Impact of artificial intelligence assessment of diabetic retinopathy on referral service uptake in a low resource setting: The RAIDERS randomized trial

Abstract

Objective

This trial was designed to determine if artificial intelligence (AI)-supported diabetic retinopathy (DR) screening with immediate results improved referral uptake in Rwanda.

Design

The RAIDERS (Rwanda Artificial Intelligence for DiabEtic Retinopathy Screening) study was an investigator-masked parallel-group, randomized controlled trial.

Participants

Have known diabetes, are aged >= 18 and require referral for DR based on AI.

Methods

The RAIDERS study screened for DR using retinal imaging with AI interpretation implemented at four facilities from March-July 2021. Eligible participants were randomly assigned (1:1) to immediate feedback of AI grading (intervention) or communication of referral advice after human grading was completed three-to-five days after the initial screening (control).

Main Outcome Measures

Difference between study groups in the rate of presentation for recommended referral services within 30 days of being informed of the need for a referral visit.

Results

Of the 823 clinic patients with diabetes who met inclusion criteria, a total of 275 participants (33.4%) screened positive for referrable DR based on AI screening and were randomized for inclusion in the trial. Study participants (mean age 50.7 years, 58.2% women) were randomized to intervention (n=136, 49.5%) or control (n=139, 50.5%). There were no significant inter-group differences at baseline, and 100% had main outcome data available for analyses. Referral adherence was statistically significantly higher in the intervention (70/136=51.5%) versus control (55/139=39.6%; p=0.048), a 30.1% increase. Older age (Odds ratio=1.04, 95% Confidence Interval: 1.02-1.05, p<0.0001), male sex (Odds ratio=2.07, 95% Confidence Interval: 1.22-3.51, p=0.007), rural residence (Odds ratio=1.79 (95% Confidence Interval: 1.07-3.01, p=0.027), and intervention group membership (Odds ratio=1.74, 95% Confidence Interval: 1.05-2.88, p=0.031) were statistically significantly associated with acceptance of recommended referral in multi-variable analyses.

Conclusion

Immediate feedback on referral status based on AI-supported screening was associated with statistically significantly higher referral adherence compared to delayed communications of results from human graders, necessitated by time for human image review in this setting. These results provide evidence for an important benefit of AI screening in promoting adherence to prescribed treatment for diabetic eyecare in sub-Saharan Africa.