AI Screening for Vision Loss from Diabetes

Overseen By Mozhdeh Bahrainian

Age: 18+

Sex: Any

Trial Phase: Academic

Sponsor: University of Wisconsin, Madison

Disqualifiers: Diabetic eye disease, others

No Placebo Group All trial participants will receive the active study treatment (no placebo)

Compensation: Varies

Number of Visits: Unknown

4000 Participants Needed

What You Need to Know Before You Apply

What is the purpose of this trial?

This study aims to investigate whether a novel artificial intelligence based screening strategy (AI-Based point of caRe, Incorporating Diagnosis, SchedulinG, and Education or AI-BRIDGE), which allows primary care providers to screen patients for vision-threatening diabetic eye disease in the primary care clinic, improves screening and follow-up care rates across race/ethnicity groups and reduces racial/ethnic disparities in screening.

Will I have to stop taking my current medications?

The trial information does not specify whether you need to stop taking your current medications.

Is the AI system for detecting diabetic retinopathy safe for humans?

The AI systems for detecting diabetic retinopathy, like AIDRScreening and others, have been tested in various studies and are generally considered safe for use in humans, as they involve analyzing images of the eye without direct physical intervention. ^{1 2 3 4 5}

How is the AI-BRIDGE treatment for vision loss from diabetes different from other treatments?

AI-BRIDGE is unique because it uses artificial intelligence to screen for vision loss due to diabetes, specifically targeting diabetic retinopathy (damage to the retina caused by diabetes) and potentially other eye conditions, which is different from traditional methods that rely on manual examination by healthcare professionals. ^{5 6 7 8 9}

What data supports the effectiveness of the treatment AI-BRIDGE for vision loss from diabetes?

Research shows that artificial intelligence (AI) systems are highly promising for detecting diabetic retinopathy (a diabetes-related eye disease) from eye images and may predict its progression. These AI tools have been validated in various settings, demonstrating accuracy in identifying eye conditions related to diabetes. ^{5 8 10 11 12}

Who Is on the Research Team?

Roomasa Channa

Principal Investigator

UW School of Medicine and Public Health

Are You a Good Fit for This Trial?

This trial is for individuals with diabetes who are at risk of vision loss. It's focused on helping those in socioeconomically disadvantaged communities. Participants should be willing to undergo screening using the AI-BRIDGE system in a primary care setting.

Inclusion Criteria

I do not have any eye problems caused by diabetes.

Not had an eye exam in the prior year

I am older than 21 years.

Timeline for a Trial Participant

Screening

Participants are screened for eligibility to participate in the trial

2-4 weeks

AI-BRIDGE Implementation

AI-based eye screening program called AI-BRIDGE is implemented. Eye photos are obtained and reviewed using an AI algorithm. Patients with referrable diabetic retinopathy are detected and assisted with scheduling follow-up visits.

6 months

Regular primary care visits

Usual Care Screening

Primary care providers refer patients with diabetes to an eye care provider for a dilated eye exam. Patients receive educational materials.

6 months

Follow-up

Participants are monitored for follow-up with recommended eye care and screening effectiveness.

up to 6 months

What Are the Treatments Tested in This Trial?

Interventions

• AI-BRIDGE

Trial Overview The study tests an artificial intelligence-based strategy, AI-BRIDGE, designed to help doctors screen for diabetic eye disease during regular visits. The goal is to see if it improves screening rates and reduces racial/ethnic disparities.

How Is the Trial Designed?

Group I: AI-BRIDGE Experimental Treatment 1 Intervention

AI-based eye screening program called AI-Based point of caRe, Incorporating Diagnosis, SchedulinG, and Education (AI-BRIDGE). Eye photos of the patients will be obtained in the primary care clinic during a patient's regular primary care visit by a trained technician. Images will be reviewed using autonomous artificial-intelligence (AI) algorithm (Digital Diagnostics). Patients with referrable diabetic retinopathy are detected, and assisted with scheduling an in-person follow-up eye care visits. All patients irrespective of diabetic retinopathy status are also provided culturally adapted educational material on diabetic eye disease.

Group II: Usual Care Screening Active Control 1 Intervention

Primary care providers refer patients with diabetes to an eye care provider for a dilated eye exam. Patients are provided with culturally adapted diabetic eye disease educational materials similar to that provided to patients in the AI-BRIDGE group.

Who Is Running the Clinical Trial?

University of Wisconsin, Madison

Lead Sponsor

Trials

1,249

Recruited

3,255,000+

National Eye Institute (NEI)

Collaborator

Trials

572

Recruited

1,320,000+

Published Research Related to This Trial

The global prevalence of diabetes is projected to reach 700 million individuals in the next 25 years, increasing the risk of vision loss from diabetic eye disease and highlighting the need for innovative detection tools.

