Ophthalmology AI Standard of Care: Diabetic Retinopathy, Glaucoma Detection, and Liability

AI Revolutionizes Eye Disease Detection

#

Ophthalmology became the proving ground for autonomous AI in medicine when the FDA cleared the first-ever fully autonomous AI diagnostic system:IDx-DR (now LumineticsCore), in 2018. Today, AI systems can diagnose diabetic retinopathy at the point of care without a specialist, detect early signs of glaucoma and age-related macular degeneration (AMD), and guide treatment decisions. But with autonomy comes unprecedented liability questions: When AI screens for diabetic retinopathy in a primary care office and misses disease, who bears responsibility?

This guide examines the standard of care for AI use in ophthalmology, the expanding landscape of FDA-cleared devices, and the complex liability framework for AI-assisted eye care.

FDA-Cleared Ophthalmology AI Devices

#

Autonomous Diabetic Retinopathy Screening

#

Ophthalmology pioneered autonomous AI diagnostics, systems that provide diagnostic results without physician interpretation:

FDA-Cleared Autonomous DR Screening Systems:

LumineticsCore (formerly IDx-DR):

• First FDA-cleared fully autonomous AI diagnostic system in any field of medicine
• Detects more-than-mild diabetic retinopathy (ETDRS level 35+)
• Also detects central-involved diabetic macular edema and clinically significant DME
• 87.4% sensitivity, 89.5% specificity in pivotal trial
• Operates at point of care in primary care settings
• European CE Mark as Class IIa medical device

EyeArt System:

• Cloud-based autonomous analysis
• Enables remote screening programs
• Integration with multiple fundus camera platforms

AEYE-DS:

• First fully autonomous AI for portable DR screening
• Ultra-rapid point-of-care analysis
• Designed for community health settings

Glaucoma AI Development

#

Unlike diabetic retinopathy, no FDA-cleared autonomous AI exists for glaucoma screening:

Current Status:

• No FDA-approved autonomous glaucoma diagnostic device
• Research shows promise but faces unique challenges
• Disease is multifaceted, requiring multiple data types

Challenges for Glaucoma AI:

• Requires combination of fundus images, OCT, IOP, and visual field testing
• Lacks standardized diagnostic criteria
• Progressive nature complicates screening vs. diagnosis
• “Black box” architecture limits interpretability

AI-Assisted Tools (Non-Autonomous):

• ANTERION (Heidelberg Engineering), AI-integrated anterior segment imaging
• Various OCT analysis algorithms, Not autonomous, require physician interpretation
• Research platforms in development

Age-Related Macular Degeneration

#

AI is advancing AMD detection and monitoring:

FDA-Cleared/Pending Devices:

Performance Data:

• iPredict achieved 86% accuracy in AREDS dataset for 2-year late AMD risk
• AI screening algorithms report 94% sensitivity, 99% specificity
• Deep learning predicts anti-VEGF treatment needs with high accuracy

The Autonomous AI Liability Framework

#

A New Legal Frontier

#

Autonomous AI creates unprecedented liability questions in medicine:

The Core Question:

When AI autonomously diagnoses diabetic retinopathy in a primary care office and misses disease that leads to vision loss, who is responsible, the AI developer, the primary care physician, or the healthcare system?

Mixed Views on Accountability:

• Industry partners: Hold physicians responsible
• Ophthalmologists: Blame developers
• Legal/ethics experts: Advocate shared responsibility

Who Administers the Test?

#

Autonomous DR screening typically occurs in primary care, not ophthalmology:

The Physician’s Dilemma:

• Primary care physicians administer the test
• They lack specialized retinal knowledge
• Should they be liable for incorrect AI results?
• Autonomous AI outputs may not constitute “medical records”

Current Legal Ambiguity:

• State Medical Boards decide what constitutes a medical record
• Autonomous AI output currently lacks equivalent medicolegal status
• AI diagnostic reports may not be part of medical record unless physician signs off

Liability Allocation Models

#

Potential Defendants in AI Ophthalmology Cases:

Product Liability vs. Medical Malpractice

#

Product Liability Framework:

• AI systems are increasingly treated as “products” (see Garcia v. Character.AI)
• Design defect claims may target algorithm accuracy
• Failure to warn claims for AI limitations
• Manufacturer strict liability possible

Medical Malpractice Framework:

• Physician duty to exercise reasonable care
• Question: Does deploying AI meet or breach standard of care?
• Failure to override incorrect AI results
• Inadequate patient counseling about AI limitations

Standard of Care Considerations

#

AAO and Professional Guidance

#

The American Academy of Ophthalmology and professional bodies are developing AI guidance:

Key Principles:

1. AI should augment, not replace, clinical judgment
2. Physicians remain responsible for patient care decisions
3. AI limitations must be understood and communicated
4. Appropriate follow-up for AI results is essential

Diabetic Eye Exam Standards:

• ADA recommends annual dilated eye exam for diabetics
• AI screening can increase access to screening
• Positive AI screens require ophthalmology referral
• Negative AI screens don’t eliminate need for periodic specialist exams

When AI Misses Disease

#

Missed Diabetic Retinopathy Scenarios:

Key Questions:

• Was the AI device appropriately selected for the patient population?
• Were AI limitations disclosed to the patient?
• Was appropriate follow-up arranged regardless of AI result?
• Did image quality meet device requirements?

