JMIR Publications

ABSTRACT

Background:

AI (artificial intelligence) systems for automating medical history-taking and triage can potentially improve patient flow through healthcare systems. Despite numerous studies of AI systems showing good performance only a small number of AI systems are integrated into healthcare practice. To better understand how AI systems for automating medical history-taking and triage could be used and create value in this context, it is important to identify the current state of knowledge including the readiness of these AI systems, the facilitators and barriers to their implementation, and how the perspectives of different stakeholders are considered in their development and implementation.

Objective:

The aim of this scoping review was to map and summarize the empirical research on AI systems for automating medical history-taking and triage in healthcare.

Methods:

The scoping review followed the principles of Arksey and O’Malley’s (2006) framework. It was reported according to the PRISMA-ScR guidelines. Research publications were systematically searched in five databases: PubMed, CINAHL, PsycINFO, Scopus and Web of Science. A protocol was established prior to conducting the scoping review.

Results:

A total of 1248 research publications were identified and screened through the database search. 86 were included. The majority of studies (n=63, 74%) were published between 2020 and 2022, with a significant number of studies focusing on emergency care (n=32, 37%). Other clinical contexts included radiology (n=12, 14%) and primary care (n=6, 7%). A large group of the studies did not specify the clinical context (n=15, 17%). Most studies were retrospective (n=31, 36%) or did not specify the study design (n=34, 29%,. The most common type of AI system was hybrid models (n=68, 79%) and the most common tasks performed by the AI systems were forecasting (n=40, 47%) and recognition (n=36, 42%). The sample population in many studies was patients (n=70, 81%). However, only one study explored patients' views on AI-based medical history-taking and triage and two studies explored healthcare professionals' perspectives. There were a few studies that demonstrated or validated the AI systems in relevant clinical settings (i.e., described real-time model testing, workflow implementation, clinical outcome evaluation, or model integration) (n=6, 7%). The facilitators and barriers to the introduction of the AI systems described in the studies were linked to four themes: technical aspects of the AI systems, contextual and cultural aspects, end-users and evaluation of AI systems.

Conclusions:

By shedding light on current trends, stakeholder perspectives, innovation development stages, and influencing factors, this review contributes to the understanding of AI system implementation in healthcare. The identified gaps in the literature concerning stakeholders’ perspectives, the research on AI systems for automating medical history-taking and triage suggest opportunities for further research.