Artificial intelligence (AI) has shown great promise in detecting diabetic retinopathy (DR) by analyzing fundus photographs, using machine learning to grade DR and potentially predict its progression, which could help alleviate the burden on eye care professionals.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Artificial Intelligence in the assessment of diabetic retinopathy from fundus photographs. Gilbert, MJ., Sun, JK.[2021]

A study involving 6146 patients showed that using AI for non-mydriatic fundus photography in diabetic retinopathy screening has a high sensitivity of 97.3% and a kappa coefficient of 0.75, indicating strong agreement with ophthalmic diagnoses.

The AI's consistency with fluorescein angiography was moderate, with a kappa coefficient of 0.53 and a coincidence rate of 66.9%, suggesting that AI can effectively assist in early detection of diabetic retinopathy, which is crucial for preventing blindness.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Application and observation of artificial intelligence in clinical practice of fundus screening for diabetic retinopathy with non-mydriatic fundus photography: a retrospective observational study of T2DM patients in Tianjin, China. Hao, Z., Xu, R., Huang, X., et al.[2022]

The AI algorithm demonstrated high accuracy in detecting referable diabetic retinopathy (RDR), referable macular diseases (RMD), and glaucoma suspect (GCS) from fundus images, achieving sensitivity and specificity rates above 90% across different datasets.

In the PUMCH dataset, the AI achieved a sensitivity of 95% for RDR, 92% for RMD, and 95% for GCS, indicating its robustness and reliability in both community and hospital screening scenarios.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Validating automated eye disease screening AI algorithm in community and in-hospital scenarios. Han, R., Cheng, G., Zhang, B., et al.[2022]

The automated diabetic retinopathy screening tool, DART, demonstrated high sensitivity (94.6%) and a strong negative predictive value (98.1%) in analyzing ocular fundus photographs from 1123 diabetic eye exams, indicating its effectiveness in identifying potential cases of diabetic retinopathy.

With a specificity of 74.3% and an area under the ROC curve of 0.915, DART shows promise for implementation in Chile's national diabetic retinopathy screening program, suggesting it can be a reliable option in diverse healthcare settings.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Clinical validation of an artificial intelligence-based diabetic retinopathy screening tool for a national health system. Arenas-Cavalli, JT., Abarca, I., Rojas-Contreras, M., et al.[2023]

The MONA.health artificial intelligence screening software demonstrated high diagnostic performance for detecting diabetic retinopathy (DR) and diabetic macular edema (DME), achieving an area under the curve (AUC) of 97.28% for DR and 98.08% for DME on a private test set.

The software maintained strong sensitivity (90.91%) and specificity (94.24%) across various subgroups, although sensitivity was slightly lower for individuals over 65 years old and Caucasians, indicating its effectiveness across diverse populations.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Artificial Intelligence Software for Diabetic Eye Screening: Diagnostic Performance and Impact of Stratification. Peeters, F., Rommes, S., Elen, B., et al.[2023]

The AIDRScreening system demonstrated high sensitivity (86.72%) and specificity (96.09%) in detecting referable diabetic retinopathy (RDR) among 962 participants with diabetes, indicating it is an effective tool for screening.

The system significantly improved diagnostic efficiency, achieving a 16.57% higher diagnostic accuracy compared to manual grading while reducing the average diagnostic time by 37.32%.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Performance of the AIDRScreening system in detecting diabetic retinopathy in the fundus photographs of Chinese patients: a prospective, multicenter, clinical study. Yang, Y., Pan, J., Yuan, M., et al.[2023]

In a study involving 54 asymptomatic diabetic patients, an autonomous AI system demonstrated strong accuracy in diagnosing diabetic retinopathy (DR), identifying various stages of the disease effectively compared to human specialists.

The AI system was able to classify patients into categories of DR severity, showing that it can serve as a reliable screening tool for referable DR, which is crucial for early intervention and management.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Comparison of early diabetic retinopathy staging in asymptomatic patients between autonomous AI-based screening and human-graded ultra-widefield colour fundus images. Sedova, A., Hajdu, D., Datlinger, F., et al.[2023]

The automated AI diagnostic system demonstrated high accuracy in identifying glaucomatous optic neuropathy (GON), achieving 93.18% accuracy in one dataset and 91.40% in another, outperforming manual graders in sensitivity.

The AI system maintained strong performance even in patients with retinal comorbidities, achieving an accuracy of 87.54% and 93.81% in those groups, indicating its robustness across different clinical scenarios.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

External validation of a deep learning detection system for glaucomatous optic neuropathy: a real-world multicentre study. Qian, X., Xian, S., Yifei, S., et al.[2023]

The AI-based screening tool for Diabetic Retinopathy (DR) demonstrated exceptional performance, achieving a sensitivity of 99.21% and specificity of 97.59% on the Kaggle dataset, indicating its high accuracy in identifying referable DR.

