Perioperative medicine & outcomes
Prediction of Intensive Care Length of Stay for Surviving and Nonsurviving Patients Using Deep Learning
Authors' conclusion
This novel LOS model represents a step forward in achieving more equitable benchmarking across diverse ICU settings with varying risk profiles.
Study context
- Population
- ICU admissions, 329 ICUs in 194 US hospitals, eICU Research Institute database, 2017–2019628,815 patients
- Participants
- 669,876 admissions
- Comparison
- Critical Care Outcomes Prediction Model (CCOPM) length of stay (LOS) model vs Acute Physiology and Chronic Health Evaluation (APACHE) IVb
- Outcome
- Prediction of ICU and hospital LOS for surviving and nonsurviving patients (R2, concordance index, mean absolute error, calibration)
Methodological detailsdeep learning, competing risk time-to-event modelling, training, validation and test sets (2:1:1), multicentre registry
- Deep learning
- Competing risk time-to-event modelling
- Training, validation and test sets (2:1:1)
- Multicentre registry
Performance
As reported in the abstract
Model performance (R2), test set
- ICU LOS, individual level, survivors
- R2 = 0.29
- ICU LOS, individual level, nonsurvivors
- 0.23
- ICU LOS, ICU level, survivors
- R2 = 0.48
- ICU LOS, ICU level, nonsurvivors
- 0.23
- Hospital LOS, individual level, survivors
- R2 = 0.46
- Hospital LOS, individual level, nonsurvivors
- 0.52
- Hospital LOS, ICU level, survivors
- R2 = 0.71
- Hospital LOS, ICU level, nonsurvivors
- 0.64
Further findings
| Outcome | CCOPM | APACHE IVb |
|---|---|---|
| R2 vs APACHE IVb (test subset with APACHE IVb predictions) | ||
| R2 ICU LOS, survivors | 0.30 | 0.16 |
| R2 ICU LOS, nonsurvivors | 0.23 | 0 |
| R2 hospital LOS, survivors | R2 = 0.39 | 0.27 |
| R2 hospital LOS, nonsurvivors | 0.40 | 0 |
Original publication
Prediction of Intensive Care Length of Stay for Surviving and Nonsurviving Patients Using Deep Learning. Crit Care Med. Published 7 February 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →