FIRST-ICU: forecasting interventions and risk stratification in the ICU using graph neural network autoencoders
Nosa Aikodon, Ivan Olier, Brian W. Johnston, Ingeborg D. Welters, Gregory Y. H. Lip, Sandra Ortega-Martorell · 2026
Read FIRST-ICU: forecasting interventions and risk stratification in the ICU using graph neural network autoencoders by Nosa Aikodon, Ivan Olier, Brian W. Johnston, Ingeborg D. Welters, Gregory Y. H. Lip, Sandra Ortega-Martorell (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13601581.txt.
Description
Nosa Aikodon and five colleagues present FIRST-ICU in 2026, a deep learning system that forecasts several intensive care treatments together rather than one by one. Critically ill patients often need many interventions at once, and those needs interact: sedation, for instance, can lead to ventilation. Older prediction models treated each intervention as independent, missing those cascades. The new framework joins a graph neural network, a sequence model, and an attention module built for intervention interactions.
Training used the MIMIC-IV database with 23,926 patients, and testing on Amsterdam's database with 12,603 patients required no retraining. Accuracy was high across all seven predicted interventions, with area-under-curve scores above 0.98 from the temporal decoder. A mapping layer also separated patients into six clinically distinct phenotypes, giving staff an interpretable way to sort risk. The authors present the system as decision support that respects how ICU treatments cluster and trigger each other.
Catalog record
| CREATOR | Nosa Aikodon, Ivan Olier, Brian W. Johnston, Ingeborg D. Welters, Gregory Y. H. Lip, Sandra Ortega-Martorell |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 184689 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |