MERIT: Multi-agent EHR-grounded Risk Identification and Tailored Care for Early Cognitive Decline in Primary Care

MERIT, an EHR-grounded multi-agent CDSS for early cognitive decline in primary care, achieved provider-facing ratings around 3.9–4.3/5 and readability aligned with college-level summaries.

• Why it matters: Dementia is frequently missed in primary care, especially at early stages, leaving clinicians without timely, actionable cognitive risk summaries and tailored next steps.
• What they did: MERIT was co-designed with primary care providers, using RAG to retrieve dementia risk factors from multimodal EHR and reflection to improve agent-generated text; it was optimized on one PCP’s data and tested on patients from four additional PCPs.
• The result: In the test set, patient virtual cognitive status summaries scored 4.0 correctness, 4.2 completeness, 4.3 readability, and 4.1 clinical usefulness, while decision support scored 3.9 correctness and 4.3 completeness/adoptability, with mean Flesch Reading Ease 18.3 and Flesch-Kincaid 13.7.

AI summary of the abstract · may contain errors · not medical advice · read the paper.