Development and Validation of a Model for Opioid Prescribing Following Gynecological Surgery.

Importance: Overprescription of opioid medications following surgery is well documented. Current prescribing models have been proposed in narrow patient populations, which limits their generalizability. Objective: To develop and validate a model for predicting outpatient opioid use following a range of gynecological surgical procedures. Design, Setting, and Participants: In this prognostic study, statistical models were explored using data from a training cohort of participants undergoing gynecological surgery for benign and malignant indications enrolled prospectively at a single institution's academic gynecologic oncology practice from February 2018 to March 2019 (cohort 1) and considering 39 candidate predictors of opioid use. Final models were internally validated using a separate testing cohort enrolled from May 2019 to February 2020 (cohort 2). The best final model was updated by combining cohorts, and an online calculator was created. Data analysis was performed from March to May 2020. Exposures: Participants completed a preoperative survey and weekly postoperative assessments (up to 6 weeks) following gynecological surgery. Pain management was at the discretion of clinical practitioners. Main Outcomes and Measures: The response variable used in model development was number of pills used postoperatively, and the primary outcome was model performance using ordinal concordance and Brier score. Results: Data from 382 female adult participants (mean age, 56 years; range, 18-87 years) undergoing gynecological surgery (minimally invasive procedures, 158 patients [73%] in cohort 1 and 118 patients [71%] in cohort 2; open surgical procedures, 58 patients [27%] in cohort 1 and 48 patients [29%] in cohort 2) were included in model development. One hundred forty-seven patients (38%) used 0 pills after hospital discharge, and the mean (SD) number of pills used was 7 (10) (median [IQR], 3 [0-10] pills). The model used 7 predictors: age, educational attainment, smoking history, anticipated pain medication use, anxiety regarding surgery, operative time, and preoperative pregabalin administration. The ordinal concordance was 0.65 (95% CI, 0.62-0.68) for predicting 5 or more pills (Brier score, 0.22), 0.65 (95% CI, 0.62-0.68) for predicting 10 or more pills (Brier score, 0.18), and 0.65 (95% CI, 0.62-0.68) for predicting 15 or more pills (Brier score, 0.14). Conclusions and Relevance: This model provides individualized estimates of outpatient opioid use following a range of gynecological surgical procedures for benign and malignant indications with all model inputs available at the time of procedure closing. Implementation of this model into the clinical setting is currently ongoing, with plans for additional validation in other surgical populations.