Clinical interpretation
What ESP360° adds to the clinical picture
Not another glucose dashboard. A structured record of the decision state surrounding an action.
What clinicians already see
CGM values, trajectories, insulin doses, time in range, variability and treatment outcomes remain essential. ESP360° does not compete with any of it.
What may be missing
The reason a correction was delayed. Whether a lower dose reflected fear after a recent hypoglycemic event. Whether fatigue, cognitive load, social pressure or uncertainty shaped the action. And what the person expected to happen, before the outcome was known.
A worked example
Over roughly two hours on one afternoon in the study window, glucose moved 8.7 → 9.2 → 11.0 → 9.9 → 10.9 mmol/L. No threshold was crossed. No alarm fired. On review, the trace is unremarkable — mild post-prandial drift, the kind of stretch that would not attract a second look in a clinic visit.
The decision record across the same three annotated events shows confidence falling 7 → 3 → 1 while stress holds at 5. The first record describes slow absorption and watchful waiting. The second describes a plateau that will not break and three competing explanations held at once — insulin failure, wrong ratio, or the body simply being difficult. By the third, the person has stopped reasoning about the physiology and is describing the act of dosing as a gamble.
What this is not
It is not a diagnosis, a risk score, or a psychological assessment. Nothing in the architecture generates a treatment recommendation, and the human decision right is preserved at every point by design. A clinician reading a decision record is reading what the patient documented at the time, not a model's interpretation of it.
Questions clinicians raise
Is this feasible outside a research setting?
Not yet at this density. The study record was produced by a single expert annotator with unusual training and motivation. Establishing what a realistic clinical subset looks like is open work, and is named as such on the Evidence page.
How would it change a consultation?
The intended use is retrospective review — being able to ask why a particular week looked the way it did and get a documented answer rather than a reconstructed one. It is not designed for real-time use.
What about recall bias?
Capture is designed to be synchronised to the moment of dosing, before the outcome is known. That is the specific bias the method is built to avoid; whether it succeeds is a question for inter-rater work that has not yet been done.