The Impact of Patient Sociodemographic Characteristics on Primary Care Clinician Empathic Communication over Patient Portal Messages

Abstract

Empathic communication in the clinical setting has been associated with lower blood sugar and cholesterol levels among diabetic patients, decreased anxiety, and increased satisfaction, treatment adherence, and trust. Despite recent growth in patient portal use, the expression of empathic communication through patient portal secure messages is not well understood. In particular, disparities may exist in empathic communication to vulnerable populations such as low socioeconomic status patients and racial minorities. To understand how empathic communication is expressed over patient portal messages and determine whether disparities in communication exist, this dissertation used a dataset of 1,367,587 message threads between adult patients and primary care clinicians within a large integrated health system in Pennsylvania to: (1) construct a coding schema to define and assess empathic opportunities initiated by patients and clinician responses to these opportunities within a patient portal messaging environment, (2) evaluate large language models (LLMs) in their capacity to identify empathic opportunities and responses in messages using the coding schema in the model prompts, and (3) investigate the association between patient sociodemographic characteristics and clinician empathic communication identified by an LLM. While the LLMs demonstrated some inconsistency in their capacity to follow the coding schema, they showed moderate to strong overall performance in identifying empathic communication in messages. After expressing an empathic opportunity, Black non-Hispanic and Hispanic patients had lower odds of receiving any clinician response, longer wait times, and lower odds of receiving empathy compared with non-Hispanic White patients. Similarly, Medicaid patients experienced disparities across these three communication outcomes when compared with non-Medicaid patients. Several of these associations substantially attenuated after accounting for clustering of responses from clinician communication teams, though most remained significant. These findings provide evidence of inequities in empathic communication over patient portal messages, which may be driven by both individual clinician biases and variations in the clinician communication teams responding to patients. Addressing disparities may require clinician-level interventions to build awareness around biases in message communication, as well as structural support for overburdened communication teams.