Classifying High-Risk Patients for Persistent Opioid Use After Major Spine Surgery: A Machine-Learning Approach
Authors' conclusion
The balanced random forest classifier was found to be the most effective model for identifying persistent opioid use after spine surgery.
Study context
- Population
- Patients undergoing major spine surgeryincluded in the analysis
- Participants
- 2,611 patients
- Comparison
- Five machine-learning classifiers: logistic regression, random forest, neural network, balanced random forest, balanced bagging
- Outcome
- Persistent opioid use (patient-reported opioid use after 3 months postoperatively)
- Follow-up
- 3 months
Methodological detailsmachine learning, training and test set split, SMOTE class balancing, SHAP feature importance
- Machine learning
- Training and test set split
- SMOTE class balancing
- SHAP feature importance
Performance
As reported in the abstract
Prediction of persistent opioid use (test set)
| Test or model | AUC | Sensitivity | Specificity | F1 | Precision | Accuracy |
|---|---|---|---|---|---|---|
| Balanced random forest | 0.877 (95% CI 0.834-0.894) | 0.700 | 0.883 | 0.758 (95% CI 0.677-0.839) | 0.836 | 0.780 |
| Balanced bagging | 0.859 (95% CI 0.814-0.905) | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract |
| Random forest | 0.855 (95% CI 0.813-0.897) | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract |
| Neural network | 0.729 (95% CI 0.672-0.787) | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract |
| Logistic regression | 0.709 (95% CI 0.652-0.767) | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract |
AUC
Balanced random forest0.877 (95% CI 0.834-0.894)
Balanced bagging0.859 (95% CI 0.814-0.905)
Random forest0.855 (95% CI 0.813-0.897)
Neural network0.729 (95% CI 0.672-0.787)
Logistic regression0.709 (95% CI 0.652-0.767)
Sensitivity
Balanced random forest0.700
Balanced bagging— not stated in the abstract
Random forest— not stated in the abstract
Neural network— not stated in the abstract
Logistic regression— not stated in the abstract
Specificity
Balanced random forest0.883
Balanced bagging— not stated in the abstract
Random forest— not stated in the abstract
Neural network— not stated in the abstract
Logistic regression— not stated in the abstract
F1
Balanced random forest0.758 (95% CI 0.677-0.839)
Balanced bagging— not stated in the abstract
Random forest— not stated in the abstract
Neural network— not stated in the abstract
Logistic regression— not stated in the abstract
Precision
Balanced random forest0.836
Balanced bagging— not stated in the abstract
Random forest— not stated in the abstract
Neural network— not stated in the abstract
Logistic regression— not stated in the abstract
Accuracy
Balanced random forest0.780
Balanced bagging— not stated in the abstract
Random forest— not stated in the abstract
Neural network— not stated in the abstract
Logistic regression— not stated in the abstract
— not stated in the abstract
Further findings
Incidence of persistent opioid use · 3 months
- Continued opioid use 3 months after surgery
- 1209 (46.3%)
Most important features (SHAP)
- Age, preoperative opioid use, preoperative pain scores and BMI had the highest impact
Original publication
Classifying High-Risk Patients for Persistent Opioid Use After Major Spine Surgery: A Machine-Learning Approach. Anesth Analg. Published 4 September 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →