Modelling Disease States Trajectories to Inform Health Outcomes and Economics Research

Chronic and complex diseases are difficult to track and predict, creating major challenges for patients and health systems. Although health records now contain rich data on diagnoses, lab results, medications, and procedures, current analytic methods fail to capture how diseases change over time. This limits our ability to define disease stages, study progression, or plan healthcare resources effectively.

This project will develop a new way to model disease progression by combining two powerful AI techniques: joint embeddings, which learn how different types of medical data relate to one another, and generative transformers, which simulate how a patient’s health may change over time.

As a start, we will apply these methods to disease states defined by Elixhauser comorbidity codes, an established set of conditions that can predict costs and health outcomes and renal progression measured by eGFR. We will test these AI methods against classical statistical methods and identify their strengths and weaknesses in its ability to forecast future events, utilisation, outcome and costs.