retrocervical and rectosigmoid lesions. The second group profile is characterized by the presence of infertility, associated with lower pain intensity and ovarian, round ligament, tubal, and pararectal lesions. The third group profile includes patients with mild-moderate pain symptoms, not related to any location of endometriosis.

CONCLUSION

MCA identified distinct clinical profiles in endometriosis patients. Symptom patterns were associated with specific anatomical sites of lesions, suggesting that symptom profiling may assist in predicting disease location and guiding surgical planning. While these associations are probabilistic and cannot replace clinical imaging or individualized judgment, they provide a structured framework that can complement diagnostic reasoning in specialized settings.