Projects

• ICareFM

ICareFM is a foundation model for clinical time series from intensive and emergency care. Together with ETH Zurich and Charité, we harmonized a large Charité cohort comprising 250,000 ICU stays and used it for external validation of the model.

Preprint

• PREEVENT-ICU

In the PREEVENT-ICU project, we develop and evaluate models for predicting clinically relevant events from routinely collected intensive care data. A particular focus is on external validation and clinically realistic evaluation. In a study on sepsis prediction, we showed that both transfer to other hospitals and the choice of evaluation strategy can substantially affect measured model performance.

Publication

• Nephrology AI

In the Nephrology AI project, we develop AI-based methods for predicting contrast-associated acute kidney injury and investigate their integration into clinical decision-making.

Selected Publications

• Burger M, Chopard D, Lichtner G, et al. A Foundation Model for Intensive Care: Unlocking Generalization across Tasks and Domains at Scale. medRxiv. 2026. Preprint. https://doi.org/10.1101/2025.07.25.25331635 
• Do DK, Rockenschaub P, Boie SD, et al. The Impact of Evaluation Strategy on Sepsis Prediction Model Performance Metrics in Intensive Care Data: Retrospective Cohort Study. Journal of Medical Internet Research. 2026;28:e72083. DOI
• Lichtner G, Balzer F, Haufe S, et al. Predicting lethal courses in critically ill COVID-19 patients using a machine learning model trained on patients with non-COVID-19 viral pneumonia. Scientific Reports. 2021;11:13205. doi.org/10.1038/s41598-021-92475-7

Contacts

• Dr. rer. nat. Gregor Lichtner
• Prof. Dr. med. Falk von Dincklage