HTNSystem: Hypertension Information Extraction System for Unstructured Clinical Notes* - Technologies and Applications of Artificial Intelligence - page 218

HTNSystem: Hypertension Information Extraction
System for Unstructured Clinical Notes *
Jitendra Jonnagaddala 1,2,3 , Siaw-Teng Liaw 2,* , Pradeep Ray 3 , Manish Kumar 1 ,
and Hong-Jie Dai 4,*
1 Translational Cancer Research Network
UNSW Australia
2 School of Public Health and Community Medicine
UNSW Australia
3 Asia-Pacific Ubiquitous Healthcare Research Centre
UNSW Australia
{z3339253,siaw,p.ray,manish.kumar}@unsw.edu.au
4 Graduate Institute of Biomedical Informatics, College of Medical Science and Technology
Taipei Medical University, Taipei, Taiwan, R.O.C.
[e-mail]
Abstract. Hypertension (HTN) relevant information has great application potential
in cohort discovery and building predictive models for prevention and surveillance.
Unfortunately most of this valuable patient information is buried in the form of
unstructured clinical notes. In this study we present HTN information extraction
system called HTNSystem which is capable of extracting mentions of HTN and in-
ferring HTN from BP lab values. HTNSystem is a rule based system which
implements MetaMap as a core component together with custom built BP value
extractor and post processing components. It is evaluated on a corpus of 514 clini-
cal notes (82.92% F-measure). HTNSystem is distributed as an open source com-
mand line tool available at https://github.com/TCRNBioinformatics/HTNSystem.
Keywords: Hypertension, Blood pressure, Information extraction, Rule based,
Apache UIMA, Apache Ruta, Text mining.
1 Introduction
Hypertension (HTN) or high blood pressure (HBP) is one of the major public health
burdens in developing and developed countries. It is estimated that there will be 60%
increase in adults with hypertension by year 2025[1, 2]. HTN is also a leading risk
factor for many cardio vascular diseases (CVD) and kidney diseases [1]. Any patient
information relevant to HTN has great application potential in cohort discovery and
building predictive models for prevention and surveillance. In general, most of
this valuable patient information is buried in the form of unstructured clinical notes
scattered across various electronic health records (EHR) or electronic medical records
* Corresponding authors.