Physician Responsibilities

#

Before AI Deployment:

• Understand AI device capabilities and limitations
• Verify FDA clearance and intended use
• Establish protocols for positive results
• Train staff on proper image acquisition

During AI Use:

• Ensure proper image quality
• Document AI results appropriately
• Arrange immediate referral for positive screens
• Counsel patients on AI limitations

After AI Results:

• Follow up on referral completion
• Track patients regardless of AI results
• Monitor for symptoms between screenings
• Maintain appropriate screening intervals

Emerging Liability Concerns

#

Bias and Health Disparities

#

AI ophthalmology systems may perform differently across populations:

Documented Concerns:

• Training data may underrepresent certain racial/ethnic groups
• Fundus pigmentation affects image analysis
• Socioeconomic factors influence image quality (equipment access)
• AI may exacerbate existing disparities in eye care access

Liability Implications:

• Disparate impact claims possible
• Failure to validate across populations
• Duty to disclose performance limitations by group

“Black Box” Transparency

#

Ophthalmology AI faces the same transparency challenges as other AI:

Challenges:

• Cannot explain how diagnosis was reached
• Difficult to establish causation in litigation
• Physicians cannot evaluate AI reasoning
• Patients cannot give fully informed consent

Scalability and Infrastructure

#

Rapid AI deployment raises systemic concerns:

Three Pressing Issues (per academic literature):

1. Transparency, Explanation and interpretation of AI models
2. Attribution, Responsibility for AI-induced harms
3. Scalability, Screening infrastructure and follow-up capacity

Systemic Liability:

• Healthcare systems may be liable for inadequate AI integration
• Referral bottlenecks may delay care after positive AI screens
• IT failures affecting AI availability

Case Examples and Analogies

#

Diabetic Retinopathy Screening Failures

#

While pure autonomous ophthalmology AI litigation is limited, analogous cases inform liability:

Traditional DR Screening Cases:

• Delayed referral after abnormal screening, Permanent vision loss
• Failure to screen diabetic patients, Progression to blindness
• Inadequate follow-up systems, Missed appointments, disease progression

Applicable Principles:

• Physician duty to arrange appropriate screening
• System duty to ensure referral completion
• Standard of care requires documented follow-up

AI Diagnostic Error Patterns

#

Lessons from AI Medical Device Experience:

• JAMA Health Forum study: 489 adverse events, 113 recalls, 1 death across 691 AI devices
• 43% of recalls occurred within first year of clearance
• Diagnostic errors most common adverse event category

Ophthalmology-Specific Risks:

• Image quality failures leading to missed disease
• Edge cases outside AI training data
• Rare presentations not in validation studies

Frequently Asked Questions

#

Practical Guidance

#

For Healthcare Systems

#

Before AI Deployment:

• Conduct thorough vendor due diligence
• Verify FDA clearance and intended use match your application
• Establish clear protocols for positive and inconclusive results
• Create referral tracking systems
• Train all staff on proper use

During Implementation:

• Monitor AI performance metrics
• Track referral completion rates
• Document any AI failures or errors
• Maintain quality assurance programs

For Risk Management:

• Review insurance coverage for AI-related claims
• Consider contractual indemnification from AI vendors
• Establish incident reporting procedures
• Create patient consent/disclosure processes

For Ophthalmologists

#

Supervision Considerations:

• If AI is used under your supervision, you may share liability
• Review protocols for AI oversight
• Ensure adequate credentialing of AI systems
• Monitor for AI failures and patterns

Clinical Integration:

• AI results are starting points, not final diagnoses
• Apply clinical judgment to all AI outputs
• Document independent clinical reasoning
• Consider AI limitations for each patient

For Patients

#

Questions to Ask:

• Is AI being used in my care?
• What are the AI’s limitations?
• Who reviews the AI results?
• What follow-up is recommended regardless of AI results?

Related Resources

#

• Medical AI Device Cases, FDA adverse events, recalls, malpractice trends
• AI Misdiagnosis Case Tracker, Radiology, pathology, and diagnostic AI failures
• AI Product Liability, AI LEAD Act, Garcia v. Character.AI, strict liability framework
• Professional Liability Insurance Gaps, AI coverage gaps and policy considerations