Related Experiment Videos
Development and external validation of a multimodal artificial intelligence mortality prediction model of critically
Behrooz Mamandipoor1,2, Chun-Nan Hsu2, Martin Krause2,3
1Department of Biomedical Informatics, University of California, San Diego Health, La Jolla, CA, USA.
Anesthesiology
|August 4, 2026
Summary
A new multimodal deep learning model accurately predicts in-hospital mortality in critically ill patients. Integrating diverse clinical data sources, including notes and imaging, improves prediction accuracy for better patient care.
Area of Science:
- Critical Care Medicine
- Artificial Intelligence in Healthcare
- Clinical Informatics
Background:
- Early prediction of in-hospital mortality is crucial for optimizing critical care.
- Developing accurate predictive models aids clinicians in treatment decisions.
Purpose of the Study:
- To develop a multimodal deep learning model for predicting in-hospital mortality risk in critically ill patients.
- To utilize structured and unstructured clinical data within the first 24 hours of intensive care unit (ICU) admission.
Main Methods:
- Utilized data from MIMIC-III, MIMIC-IV, eICU, and HiRID datasets.
- Developed a multimodal model incorporating time-series variables, clinical notes, and chest X-ray images.
- Externally validated the model on independent datasets and multiple institutions.
Main Results:
- The model achieved an AUROC of 0.92 and AUPRC of 0.53 using structured data.
- External validation demonstrated robust performance with AUROCs ranging from 0.84-0.92.
- Incorporating clinical notes and imaging data significantly improved prediction performance (AUROC increased from 0.87 to 0.89).
Conclusions:
- Multimodal data integration is vital for accurate in-hospital mortality prediction.
- External validation confirms the generalizability and reliability of the developed model.
- This approach can enhance clinical decision-making for critically ill patients.