A tablet drawing task scored by AI separated Alzheimer's disease from controls in a small study

Design

Cross-sectional diagnostic validation study

Population

111 adults aged 65 or older: 55 with Alzheimer's disease and 56 healthy controls

Primary outcome

Diagnostic accuracy of the AI drawing score against clinical diagnosis

Effect

AUC 0.868; at cut-off below 27, sensitivity 83.6% and specificity 75.0%

A cross-sectional validation study enrolled 111 adults aged 65 or older, 55 with clinically diagnosed Alzheimer's disease and 56 healthy controls. Participants did manual clock drawing tests and an open-source tablet task using simple pre-linguistic symbols, scored by AI.

The AI score distinguished the groups with an area under the curve of 0.868, comparable to the 10-point manual clock scale. A cut-off below 27 gave sensitivity of 83.6% and specificity of 75.0%. The platform also flagged 14 of 56 controls (25%) as at risk who passed the manual test. That could be early detection or false positives, and the study cannot say which.

The cut-off was chosen in the same sample, which tends to flatter performance, and participants were clinically diagnosed people versus healthy controls, not the mixed group seen in primary care. It is a promising tool at an early stage, not one to adopt for screening yet.

• Keep using a validated cognitive screen, and do not replace it with this tool on current evidence
• A tablet task could help where literacy or language limits a written clock test, but needs validation in primary care
• Treat a flagged result as a prompt for assessment, never a diagnosis
• Look for reversible contributors before attributing impairment to dementia

Why it matters

Literacy and language limit clock drawing in many Indian clinics, so a language-free screen would be useful if validated.

Don't overread it

The sample was small, the cut-off was derived in the same sample, and the comparison was with clinically diagnosed disease rather than a mixed primary care population.

The statistics, in plain English

An area under the curve of 0.868 means that if you pick one person with Alzheimer's and one control at random, the tool scores the patient as worse about 87% of the time. A cut-off derived and tested in the same group overstates real-world accuracy, and 25% of healthy controls being flagged shows a high false-positive rate if those are truly healthy.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.