Stakeholder perspectives on using AI for diagnosis: a co-produced qualitative evidence synthesis

Introduction:

Diagnosis is a cornerstone of medical practice. Worldwide, there is an increase in demand for diagnostic services, and a shortage of healthcare workers. There is diagnostic inequity: patients in low-resource settings have reduced access to diagnostic services.

Artificial intelligence (AI) holds promise to improve diagnostic efficiency, accuracy, and access. However, there is limited uptake of diagnostic AI. Understanding stakeholder perspectives is key to developing consensus, and informing the acceptability, design, and delivery of AI tools across diverse settings.

Goals and methods:

This qualitative evidence synthesis (QES) aims to explore and synthesise stakeholder perspectives on the use of diagnostic AI, following the PRISMA statement.

This QES was co-produced with five patient and public involvement contributors, with lived experience of healthcare services, broadening our group perspective. Articles were appraised using the CASP checklist. Data was managed using Covidence and NVivo. We have taken a ‘Best-fit’-framework approach, using the Non-adoption, Abandonment, Scale-up, Spread, Sustainability (NASSS) framework.

Results:

Searches were conducted in PubMed, OvidMEDLINE, Embase, Scopus, CINAHL and Web-of-Science on 22/02/2023. 29 studies across 30 publications were included. We adapted the NASSS framework for diagnostic AI, developing two new subdomains, 43 themes and 33 subthemes. We identified four distinct stakeholder groups: patients, clinicians, researchers, and healthcare leaders.

We present a conceptual model describing and linking each group, hinging on the establishment of control/trust, engagement, and collaboration. For patients and clinicians, the key question was, “Will I accept AI?”. For researchers, “How can I develop AI that people will use?”. For leaders, “How can I implement AI?”.

Conclusions:

We provide a contextualised framework for diagnostic AI implementation, and a conceptual model describing the stakeholder relationships. These results could direct future implementation and research in diagnostic AI and highlight the importance of qualitative research in providing a voice to each stakeholder group.