MDP — Mayo Delirium Prediction tool

What this tool is and how to interpret its evidence

The Mayo Delirium Prediction (MDP) tool is an electronic risk-stratification model for predicting delirium during hospitalisation; it is not a test for delirium that is already present. It uses information available during the first hospital day for medical and surgical patients aged 50 years or older. Nineteen structured EHR variables cover age, previous delirium and cognitive or psychiatric diagnoses, care setting and admission route, selected treatments and comorbidities, fall risk, laboratory results, pressure ulcer and fracture. The model returns a predicted probability classified as low at 5% or less, moderate at 6-29%, or high at 30% or more. Admissions for acute substance intoxication or withdrawal were treated as high risk outside the calculated model.

The retrospective development study included 120,764 Mayo Clinic admissions: 80,000 for derivation and 40,764 allocated to validation. AUROC was .85 and .84 respectively, with similar observed delirium rates across risk groups. Subsequent modified and external validations support the approach, including a recent cancer-centre cohort, but versions and calibration should not be assumed interchangeable.

The original outcome came from electronic physician diagnoses of delirium or encephalopathy and nursing CAM-ICU documentation, so missed or undocumented cases could affect the model. The cohort was predominantly White, and implementation depends on local EHR definitions, data completeness and recalibration. MDP identifies people who may merit prevention and systematic assessment; it neither confirms nor excludes delirium and should never replace bedside evaluation.