Back

From Code to Critical Care Time: Implementing an AI-Driven ICU Length-of-Stay Clinical Decision Support System Under European Governance Constraints

Althammer, A.; Hummel, A.; Steghoefer, J.-P.; Reichel, F.; Kolonko, J.; Hartfield, S.; Fischer, M.; Schloegl-Flierl, K.; Ziethmann, P.; Weiss, M.; Simon, P.; Moegerlein, M.; Mamtschur, E.; Spring, O.; Shmygalev, S.; Ortmann, N.; Raffler, J.; Hinske, L. C.; Brunner, J. O.; Heller, A. R.; Bartenschlager, C.

2026-02-05 intensive care and critical care medicine
10.64898/2026.02.04.26345355 medRxiv
Show abstract

BackgroundML models in critical care often perform well retrospectively but deliver limited bedside value once deployed under European governance constraints. ObjectiveEvaluate feasibility and early sociotechnical lessons from an offline Clinical Decission Support System (CDSS) predicting ICU LoS on a surgical ICU in practice MethodsWe conducted a prospective implementer study. Residents used AI and recorded estimates (n=136), consultants provided blinded estimates (n=162), AI outputs were logged (n=221). Version 1 showed LoS prediction, Version 2 updated the model and added a compact feature importance panel by using TreeSHAP. Human factors were assessed with Psychological Assessment of AI-based Decision Support (PAAI) and an embedded ethicist informed design and onboarding. Ethics Projekt.Nr 24-0336-KB, registry: DRKS00037851. ResultsOffline deployment was feasible but caused coordination burden. Version 2 reduced MAE for AI (5.80[-&gt;]4.92) and resident+AI estimates (6.18[-&gt;]3.84; both p<0.05). PAAI identified user groups. ConclusionsEarly embedding exposed governance-driven costs and highlighted iterative upgrading to address the translation gap.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.