Generation of Patient After-Visit Summaries to Support Physicians

Proceedings of the 29th International Conference on Computational Linguistics, pages 6234–6247 October 12–17, 2022. 6234 Generation of Patient After-Visit Summaries to Support Physicians Pengshan Cai,1 Fei Liu,2 Adarsha Bajracharya,3 Joe Sills,3 Alok Kapoor,3 Weisong Liu,4 Dan Berlowitz,4 David Levy,4 Richeek Pradhan,5 Hong Yu1,3,4 1University of Massachusetts, Amherst 2Emory University 3UMass Chan Medical School 4University of Massachusetts, Lowell 5McGill University {pengshancai, hongyu}@cs.umass.edu fei.liu@emory.edu {weisong_liu, dan_berlowitz, david_levy}@uml.edu {adarsha.Bajracharya, alok.kapoor}@umassmemorial.org joe.sills@baystatehealth.org richeekp@gmail.com Abstract An after-visit summary (AVS) is a summary note given to patients after their clinical visit. It recaps what happened during their clinical visit and guides patients’ disease self-management. Studies have shown that a majority of patients found after-visit summaries useful. However, many physicians face excessive workloads and do not have time to write clear and informative summaries. In this paper, we study the prob- lem of automatic generation of after-visit sum- maries and examine whether those summaries can convey the gist of clinical visits. We report our findings on a new clinical dataset that con- tains a large number of electronic health record (EHR) notes and their associated summaries. Our results suggest that generation of lay lan- guage after-visit summaries remains a challeng- ing task. Crucially, we introduce a feedback mechanism that alerts physicians when an au- tomatic summary fails to capture the important details of the clinical notes or when it contains hallucinated facts that are potentially detrimen- tal to the summary quality. Automatic and hu- man evaluation demonstrates the effectiveness of our approach in providing writing feedback and supporting physicians.1 1 Introduction Studies have shown that the majority of patients do not understand their clinical visits (O’Leary et al., 2010). After-visit summary note (AVS) is a sum- mary given to patients after their clinical visit, it is intended to summarize patients’ clinical visits and help their disease self-management (Federman et al., 2018). Compared to clinical notes, an after- visit summary has the following characteristics: 1) it is written in lay person language thus is easy for patients to read and comprehend; 2) it only contains information that patients should be aware of, leaving out redundant details, e.g. unimportant 1Our project page is available at: https://github.com/ pengshancai/AVS_gen lab results, etc. Studies have shown that around 36% of American adults have limited health liter- acy (Kutner et al., 2006), and 94.4% of patients found that lay language after-visit summary helps them understand their clinical visits (Pathak et al., 2020). However, the implementation of after-visit summary is challenging. Many physicians face ex- cessive workloads (West et al., 2018) and do not have time to complete the summaries in a timely manner (Hong et al., 2013). Thus, there is a real need for—and this study contributes to—automatic generation of after-visit summaries to unburdening physicians with complex information workflows. We explore best-performing neural abstractive summarizers to generate after-visit summaries from EHR notes. The summaries are rated by physicians as concise and easy to read. However, they can not be presented directly to patients, as they frequently contain two types of errors: 1) Missing …

Content truncated to 500 words