Natural Language Processing For Analysing Notes In Heart Failure Telehealth Patients

Research output: ThesisBachelor's Thesis

Abstract

Nowadays, due to an ever-increasing life expectancy, chronic illnesses, like e.g. diabetes mellitus or cardiovascular diseases, are on the rise. Amongst these health issues, heart failure is of particular interest, because it is one of the number one reasons for hospitalizations of elderly people. Therefore, the heart failure telehealth network \HerzMobil", which particularly aims to enhance the quality of life of heart failure patients, was established in cooperation with the AIT Austrian Institute of Technology. Within HerzMobil, health care professionals communicate via clinical notes in the form of unstructured free text. Most of these notes are directly related to the patient's treatment and refer to e.g. a change in medication or the general well-being of the patient. Thus, valuable information could be gained from analysing them. However, with more than 20.000 German notes collected in the system, manual screening of the free text notes would be a tedious task. Therefore, over the course of this thesis, three di erent Python scripts were developed to extract time references from the clinical notes, to calculate a corresponding date and to match these dates with certain events, which occur in the same note. To achieve this goal, regular expressions from the eld of natural language processing were applied. The results of the rst processing step were compared to a pre-existing Python library called \parsedatetime". The developed script achieved superior results in all analysed key gures (Accuracy, Precision, Recall, F1 score). The calculation of corresponding dates achieved an accuracy of 78.59%. The detection of events was limited to two di erent types, namely \death" and \threshold adaptation". This third algorithm works nearly perfect in the context of the available HerzMobil notes. Implementing these scripts could be used to analyse notes in retrospect or even to give real-time feedback to health care professionals. Thus, applying them in HerzMobil or, after minor changes, in other telehealth networks of the AIT, would potentially spare a lot of manual work and further improve these systems. v
Original languageEnglish
Awarding Institution
  • FH Joanneum University of Applied Sciences
Supervisors/Advisors
  • Hanke, Sten, Supervisor, External person
  • Eggerth, Alphons, Supervisor
Publication statusPublished - 2020

Research Field

  • Exploration of Digital Health

Fingerprint

Dive into the research topics of 'Natural Language Processing For Analysing Notes In Heart Failure Telehealth Patients'. Together they form a unique fingerprint.

Cite this