Hemodynamics, fluids & transfusion
Early prediction of shock in intensive care unit patients by machine learning using discrete electronic health record data
Authors' conclusion
Machine learning prediction models can achieve very good discrimination and accuracy for new-onset shock in ICU patients using vital sign data within 4 h after ICU admission.
Study context
- Population
- Adult cardiac, medical, neurological and surgical intensive care unit (ICU) patients without shock within 4 h of ICU admission; routine vital sign data from first 4 hvalidation cohort n = 2826
- Participants
- 11,305 participants
- Comparison
- Logistic regression vs ML models
- Outcome
- Prediction of new-onset shock
- Follow-up
- Within the subsequent 4 h
Methodological detailsmachine learning, validation cohort
- Machine learning
- Validation cohort
Performance
As reported in the abstract
Discrimination for new-onset shock (validation cohort)
| Test or model | Value | NPV | Accuracy | PPV |
|---|---|---|---|---|
| Elastic net | 0.76 | ≥ 98 % | ||
| Boosted tree | 0.76 | ≥ 98 % | ||
| Random forest | 0.79 | ≥ 98 % | ||
| XGB | 0.82 | ≥ 98 % | 81 % | 14 % |
Value
Elastic net0.76
Boosted tree0.76
Random forest0.79
XGB0.82
NPV
Elastic net≥ 98 %
Boosted tree≥ 98 %
Random forest≥ 98 %
XGB≥ 98 %
Accuracy
Elastic net— not stated in the abstract
Boosted tree— not stated in the abstract
Random forest— not stated in the abstract
XGB81 %
PPV
Elastic net— not stated in the abstract
Boosted tree— not stated in the abstract
Random forest— not stated in the abstract
XGB14 %
Further findings
Incidence of new-onset shock after 4 h
- New-onset shock
- 483 (4.3 %)
Mortality, new-onset shock vs no shock
- ICU mortality
- 8.5 % vs. 1.9 %
- In-hospital mortality
- 14.3 % vs. 5.0 %
In-hospital mortality prediction (XGB)
- XGB predicted in-hospital mortality with good discrimination
Original publication
Early prediction of shock in intensive care unit patients by machine learning using discrete electronic health record data. J Crit Care. Published 22 April 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →