Generating Accurate Electronic Health Assessment from Medical Graph

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 3764–3773 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 3764 Generating Accurate Electronic Health Assessment from Medical Graph Zhichao Yang1, Hong Yu1,2 1 College of Information and Computer Sciences, University of Massachusetts Amherst 2 Department of Computer Science, University of Massachusetts Lowell zhichaoyang@cs.umass.edu hong yu@uml.edu Abstract One of the fundamental goals of artificial in- telligence is to build computer-based expert systems. Inferring clinical diagnoses to gener- ate a clinical assessment during a patient en- counter is a crucial step towards building a medical diagnostic system. Previous works were mainly based on either medical domain- specific knowledge, or patients’ prior diag- noses and clinical encounters. In this paper, we propose a novel model for automated clin- ical assessment generation (MCAG). MCAG is built on an innovative graph neural network, where rich clinical knowledge is incorporated into an end-to-end corpus-learning system. Our evaluation results against physician gen- erated gold standard show that MCAG signif- icantly improves the BLEU and rouge score compared with competitive baseline models. Further, physicians’ evaluation showed that MCAG could generate high-quality assess- ments. 1 Introduction Electronic health record (EHR) is widely used by hospitals in the United States and other countries, resulting in an unprecedented amount of digital data or EHRs associated with patient encounters. In recent years, secondary use of EHRs has helped advance EHR-related computational approaches to foster precision medicine and a learning health system (Evans, 2017). Rich clinical information is documented in the EHRs. Among many structures and formats in EHRs, a problem-oriented SOAP (Subjective, Ob- jective, Assessment, and Plan) structure is com- monly used by providers (Podder et al., 2020). Fig- ure 1 illustrate an example of a SOAP note for an outpatient encounter. Typically, Chief Complaint includes a brief description of a patient’s condi- tions and the reasons for the visit. The Subjec- tive section is a detailed report of the patient’s cur- rent conditions, such as source, onset, and duration of symptoms, mainly based on the patient’s self- report. This section usually includes a history of present illness and symptoms, current medications, and allergies. The Objective section documents the results of physical exam findings, laboratory data, vital signs, and descriptions of imaging re- sults. The Assessment section typically contains medical diagnoses and reasons that lead to medi- cal diagnoses. The assessment is typically based on the content from the chief complaint, and the subjective and objective sections. The Plan section addresses treatment plans based on the assessment. Inferring clinical diagnosis to generate an assess- ment is a crucial step during the patient encounter. Earlier expert systems were mainly knowledge- based, typically using decision rules. Later, ma- chine learning approaches were developed, mainly used longitudinal electronic health records (EHR) to predict ICD codes (Subotin and Davis, 2014; Amoia et al., 2018), the diagnostic codes assigned to EHRs after each patient’s visit or encounter. However, ICD codes are used mainly for billing purposes and have limitations (e.g., incomplete as- signment) when used …

Content truncated to 500 words