ClinIDMap: Towards a Clinical IDs Mapping for Data Interoperability

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3661–3669 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3661 ClinIDMap: Towards a Clinical ID Mapping for Data Interoperability Elena Zotova12, Montse Cuadros1, German Rigau23 1 SNLT group at Vicomtech Foundation, Basque Research and Technology Alliance (BRTA), Mikeletegi Pasealekua 57, Donostia/San-Sebasti´an, 20009, Spain, 2 Department of Languages and Computer Systems, University of the Basque Country (UPV-EHU) 3 HiTZ Basque Center for Language Technologies Paseo Manuel de Lardiz´abal, 1, 20018, Donostia/San-Sebasti´an, Spain {ezotova, mcuadros}@vicomtech.org, german.rigau@ehu.eus Abstract This paper presents ClinIDMap, a tool for mapping identifiers between clinical ontologies and lexical resources. ClinIDMap interlinks identifiers from UMLS, SMOMED-CT, ICD-10 and the corresponding Wikipedia articles for concepts from the UMLS Metathesaurus. Our main goal is to provide semantic interoperability across the clinical concepts from various knowledge bases. As a side effect, the mapping enriches already annotated medical corpora in multiple languages with new labels. For instance, spans manually annotated with IDs from UMLS can be annotated with Semantic Types and Groups, and its corresponding SNOMED-CT and ICD-10 IDs. We also experiment with sequence labelling models for detecting Diagnosis and Procedures concepts and for detecting UMLS Semantic Groups trained on Spanish, English, and bilingual corpora obtained with the new mapping procedure. The ClinIDMap tool is publicly available. Keywords: Clinical Coding, Concept Classification, Sequence Labeling, Interoperability 1. Introduction Sequence labeling (SL)—a task which assigns a class or label to each span of text in a given input sequence— is a crucial first step to a number of natural language processing (NLP) tasks in the clinical domain, such as relation extraction, entity linking, and automatic clini- cal coding. In particular, the automatic clinical coding task consists of detecting an entity span, classifying the span, and linking it to a clinical knowledge base. This paper focuses on the first part of a clinical coding pipeline, entity detection and recognition. Building supervised machine learning and deep learn- ing models for this task require a large amount of man- ually annotated data. Manual annotation of a particular textual corpora is one of the most valuable and expen- sive part when building a new NLP system. The data in the biomedical domain is especially difficult to ob- tain because of two main reasons. First, clinical reports and other medical documents usually have privacy is- sues. Even after de-identification, it is complicated to obtain free access to this kind of medical data and, for this reason, there are very few clinical corpora freely available for research. Second, the process of manual annotation requires high level expertise which makes the use of crowd-sourcing platforms almost impossible, making it more expensive than general purpose NLP- corpora. The problem is even more difficult in multi- lingual setting since very few resources are available for languages other than English. Fortunately, most clinical concepts are transferable across the various knowledge bases and languages. The goal of our mapping tool is to align different types of clinical identifiers (IDs, …

Content truncated to 500 words