Predicting opioid consumption after surgical discharge: a multinational derivation and validation study using a...

Area of Science:

• Medical informatics
• Artificial intelligence in healthcare
• Pain management and pharmacology

Background:

• Opioid overprescription post-surgery is a significant public health concern.
• Accurate prediction of post-discharge opioid consumption is crucial for optimizing pain management and reducing misuse.
• Existing methods for risk stratification may not fully capture the complexity of opioid use patterns.

Purpose of the Study:

• To develop and validate a tabular foundation model for predicting the risk of post-discharge opioid consumption in surgical patients.
• To assess the model's performance in both internal and external validation cohorts.
• To estimate the potential impact of implementing such a predictive model on global opioid prescribing rates.

Main Methods:

• A tabular foundation model was trained and validated on a large surgical cohort (n=4267) from the 'Opioid PrEscRiptions and usage After Surgery' study.
• Internal validation used an 80:20 training/test split, while external validation involved a separate cohort of general surgery patients (n=826).
• Model performance was evaluated using the area under the receiver operator curve (AUC) and Brier scores.

Main Results:

• The model achieved an AUC of 0.84 (95% CI 0.81-0.88) in internal testing and 0.77 (95% CI 0.74-0.80) in external validation.
• Brier scores were 0.13 (95% CI 0.12-0.14) internally and 0.19 (95% CI 0.17-0.2) externally, indicating good predictive accuracy.
• Patients with <50% predicted risk consumed a median of 0 oral morphine equivalents in the first week post-surgery.
• Global opioid prescriptions could be reduced by 4.5% without increasing severe pain, according to counterfactual modeling.

Conclusions:

• A tabular foundation model effectively predicts post-discharge opioid consumption risk in surgical patients.
• The model demonstrates robust performance across internal and external validation datasets.
• Implementation of this AI tool holds significant potential for reducing global opioid prescriptions and mitigating opioid-related harms while maintaining adequate pain control.