Prediction of ten intraoperative physiologic derangements using routine monitoring and drug administration data

Abstract Intraoperative hypotension, hypoxemia, and related derangements are associated with postoperative organ injury, yet existing predictors target a single event and require invasive or specialized monitoring. A temporal convolutional model using only flowsheet vital signs and the drug administration record was developed on 3,724 noncardiac surgical cases from the open VitalDB registry and evaluated, frozen, on 9,720 cases from an independent academic medical center. Ten hemodynamic, respiratory, and thermal events were predicted at 3-, 5-, and 10-min horizons. Mean area under the receiver operating characteristic curve at 5 min, restricted to windows in which the patient was not already breaching the threshold, was 0.887 internally and 0.843 externally. At two alarms per case-hour, sensitivity was 0.61 and positive predictive value 0.27 internally, with a median lead time of 3.8 min that was unchanged externally. Alarm precision was modest and unusable externally for hypoxemia, supporting feasibility rather than clinical readiness.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-12
DOI
https://doi.org/10.21203/rs.3.rs-11446566/v1
Primary Topic
Healthcare Technology and Patient Monitoring
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Prediction of ten intraoperative physiologic derangements using routine monitoring and drug administration data

Jonathan W. Pillow, A Kazemi, Mitchell Ng, Jake Stephen et al.
Research Square
Healthcare Technology and Patient Monitoring
preprint

Prediction of ten intraoperative physiologic derangements using routine monitoring and drug administration data

Jonathan W. Pillow, A Kazemi, Mitchell Ng, Jake Stephen, Kenneth Ng
preprint en

Abstract

Abstract Intraoperative hypotension, hypoxemia, and related derangements are associated with postoperative organ injury, yet existing predictors target a single event and require invasive or specialized monitoring. A temporal convolutional model using only flowsheet vital signs and the drug administration record was developed on 3,724 noncardiac surgical cases from the open VitalDB registry and evaluated, frozen, on 9,720 cases from an independent academic medical center. Ten hemodynamic, respiratory, and thermal events were predicted at 3-, 5-, and 10-min horizons. Mean area under the receiver operating characteristic curve at 5 min, restricted to windows in which the patient was not already breaching the threshold, was 0.887 internally and 0.843 externally. At two alarms per case-hour, sensitivity was 0.61 and positive predictive value 0.27 internally, with a median lead time of 3.8 min that was unchanged externally. Alarm precision was modest and unusable externally for hypoxemia, supporting feasibility rather than clinical readiness.

Research Square
Princeton University (US), Thomas Jefferson University Hospital (US), Hackensack Meridian Health (US)
Good health and well-being
Healthcare Technology and Patient Monitoring
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon