Hemodynamics, fluids & transfusion
Methodologic Insights on Building and Evaluating Models for Early Warning of Hypotension during Surgery
Authors' conclusion
Data selection is critical when building and evaluating hypotension prediction models.
Study context
- Population
- Patients undergoing noncardiac and cardiac surgery
- Exposure
- Models trained and tested with vs without segments containing ongoing hypotension or classification-altering interventions
- Outcome
- Hypotension prediction model performance (AUPRC, AUC, dedicated metric)
Methodological detailsretrospective datasets, forward framing, bootstrap comparison
- Retrospective datasets
- Forward framing
- Bootstrap comparison
Main finding
As reported in the abstract
AUPRC by data selection
Mean difference (95% CI)P value
Training without classification-altering interventions vs Training with classification-altering interventionsbootstrap P
0.01(0.007–0.012)
P < 0.01Training without ongoing hypotension vs Training with ongoing hypotension
0.000(−0.003 to 0.004)
- Cardiac set: filtering classification-altering interventions increased area under the precision recall curve (AUPRC) on average
- 15.5% (0.54; 95% CI, 0.53 to 0.55 vs 0.47; 95% CI, 0.45 to 0.48)
- Cardiac set: including ongoing hypotension in evaluation data increased AUPRC on average
- 72.2% (0.47; 95% CI, 0.45 to 0.48 vs 0.80; 95% CI, 0.76 to 0.84)
Observational data: associations, not effects.
Original publication
Methodologic Insights on Building and Evaluating Models for Early Warning of Hypotension during Surgery. Anesthesiology. Published 30 March 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →