Making decisions in intensive care units (ICU) is a complex, demanding task, as patients’ conditions are rapidly changing. Clinicians typically make choices based on their understanding of the patient, their own experiences, or oftentimes both. Making decisions from data isn’t always the clearest path: clinical signals are irregularly sampled and often riddled with human error. In these high pressure environments, interventions can be ineffective, harmful, or costly. We’re looking to change this process by creating models that help predict when interventions need to be made. Unlike previous work, we’re leveraging data to focus on predictable action rather than mortality. These include factors like immediate need for an intervention, a need for an intervention in the near future, or when a patient is ready to be weaned from an intervention.
“Understanding vasopressor intervention and weaning: Risk prediction in a public heterogeneous clinical time series database.”
“Predicting intervention onset in the ICU with switching state space models.”