Septic Shock Prediction
Septic is a whole-body infection that affects thousands of people annually. This condition is often fatal if it progresses to septic shock, sepsis with organ failure that does not respond to fluid resuscitation. Developing methods to predict when a patient is transitioning to septic shock can lead to earlier clinical interventions and improved patient outcomes.
Clinical Decision Support
The wide-spread adoption of electronic health records presents exciting opportunities to use machine learning on large datasets of clinical records to develop new decision support tools. However, these clinical decision support tools have to function in the context of treatments. For instance, a patient on a medication to raise their blood pressure, may appear have normal blood pressure, but actually their underlying state is severe because they require such medication to maintain a normal blood pressure. Additionally it is important to understand treatment effects so that we can predict to which treatment regime a patient is most likely to respond. By understanding treatment-effect, we can hope to build tools that provide more reliable clinical decision support.
Physiological time-series like heart rates and respiratory rates have been shown to contain information about whether a patient has a given disease. However, how precisely to represent these signals and the relationships between different signals, is still in question. Developing methods to automatically learn represenations for physiological time-series is critical for implementing disease detection systems.