Automated clinical coding using semantic atoms and topology

Automated clinical coding (ACC) transforms narrative text in clinical records into a structured form. ACC may assign unique identifiers from standard terminologies. Previous ACC research focuses on mapping narrative phrases to terminological descriptions (e.g. concept descriptions). These methods make little or no use of the additional semantic information available through topology. We investigate a token based (non-phrase based) ACC approach that exploits additional semantic information available in SNOMED CT. Our method codes tokens as SNOMED CT concepts and manipulates concepts according to linguistic structures present in narrative text. It performed well (85.6%) in a recall test and performed significantly better than MetaMap in a precision test. MetaMap correctly coded 16% and our ACC method correctly coded 30%. This paper demonstrates the viability of non-phrase based ACC methods.