Perioperative medicine & outcomes
Effect of machine learning models on clinician prediction of postoperative complications: the Perioperative ORACLE randomised clinical trial
Authors' conclusion
Clinician performance was not improved by machine learning assistance.
Study context
- Population
- Surgical patients aged ≥18 yr; predictions by anaesthesiologists and nurse anaesthetists providing remote telemedicine support89 clinicians
- Participants
- 5,071 patients
- Comparison
- Clinician risk prediction with vs without machine learning predictions
- Primary endpoint
- AUROC for clinician predictions of 30-day all-cause mortality and AKI within 7 days, assisted vs unassisted
Methodological detailssingle-centre, prospective
- Single-centre
- Prospective
Primary result
As reported in the abstract
Assisted
0.793
Unassisted
0.780
Assisted vs UnassistedEstimate (95% CI)P value
AUROC for death
mean difference 0.013(−0.070 to 0.097)
P = 0.76AUROC for AKI
difference 0.046(−0.003 to 0.091)
P = 0.06Further findings
Weighted kappa
| Outcome | Assisted | Unassisted | Mean difference (95% CI) | |
|---|---|---|---|---|
| Agreement of clinician predictions with models (weighted kappa), death | 0.75 | 0.62 | 0.13 (0.10–0.17) | |
| Agreement of clinician predictions with models (weighted kappa), AKI | 0.79 | 0.54 | 0.25 (0.21–0.29) | |
| −0.200.20.4Difference (linear scale) |
Observed outcomes (all patients)
- Died within 30 days of surgery
- 98 (2.2%)
- Sustained acute kidney injury (AKI)
- 450 (11.1%)
Original publication
Effect of machine learning models on clinician prediction of postoperative complications: the Perioperative ORACLE randomised clinical trial. Br J Anaesth. Published 10 September 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →