Precision Opioid Prescription in ICU Surgery: Insights from an Interpretable Deep Learning Framework

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Pharmacology

Background:

• Opioid management is critical for ICU surgical patients to mitigate overdose risks and improve recovery.
• Accurate prediction of postoperative opioid requirements is essential for patient safety and effective treatment.
• Current machine learning models for opioid prediction lack interpretability, limiting clinical adoption.

Purpose of the Study:

• To develop an interpretable deep learning framework for evaluating feature impact on postoperative opioid use.
• To identify significant factors influencing opioid consumption in ICU surgical patients.
• To enhance clinical decision-making for precise opioid prescribing.

Main Methods:

• Developed an interpretable deep learning framework incorporating a Permutation Feature Importance Test (PermFIT).
• Evaluated Support Vector Machines, eXtreme Gradient Boosting, Random Forest, and Deep Neural Networks (DNN).
• Utilized Mean Squared Error (MSE) and Pearson Correlation Coefficient (PCC) for model performance assessment.

Main Results:

• The DNN model demonstrated superior performance with the lowest MSE (7889.2 mcg) and highest PCC (0.283) in 10-fold cross-validation.
• Analysis of 4,912 ICU surgical patients' electronic health records identified 13 significant predictors of opioid use (p < 0.05).
• Key predictors included age, surgery type, and other factors influencing postoperative opioid consumption.

Conclusions:

• The DNN model, combined with the PermFIT framework, effectively predicts postoperative opioid consumption and identifies significant influencing factors.
• This interpretable approach provides a valuable tool for tailoring opioid prescriptions to individual ICU surgical patient needs.
• The findings contribute to improved patient outcomes and enhanced safety through precise opioid management.