Conversational artificial intelligence intervention to support patients with atrial fibrillation: process evaluation of a randomised controlled trial

Abstract

Abstract Background/Introduction Atrial fibrillation (AF) is a multifaceted disease with patients needing ongoing support and care. European Society of Cardiology AF guidelines recommend the use of digital tools to support AF self-management and the recent popularity of conversational artificial intelligence (AI) suggests its potential in engaging consumers. However, use of these interventions in AF and associated patient perspectives remain underexplored. Purpose Evaluate the acceptability, engagement, barriers, and enablers to the use of a conversational AI intervention designed to support patients with AF in the community. Methods Mixed methods study involving the process evaluation of a 6-month 2-arm randomised controlled trial (RCT) and post-intervention semi-structured interviews. The trial evaluated an intervention comprising seven outreaches with automated telephone calls (conversational AI technology and voice recognition), text messages, emails; and an education website. The 1st outreach was at 24 to 48 hours after discharge from hospital services. Outreaches provided education and behaviour support, assessed AF symptoms and other aspects of self-management, with participants’ responses triggering different system responses, including the identification of red flags requiring escalation. User satisfaction, completeness rate, and other engagement metrics were tracked. Purposive sampling techniques were used to invite patients from the intervention arm to participate in interviews. Interviews were recorded, transcribed verbatim, and analysed in NVivo using thematic analysis. Results 82 participants were allocated to receive the intervention (81 completed, 1 withdrew), with 69% males and a mean age of 64 (SD 11) years. 30 participants completed semi-structured interviews [21 male; mean age 66 (SD 12) years]. On average, across the 7 outreaches, 88% of patients replied "yes" to the question "Did you find the information in this call helpful"? The average outreach completion rate was 61%, starting at 76% for the first call. Participant interviews revealed that the digital health intervention was perceived to address gaps in AF post-discharge care, providing important information that both increased patients’ knowledge of AF and assisted them with modifying lifestyle and self-management. However, participants noted that the limited dialogue flexibility in phone calls, limited personalisation of content and contact (mode and frequency) were barriers to engagement. Participants also reported the need for ongoing adaptation of the intervention to account for the episodic nature of their AF symptoms. Conclusion To date, this is the first mixed methods study exploring user perspectives on a conversational AI intervention to support patients with AF. The capabilities of this technology to address the current gaps in AF models of care remains promising, however, requires further optimisation, such as additional dialogue flexibility and personalisation.