This study will test whether anesthesiology clinicians working in a telemedicine setting can predict patient risk for postoperative complications (death and acute kidney injury) more accurately with access to a machine learning display than without it.
The Perioperative Outcome Risk Assessment with Computer Learning Enhancement (Periop ORACLE) study will be a sub-study nested within the ongoing TECTONICS trial (NCT03923699). TECTONICS is a single-center randomized clinical trial assessing the impact of an anesthesiology control tower (ACT) on postoperative 30-day mortality, delirium, respiratory failure, and acute kidney injury. As part of the TECTONICS trial, investigators in the ACT perform medical record case reviews during the early part of surgery and document how likely they feel each patient is to experience postoperative death and acute kidney injury (AKI). In Periop ORACLE, these case reviews will be randomized to be performed with or without access to machine learning (ML) predictions. Investigators in the ACT will conduct all case reviews by viewing the patient's records in AlertWatch (AlertWatch, Ann Arbor, MI) and Epic (Epic, Verona, WI). AlertWatch is an FDA-approved patient monitoring system designed for use in the operating room. The version of AlertWatch used in this study has been customized for use in a telemedicine setting. Epic is the electronic health record system utilized at Barnes-Jewish Hospital. Each case review will be randomized in a 1:1 fashion to be completed with or without ML assistance. If the case review is randomized to ML assistance, the investigator will access a display interface (currently deployed as a web application on a secure server) that shows real-time ML predicted likelihood for postoperative death and postoperative AKI. If the case review is not randomized to ML assistance, the investigator will not access this display. After viewing the patient's data, the investigator will predict how likely the patient is to experience postoperative death and postoperative AKI and will document this prediction. The area under the receiver operating characteristic curves for predictions made with ML assistance and without ML assistance will be compared.
Study Type
INTERVENTIONAL
Allocation
RANDOMIZED
Purpose
SCREENING
Masking
SINGLE
Enrollment
5,114
The machine learning display uses data from the electronic health record to predict the likelihood of postoperative death and postoperative acute kidney injury.
Washington University School of Medicine
St Louis, Missouri, United States
Area under receiver-operating characteristic curve of clinician prediction for postoperative death
Clinicians will predict the likelihood of postoperative death for each case using a categorical scale. A logistic regression will be constructed using the clinician predictions as inputs, and the area under the receiver-operating characteristic curve will be determined.
Time frame: 30 days
Area under receiver-operating characteristic curve of clinician prediction for postoperative acute kidney injury
Clinicians will predict the likelihood of postoperative acute kidney injury for each case using a categorical scale. A logistic regression will be constructed using the clinician predictions as inputs, and the area under the receiver-operating characteristic curve will be determined.
Time frame: 7 days
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.