Beyond Binary Onset Prediction

Background: Modeling longitudinal electronic health records (EHRs) often involves sequential prediction tasks, where models must distinguish between the first occurrence (onset) of a clinical event and its subsequent recurrence.

Question / Future Work: Further exploration is needed to determine the optimal methods for moving beyond simple next-visit onset prediction to model more complex clinical progression states, such as the progression of a disease, remission, or the response to specific treatments. These nuanced states inherently involve timing and repetition in ways that require more sophisticated modeling than single-step onset prediction.