Hemodynamics, fluids & transfusion
The effect of selection bias on the performance of a deep learning-based intraoperative hypotension prediction model using real-world samples from a publicly available database
Authors' conclusion
Both the DL model and the MAP-only model demonstrated worse predictive performance when tested on the unbiased dataset compared with the biased dataset. Although the DL model statistically performed better than the MAP-only model, the difference between the two models was not clinically meaningful.
Study context
- Population
- Surgical cases in the VitalDB open dataset; hypotensive event = mean arterial pressure (MAP) <65 mm Hg for 1 min
- Comparison
- Unbiased vs biased testing dataset; Deep learning (DL) model vs MAP-only model
- Outcome
- Alarms per hour and PPV of intraoperative hypotension prediction (DL vs MAP-only models)
Methodological detailsdeep learning, publicly available database (VitalDB)
- Deep learning
- Publicly available database (VitalDB)
Performance
As reported in the abstract
DL model, prediction 5 min before the event · 5 min before the event
| Test or model | Value | PPV |
|---|---|---|
| Unbiased dataset | 18.1 alarms per hour | 0.068 |
| Biased dataset | 10.8 alarms per hour | 0.937 |
Value
Unbiased dataset18.1 alarms per hour
Biased dataset10.8 alarms per hour
PPV
Unbiased dataset0.068
Biased dataset0.937
- Unbiased dataset vs Biased dataset: P < 0.001
- Unbiased dataset vs Biased dataset: P < 0.001
Further findings
DL vs MAP-only model performance
- DL model generally performed better than the MAP-only model
Original publication
The effect of selection bias on the performance of a deep learning-based intraoperative hypotension prediction model using real-world samples from a publicly available database. Br J Anaesth. Published 22 May 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →