Related Experiment Videos
Identifying subgroups of ICU patients with high mortality rates using machine learning: A nationwide,
Julian van Gemert1, Mark van den Boogaard2, Cornelia Hoedemaekers1
1Radboud University Medical Center, Department of Intensive Care Medicine, Nijmegen, the Netherlands.
Journal of Critical Care
|June 27, 2026
Summary
Machine learning identified intensive care unit (ICU) patient subgroups with high mortality. Impaired consciousness and reduced urine output were key indicators in these high-risk groups.
Area of Science:
- Critical Care Medicine
- Machine Learning Applications
- Health Services Research
Background:
- Identifying high-risk patient subgroups in intensive care units (ICUs) is crucial for targeted medical interventions and resource allocation.
- Understanding the characteristics of patients with elevated mortality rates can inform policy decisions in intensive care medicine.
Purpose of the Study:
- To identify demographic and clinical characteristics of intensive care unit (ICU) patient subgroups exhibiting high (≥80%) mortality rates six months post-admission.
- To leverage machine learning for the discovery of distinct patient profiles associated with critical illness outcomes.
Main Methods:
- Utilized data from the Dutch National Intensive Care Evaluation (NICE) registry, encompassing 807,727 ICU admissions across 84 hospitals (2013-2023).
- Trained a machine learning model on 70% of hospital data (2013-2022) and validated it on the remaining 30%, with temporal validation using 2023 data.
- Employed interpretable machine learning techniques to analyze patient data at different stages of ICU admission.
Main Results:
- Identified ten distinct subgroups of ICU patients with mortality rates exceeding 80%.
- Key defining factors for high-mortality subgroups included reduced urine output and low Glasgow Coma Scale (GCS) scores (eye and motor components).
- External validation demonstrated high accuracy, with minor deviations in predicted mortality (median absolute deviations of -1% and -2%).
Conclusions:
- Interpretable machine learning effectively identifies high-mortality ICU patient subgroups (≥80% 6-month mortality) using routinely collected clinical data.
- Impaired consciousness and reduced urine output are significant indicators characterizing these vulnerable patient groups.
- Findings support the integration of these insights into patient-centered care frameworks to optimize intensive care management and ethical decision-making.