Researchers have proposed a new framework called KREL for automatic medical coding, which involves assigning standardized ICD codes to clinical notes. Current methods use pre-trained language models or large language models to frame the task as classification, generation, or multi-step reasoning. However, these approaches struggle with the length of clinical notes and complex coding rules. The proposed KREL framework integrates external ICD coding guidelines into a large language model's reasoning process to improve accuracy and reduce hallucinations. Experiments show that KREL outperforms existing methods on benchmark datasets. --- Why it matters: KREL matters because it can help improve the accuracy of medical coding, which is crucial for medical reimbursement, quality reporting, and clinical research. By reducing hallucinations and improving compliance with coding standards, KREL has the potential to make a significant impact in the field of healthcare. Source: https://arxiv.org/abs/2608.20887