SESSION TYPE: ICU: Improving Outcomes
PRESENTED ON: Sunday, October 21, 2012 at 10:30 AM - 11:45 AM
PURPOSE: Clinical deterioration of ward patients can result in cardiac arrest (CA) or intensive care unit (ICU) transfer. We developed and validated a prediction model using electronic health record (EHR) data to detect both these events simultaneously and compared it to the ViEWS, the best performing risk score from a recent comparison study.
METHODS: Patients hospitalized on the wards at our hospital from November 2008 until August 2011 were divided into three groups: those who suffered a ward CA, those transferred to the ICU without suffering a ward CA, and those discharged alive without experiencing either event. A person-time multinomial logistic regression model was used to predict CA and ICU transfer simultaneously utilizing vital sign, demographic, and laboratory data. The prediction model was compared to the ViEWS using the area under the receiver operating characteristic curve (AUC) and was validated by performing 100 repetitions of a 3-fold cross validation procedure.
RESULTS: A total of 56649 controls, 109 CA patients, and 2543 ICU transfers were included. The final model contained age, prior ICU admission, heart rate, diastolic blood pressure, respiratory rate, oxygen saturation, supplemental oxygen use, mental status, temperature, blood urea nitrogen, potassium, anion gap, hemoglobin, platelet count, and white blood cell count. The model more accurately detected CA (AUC 0.88 vs. 0.78; P<0.001) and ICU transfer (AUC 0.77 vs. 0.73; P<0.001) than the ViEWS, and accuracy was similar with cross-validation. At a specificity of 93%, our model had a higher sensitivity than the ViEWS for CA patients (65% vs. 41%) and could detect the event a median of 50 hours before it occurred.
CONCLUSIONS: We developed and validated a prediction tool for ward patients utilizing vital sign, demographic, and laboratory data to simultaneously predict the risk of CA and ICU transfer. Our model performed better than the ViEWS for detecting both outcomes.
CLINICAL IMPLICATIONS: Our risk score could be implemented in the EHR to alert caregivers with real-time information regarding patient deterioration.
DISCLOSURE: Dana Edelson: Grant monies (from industry related sources): Philips Healthcare research support, Consultant fee, speaker bureau, advisory committee, etc.: Philips Healthcare honoraria and consultation fee, Grant monies (from sources other than industry): National Heart, Lung, and Blood Institute (K23 HL097157-01)
The following authors have nothing to disclose: Matthew Churpek, Trevor Yuen, Seo Young Park, Robert Gibbons
No Product/Research Disclosure InformationUniversity of Chicago, Chicago, IL