Human study · Aging · 4 Nov 2026
On two questionnaire datasets, a language model screened Parkinson's disease with AUROCs of 0.977 and 0.974
Source: Nature Aging
What the research found
Nature Aging published a paper on 4 November 2026. Reyes and colleagues present Quest2Dx, a language model that reads the wording of health-questionnaire items and answers so the same model can move across forms when some answers are missing. The authors describe Parkinson's disease as a growing public-health problem as populations get older, because diagnosis still leans on motor signs and invasive tests. On the Parkinson's Progression Markers Initiative Online dataset the area under the receiver operating characteristic curve was 0.977. On Fox Insight it was 0.974. Trained on one questionnaire and tested on the other, the figures were 0.920 and 0.952. The abstract does not give the number of people in either dataset. The authors say the model points toward low-cost screening in primary care. A questionnaire score is not a diagnosis and does not replace a clinician exam. DOI 10.1038/s43587-026-01260-6. Not medical advice.
Evidence level: Human study. A study in people, but an association isn't proof of cause.
More aging news
- The Dog Aging Project described its Pack: 50,188 companion dogs enrolled through the end of 2024, with 14,978 recorded deaths
- In middle-aged male mice, an earlier loss of new tumor-fighting immune cells lined up with weaker melanoma control
- A commentary says telomere RNA loops can stall DNA copying and, in some tumors, get used to keep chromosome ends going
- After a long decline, heart-valve death rates in U.S. older adults rose from 2018 to 2023
- In adults over 50 followed for 24 years, a higher body mass index lined up with a faster drop in memory and thinking