Integrating AI-based triage in primary care: a qualitative study of Swedish healthcare professionals’ experiences applying normalization process theory

Abstract

Given the growing challenges in primary care, including high demand and workforce shortages, artificial intelligence (AI)-based triage applications are being explored as a means of alleviating workloads. While the potential of AI in this context is widely acknowledged, there is still limited empirical research on how such tools become embedded in routine practice, especially from healthcare professionals’ perspectives. This study focused on exploring healthcare professionals’ experiences of using an AI-based triage application in primary care. The study had a qualitative design with a deductive approach, involving 14 healthcare professionals (physicians, nurses, psychologists, and a social worker). Data were collected through semi-structured interviews. The data were analyzed through directed qualitative content analysis and categorized in accordance with normalization process theory (NPT). The results of this study were framed by the NPT constructs: Coherence, Cognitive Participation, Collective Action, and Reflexive Monitoring. Professionals aimed to achieve Coherence by making sense of the AI triage application’s purpose and potential role in practice; however, insufficient initial information was reported to hinder a full understanding and meaningful engagement with the tool. The work of building and sustaining engagement (Cognitive Participation) was challenged by staff’s perceptions that use of the triage application was optional: this hindered the development of a “community of practice”. During Collective Action, professionals tended to rely more on patients’ free-text descriptions than on the AI-generated summaries, reflecting concerns about the application’s adequacy, compared to clinical judgment. Finally, Reflexive Monitoring revealed persistent uncertainty about the application’s value, with professionals questioning its usefulness, effectiveness, and equitable accessibility across patient groups. This study found that, although the AI-based triage application appeared, at first, to be integrated into primary care practice, it was not embedded fully within professional and organizational routines. Despite a broad acceptance of digitalization among healthcare professionals, several barriers to meaningful use were identified. These included concerns about insufficient organizational and policy support, which hindered the application’s full integration into everyday workflows. The study results suggest that further efforts are needed to overcome these barriers and support the successful normalization of the AI-based triage application into routine practice.

Ähnliche Arbeiten