In an Intensive Care Unit, nurses can be overwhelmed with alarms which may not be warranted. Reasons such as movement, disconnected sensors, or general interference can trigger what is known as false alarms (alarms that seem to require urgent care but really do not). These can cause alarm fatigue in nurses leading to desensitization and thus, nurses are more likely to react poorly to true alarms when they occur. We can mitigate this poor response to true alarms by reducing false alarms, resulting in minimizing the response time patients wait for care. Previous attempts to reduce false alarms have mainly focused on improving hardware, which not only comes with an expensive price tag but is also useless in cases where sensor disconnection is involved. Our application has the ability to reduce false alarms in more cases while avoiding the large cost of expensive hardware. To accomplish this, analysis is done on a patient’s signals to give a set of features that are used to describe a patient's vital signs. By integrating machine learning techniques, we are able to extract the correlation among multiple different collected signals in ICUs that give us better accuracy to differentiate between noise, patient movement or sensor disconnection and true alarms. Using the correlations obtained through machine learning in conjunction with health condition definitions we are able to distinguish between potential alarms and suppress more of the false alarms that cause alarm fatigue.
* The initial concept for this project provided by Dr. Fatemeh Afghah in the form of a capstone project proposal.
** Subset of ongoing research supported by the National Science Foundation.