TY - JOUR
T1 - Predicting blood transfusion using automated analysis of pulse oximetry signals and laboratory values
AU - Shackelford, Stacy
AU - Yang, Shiming
AU - Hu, Peter
AU - Miller, Catriona
AU - Anazodo, Amechi
AU - Galvagno, Samuel
AU - Wang, Yulei
AU - Hartsky, Lauren
AU - Fang, Raymond
AU - Mackenzie, Colin
AU - Barker, Steven
AU - Blenko, John
AU - Chang, Chein I.
AU - Chen, Hegang
AU - Dinardo, Theresa
AU - DuBose, Joseph
AU - Fouche, Yvette
AU - Goetz, Linda
AU - Grissom, Thomas
AU - Giustina, Victor
AU - Hagegeorge, George
AU - Herrera, Anthony
AU - Hess, John
AU - Imle, Cris
AU - Menaker, Jay
AU - Murdock, Karen
AU - Narayan, Mayur
AU - Oates, Tim
AU - Pasley, Jason
AU - Saccicchio, Sarah
AU - Scalea, Thomas
N1 - Publisher Copyright:
© 2015 Copyright Wolters Kluwer Health, Inc. All rights reserved.
PY - 2015
Y1 - 2015
N2 - BACKGROUND: Identification of hemorrhaging trauma patients and prediction of blood transfusion needs in near real time will expedite care of the critically injured. We hypothesized that automated analysis of pulse oximetry signals in combination with laboratory values and vital signs obtained at the time of triage would predict the need for blood transfusion with accuracy greater than that of triage vital signs or pulse oximetry analysis alone. METHODS: Continuous pulse oximetry signals were recorded for directly admitted trauma patients with abnormal prehospital shock index (heart rate [HR]/systolic blood pressure) of 0.62 or greater. Predictions of blood transfusion within 24 hours were compared using Delong's method for area under the receiver operating characteristic (AUROC) curves to determine the optimal combination of triage vital signs (prehospital HR + systolic blood pressure), pulse oximetry features (40 waveform features, O2saturation, HR), and laboratory values (hematocrit, electrolytes, bicarbonate, prothrombin time, international normalization ratio, lactate) in multivariate logistic regression models. RESULTS: We enrolled 1,191 patients; 339 were excluded because of incomplete data; 40 received blood within 3 hours; and 14 received massive transfusion. Triage vital signs predicted need for transfusion within 3 hours (AUROC, 0.59) and massive transfusion (AUROC, 0.70). Pulse oximetry for 15 minutes predicted transfusion more accurately than triage vital signs for both time frames (3-hour AUROC, 0.74; p = 0.004) (massive transfusion AUROC, 0.88;p < 0.001). An algorithm including triage vital signs, pulse oximetry features, and laboratory values improved accuracy of transfusion prediction (3-hour AUROC, 0.84; p < 0.001) (massive transfusion AUROC, 0.91; p < 0.001). CONCLUSION: Automated analysis of triage vital signs, 15 minutes of pulse oximetry signals, and laboratory values predicted use of blood transfusion during trauma resuscitation more accurately than triage vital signs or pulse oximetry analysis alone. Results suggest automated calculations from a noninvasive vital sign monitor interfaced with a point-of-care laboratory device may support clinical decisions by recognizing patients with hemorrhage sufficient to need transfusion. LEVEL OF EVIDENCE: Epidemiologic/prognostic study, level III.
AB - BACKGROUND: Identification of hemorrhaging trauma patients and prediction of blood transfusion needs in near real time will expedite care of the critically injured. We hypothesized that automated analysis of pulse oximetry signals in combination with laboratory values and vital signs obtained at the time of triage would predict the need for blood transfusion with accuracy greater than that of triage vital signs or pulse oximetry analysis alone. METHODS: Continuous pulse oximetry signals were recorded for directly admitted trauma patients with abnormal prehospital shock index (heart rate [HR]/systolic blood pressure) of 0.62 or greater. Predictions of blood transfusion within 24 hours were compared using Delong's method for area under the receiver operating characteristic (AUROC) curves to determine the optimal combination of triage vital signs (prehospital HR + systolic blood pressure), pulse oximetry features (40 waveform features, O2saturation, HR), and laboratory values (hematocrit, electrolytes, bicarbonate, prothrombin time, international normalization ratio, lactate) in multivariate logistic regression models. RESULTS: We enrolled 1,191 patients; 339 were excluded because of incomplete data; 40 received blood within 3 hours; and 14 received massive transfusion. Triage vital signs predicted need for transfusion within 3 hours (AUROC, 0.59) and massive transfusion (AUROC, 0.70). Pulse oximetry for 15 minutes predicted transfusion more accurately than triage vital signs for both time frames (3-hour AUROC, 0.74; p = 0.004) (massive transfusion AUROC, 0.88;p < 0.001). An algorithm including triage vital signs, pulse oximetry features, and laboratory values improved accuracy of transfusion prediction (3-hour AUROC, 0.84; p < 0.001) (massive transfusion AUROC, 0.91; p < 0.001). CONCLUSION: Automated analysis of triage vital signs, 15 minutes of pulse oximetry signals, and laboratory values predicted use of blood transfusion during trauma resuscitation more accurately than triage vital signs or pulse oximetry analysis alone. Results suggest automated calculations from a noninvasive vital sign monitor interfaced with a point-of-care laboratory device may support clinical decisions by recognizing patients with hemorrhage sufficient to need transfusion. LEVEL OF EVIDENCE: Epidemiologic/prognostic study, level III.
UR - http://www.scopus.com/inward/record.url?scp=84953345184&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84953345184&partnerID=8YFLogxK
U2 - 10.1097/TA.0000000000000738
DO - 10.1097/TA.0000000000000738
M3 - Article
C2 - 26406427
AN - SCOPUS:84953345184
SN - 2163-0755
VL - 79
SP - S175-S180
JO - Journal of Trauma and Acute Care Surgery
JF - Journal of Trauma and Acute Care Surgery
IS - 4
ER -