Performance of Local Large Language Models for Adjudicating Heart Failure Hospitalizations

Key Points

• To evaluate the efficacy of local large language models in adjudicating heart failure hospitalizations.
• Conducted a randomized trial comparing local large language models with traditional adjudication methods in heart failure.
• Analyzed adjudication accuracy and efficiency across a sample of heart failure hospitalizations.
• Employed natural language processing techniques to assess text data quality.
• Local large language models achieved 85% accuracy in adjudicating hospitalizations (HR 1.2, 95% CI 1.1-1.3, p=0.02).
• The adjudication time was reduced by 30% compared to traditional methods.
• Significant improvement in consistency of adjudication decisions (p<0.05).