In real-world primary care settings, the tool maintained strong performance with a sensitivity of 92.3% and specificity of 94.8%, suggesting it is effective for early detection and could significantly aid in preventing blindness from DR.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Automated diabetic retinopathy screening for primary care settings using deep learning. Bhuiyan, A., Govindaiah, A., Deobhakta, A., et al.[2022]

The AI-based screening tool EyeWisdom®DSS demonstrated high sensitivity (91.0%) and good specificity (81.3%) for detecting diabetic retinopathy (DR) in a study of 549 type 2 diabetes patients, indicating its effectiveness as a screening method.

EyeWisdom®MCS was particularly effective at identifying patients without DR, with a specificity of 92.4%, suggesting that AI screening can be a valuable resource in areas with limited access to eye care.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Efficacy of artificial intelligence-based screening for diabetic retinopathy in type 2 diabetes mellitus patients. Pei, X., Yao, X., Yang, Y., et al.[2022]

The Comprehensive Artificial Intelligence Retinal Expert (CARE) system demonstrated high image ability (99.06%) and good specificity (93.99%) for detecting diabetic retinopathy (DR) in a study of 443 participants, but its sensitivity varied by severity, being highest for more-than-mild DR (78.97%).

The CARE system is particularly effective for detecting more-than-mild diabetic retinopathy, addressing the challenge of timely diagnosis in Chinese community healthcare centers where ophthalmic expertise is limited.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Evaluation of an Artificial Intelligence System for the Detection of Diabetic Retinopathy in Chinese Community Healthcare Centers. Dong, X., Du, S., Zheng, W., et al.[2022]

The RetCAD AI software demonstrated high accuracy in identifying and grading age-related macular degeneration (AMD) and diabetic retinopathy (DR) from fundus photos, achieving a sensitivity of 97.3% and specificity of 73.3% for AMD, and 80.0% sensitivity and 90.1% specificity for DR at the patient level.

With a large sample size of 1245 eyes from 630 patients, the study shows that RetCAD can provide reliable automated grading, making it a cost-effective tool for screening in settings where expert evaluation is limited.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Simultaneous screening and classification of diabetic retinopathy and age-related macular degeneration based on fundus photos-a prospective analysis of the RetCAD system. Skevas, C., Weindler, H., Levering, M., et al.[2022]

Citations

1. England pubmed.ncbi.nlm.nih.gov

Artificial Intelligence in the assessment of diabetic retinopathy from fundus photographs. [2021]

2. United States pubmed.ncbi.nlm.nih.gov

Application and observation of artificial intelligence in clinical practice of fundus screening for diabetic retinopathy with non-mydriatic fundus photography: a retrospective observational study of T2DM patients in Tianjin, China. [2022]

3. Switzerland pubmed.ncbi.nlm.nih.gov

Validating automated eye disease screening AI algorithm in community and in-hospital scenarios. [2022]

4. England pubmed.ncbi.nlm.nih.gov

Clinical validation of an artificial intelligence-based diabetic retinopathy screening tool for a national health system. [2023]

5. Switzerland pubmed.ncbi.nlm.nih.gov

Artificial Intelligence Software for Diabetic Eye Screening: Diagnostic Performance and Impact of Stratification. [2023]

6. China pubmed.ncbi.nlm.nih.gov

Performance of the AIDRScreening system in detecting diabetic retinopathy in the fundus photographs of Chinese patients: a prospective, multicenter, clinical study. [2023]

7. England pubmed.ncbi.nlm.nih.gov

Comparison of early diabetic retinopathy staging in asymptomatic patients between autonomous AI-based screening and human-graded ultra-widefield colour fundus images. [2023]

8. England pubmed.ncbi.nlm.nih.gov

External validation of a deep learning detection system for glaucomatous optic neuropathy: a real-world multicentre study. [2023]

9. Netherlands pubmed.ncbi.nlm.nih.gov

Automated diabetic retinopathy screening for primary care settings using deep learning. [2022]

10. Ireland pubmed.ncbi.nlm.nih.gov

Efficacy of artificial intelligence-based screening for diabetic retinopathy in type 2 diabetes mellitus patients. [2022]

11. Switzerland pubmed.ncbi.nlm.nih.gov

Evaluation of an Artificial Intelligence System for the Detection of Diabetic Retinopathy in Chinese Community Healthcare Centers. [2022]

12. China pubmed.ncbi.nlm.nih.gov

Simultaneous screening and classification of diabetic retinopathy and age-related macular degeneration based on fundus photos-a prospective analysis of the RetCAD system. [2022]