Opal: an implementation science tool for machine learning clinical decision support in anesthesia.

Area of Science:

• Anesthesiology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Leveraging machine learning (ML) for clinical decision support in anesthesia is challenging.
• Existing platforms lack comprehensive infrastructure for implementation science in this domain.
• There is a need for tools that streamline ML model development and deployment in the perioperative setting.

Purpose of the Study:

• To introduce Opal, a full-stack platform infrastructure for ML in anesthesia.
• To demonstrate Opal's capability in predicting post-operative acute kidney injury (AKI).
• To showcase Opal's utility for unsupervised learning tasks, such as patient clustering.

Main Methods:

• Opal platform used for data extraction from 29,004 operating room (OR) cases for AKI prediction.
• Pre-operative data including demographics, medical history, medications, and flowsheet information were utilized.
• Unsupervised learning (PCA, k-means clustering) applied to intra-operative flowsheet data from 2995 OR cases.
• A gradient boosting machine model was developed and evaluated using an 80/20 train-test split.

Main Results:

• The gradient boosting model achieved an ROC-AUC of 0.85 (95% CI [0.80-0.90]) for AKI prediction.
• At a 0.5 probability threshold, the model demonstrated 0.9 sensitivity and 0.8 specificity.
• K-means clustering identified two patient clusters, facilitating hypothesis generation for intra-operative outcomes.

Conclusions:

• Opal provides a streamlined ML infrastructure for perioperative researchers and clinicians.
• The platform facilitates ML implementation for clinical decision support and predictive analytics.
• Opal enables future applications in data mining, clinical simulation, high-frequency prediction, and quality improvement.