Delirium is common, dangerous, and frequently missed—particularly when cognitive change is misattributed to pre-existing dementia, medication side-effects, fatigue, or “normal ageing”.
This predictor aims to estimate the probability of delirium onset 24–72 hours in advance and to surface the specific risk of underdiagnosis due to misattribution. The design positions the predictor as a cognitive feature within an existing cognition model: it operates as a calibrated early warning system plus a parallel “misattribution
detector” that flags likely diagnostic drift.
The output is intentionally decision-related rather than decision-making: it produces
- a forecast risk curve with uncertainty,
- an explanation bundle (drivers, recent change-points, medication context), and
- a misattribution risk score with concrete prompts for confirmation steps.
Why delirium is systematically underdetected
Delirium is an acute fluctuating disturbance in attention and awareness, commonly precipitated by infection, dehydration, metabolic derangement, pain, polypharmacy, sleep disruption, hypoxia, and environmental change. In practice it is underdetected because:
- Baseline ambiguity: patients with dementia, depression, Parkinsonism, aphasia, or frailty have
variable baseline cognition, making change hard to define. - Attribution bias: clinicians (and families) attribute new confusion to “their dementia”, “opioids”,
“benzodiazepines”, or “they are just tired”, delaying formal assessment. - Signal fragmentation: early physiological deterioration (tachycardia, hypotension, hypoxia, fever)
and subtle behavioural changes are distributed across notes, charts, and shifts. - Temporal mismatch: the earliest physiological deviation can precede overt delirium by 24–72h,
but care processes often react only once behaviour becomes disruptive.
mike_tremblay@skythunder.net