JMIR Publications

ABSTRACT

Background:

Primary care providers (PCPs) must consolidate diverse sources of data (clinical, laboratory, administrative etc.) to make clinical decisions. As these sense-making tasks become increasingly challenging, artificial intelligence (AI) offer potential to prioritise guideline-recommended tasks based on clinical benefit.

Objective:

To demonstrate a participatory approach to the conception, design, and development of such AI-assisted clinical decision support (CDS) systems for primary care.

Methods:

A mixed methods study was performed, including in-clinic observations at primary care clinics and a focus group involving 20 PCPs in Singapore. The design thinking double diamond process model was applied to define care delivery challenges and conceptualise digital tools. Participants periodically evaluated data saturation, defined as saturation ratio <5% on two consecutive occasions.

Results:

In-clinic observations produced a patient journey map (Figure 1) highlighting current workflows, data sources and challenges. PCPs described consolidating patients’ medical records, presenting complaint, financial and sociobehavioral considerations before formulating a management plan based on multiple guidelines and the latest literature. PCPs also reported that core challenges included rapid guideline adaptation, repeated manual entry across multiple systems, complex claims processes, and limited patient health ownership (Figure 3). Participants further conceived AI tools that could automate eligibility checks for recommended interventions (e.g. screening and vaccinations), deliver just-in-time reminders at the point-of-care, consolidate actionable sociobehavioral data, contextualise relevant literature, and develop personalised risk-based action lists (Figure 5).

Conclusions:

This study describes Singapore’s primary care delivery challenges and identifies parallels from international reports in the United States and Europe. Key providers’ considerations for AI-assisted CDS tools to best support care delivery are described. Additional findings include provider concerns over AI-scribes, highlighting a need for robust evaluation and privacy-preserving approaches. A blended implementation strategy for developed countries was developed using AI agents to aggregate and analyse data, suggest “next best action” lists, and prioritise recommended tasks based on AI-predicted health benefit.