A Factored Generalized Additive Model for Clinical Decision Support in the Operating Room

Area of Science:

• Medical Informatics
• Machine Learning
• Clinical Prediction

Background:

• Logistic regression (LR) is a common but limited tool for clinical prediction due to its linear nature.
• Generalized additive models (GAMs) offer improvements but may lack feature interaction capabilities.
• Existing models often struggle to balance predictive power with interpretability.

Purpose of the Study:

• To introduce the Factored Generalized Additive Model (F-GAM) for enhanced clinical prediction.
• To evaluate F-GAM's performance against established models in predicting postoperative complications.
• To assess F-GAM's ability to maintain model interpretability for targeted features.

Main Methods:

• Development of the Factored Generalized Additive Model (F-GAM).
• Evaluation of F-GAM on predicting acute kidney injury and acute respiratory failure using a single-center database.
• Comparative analysis against logistic regression, other GAMs, random forests, support vector machines, and deep neural networks.

Main Results:

• F-GAM demonstrated superior performance in predicting postoperative acute kidney injury and acute respiratory failure.
• The model achieved higher AUPRC and AUROC scores compared to all benchmark models.
• F-GAM maintained good interpretability for targeted features, yielding results with high face validity.

Conclusions:

• F-GAM represents a significant advancement in clinical prediction models.
• The model effectively balances high predictive accuracy with essential interpretability.
• F-GAM shows promise for improving patient outcomes through more accurate risk stratification.