Limited health literacy places individuals at greater risk of type 2 diabetes and its complications, making limited health literacy a critical clinical and publi health problem. As healthcare becomes increasingly dependent on electronic communications, patients with limited health literacy may have difficulty communicating by email with their clinician or understanding the clinician's emailed replies or instructions. This proposal will use computational linguistics to examine how diabetes patients with a variety of health literacy levels interact with their clinicians via patient portals, will explore whether linguistic gaps between patients and clinicians are associated with diabetes outcomes, and will create a feedback tool to assist clinicians to better accommodate diabetes patients' communication needs.
Limited health literacy (HL) places individuals at greater risk of type 2 diabetes (DM2) and its complications, is a marker of vulnerability, and presents a critical clinical and public health problem. To be health literate in the 21st century, patients will need a certain level of linguistic facility, in combination with technical skills, to access services via online patient portals. Our research has shown that DM2 patients with limited HL are actively using patient portals. However, as healthcare becomes increasingly dependent on electronic communications (e.g., secure messages via internet-based patient portals), patients with limited HL may have difficulty communicating electronically with their clinician or understanding their clinician's secure message responses or instructions. For clinicians to electronically provide meaningful and actionable information and support, their secure messages must be written in an easily comprehended style. Few studies have examined how patients with limited HL interact with their healthcare providers via patient portals. This trans-disciplinary proposal, involving a team of health services researchers, health communication scientists, and computational linguists, will focus on a population of ethnically diverse DM2 patients and their primary care providers from 1) a large, integrated group model HMO with a well-developed patient portal and 2) a county-run, integrated public (safety net) delivery system with a newly launched electronic health record and patient portal. Our study is designed around a conceptual framework promoted most recently by the Institute of Medicine: overcoming the challenges LHL patients face in managing DM2 requires that healthcare systems, and their clinicians, make accommodations to meet patients' communication needs. The degree of linguistic "mismatch" observed in secure message exchanges between DM2 patients and their providers, measured using computational linguistics, will serve as one indicator of the extent to which providers are, or are not, making such accommodations. Our specific aims are to (Aim 1) develop and validate a novel, automated linguistic complexity profile (LCP) to assess secure message content generated by DM2 patients and their providers via patient portals. We will employ natural language processing (NLP) to develop and validate the LCP, based on secure messages and data from >200,000 DM2 patients. The LCP will demonstrate construct validity with patient HL and patient reports of provider communication, and will be associated with DM2 outcomes; (Aim 2) examine whether concordance between provider and patient LCP is associated with adherence among DM2 patients newly prescribed insulin or antidepressants; (Aim 3) characterize the collaborative nature of exchanges between providers and low LCP patients, using mixed methods, to enhance our understanding of communication in the critical period surrounding initiation of insulin or antidepressants; (Aim 4) create an automated, LCP-based prototype to provide real-time feedback to providers while writing secure messages to reduce linguistic complexity and better accommodate DM2 patients' linguistic skills and HL.