Skip to content

ICU Foundation Model: Evaluation and further development of a universal intensive care data prediction model based on routine data from Charité – Universitätsmedizin Berlin

ICU Foundation Model: Evaluation and further development of a universal intensive care data prediction model based on routine data from Charité – Universitätsmedizin Berlin - ICareFM-Charite

Status
Recruiting
Phases
Unknown
Study type
Observational
Source
DRKS
Registry ID
DRKS00038236
Enrollment
220000
Registered
2026-02-25
Start date
2026-02-25
Completion date
Unknown
Last updated
2026-03-02

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

Intensive care patients

Interventions

Group 1: Retrospective analysis of adult intensive care patients (=18 years) who completed their intensive care stay during the study period. A secondary evaluation of routine intensive care data (e.g

Sponsors

Charité - Universitätsmedizin Berlin
Lead Sponsor

Eligibility

Sex/Gender
All
Age
18 Years to No maximum

Inclusion criteria

Inclusion criteria: - Intensive care patients - Completed intensive care stay during the study period

Exclusion criteria

Exclusion criteria: - Intensive care stay of less than 4 hours

Design outcomes

Primary

MeasureTime frame
The predictive performance (AUROC) of the model trained on Charité data in external validation on publicly available intensive care datasets for the occurrence of the following events relevant to intensive care medicine: • Sepsis • Circulatory failure (shock) • Renal failure • Hyperglycaemia • Severe liver dysfunction (high MELD score) • Respiratory insufficiency • Decompensation (e.g. in chronic disease) The prediction horizons – i.e. how far in advance of an event its occurrence is to be predicted – are based on clinically relevant time frames (e.g. 8–48 hours, depending on the event).

Secondary

MeasureTime frame
Performance of the prediction of the occurrence of the aforementioned intensive care-relevant endpoints, measured as AUROC, area under the precision-recall curve (AUPRC), Brier score, calibration curve, and other common performance measures of binary prediction such as sensitivity, specificity, precision/positive predictive value, negative predictive value, each at clinically relevant thresholds, as well as measures that quantify the timeliness of events.

Countries

Germany

Contacts

Public ContactFelix Balzer

Charité - Universitätsmedizin Berlin

icu-foundation-model@charite.de+49 30 450 570 425

Outcome results

None listed

Source: DRKS (via WHO ICTRP) · Data processed: Mar 14, 2026