Automatic Placement Within a Hierarchical Clinical Decision Support Terminology

Hierarchical terminologies are often used to trigger clinical decision support within electronic health record software systems. This study utilized two algorithms to automate the process of placing clinical terms within a widely used hierarchical decision support terminology. The first is a feature ranking system similar to the TF-IDF algorithm. The second uses the clinical bidirectional encoder representations from transformers machine learning system. In a test set of 100 terms that were originally assigned a placeholder grouper, the correct grouper was identified by one of the two algorithms in 93% of cases. For the 48 cases in which both algorithms agreed with their top grouper, this option was correct 97.9% of the time. These methods can drastically reduce clinician burden.