Development and Validation of Machine Learning Algorithms for Predicting Prolonged Postoperative Opioid Use in Spinal Metastatic Disease

Abstract

Introduction

Operative management of spinal metastatic disease is largely for symptom palliation and revolves around the expectation that postoperative survival will exceed the recovery period. Long-term postoperative opioid use is a clinically useful indicator of recovery. Few studies have developed machine learning models to predict this outcome in spinal metastatic disease patients.

Methods

The Merative™ Marketscan® Commercial Database and Medicare Supplement were analyzed to identify adult patients who underwent surgery for extradural spinal metastatic disease between 2006 and 2023. Patients were required to have at least 6 months of continuous preadmission data, and 6 months of continuous post-discharge follow-up. The primary outcome was prolonged opioid use, defined as filling a perioperative prescription followed by another between 90- and 180-days post-discharge. Cumulative days of postoperative opioid supply was assessed as a secondary outcome. Five models (stochastic gradient boosting, support vector machine, neural network, random forest and penalized logistic regression) were trained on a 70% training sample and validated on the withheld 30%.

Results

A total of 732 patients were included, of which 341 (46.6%) had prolonged post-discharge opioid use. The random forest algorithm had the best predictive performance in terms of discrimination (area under the curve [AUC]: 0.611), calibration (intercept: 0.18, slope: 0.613) and overall accuracy (Brier score: 0.24).

Conclusion

We developed and validated parsimonious predictive models to estimate risk of prolonged opioid use after surgery for extradural spinal metastatic disease. Integrating these models into physician- and patient-facing interfaces may improve prognostication, enhance clinical decision-making, and ultimately optimize pain management to support more tailored postoperative care strategies.