JMIR Publications

ABSTRACT

Background:

Preventing the onset and consequences of diabetes is a top priority for governments and health systems worldwide due to the rising burden and cost of diabetes management. Artificial intelligence (AI) offers immense potential to address public health issues at scale. Our team developed and validated two machine learning models for predicting the 5-year risk of type 2 diabetes onset and the 3-year risk of diabetes-related complications, with the intention to deploy these models at a population-level- in a large and high burden region of Canada.

Objective:

To identify operational principles for meaningful community engagement to responsibly deploy machine learning models for diabetes prevention and management within a diverse population.

Methods:

To achieve study objectives, we organised a co-design workshop with a wide range of stakeholders including patients, caregivers, community organisation representatives, clinicians, and policymakers. Prior to the workshop, we conducted a rapid scan of literature on frameworks and approaches for engaging patients and communities in AI or digital health initiatives, and we carried out purposive outreach activities to ensure diverse perspectives were represented in the workshop. We used a modified nominal group technique to facilitate the co-design session, from which we identified and ranked top principles to guide community engagement in AI for population health, alongside considerations for operationalizing these principles.

Results:

We identified 27 articles in the rapid literature scan, six of which offered specific insights on frameworks and approaches for patient and community engagement in AI in health care. We identified 10 principles from these articles which were presented during the co-design workshop, attended by thirty diverse participants. Through a ranking exercise, workshop participants identified the top six principles of community engagement - trust, equity, accountability, transparency, co-design, and value alignment. Participants expressed that meaningful community engagement at different phases of the AI life cycle, requires inclusivity and diversity, leveraging existing resources, and a centralized and structured approach to decision-making and AI governance.

Conclusions:

Using participatory approaches, our study offered insights into strategies and practices for deploying AI at a population level in a way that centers the values and perspectives of patients and community members. When operationalised, the identified principles can guide meaningful patient engagement as AI continues to be adopted at a rapid pace within health systems. Implementation frameworks for AI that incorporate these principles can advance the responsible and equitable deployment of AI for health conditions of public health importance.