Text Mining in Healthcare: Bringing Structure to Electronic Health Records

Electronic health records (EHRs) are rich in data with the potential to leverage applications that provide safer care, reduce medical errors, reduce healthcare expenditure, and enable providers to improve their productivity and efficiency. A major portion of this data is inside free text in the form of physicians’ notes, discharge summaries, and radiology reports among other types of clinical narratives. Despite many attempts to encode text in the form of structured data, free text continues to be used in EHRs. Therefore, text mining techniques can be applied to create a more structured representation of a text, making its content more accessible for data science, machine learning and statistics. To this end, this thesis is concentrated on providing solutions for some of the challenges in the analysis of the free text element frequently included in the clinical domain. Text mining techniques have been applied to numerous health applications involving clinical decision support systems, patient identification, disease classification, and the prediction models. Many of these clinical applications depend on some form of text classification. Text classification is a task that consists in automatically assigning a document to a predefined set of labels. The focus of the work described in this thesis is the analysis and classification of the clinical free text found in heath records used in the University Medical Center (UMC) Utrecht.