Employing Computational Linguistics to Improve Patient-Provider Secure Email Exchange: The ECLIPPSE Study

Abstract

Electronic patient portals (web-based platforms providing patients immediate access to their health records) are a popular channel that are increasingly available for patients' and providers' communication using secure messaging (SM). These exchanges offer an unprecedented possibility of employing computational linguistics to estimate patients’ health literacy (HL) as well as providers’ language complexity. This chapter describes the NLM-funded ECLIPPSE project and discusses various approaches employed to develop the HL and language complexity measures, the results to date. ECLIPPSE represents an interdisciplinary effort that brought together primary care clinicians with computational linguistics, computers, communication, and health services scientists to harness online SMs exchanged between the patients and the providers to 1) create an automated and valid patient Literacy Profile (LP) that characterizes a patients’ HL, 2) develop an automated and valid physicians’ Complexity Profile (CP) to determine their SMs linguistic complexity, 3) assess whether physicians tailor their SM responses to accommodate limited HL patients communication needs, and 4) design a CP-based automated aid that provides clinicians with real-time feedback to curtail their SMs complexity. Herein, the progress related to Aims 1 and 2 is reported. During the course of LP and CP development, several natural language processing (NLP) tools measuring different language aspects were used and various challenges encountered. To enable successful implementation, solutions were devised for challenges related to data mining, analyses related to NLP and machine learning (ML), and interdisciplinary collaboration. Ultimately, patients’ self-reported HL, physician linguistic complexity, and patients’ HL rated by experts scores were used to train and evaluate ML classification models for building the patients’ LP and physicians’ CP. The ability to rapidly assess patients’ HL at scale could improve clinical care and population health among diverse populations, and determining whether physicians’ CP can be modified to mitigate the untoward effects of limited patient HL is of interest to healthcare delivery systems focused on promoting health equity. This research can reduce HL-related disparities in healthcare by supporting the health systems and healthcare organizations in becoming “health literate”.