LLM Identifies Stroke Innovation Patent Trends

Marcus et al. (2026) analyze patent and publication records from 1993–2023 to quantify innovation in stroke care, using a 13-billion-parameter Llama LLM to filter patents. The LLM achieved 99.2% accuracy (96.5% sensitivity, 99.6% specificity), reducing 237,035 patents to 28,225 stroke-related filings grouped into seven innovation clusters. Pharmacological treatments dominated earlier patents but plateaued, while AI methods, rehabilitation devices, and imaging show rapid patent growth.