Cognitive Assessment of Dementia Patients Using Large Language Models

Topic: Risk, prevention and care in population studies

The study included 685 patients who underwent MDS-cognitive and 3,630 patients assessed using OASIS. Overall, LLMs demonstrated strong cross-dataset generalizability. Among them, Llama-3.3-70B achieved the highest F1-score (66.31%) in OASIS assessments, while QWEN2.5-72B performed best in MDS (45.23%). The larger LLMs provided significant performance gains, whereas the CoT and summary strategies enhanced interpretability but did not consistently improve accuracy. This may be due to the models' limited understanding of clinical text, making precise information extraction and reasoning challenging. This study highlights the potential of LLMs in predicting cognitive function for dementia patients using EHRs. Larger LLMs showed superior performance, while the CoT and summary strategies improved interpretability but may slightly impair performance. The results highlight the necessity for further adoption and optimization of LLM modeling and reasoning strategies to better handle clinical text, minimizing hallucinations and improving reliability. Integrating LLMs into cognitive assessment workflows could enhance efficiency and accessibility, supporting more effective dementia care and management.

@article{wu2025-41449636,
  author  = {Wu, Jiageng and Wyss, Richard and Lin, Josh C. H. and Yang, Jie},
  title   = {{Cognitive Assessment of Dementia Patients Using Large Language Models}},
  journal = {Alzheimer s \& Dementia},
  year    = {2025},
  doi     = {10.1002/alz70858_103769},
  pmid    = {41449636},
  url     = {https://doi.org/10.1002/alz70858_103769}
}

Key findings

• The study included 685 patients who underwent MDS-cognitive and 3,630 patients assessed using OASIS.
• Overall, LLMs demonstrated strong cross-dataset generalizability.
• Among them, Llama-3.3-70B achieved the highest F1-score (66.31%) in OASIS assessments, while QWEN2.5-72B performed best in MDS (45.23%).
• The larger LLMs provided significant performance gains, whereas the CoT and summary strategies enhanced interpretability but did not consistently improve accuracy.
• This may be due to the models' limited understanding of clinical text, making precise information extraction and reasoning challenging.

Why it matters

This study found that large language models could predict cognitive function for dementia patients from electronic health records, with larger models performing better, though accuracy was limited and reasoning strategies did not consistently improve it. The authors call for further optimisation to reduce hallucinations; it concerns assessment, not treatment or cure.

New to this subject? Start with our plain-English guide: Getting a diagnosis →

Study details

• Study designcohort study
• Participants685
• Effect size66.31%; 45.23%
• Evidence strengthModerate

Read the original

• DOI: 10.1002/alz70858_103769
• PubMed: 41449636
• Original source →
• Free full text: Alzheimer s & Dementia →

Full text: free to read — OpenAlex lists this copy as free to read: the link above (checked 6 Feb 1020).

Related studies in Getting a diagnosis

• Large Language Models: A Paradigm Shift for Dementia Diagnosis and Care British Journal of Hospital Medicine · 2025-10-09 · usefulness score 37/100
• Loneliness, social isolation, and effects on cognitive decline in patients with dementia: A retrospective cohort study using natural language processing Alzheimer s & Dementia Diagnosis Assessment & Disease Monitoring · 2025-07-01 · usefulness score 42/100
• The measurement of cognition in large-scale cross-national surveys: lessons from the Health and Retirement Study International Network of Studies and the Harmonized Cognitive Assessment Protocol The Journals of Gerontology Series B · 2025-10-16 · usefulness score 51/100
• Language, literacy, and sensory impairments and missing cognitive test scores in the Harmonized Cognitive Assessment Protocol of the China Health and Retirement Longitudinal Study Aging Clinical and Experimental Research · 2025-05-12 · usefulness score 57/100
• Measurement differences in the Harmonized Cognitive Assessment Protocol across ethnicity and language in the United States Neuropsychology · 2026-02-09 · usefulness score 39/100
• The Impact of Mild Cognitive Impairment on Gait and Balance: A Systematic Review and Meta-Analysis of Studies Using Instrumented Assessment Gerontology · 2016-05-13 · usefulness score 64/100

Also tagged “prevalence”

1,224 studies in the library carry the topic tag “prevalence”; these are this study's neighbours in a fixed, alphabetical-by-hash tour of them, not the closest in content. All 1,224 tagged studies.

• Outcome measures in clinical trials on medicinal products for the treatment of dementia: a European regulatory perspective International psychogeriatrics · 2007-04-16 · usefulness score 37/100 · Treatments
• Pathways to dementia diagnosis: evidence for cross-ethnic differences Alzheimer disease and associated disorders · 2004-07-01 · usefulness score 40/100 · Disparities
• The potential for dementia prevention in Brazil: a population attributable fraction calculation for 14 modifiable risk factors The Lancet Regional Health - Americas · 2025-08-07 · usefulness score 58/100 · Prevention and risk

Similar studies

• Predicting dementia from spontaneous speech using large language models 2022 · Digital health tools
• An analysis of communication in conversation in patients with dementia 2010 · Digital health tools
• Review of brief cognitive tests for patients with suspected dementia 2014 · Cognitive tests and assessment
• Evaluation of Reminiscence Therapy on Language Outcomes Among People With Dementia 2025 · Getting a diagnosis
• Evaluation of dementia in the cardiovascular health cognition study 2003 · Disparities
• Assessment scales in dementia 2012 · Caregiving

More on Getting a diagnosis

• A review that covers this ground Animal models of Alzheimer's disease and frontotemporal dementia Nature reviews. Neuroscience · 2008-07-01 · usefulness score 47/100
• Newest in this subject The Contribution of Routinely Collected Clinical Data to Dementia Identification in Primary Care: A Cross-Sectional Study Research Square · 2026-10-28 · Preprint — not yet peer-reviewed · usefulness score 16/100
• Newest in this subject Beyond “Denial”: Understanding and Responding to Help-Refusal in Dementia Preprints.org · 2026-10-26 · Preprint — not yet peer-reviewed · usefulness score 16/100

See all studies in Getting a diagnosis →