Development, validation, and feature extraction of a deep learning model predicting in-hospital mortality using Japan’s largest national ICU database: a validation framework for transparent clinical Artificial Intelligence (cAI) development
Authors' conclusion
Our DNN model consistently generated AUCs between 0.91-0.95 regardless of days since ICU discharge. The 20 most important variables to our DNN, also generated higher AUCs than traditional severity scores regardless of days since ICU discharge.
Study context
- Population
- ICU patients in Japan's largest ICU database
- Comparison
- Deep neural network (DNN) model vs traditional severity scores
- Outcome
- In-hospital mortality, including ICU and post-ICU mortality by days since ICU discharge
Methodological detailsdeep neural network, SHAP feature extraction, external validation, national ICU database
- Deep neural network
- SHAP feature extraction
- External validation
- National ICU database
Performance
As reported in the abstract
| Test or model | AUC, ICU mortality | AUC, in-hospital mortality | AUC range, throughout one year since ICU discharge |
|---|---|---|---|
| DNN | 0.94 [0.93-0.95] | 0.91 [0.90-0.92] | 0.91-0.95 |
DNN
AUC, ICU mortality0.94 [0.93-0.95]
AUC, in-hospital mortality0.91 [0.90-0.92]
AUC range, throughout one year since ICU discharge0.91-0.95
Further findings
External validation, top 20 variables
- Higher AUCs with the top 20 variables than traditional severity scores
Original publication
Development, validation, and feature extraction of a deep learning model predicting in-hospital mortality using Japan’s largest national ICU database: a validation framework for transparent clinical Artificial Intelligence (cAI) development. Anaesth Crit Care Pain Med. Published 24 October 2022. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →