Identifying Drug-Induced Liver Illness (DILI) with Computerized Information Extraction: No More Dilly-Dallying

Excerpt

The rapid adoption of electronic medical record (EMR) has provided health-care professionals with better access to patient records while also improving the quality of medical care, reducing medical errors, and lowering medical costs. As a result, the EMR has produced a parallel growth of digitized clinical data, an important medical resource. Clinical data extracted from EMRs have helped health-care professionals support their decisions and have also aided biomedical research, clinical trial screening, adverse drug reaction monitoring, and drug–drug interaction assessment. Nevertheless, a major feature of each EMR is the inclusion of a large amount of clinical narrative text, including medical histories, social histories, laboratory studies, progress notes, discharge summaries, nursing and consultation notes, and pathology, radiology, surgery, and medical imaging reports. Such information is often presented in an unstructured format not immediately suitable for computer analysis. In order to best utilize the vast amount of medical information included in the EMR, data have to be properly extracted and encoded into a structured format suitable for predefined templates. Therefore, effective tools and techniques are required to retrieve and organize these huge volumes of clinical narrative text data in order to make this information useful for supporting medical practice, project management, research, and policy-making. Natural language processing (NLP), and more specifically information extraction (IE), is the most popular and useful technique/tools to date. IE, a subdomain of NLP, is aimed at better understanding the human process of language comprehension in order to develop tools and techniques in order to enable computer systems to manipulate natural languages and perform desired tasks [1]. One of the NLP’s major tasks is the extraction of semantic information from text [2]. As a result, large amounts of text can be automatically analyzed by effective extraction tools in order to gather useful information, which can then be represented in a tabular/structured format. In development since the 1950s–1960s [3, 4], the recent literature has reported significant advances in IE, particularly in the last 30 years [5]. Nevertheless, IE has mostly been developed outside of the biomedical domain, being adapted to the biomedical field much later than for other fields. …