Duke Takes Dementia AI Beyond the Lab

2026-10-21

Dementia AI needs a bedside test. A Duke initiative aims to move artificial intelligence tools into routine clinical practice, addressing a stubborn gap between research promise and the care available to older adults and people living with dementia. Publication is not adoption. Few tools have successfully crossed that divide.

Accuracy alone proves little. For dementia tools intended to flag possible disease, sensitivity measures how often affected patients are identified, while specificity measures how often people without the condition are correctly excluded. Both carry consequences. False positives can prompt unnecessary assessments and anxiety, while false negatives can delay further evaluation and support. Context changes the calculation. External validation, which tests a model on data beyond those used to develop it, helps assess whether reported performance holds across different patient populations and care settings rather than merely reflecting the conditions of a research dataset.

The harder problem is use. An AI output has limited value if clinicians cannot interpret it, determine what action it warrants or fit that action into an appointment already crowded with competing needs. Responsibility must stay visible. Bringing these tools into practice therefore requires evidence not just that predictions are accurate, but that acting on them improves care without creating disproportionate burdens for patients, caregivers or staff. Duke's stated ambition addresses that research-to-practice gap; the supplied information does not establish that any particular tool has already cleared it. The bedside remains the examiner. A prediction can arrive in an instant; its meaning must survive the encounter.

Recommendations