Modern Clinical Text Mining: A Guide and Review

Abstract

Electronic health records (EHRs) are becoming a vital source of data for healthcare quality improvement, research, and operations. However, much of the most valuable information contained in EHRs remains buried in unstructured text. The field of clinical text mining has advanced rapidly in recent years, transitioning from rule-based approaches to machine learning and, more recently, deep learning. With new methods come new challenges, however, especially for those new to the field. This review provides an overview of clinical text mining for those who are encountering it for the first time (e.g. physician researchers, operational analytics teams, machine learning scientists from other domains). While not a comprehensive survey, it describes the state of the art, with a particular focus on new tasks and methods developed over the past few years. It also identifies key barriers between these remarkable technical advances and the practical realities of implementation at health systems and in industry.

Can you hear me now? Clinical applications of audio recordings

“…For details of studies from before 2010 and/or focusing on clinical speech, we recommend other excellent reviews. 1,[3][4][5]9 The PRISMA flow chart in Figure 2 details the search results and inclusion/exclusion selection criteria. From the initial 1812 records retrieved from the databases, 561 duplicates were automatically removed, and a further 880 were removed after manual screening of titles and abstracts based on specific inclusion/exclusion criteria, leaving 371 in-scope articles.…”

GERNERMED: An open German medical NER model

“…For example, large general purpose language models from the kind of BERT [1]-or GPT [2,3]-inspired architectures are commonly trained on large corpora such as Common Crawl [4] or The Pile [5] that are composed of 320 TiB (Common Crawl) or 825 GiB (The Pile) raw text data. Since any kind of such large-scale data is infeasible to annotate, these datasets are mainly purposed for unsupervised methods such as pretraining [6]. However, when facing case-specific downstream tasks, well-suited datasets are used for fine-tuning in a supervised fashion [6].…”

“…Since any kind of such large-scale data is infeasible to annotate, these datasets are mainly purposed for unsupervised methods such as pretraining [6]. However, when facing case-specific downstream tasks, well-suited datasets are used for fine-tuning in a supervised fashion [6]. In this context the dataset is required to be annotated for a certain task accordingly.…”

“…Processing of unstructured German clinical data remains an ongoing area of research. Common tasks in NLP, such as named entity recognition (NER), are used for determining key elements from texts like medication information and various related information like dosage and duration [6].…”

The Scope, Methods and Applications of Biomedical Data Mining

“…Nevertheless, analyzing and processing this data is difficult and necessitates the use of specialized software. DM methods have a tremendous potential for analyzing such huge amounts of saved biomedical data in attempt to uncover expertise, as demonstrated by the positive stories discussed Percha in [2]. KDD is a broad term for the procedure of extracting usable, tacit, and previously undiscovered knowledge from massive data sets.…”