JMIR Publications

ABSTRACT

Background:

In recent years, the health data collected during clinical care are often repurposed for various secondary use through Clinical data warehouses (CDWs), which interconnect disparate data from different sources. The majority of clinical data is stored in unstructured text format. NLP, which implements algorithms that can operate on a scale as massive as unstructured textual data, have the potential to make clinical data more accessible.

Objective:

The objective of this paper is to provide an overview of studies applying Natural Language Processing (NLP) to clinical text from CDWs.

Methods:

We searched relevant articles in three bibliographic databases: PubMed, Association of Computational Linguistics (ACL) Anthology and Google Scholar. We reviewed the titles and abstract and included articles that focused on NLP applied to data from CDWs. Included articles were reviewed a second time, by reading the whole article, to retrieve the following information: NLP tasks, NLP methods used, data language and CDW involved in the studies.

Results:

We identified a total of 1,251 articles, published between 1995 and 2021, of which 196 met inclusion criteria. For studies that applied NLP to text, information extraction was the most predominant task (55.33%). Symbolic methods were the most common NLP methods (52.7%), followed by machine learning (32.3%) and deep learning (15%). These NLP methods were mostly applied to data written in English (75.63%).

Conclusions:

Although the use of NLP in CDWs is growing, CDWs are still underexploited for clinical NLP researches and there is remaining challenges in this field. Clinical NLP is an effective strategy for accessing, extracting and transforming data from CDWs. Information retrieved with NLP can assist in clinical research and impact clinical practice.