Perioperative medicine & outcomes
Utilising intraoperative respiratory dynamic features for developing and validating an explainable machine learning model for postoperative pulmonary complications
Authors' conclusion
These findings suggest that real-time identification of surgical patients' risk of postoperative pulmonary complications could help personalise intraoperative ventilatory strategies and reduce postoperative pulmonary complications.
Study context
- Population
- Surgical patients aged 65 yr and older at an academic medical centre, 2019 to 2023
- Participants
- 10,284 patients · 10,484 operations
- Comparison
- Machine learning models vs ARISCAT
- Outcome
- PPCs (prediction performance, AUROC)
Methodological detailsmachine learning, single academic medical centre, prospective validation cohort, internal testing cohort, Shapley additive explanation
- Machine learning
- Single academic medical centre
- Prospective validation cohort
- Internal testing cohort
- Shapley additive explanation
Performance
As reported in the abstract
PPC prediction (AUROC)
| Test or model | AUROC, validation cohort | AUROC, prospective cohort | AUROC |
|---|---|---|---|
| XGBoost | 0.878 (0.866-0.891) | 0.881 (0.879-0.883) | — not stated in the abstract |
| ARISCAT | — not stated in the abstract | — not stated in the abstract | 0.496-0.533 |
AUROC, validation cohort
XGBoost0.878 (0.866-0.891)
ARISCAT— not stated in the abstract
AUROC, prospective cohort
XGBoost0.881 (0.879-0.883)
ARISCAT— not stated in the abstract
AUROC
XGBoost— not stated in the abstract
ARISCAT0.496-0.533
— not stated in the abstract
Further findings
- Simplified 20-variable XGBoost model, internal testing cohort (AUROC)
- 0.864 (0.852-0.875) · AUROC, simplified 20-variable model
Key modifiable contributors to postoperative pulmonary complications (PPCs)
- Intraoperative respiratory dynamic system compliance, mechanical power and driving pressure identified as key modifiable contributors
Original publication
Utilising intraoperative respiratory dynamic features for developing and validating an explainable machine learning model for postoperative pulmonary complications. Br J Anaesth. Published 18 April 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →