Researchers have developed a method using large language models to identify protected health information (PHI) in electronic medical records that existing de-identification systems miss. They tested eight different language models on 100 pediatric oncology notes from Texas Children's Hospital and found that the best single prompt achieved a recall of 98.1% and an F1 score of 90.7%. The study suggests that institution-specific prompting can be an effective alternative to purpose-built de-identification systems, but may require more computational resources. --- Why it matters: This research matters because it provides a new approach to identifying protected health information in electronic medical records, which is essential for maintaining patient confidentiality and complying with regulations such as HIPAA. The method's ability to adapt to specific institutions' needs could also improve the accuracy of de-identification systems. Source: https://arxiv.org/abs/2608.17051