Outlier Detection for Clinical Monitoring and Alerting
In this talk, I present a clinical monitoring and alerting framework that aims to identify unusual patient management actions in electronic health record data. Our hypothesis is that patient-management actions that are unusual with respect to past patients may be due to medical errors and that it is worthwhile to raise an alert if such a condition is encountered prospectively. The current version of the framework relies on a collection machine learning models built for the different actions from past EHR data and supports alerting in real-time. To demonstrate the promise of our framework I present (a) results of an evaluation study in which the quality of alerts raised by our framework were evaluated retrospectively by a panel clinicians, and (b) preliminary results of the new (ongoing) study where alerts are raised in real-time.