Declined or refused?: Electronic health record documentation of patients who do not proceed with recommended care by provider characteristics.

Abstract

e13725 Background: Labeling a patient as "refusing" care can be stigmatizing; alternative terms more accurately and neutrally capture a patient's barriers, preferences, and decision-making. This study investigates if the use of “refused” as opposed to “declined” to describe these scenarios in cancer care electronic health record (EHR) notes varies by provider training and demographic characteristics. Methods: We developed a natural language processing model to identify documentation of “refusing” and “declining” care from a population of nearly 200,000 EHR notes of patients with cancer from 2022-2023 in a large upper Midwest health system containing both academic and community sites. Provider information was abstracted from publicly available licensing data. Providers in cancer treatment specialties (e.g. medical oncology, radiation oncology, surgical subspecialties) were included if they wrote a minimum of three notes between 2022-2023 that described a patient “refusing” or “declining” care. The percentage of notes containing “refuse” terminology per provider was analyzed by univariate linear regression models with provider characteristic predictors (age, gender, specialty, type [physician/NP/PA], time since training completion, and geographic region of terminal training [physician/PA only]). Results: 247 providers were represented: 68.8% physicians, 15.4% PAs, and 15.8% NPs. Mean age and time since training completion were 44 and 11 years, respectively. Provider gender was 56.7% female and 43.3% male. Medical oncology/hematology (46.2%), radiation oncology (9.3%), and general surgery (9.3%) specialties were most common, and the majority completed training in the Midwest (119, 57.2%). Average use of refusal language varied significantly by medical specialty: Compared to medical oncology/hematology (25.2%), use was highest among those in general surgery (39.0%, p = 0.02) and lowest among those in gynecologic oncology (6.9%, p = 0.01). Use of refusal language was also higher among male providers than female providers (29.8% vs. 20.0%, p = 0.003). Conclusions: Provider specialty and gender were associated with differing frequency of refusal language use in cancer care notes. This suggests that provider lived experiences, training, and/or specialty-specific culture or resources may play a role in the way stigmatizing language is used in EHR notes as well as its potential impact on patients and other providers.