Machine learning model Endometriosis Pain Index (EPI) may predict poor pain-related outcomes and help in preoperative counseling for endometriosis surgery, according to study results published in Pain.
Approximately 10% of individuals of reproductive age are affected by endometriosis; however, resources are limited for predicting pain-related outcomes after endometriosis surgery.
To develop a predictive machine learning-based model for pain-related quality of life (QOL) after endometriosis surgery, researchers analyzed data from participants prospectively enrolled in the Endometriosis Pelvic Pain Interdisciplinary Cohort (ClinicalTrials.gov Identifier: NCT02911090) at a tertiary referral center from December 2013 to 2020.
Pending external validation, the EPI may support presurgical counseling alongside other factors (including fertility goals and response to hormonal therapy).
Eligible participants were aged 50 years and younger, were preoperatively diagnosed with endometriosis, underwent index surgery for endometriosis pain at the center, and completed the pain subscale of the Endometriosis Health Profile-30 (EHP-30), evaluating 11 key behaviors of daily functioning and QOL, at both baseline and follow-up.
Patients with previous hysterectomy were excluded from the analysis.
Participant data were randomized into training and test cohorts. The training cohort data was used for model development and performance assessment and the test cohort data was used as a control.
The primary outcome of the study was self-reported pain-related QOL in the top 25% of EHP-30 scores in North America at 1 to 2 years after surgery.
Data from 650 participants (488 in the training cohort and 162 in the control cohort) were included in the study. Of the machine learning models analyzed, the random forest model was found to be the most consistent, with the highest net benefit based on discrimination, integrated calibration (0.029), and decision curve analyses between the training cohort (area under the curve [AUC], 0.768; 95% CI, 0.690-0.837) and test cohort (AUC, 0.766; 95% CI, 0.676-0.863).
While results were comparable with those in the multilayer perception neural network, the random forest model was simpler and had more clinical usefulness.
Researchers also found that preoperative predictions of poor pain outcomes after surgery were highly associated with demographic factors, such as race, ethnicity, and younger age (P P <.001 surgery="" type="" vs="" hysterectomy="" and="" psychologic="" factors="" including="" anxiety="" depression="" pain="" catastrophizing="" wp_automatic_readability="8.8313817330211">P <.001 wp_automatic_readability="17.661971830986">
Limitations included the single-center design; the study population lacking diversity for granular comparison, leading to lack of generalizability; the small number of bilateral salpingo-oophorectomy cases; and limitations in the EHP-30 pain subscale that may have presented classification imbalances.
“Pending external validation, the EPI may support presurgical [counseling] alongside other factors (including fertility goals and response to hormonal therapy),” the study authors concluded. “Future research could include clinical trial randomizing patients to EPI-guided [counseling] vs standard care, evaluating whether the EPI reduces unnecessary surgeries and improves postoperative pain outcomes.”
Some of the authors declared affiliations with pharmaceutical industry, please see the original reference for the full list of disclosures.
Leave a comment