Perception of AI Symptom Models in Oncology Nursing: Mixed Methods Evaluation Study

Abstract

Background: Patients undergoing cancer treatment experience a significant symptom burden. The standard process of symptom management includes patient reporting and clinical response following symptom escalation. Emerging predictive symptom models use artificial intelligence (AI) components of machine learning and deep learning to identify the risk of symptom deterioration, facilitating earlier intervention to prevent downstream effects. However, integrating predictive symptom models into clinical practice will require oncology nurses to adopt innovative approaches. Objective: This study aims to explore oncology nurses' perceptions of the use of predictive symptom models in cancer care and the factors influencing the adoption of this symptom care innovation. Methods: The evaluation was guided by the Rogers Diffusion of Innovation Theory, which describes the process of how individuals adopt new technologies. The investigators developed an interview guide that asked oncology nurses to rate their perceptions of AI symptom models on a Likert scale. Participants were also asked to provide qualitative comments to support their ratings for each question, in order to better understand the key factors that would influence AI predictive model adoption. Investigators analyzed demographic data and Likert ratings with descriptive statistics. Qualitative analysis of participant comments included content analysis and inductive coding to identify themes. Nurses' perception of factors that would influence the adoption of AI symptom models, based on the Rogers theory, included relative advantage, compatibility, complexity, trialability, and observability. Results: Responses of 15 oncology nurses with more than 1 year of experience in oncology were analyzed. There was high agreement among nurse participants that an AI model could improve symptom management for patients with cancer (n=10, 67%) and increase early intervention to prevent the escalation of symptoms (n=12, 86%). All participants (N=15) agreed that receiving symptom information would be helpful. Nearly three-quarters of participants (n=11, 73%) endorsed that the information would save time. Most (n=12, 80%) recommended that clinicians receive information about the predicted symptom deterioration of their patients. Among open-ended responses, key themes were consistent with factors identified in the Diffusion of Innovation theory including: (1) perceptions related to the AI model (compatibility or complexity), (2) nurses' perception of patients' benefit (observability), (3) improved clinical processes (relative advantage or observability), (4) apprehension over model accuracy and impact (compatibility or trialability or observability), and (5) implementation or adoption (trialability or complexity or observability). Conclusions: Oncology nurses agree that predictive symptom models could help improve symptom management for patients undergoing cancer treatment. However, nurses noted that transparency in the factors included in the AI model was essential, that nurses should be involved in the development and testing of models, and that the observability of the benefit in symptom care would need to be evident for ultimate adoption.

Ähnliche Arbeiten