Critical Incident Reporting System

Access and retrieval of relevant information for patient safety and quality assessment is important in clinical contexts. The objective of critical incident reporting systems (CIRS) is to enable users, e.g. health care professionals working for a hospital, to report in an anonymous manner critical events that occurred in their working environment. Incident reporting has been instituted in healthcare systems in many countries for some time now, e.g. in Switzerland in 1997 [1], but not in all healthcare systems it is obligatory to report critical incidents. However, it has been shown that those anecdotal reports bear important information on limitations of systems and processes [2]. On the one hand, critical situations or even systematic errors

2. Material and Methods

2.1. Requirements

In previous work, we studied user requirements and challenges of incident report retrieval [10]. In summary, a future incident report retrieval system: (1) provides a keyword search on the entire data set, (2) enables coordinated queries (AND, OR), (3) performs a semantic enrichment or automatic query expansion with synonyms (e.g. when only a drug name is mentioned in a text, the text should be anyway retrieved when searching for the keyword “medication”) and lexical variants, and (4) identifies matches that sound similarly (SoundEx, phonetic algorithm [17]).

2.2. Material

The basis of the analysis and retrieval experiment are 581 randomly selected incident reports from a university hospital. They originate from different clinics of the hospital and consist of at least of a date and a free-textual event description. Most of them have a title that is summarizing the critical event consisting mainly of one to three keywords. Sometimes, a potential measure for addressing the problem is suggested in a separate data field.

2.3. Retrieval

Each sentence is then broken up by a chunking method. The resulting chunks contain noun, verb or adverbial phrases. In each of these phrases the clinical and additional concepts are identified. If one potential word is found, it is mapped onto the respective concept of the Wingert Nomenclature. The latter is a German derivate of an early version of SNOMED [12]. It is a polyaxial nomenclature that contains ten axes of different categories of concepts. For example, the Topology-axis contains topological concepts, the Morphology-axis contains morphological concepts and the Procedure-axis contains concepts referring to medical procedures. In addition, the G-axis contains helpful concepts for certain adjectives and verbs and linguistic meta-information (e.g. „negated