Data-Driven
Approaches in
Health & Medicine.

When does disease actually begin?

In biomedical data and machine learning, the answer is rarely straightforward.

In sepsis, conclusions may change depending on whether a model is anchored to T0, T1, or T2. After surgery, the intervention itself provides a precise timestamp, while the true onset of an infection or complication may be much harder to capture. In oncology, disease can evolve over months or years, with different timelines emerging from clinical, molecular, and patient perspectives.

These are examples of a broader challenge: how we represent time can fundamentally shape the questions we ask, the models we build, and the conclusions we draw.

Temporal representation remains a shared methodological challenge across biomedical domains. Baseline definitions differ between cohorts, clinically meaningful events are often incompletely documented in routine data, and disease trajectories unfold across very different timescales.