Modeling patient trajectories using different representation learning techniques

Title Modeling patient trajectories using different representation learning techniques
Summary Modeling Electronic Health Record (EHR) data and predict future events for specific patients
References Attention is all you need:

Bert: Pre-training of deep bidirectional transformers for language understanding: BEHRT: transformer for electronic health records: MIMO: Mutual Integration of Patient Journey and Medical Ontology for Healthcare Representation Learning: Heterogeneous Similarity Graph Neural Network on Electronic Health Records: Learning the Graphical Structure of Electronic Health Records with Graph Convolutional Transformer: Variationally Regularized Graph-based Representation Learning for Electronic Health Records:

Supervisor Stefan Byttner, Kobra Etminani, Amira Soliman
Level Master
Status Open

The importance of effective modeling of Electronic Health Records (EHR) is increasing in both research and clinical practice to enhance the ability to provide the needed care to patients without posing additional economic burdens. Patient data is considered temporal data that is being measured over time. A patient trajectory describes the evolution of a quantity, behavior, biomarkers from lab tests, or some other repeated measure of interest over time. Modeling patient trajectories is a hot topic and there are many state-the-art deep learning techniques used for such purpose.

In this master thesis, the objective is to compare state-of-the-art representation learning techniques for modeling EHR data and predicting patient trajectories. For this comparative analysis we plan to consider transformers, recurrent neural architectures (RNNs), and graph neural networks (GNNs). Patients with congestive heart failure can be used as a use-case, specifically predicting the readmission of those patients is one of the challenging tasks for such group.