Computing Reviews

Evaluation of an automated knowledge-based textual summarization system for longitudinal clinical data, in the intensive care domain
Goldstein A., Shahar Y., Orenbuch E., Cohen M. Artificial Intelligence in Medicine82 20-33,2017.Type:Article
Date Reviewed: 02/22/18

This paper examines the potential of the Clinitext system [1] for the production of clinical summaries. Clinitext is designed to be general and not specific to any clinical domain; details on Clinitext are provided in [1]. In particular, the authors consider the case of “handoff” summaries, which are required when a patient is transferred from one department to another. The study compares summaries written by physicians with those prepared by Clinitext. Absent a “gold standard” for such summaries, the authors chose two evaluative approaches. In one, the summaries were examined by physicians; in the other, the summaries were assessed for completeness and correctness. For the comparison, 31 MD-written letters and 31 Clinitext-generated ones were compared. The letters, all about one page in length, related to the release of patients from an intensive care unit (ICU).

For the functional evaluation of the letters, two ICU experts were asked to generate a list of five questions about clinical decisions that should be addressed in the handoff letter. The authors found that Clinitext letters were more complete than MD written ones, though the omissions in the MD letters were “non-important.”

The authors do not conclude that Clinitext-generated letters can replace MD-written ones, mainly because Clinitext “lacks the general understanding of the patient’s clinical course.” However, the authors do feel that a Clinitext-generated letter can form a useful template for the final letter. The paper is meticulously written and the reader is able to follow the authors’ reasoning with no difficulty. The paper is highly recommended to anyone interested in text generation.


1)

Goldstein, A.; Shahar, Y. Implementation of a system for intelligent summarization of longitudinal clinical records. In: Process support and knowledge representation in health care (AIME 2013 Joint Workshop, KR4HC 2013/ProHealth 2013). 68-82, Springer, 2013, https://link.springer.com/chapter/10.1007/978-3-319-03916-9_6.

Reviewer:  J. P. E. Hodgson Review #: CR145875 (1806-0345)

Reproduction in whole or in part without permission is prohibited.   Copyright 2024 ComputingReviews.com™
Terms of Use
| Privacy Policy