Automated Medical Coding

Automated medical coding maps clinical notes and other patient data to standardized codes such as ICD, supporting billing, health-system administration, and epidemiological research. Common approaches include multi-label classifiers with label-wise attention, hierarchy-aware prediction, and large language models fine-tuned or augmented with coding guidelines, ontologies, and retrieval. Research also explores synthetic training data, combining structured and narrative information, and evidence-linked predictions that clinicians can audit. Performance remains uneven for rare codes and across coding policies or settings, with an ongoing tension between improving recall and avoiding false-positive codes.