What are the perceptions and concerns of people living with diabetes and NHS staff about potential implementation of AI-assisted screening for Diabetic Eye Disease?

Abstract

Purpose : The English Diabetic Eye Screening Programme (DESP) generates >12 million retinal images annually, requiring grading for diabetic retinopathy by human graders. Some AI systems can identify images with diabetic retinopathy as well as human graders and could reduce grading workload. It is important to understand the attitudes of people living with diabetes (PLD) and healthcare professionals (HCPs) to help aid the implementation process, should this technology be approved for use.

Methods : Two surveys were co-designed with PLD and HCPs from the North East London (NEL) DESP to measure perceptions / concerns around the potential introduction of AI-assisted screening. Likert scale style questions from strongly agree to strongly disagree were used. Questions were categorised into themes, including: general questions about AI-assisted screening, efficiency, data security, trust, impact on workforce, screening experience and patient-practitioner relationship. The PLD survey underwent additional validation through individual interviews at three DESP centres. Validated surveys were distributed from September 1^st 2023 among four DESP centres, via patient groups, relevant charities and the British Association of Retinal Screeners.

Results : To date, 1,205 PLD (mean age 62.7 years, 40.6% female; 65.9% of White, 16.0% Black, and 11.9% Asian ethnicity) and 215 healthcare professionals (43.4 years, 69.7%; 74.0%, 3.3%; 15.4%, respectively) have responded. Overall, 59% (95%CI 54-60%) of PLD believed AI could detect diabetic eye disease equally well in people of different ethnic groups and different ages, compared with 32% (95%CI 26-39%) of HCPs. Two-thirds of PLD (67%, 95%CI 65-70%) and HCPs (67%, 95%CI 61-73%) believed use of AI would save the UK National Health Service money. Trust in AI was lower in HCPs. Both PLD and HCPs were concerned around issues of data security, data regulation, who takes responsibility if errors occur (HCPs 87.9%, 95%CI 83-92%; PLD 72%, 95%CI 70-75%). Differences in Likert scores overall and within each theme by subgroups of age, ethnicity, sex and Townsend scores were examined.

Conclusions : While there was general acceptance of AI technology use within the DESP, there was a need to educate about issues surrounding use. Findings will inform co-designed outreach to support the implementation process.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.