Large language models

1. Large language models improve data extraction accuracy in silico (100% accuracy).

   “we achieved 100% accuracy for both total CCS and CT-FFR.”

   In silicofrom free-text CTA reports

   Leveraging Large Language Models to Extract Unstructured Cardiovascular Data from the EHR for Preventive Cardiology

   AHA Scientific Sessions 2025 abstract·3 Nov 2025

More on Data extraction accuracy

1. Large language models predict mild cognitive impairment in silico (AUC=0.765, p = .013).

   “LLM scoring showed superior discrimination of cognitive status for immediate recall (t(26)=2.64, AUC=0.765, p = .013) compared to human scored responses which did not find any difference”

   In silicoimmediate recalln = 28

   Automated Scoring of Narrative Recall Assessments Using Large Language Models Enables Exploration of Alternate Scoring Criteria

   AAIC 2025 abstract·24 Dec 2025

More on Mild cognitive impairment

1. Large language models are associated with human scoring agreement in silico (r = 0.654, p < .001).

   “LLM scoring demonstrated moderately high correlations with human raters for both immediate ( r = 0.654, p < .001) and delayed recall ( r = 0.655, p < .001).”

   In silicoimmediate recalln = 28

   Automated Scoring of Narrative Recall Assessments Using Large Language Models Enables Exploration of Alternate Scoring Criteria

   AAIC 2025 abstract·24 Dec 2025