Patient-Reported Data Augment Prediction Models of Persistent Opioid Use after Elective Upper Extremity Surgery

Area of Science:

• Machine Learning in Healthcare
• Pain Management Research
• Patient-Reported Outcomes

Background:

• Prolonged opioid use post-surgery can lead to opioid use disorder.
• Predictive models for prolonged opioid use often lack patient-reported data (PRD).
• The utility of PRD in predicting postoperative opioid use remains unclear.

Purpose of the Study:

• To develop and evaluate a machine learning model predicting persistent opioid use after upper extremity surgery.
• To assess the contribution of preoperative PRD to predictive accuracy.
• To identify factors associated with prolonged postoperative opioid use.

Main Methods:

• A machine learning approach was used, combining preoperative PRD and electronic health record data.
• Models were trained on a 2018 cohort and validated on a 2019 cohort.
• Opioid use was determined by patient self-report and prescription fills up to 6 months post-surgery.

Main Results:

• The XGBoost model integrating PRD and EHR data achieved an AUC of 0.73 at 6 months.
• Predictive factors for prolonged opioid use included socioeconomic status, substance abuse history, and comorbidities.
• Preoperative Patient-Reported Outcomes Measurement Information System (PROMIS) scores were identified as protective factors.

Conclusions:

• A machine learning model using preintervention data demonstrated good performance in predicting postsurgical opioid use.
• PRD significantly enhanced the predictive capabilities of EHR-based algorithms.
• Incorporating PRD into future models is recommended for effective opioid stewardship.