Enhancing clinical decision support with physiological waveforms - A multimodal benchmark in emergency care

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Emergency Medicine Analytics

Background:

• AI algorithms can enhance emergency medicine decision-making.
• Multimodal data integration, including raw waveform signals, is underexplored in clinical decision support.

Purpose of the Study:

• To advance multimodal decision support in emergency care.
• To develop and benchmark AI models using diverse patient data.
• To predict patient discharge diagnoses and deterioration.

Main Methods:

• Utilized demographics, biometrics, vital signs, lab values, and ECG waveforms as model inputs.
• Developed models for predicting discharge diagnoses and patient deterioration.
• Established a benchmarking protocol for multimodal AI models.

Main Results:

• Diagnostic model achieved AUROC > 0.8 for 609/1428 conditions (cardiac and non-cardiac).
• Deterioration model achieved AUROC > 0.8 for 14/15 critical events (e.g., cardiac arrest, mortality).

Conclusions:

• Incorporating raw waveform data significantly improves AI decision support model performance.
• A novel, publicly available dataset and baseline models are provided.
• Foundation laid for measurable progress in AI for emergency care.