Methodology to ease the construction of a terminology of problems

Problem lists summarize an aspect of the patient's medical history and provide an important way to implement entry points for clinical pathways and guideline-oriented care. However, in order to automate processes based on problem lists, the use of controlled vocabularies is required. We developed a methodology to extract a collection of standardized problem-related terms from medical documents entered in free text by physicians.

We extracted a corpus of sentences describing problems from a randomized selection of admission notes collected at the University Hospitals of Geneva. Theses sentences underwent manual and automatic normalization processes, and a statistical clustering, in order to build a set of terms.

We obtained 17,805 sentences from 5000 admission notes. We refined them into 1546 terms, 88.6% of which could be related to a relevant problem statement.

A clinically relevant problems terminology was derived from clinical admission notes in free-text using a few methodical steps with a reasonable investment of human resources. Such an approach will ease the development and the use of problem lists better suited to user needs.