Concept-Based Retrieval from Critical Incident Reports

Background: Critical incident reporting systems (CIRS) are used as a
means to collect anonymously entered information of incidents that occurred for
example in a hospital. Analyzing this information helps to identify among others
problems in the workflow, in the infrastructure or in processes. Objectives: The
entire potential of these sources of experiential knowledge remains often
unconsidered since retrieval of relevant reports and their analysis is difficult and
time-consuming, and the reporting systems often do not provide support for these
tasks. The objective of this work is to develop a method for retrieving reports from
the CIRS related to a specific user query. Methods: atural language processing
(NLP) and information retrieval (IR) methods are exploited for realizing the
retrieval. We compare standard retrieval methods that rely upon frequency of words
with an approach that includes a semantic mapping of natural language to concepts
of a medical ontology. Results: By an evaluation, we demonstrate the feasibility of
semantic document enrichment to improve recall in incident reporting retrieval. It is
shown that a combination of standard keyword-based retrieval with semantic search
results in highly satisfactory recall values. Conclusion: In future work, the
evaluation should be repeated on a larger data set and real-time user evaluation need
to be performed to assess user satisfactory with the system and results.
Keywords. Information Retrieval, Data Mining, Natural Language Processing,
Critical Incidents Reporting.