In 1,091 people at five hospitals, voice recorded during a dementia screen classified dementia better than a drawing

What the research found

A multicenter model study was published on 11 November 2026. 1,091 people at five university hospitals in South Korea had voice and pentagon drawings recorded during dementia screening, alongside the usual clinical scores. The task was to separate cognitively normal, mild cognitive impairment, and dementia. In the three-class task the combined model reached 67.9 percent accuracy (95 percent CI 65.9 to 70.0) and an area under the curve of 0.84 (95 percent CI 0.82 to 0.85). Most mistakes were between normal and mild cognitive impairment. Voice models beat drawing-only models. For normal versus dementia, accuracy was 90.1 percent (95 percent CI 88.4 to 91.8) and the area under the curve was 0.97 (95 percent CI 0.95 to 0.97). Voice plus the clinical scores discriminated better than the clinical scores alone. Drawings added only a modest gain, mainly for mild cognitive impairment. The abstract does not give age or sex. This is a classification study, not a home test. DOI 10.1007/s11357-026-02647-4. Not medical advice.

Evidence level: Human study. A study in people, but an association isn't proof of cause.

Read the original at GeroScience