Diabetic eye disease (DED) is a leading cause of blindness globally, necessitating annual screening for adults with diabetes. However, adherence to this guideline is historically low, particularly among underserved populations. Barriers include misinformation, logistical challenges, and anxiety. The FDA-approved autonomous AI system, Luminetics® Core, offers a potential solution by enabling point-of-care retinal image analysis. This study aimed to evaluate the association between the implementation of autonomous AI and changes in DED testing adherence rates, examining disparities across patient populations within a large integrated healthcare system, Johns Hopkins Medicine (JHM). The study’s significance lies in its potential to improve access to critical eye care and reduce health disparities in DED screening, ultimately impacting patient outcomes and reducing blindness rates in vulnerable communities. The increasing accessibility and affordability of AI-driven solutions provides a unique opportunity to address these longstanding barriers to care, and thus this research was conducted to asses the real-world effectiveness of such tools in increasing health equity.

Previous studies have reported low adherence rates to annual DED testing guidelines in the United States, ranging from 50-60% among Medicare beneficiaries with diabetes to even lower rates (30-40%) in smaller systems and low-income populations. Identified barriers include misinformation, scheduling difficulties, and patient anxiety. The introduction of autonomous AI for DED diagnosis, specifically the Luminetics® Core system, presented an opportunity to address these challenges by simplifying and improving access to testing. The system's accuracy and ease of use in primary care settings held the potential to increase screening rates and improve early detection and treatment of DED.

This retrospective study, approved by Johns Hopkins School of Medicine's Institutional Review Board, included adult patients with diabetes (n=17,000) managed at JHM primary care sites. Sites were categorized as "AI-switched" (autonomous AI deployment by 2021) or "non-AI." Patient demographic data (gender, age, race, ethnicity, language, insurance, geographic location, and national Area Deprivation Index – ADI) were extracted from electronic health records. A propensity score weighting analysis was employed to compare changes in adherence rates from 2019 (pre-AI) to 2021 (post-AI) between the two site categories. This method accounted for variations in sociodemographic characteristics to ensure comparability. Statistical analyses, including chi-squared tests, difference-in-differences analysis, and Poisson regression models, were conducted using SAS and Stata software to assess changes in adherence rates and identify disparities across subgroups. Inverse probability weighting was used to adjust for confounding variables and achieve comparable groups across sites and years. Robust standard errors were used to account for clustering among patients seen in both 2019 and 2021. The primary outcome was the change in DED testing adherence rates between 2019 and 2021, stratified by patient demographics and geographical location.

The study revealed three key findings. First, AI-switched sites experienced a 6.7 percentage point greater increase in DED testing than non-AI sites from 2019 to 2021 (p < 0.001). Second, Black/African Americans showed a 12.2 percentage point increase in adherence at AI-switched sites, contrasting with a 0.6 point decrease in non-AI sites (p < 0.001). This highlights the potential of autonomous AI to improve access to retinal evaluation for historically disadvantaged groups. Third, autonomous AI was associated with improved health equity; the adherence rate gap between Asian Americans and Black/African Americans decreased from 15.6% in 2019 to 3.5% in 2021. The propensity score weighting analysis showed a 7.6 percentage point higher increase in adherence at AI-switched sites compared to non-AI sites (p < 0.001). Subgroup analyses revealed substantial improvements in adherence rates among Black/African Americans, Medicaid patients, and patients with high ADI scores in AI-switched sites. While some subgroups (e.g., Asian, American Indian or Alaska Native, non-English speakers, military insurance) showed less improvement, the overall impact of AI on increasing access to care and reducing disparities was significant. The overall adherence rate increased from 42.2% in 2019 to 48.0% in 2021, with AI-switched sites reaching 54.5% adherence in 2021. Table 1 provides baseline demographic data, showing significant differences between AI-switched and non-AI sites in 2019 and 2021 for variables like gender, race, ethnicity, insurance, and ADI quartile. Table 2 details changes in DED adherence rates by subgroup from 2019 to 2021. The large disparities in adherence rates observed between different demographic groups in 2019 were significantly reduced by 2021 in the AI-switched sites.

This real-world study demonstrates a strong association between the implementation of autonomous AI for DED testing and increased adherence rates, particularly among historically underserved populations. The significant improvements in adherence, especially within the Black/African American community, highlight the potential of AI to address health equity challenges. The reduction in adherence rate gaps between various demographic groups further emphasizes this effect. These findings suggest that the accessibility and ease of use of autonomous AI in primary care settings can overcome some of the barriers that prevent timely DED screening. The improvement in HEDIS metrics and potential for enhanced payer reimbursements adds further value to the integration of this technology. Further research is needed to fully understand the mechanisms behind the observed improvements in health equity and to explore patient perceptions and acceptance of AI-driven healthcare. Although this study addresses some limitations inherent in observational studies through the use of propensity score weighting, a prospective randomized clinical trial could provide further confirmation of a causal relationship between autonomous AI and improved adherence.

The study's results indicate a strong association between autonomous AI implementation for DED testing and increased adherence rates and health equity in a large integrated health system. The observed improvement, particularly among historically disadvantaged populations, underscores the technology's potential to expand access to crucial eye care. Future research should focus on prospective randomized clinical trials to establish causality and on qualitative studies to understand patient perspectives and address potential barriers to adoption. Further investigation into the cost-effectiveness and long-term impact of this technology is warranted.

This study's retrospective nature and the focus on a largely metropolitan population limit the generalizability of findings to rural settings. The study's observation of an association, rather than establishing causality, necessitates further research, such as a prospective randomized controlled trial. The lack of exploration into patient trust and acceptance of AI technologies represents another limitation.