The rate of false alarms associated with continuous vital signs monitoring creates the condition that leads to alarm fatigue. Alarm fatigue increases the potential to miss critical situations in hospitalized patients. In pediatric acute care patients, the risk is the greater due to the reliance on an alerting model that relies on threshold alarms that are generated when a patient's individual vital sign exceeds an upper or lower limit. The hypothesis in this research is that trends exist in specific features of vital sign behavior prior to a clinically significant event (CSE) such as a code blue event. The significance of this research aims to illustrate that using an individual's prior vital sign data for alerting is more effective at identifying a CSE in advance versus a reliance on alerts triggered by threshold alarms. The first aim is to show that the mean and standard deviation of individual vital signs vary significantly in the one hour prior to a code blue. The second aim is to show that a more accurate model that uses the trending information can increase accuracy in identifying code blue events in the one hour prior to the event. The vital signs that were assessed included the continuous monitoring of heart rate, respiratory rate, and pulse oximetry. The CSE was a code blue. A retrospective analysis of two sets of patients is used for both studies. The first data set includes patients who did not experience a code blue (control group) and the second are patients who experienced a CSE.

In Aim 1, six hours of vital signs for heart rate, respiratory rate, and pulse oximetry were obtained from records of de-identified patients approved by the Children's IRB. All study patients were housed in a pediatric acute care ward for a respiratory related illness. The data was analyzed for trends in mean and standard deviation. The results indicate that there is a significant (p<.05) difference in the mean and standard deviations of all three vital signs in the final hour for the studies who experienced a CSE compared to prior periods of time and control data.

In aim 2, the features identified related to the change in mean values in the final hour prior to a CSE were used to create test algorithms to predict the event in that final one-hour period prior to the CSE. The same data set was used from study 1 to test six predictive models: measuring sensitivity and specificity test with a receiver operator curve (ROC) analysis. The models tested the impact of the change of mean values during a period in the one hour prior to the event and the mean value one-hour prior. The current model that relies on healthy ranges of vital signs across a patient age group was the least accurate model. The optimal algorithm of the results of all models relies on a change in the mean of the Heart rate in the final hour over a moving ten-minute window comparing against the mean in the previous hour. The algorithm was tested with a separate data set of similar size and patient diagnosis and predicted the code blue with 0.94 accuracy (0.93 sensitivity and 0.95 specificity).

This research proves the hypothesis of the presence of trends in vital signs features in the one hour prior to a CSE. In addition, the data adds value to the theory that using an individual's own vital sign data and focusing on the change in mean values compared to prior periods is an effective method of identifying a CSE in the final hour. Testing larger data sets across multiple patient populations focusing on different CSE will add validity to this research.