JMIR Publications

ABSTRACT

Background:

Information overflow is a common problem of today’s clinical environment which can be mitigated by summarizing clinical data. While there are several solutions for clinical summarization, there is a lack of reviews synthesizing the available knowledge scattered in the field.

Objective:

The study aims to identify the state-of-the-art solutions of clinical summarization, to analyze their capabilities, and to identify what properties a clinical summarization system should ideally have.

Methods:

A scoping review is concluded over two databases using well-defined inclusion and exclusion criteria extended by articles found through a chain of citations retrieved from the originally included articles. All included articles are analyzed through the lens of pre-defined aspects.

Results:

127 articles are included in the analyses, among which several application fields are represented. While a slight majority (n=72) of the methods take structured data as input, there is an increasing trend for using textual sources (n=65) or including both types of input (n=13). The review could identify a large variety of proposals for summarization in healthcare, including extractive (n=13) and abstractive summarization (n=19), topic modeling (n=5), summary specification (n=11), concept and relation extraction (n=29), visual design considerations (n=59), and complete pipelines (n=7) using information extraction, selection, and communication. Regarding summary communication, graphical displays (n=53), short text (n=41), static reports (n=7), and problem-oriented views (n=7) are the most common types. While temporality and uncertainty information is usually not conserved in the majority of the works (n=74 and n=113), some works present solutions to treat this information as well. The reviewed works present n=11 systems in real clinical settings and a variety of evaluation techniques and metrics, including quantitative measures and qualitative evaluations with a median of 15 human subjects.

Conclusions:

While there are proposals to help healthcare professionals (HCPs) in summarizing patient data, not many of them are actively applied in healthcare. A bottleneck for their application might be the lack of reliable and realizable evaluation schemes, the loss of important information (e.g., temporality, medical knowledge, and uncertainty), and that research is fragmented in the domain. The existing summarization methods which received the most attention in research are extractive and abstractive text summarization techniques and visual design. The review also defines the “collect – select – communicate” framework describing information extraction, information selection, and summary communication as key steps of summarization, and finds very few article addressing all these steps.