O discurso na prática clínica e as terminologias de padronização: investigando a conexão

Abstract

The Electronic Healthcare Record (EHR) is an important source of real healthcare information. In general, information in EHRs is made available as unstructured data, that is, in free text format extracted from natural language samples. Healthcare professionals who fill EHRs often use jargon, acronyms and expressions of their routine. Although such expressions are known within the medical field and allow a quick typing of EHRs, they are not standardized and may vary between different professionals. Advances in healthcare information technologies have made it essential to standardize terminologies in clinical texts aiming improvements in information retrieval and interoperability. The unstructured data of EHRs, due to their variety of terminology and idiosyncrasy, do not correspond to standardized clinical terminologies. This fact results in difficulties in the information retrieval and in the integration between systems healthcare units, and even within the same unit. Improvements are needed in communication between professionals involved in care, mainly in the discovery and production of knowledge, to mention a few, for the benefit of healthcare and, consequently, better life quality of patients. This requires some kind of harmonization between the terms registered colloquially by professionals and terminologies. This research seeks to fill these gaps, by addressing the lack of terminological standardization of EHRs that greatly impacts information retrieval. To this end, our goal is to define a mechanism for connecting clinical terms - natural language versus standardized language - in verifying the percentage of terms that correspond to a set of data from a medical specialty, in order to establish the connection between clinical terminologies. Within an interdisciplinary approach - involving Librarianship and Information Science, information technology and healthcare fields - we developed an applied research, with a qualitative, quantitative and descriptive approach. The methodology includes Natural Language Processing techniques for the extraction and analysis of clinical texts to, ultimately, verify the level of connection between ABNT standard terminological resources for mapping clinical terminologies. Concerning the results, from 18,256 anamnesis and 14,035 patient evolution records in the sample, we obtain 1,364,364 terms and the results indicate that the connection between clinical terminologies is it still needs to be worked on, because even with a sample of terms a significant number has not obtained equivalence in the Reference and Aggregation terminologies. However, this sample demonstrated the richness of terms in Interface Terminology, which will be useful in enriching Reference Terminology. An additional contribution was the creation of a computational lexicon (corpus in healthcare) in Portuguese that can help to create algorithms for the domain of Gynecology. The main problems during the natural language processing were: grammatical ambiguity, synonyms, abbreviations, spelling errors or negation expressions. In the mapping between the terminologies, the main difficulties were related to semantics: different terms with the same meaning, absence of the corresponding terms, and synonyms not identified.