Delirium Onset Predictor Demo (24–72h) — with Misattribution / Underdiagnosis Detection

Standalone interactive prototype (no server). This illustrates the three outputs: forecast risk (24/48/72h), explanations, and misattribution risk (diagnostic drift / screening gap). It does not provide medical advice and is for demonstration only.

Patient and context

Risk modifiers (demo sliders)

Clinician narrative (misattribution signal)

Tip: include terms like dementia, opioids, benzodiazepine, just tired, baseline without a formal screen to raise misattribution risk.
Prototype logic: a transparent scoring model that mimics a calibrated ML layer. In production, this would be replaced by a trained, calibrated time-series model with uncertainty and drift monitoring.

Outputs

Delirium risk within 24h
—
—
Delirium risk within 48h
—
—
Delirium risk within 72h
—
—
— — —

Forecast curve (next 72h)

The shaded band represents uncertainty (demo: wider when inputs are noisy / contradictory).

Explanation bundle

Prompt Rationale (demo)
Underdiagnosis alert logic (conceptual): triggers when misattribution risk is high and time-since-screening is high. It should prompt a formal screen and reversible-cause review, not assert a diagnosis.

Audit panel (demo)

This panel illustrates what would be logged for quality improvement: alert timing, screening actions, and narrative drift.
Signal Value Interpretation