Patient Safety and Quality
Background: Rapid response and code team activations (RR/CTA) happen on a daily basis in hospitals. The number of these sentinel events occurring on admitted emergency department (ED) patients is unknown. More importantly if these activations occur within 24 hours of admission, it raises the question as to whether there were identifiable indicators of physiological decline that may have been overlooked. The study objective was to identify early patient- and system-level factors observed in admitted ED patients requiring a RR/CTA, and to predict time-to-event.
Methods: Data was collected retrospectively for ED patients admitted to the hospital who experienced a RR/CTA during hospitalization. Study setting was a single large urban academic institution. Data included demographic, physiological (chief complaint [CC], admitting diagnosis, vital signs [VS], past medical history [PMH]) and system-level components (arrival day, time, length of stay, staff shift times). Analysis included descriptive and machine learning algorithms such as Decision and Regression Trees. Model validation was conducted by splitting the data between training set (75% of the observations were randomly selected, n=185 patients) and validating the models on testing set (25% of the data, n= 62). Outcomes of interest were twofold: binary categorization of patients depending on time of event (within or after 24 hours), and time to event.
Results: A total of 14,254 ED patients were admitted in 2017. 247 (1.7%) of these patients met the inclusion criteria. Of these 62 had their sentinel event within 24 hours of admission. Patient factors such as PMH, CC and VS distributions did not show significant association with categorization of or time to sentinel event. 68% of sentinel events and 67% of deaths occurred in either the spring or fall, respectively. Lastly, shift time and week day were identified as indicators associated with higher probability of sentinel event occurrence.
Conclusion: Patient factors did not show significant association with categorization of or time to sentinel event. Instead, temporal factors, such as time of the day, months, and system level factor, such as shift times, showed significant potential for predicting occurrence of and time to events. These indicators provide insight into developing ED-based decision support systems for prediction of deterioration.