Physician Experience Design (PXD): More Usable Machine Learning Prediction for Clinical Decision Making

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Neuroscience

Background:

• Delirium is a challenging acute neurocognitive disorder to identify and predict.
• Accurate delirium detection is crucial for patient outcomes and hospital resource management.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for delirium identification.
• To create an Explainable Artificial Intelligence (XAI) framework using physician experience design (PXD) to enhance ML model interpretability and clinical adoption.

Main Methods:

• Utilized a labeled dataset of approximately 4,000 cases from Canada's GEMINI hospital data study.
• Developed ML models for delirium prediction and employed a participatory design process with physicians.
• Created a PXD framework incorporating conceptual investigation and an interactive dashboard for ML result visualization.

Main Results:

• Successfully developed ML models for delirium identification using a large, labeled hospital dataset.
• The PXD approach, including the interactive dashboard, enhanced the transparency of ML model predictions.
• Physician feedback and interaction were integral to selecting a preferred ML model for clinical decision-making.

Conclusions:

• Explainable AI integrated with physician experience design can significantly improve the trust and uptake of ML models in clinical practice.
• This approach facilitates better clinical decision-making for acute neurocognitive disorders like delirium.
• The study demonstrates a viable pathway for integrating advanced AI tools into routine healthcare settings.