Representation Learning for Patients in the Intensive Care Unit

My thesis focuses on representation learning for patients in intensive care, aiming to improve patient outcomes and healthcare system efficiency. It addresses predicting patient deaths and estimated discharge dates, essential for managing hospital beds effectively. The research incorporates clinical knowledge, periodic signals, systematic biases, and graph neural networks to enhance length of stay prediction, mortality prediction, and patient outcome models for mechanically ventilated patients, with the goal of discovering hidden patient phenotypes and creating real-world deployable representations.