Ontology-based Normalization for Disease- Lab test Relation Extraction

Abstract

-This poster describes our preliminary work on ontology-based normalization for diseases and lab tests, as a fundamental step toward disease-lab test relation extraction. Multiple ontologies are leveraged for this aim. Specifically, diseases and lab tests are first extracted and mapped to the Concept Unique Identifier (CUI) of the Unified Medical Language System (UMLS) by MetaMap. Codes of International Classification of Diseases, Version 9 - Clinical Modification (ICD-9CM) are then employed to further normalize diseases; while the Logical Observation Identifiers Names and Codes (LOINC) are used to normalize lab tests.

Keywords

• ontology-based normalization
• normalization
• lab test normalization
• relation extraction

I. INTRODUCTION

Disease-labtest relation extraction plays an important role in various medical appliations such as clinical decision-support systems and phenotype information extraction. However, mentions of diseases and labtests in text contains diverse nonstandard variations. Those variants need to be normalizad into standard codes first to facilitate more universal computational applications. This poster describes our preliminary work on ontology-based normalization for diseases and lab tests, as a fundamental step toward disease-lab test relation extraction.

Three existing standard ontologies UMLS[1], ICD-9CM[3] and LOINC[4] are leveraged for this aim.

II. ONTOLOGY OVERVIEW

Overview

A. UMLS

UMLS[1] is a thesaurus re-orgizaing many controlled

vocabularies in the biomedical sciences. It provides a mapping structure among viraous vocabularies and serves as a comprehensive ontology of biomedical concepts. For each concept in UMLS, a synonym list consisting of terms from multiple vocabularies is collected. For example, “Diabetes

Mellitus” and “dm” are synonyms of the same CUI C0011849.

Various ontological relations between CUIs are defined, such as “isa”, “broader”, and “sibling”, etc.

B. ICD-9CM

The International Statistical Classification of Diseases and

Related Health Problems (ICD) is the international "standard diagnostic tool for epidemiology, health management and clinical purposes"[ 3 ] . The ICD provides a hierarchical system of diagnostic codes for classifying diseases. Major categories are designed to include a set of similar diseases as subcategories. For example, “(050.0) Variola major” is a subcategory of “(050) Smallpox”. Health conditions can be mapped corresponding generic categories or more specific subcategories.

C. LOINC

The Logical Observation Identifier Names and Codes

(LOINC) is the only publicly available universal standard for laboratory test codes and names[4 5].The current version of the

LOINC code set (released in June 2014) contains 73,889 terms for lab tests, measurements and clinical observations. Lab tests are orgazied hierarchically into 14 top classes including “Microbiology”, “Blood Bank”, etc.

III. NORMALIZATION METHOD

The original mention of disease/lab test is first recognized

and mapped to UMLS concepts by MetaMap[ 2 ]. If ICD

9CM/LOINC is among the sources of terms for the mapped

UMLS concept, then the mention can be normalized to the

corresponding code and name from ICD-9CM/LOINC directly. If not, for disease, if SNOWMED CT is one source of terms for the UMLS concept, the corresponding SNOWMED CT concept can be mapped to ICD-9CM code by the rule-based mapping provided by NIH[6]. For lab test, the RELMA software is employed to map labtest to LOINC code and name[7]. Fig. 1 and Fig. 2 illustrate the workflow of disease normalization and lab test normalization, respectivley.

Multiple variations of diseases and lab tests are normalized

into standard codes following the workfolow in Fig 1 and Fig

2. Fig. 3 and Fig. 4 show the examples of disease normalization and lab test normalization, respectivley. The original mentions are mapped to UMLS concept first, and then to ICD-9 CM and LOINC code.

Bodenreider O. The unified medical language system (UMLS): integrating biomedical terminology. Nucleic acids research 2004;32(suppl 1):D267-D70. http://www.nlm.nih.gov/research/umls/mapping_projects/snomedct_to_i cd9cm_reimburse.html. http://loinc.org/downloads/relma. The School of Biomedical Informatics |The University of Texas Health Science Center at Houston

Introduction

Disease-lab test relation extraction plays an important role in various medical appliations such as clinical decision-support systems and phenotype information extraction. However, mentions of diseases and labtests in text have diverse non-standard variations. Those variants need to be normalizad into standard codes first to facilitate more universal computational applications. This poster describes our preliminary work on ontology-based normalization of diseases and lab tests, as a fundamental step toward disease-lab test relation extraction.

Ontology Overview •  UMLS [ 1 ]: Re-organized many controlled vocabularies in the biomedical sciences; a comprehensive ontology of biomedical concepts. E.g., “Diabetes Mellitus” and “dm” are synonyms of the same concept . •  ICD-9CM [ 2 ]: A hierarchical system of diagnostic codes for classifying diseases. E.g., “(050.0) Variola major” is a subcategory of “(050) Smallpox”. •  LOINC [ 3 ]: The only publicly available universal standard for laboratory test codes and names. E.g., the test of “blood culture” is under the general category “Microbiology” .

Method

Results

•  Precision: 1) General concepts of diseases/lab tests not so valuable practically. E.g., in relations between “Heart Diseases” and lab tests, “Heart Diseases” include “coronary artery disease”, “arrhythmias” and “congenital heart defects”, etc. 2) Fail to normalize to L O N I C b y R E L M A . E . g . , “ a c a n t h o c y t e c o u n t ” - > 5 6 5 - 2 : C O L O N Y [COUNT]:NUM:PT:XXX:ORD:VC. •  Coverage: Fail to recognize variants of diseases/lab tests. E.g., “blood film” refers to “blood smear” .

This poster presents the preliminary results of our ontology-based normalization of diseases and lab tests. In the next stage, machine learning methods will be employed for disease and lab test recognition. General concepts of diseases and lab tests need to be filtered. The precision of lab test normalization to LOINC also need to be further improved.

Conclusion

Figure 1 Diagrams of Disease and Lab Test Normalization

References

1.  Bodenreider O. The unified medical language system (UMLS): integrating biomedical terminology . Nucleic acids research 2004 ; 32 ( suppl 1 ): D267 - D70 .

2.  National Center for Health, S. ICD-9-CM: International Classification of Diseases 9th Revision Clinical Modification . US Department of Health and Human Services , Public Health Service, Health Care Financing Administration 2008 .

3.   McDonald CJ , Huff SM , Suico JG , et al. LOINC, a universal standard for identifying laboratory observations: a 5-year update . Clinical chemistry 2003 ; 49 ( 4 ): 624 - 33 .