Evaluation of Large Language Models for Automated HAI Case Definition Matching for Surveillance
Hospital patient records are long, free-text documents. We use large language models (open-weight models run on premises, and commercial ones, under one shared configuration) to extract over a hundred clinical variables per infection episode, then apply the ECDC case definitions for healthcare-associated infections with a deterministic, auditable rule engine. Results are compared with physicians’ independent labels, case by case.