Unmet Need
Delirium is a potentially preventable and often reversible disorder of impaired cognition. It is common in the intensive care unit (ICU), with an overall prevalence of 32.3%. However, the prevalence is dependent on the type of patients. For example, among ventilated burn patients, the prevalence is 77%, and among mechanically ventilated patients, the prevalence is 83%. Delirium in ICU patients is also associated with increased duration of mechanical ventilation, prolonged hospitalization, increased rate of self-extubation, and increased risk of mortality. Therefore, there is a need for a method that can detect delirium early, so that it can be managed properly in order to improve patient outcome.
Technology Overview
Johns Hopkins researchers have developed a series of deep learning based predictive models that can predict a patient’s risk of delirium at any point during their ICU stay and the patient’s risk of becoming delirious in the next 1, 3, 6, and 12 hours. These gradient boosting models were trained by pulling electronic health record data from a multi-center database of patient data gathered from ICUs. With this technology, patients with a high risk of being or becoming delirious can be properly assessed for delirium by the emergency room physicians and intervened with early, thus reducing the prevalence and associated negative effects of delirium in the intensive care unit.
Stage of Development
Retrospective validation has been completed on a large clinical dataset.
