ICU sedation, delirium & recovery
Improving ICU Risk Predictive Models Through Automation Designed for Resiliency Against Documentation Bias
Authors' conclusion
These mortality risk models demonstrated excellent performance compared with APACHE while appearing to mitigate bias introduced through major shifts in GCS documentation at two large health systems.
Study context
- Population
- Adult ICU stays discharged from hospital 2017–2018, eICU Research Institute database (Philips eICU telecritical care program)ICU stays in final cohort
- Participants
- 509,586 stays
- Exposure
- Critical Care Outcome Prediction Model (CCOPM) vs Acute Physiology and Chronic Health Evaluation (APACHE) IVa/IVb
- Outcome
- Model discrimination (AUROC) for ICU and hospital mortality on the validation cohort
Methodological detailsprediction model development and validation, multicentre, 3:2 random split into development and validation cohorts, generalized additive models
- Prediction model development and validation
- Multicentre
- 3:2 random split into development and validation cohorts
- Generalized additive models
Main finding
As reported in the abstract
ICU mortality discrimination
| Group | Value | Area under the receiver operating characteristic (AUROC) |
|---|---|---|
| CCOPM | 0.925 | 0.925 |
| APACHE IVa/IVb | 0.88 | 0.88 |
Value
CCOPM0.925
APACHE IVa/IVb0.88
Area under the receiver operating characteristic (AUROC)
CCOPM0.925
APACHE IVa/IVb0.88
Observational data: associations, not effects.
Further findings
| Outcome | CCOPM | APACHE IVa/IVb |
|---|---|---|
| Hospital mortality discrimination | ||
| AUROC | 0.90 | 0.86 |
Calibration by admission diagnosis, ICU type and ICU-year (subgroups)
- Better calibration with CCOPM
Prediction stability with changes in Glasgow Coma Scale (GCS) documentation (external cohort, 2 health systems)
- More stable predictions with CCOPM vs APACHE IVa
- External cohort of greater than 120,000 patients
Original publication
Improving ICU Risk Predictive Models Through Automation Designed for Resiliency Against Documentation Bias. Crit Care Med. Published 27 December 2022. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →