Team Blue Jays

Delirium is an acute onset of global brain dysfunction. It is extremely prevalent in hospitals, particularly in ICUs, affecting up to 80% of critically ill patients and costing the American health care system $164 billion annually. Delirium is independently associated with poor health outcomes in patients and is often under-recognized and misdiagnosed. As 30-40% of delirium cases are considered preventable, a high performing prediction model would allow early intervention strategies to reduce patient risks of delirium. Additionally, given the variability in root cause and clinical trajectory, clustering delirium into endotypes can provide further insight into treatment options based on the underlying pathology.

We have developed a prediction model that uses the first 24 hours of data to estimate the risk of delirium during the ICU stay. Our preliminary model outperforms PRE-DELIRIC, the current gold-standard delirium prediction algorithm. We are extending this to develop a model with a real-time risk score for delirium. We are also using clustering to analyze the outcomes and features of patients, which may yield a better understanding of delirium physiology and trajectory.