For a study, researchers sought to create and verify a prediction model for post-discharge opioid usage in cesarean delivery patients. They did a prospective cohort study of patients with cesarean delivery. Patients were enrolled after surgery and completed pain and opioid usage questionnaires 14 days later. Clinical information was extracted from the electronic health record (EHR). At discharge, participants were given 30 pills of hydrocodone 5 mg–acetaminophen 325 mg and asked about their post-discharge opioid usage. The main result was the total number of morphine milligram equivalents consumed. They built three proportional odds prediction models of postdischarge opioid use: a comprehensive model with 34 variables available before hospital discharge, an EHR model that removed questionnaire data, and a simplified model. The reduced model employed forward selection to add predictors incrementally until 90% of the full model performance was obtained. A priori, predictors were graded based on data from the literature and past research. Discrimination was used to calculate predictive accuracy (concordance index).

Between 2019 and 2020, 459 people were registered, with 279 completing the standardized study prescription. Participants used a median of eight pills (interquartile range 1–18 tablets) after discharge of the 398 with outcome assessments, 23.5% used no opioids, and 23.0% used all opioids. Each of the models predicted post-discharge opioid usage with great accuracy (concordance index range 0.74–0.76 for all models). They chose the reduced model as the final model since it had an identical model performance with the fewest number of predictors, all of which were collected from the EHR (inpatient opioid use, tobacco use, and depression or anxiety).

A model based on three variables revealed in the EHR—inpatient opioid usage, cigarette use, and depression or anxiety—exactly predicted post-discharge opioid use. It represented a chance to tailor opioid prescriptions following cesarean delivery.

Reference:journals.lww.com/greenjournal/Fulltext/2022/05000/Development_and_Validation_of_a_Model_to_Predict.20.aspx