The Evidence Behind HEDIS EED: Eye Exam for Patients With Diabetes

Key Findings

MY 2026 Specification and Age Cohort: EED targets members aged 18–75 as of December 31 who are diagnosed with diabetes. This cohort is triggered via two distinct diabetes diagnosis codes on different dates within the measurement year or the year prior, or via a single diagnosis intersecting with a documented diabetes medication dispensing event. Continuous enrollment spans the current calendar year with an allowable gap of up to 45 days. The measure applies to Commercial, Medicaid, and Medicare lines of business.

Numerator Compliance Pathways: Denominator compliance is satisfied via multiple distinct avenues: a comprehensive dilated retinal exam performed by an optometrist or ophthalmologist during the current measurement year; a documented negative retinal exam by an eye care professional from the preceding year (paired alongside a complication-free diabetes profile); remote-read fundus photography interpreted by a centralized reading center; or an autonomous AI eye examination containing a validated result flag.

Mandatory Exclusions: Members are permanently removed from the denominator if they meet exclusions during the eligible timeframe: death, hospice enrollment, ongoing palliative care (ICD-10 Z51.5), institutional residence (I-SNP/LTI), or individuals aged 66+ exhibiting concurrent frailty and advanced illness indicators. The structural anatomical exclusion for this measure is documented bilateral eye loss or bilateral enucleation.

Alignment with the ADA 2025 Standards of Care: The measure directly reflects the American Diabetes Association (ADA) framework, which recommends a baseline dilated eye evaluation immediately upon diagnosis for Type 2 patients, or within 5 years of onset for Type 1 individuals. Following a negative baseline screening under stable glycemic control, a biennial (1–2 year) interval may be considered, while any active retinopathy signals a strict annual or greater monitoring frequency. The ADA explicitly endorses remote fundus photography and FDA-cleared autonomous AI screening.

The Microvascular Treatment Evidence Base: The structural backbone of diabetic eye care rests on legacy treatment studies rather than screening interval RCTs. The Diabetic Retinopathy Study (DRS) demonstrated that panretinal photocoagulation significantly drops severe vision loss risks in proliferative disease states. Concurrently, the Early Treatment Diabetic Retinopathy Study (ETDRS) proved that focal/grid laser treatments mitigate moderate vision loss stemming from diabetic macular edema. Retinal screening is designed entirely to flag the early asymptomatic lesions validated by these treatment trials.

Autonomous AI Grading and Clinical Triage: The FDA’s 2018 De Novo clearance of autonomous point-of-care diagnostic devices established that computer-vision algorithms can successfully identify referable diabetic retinopathy with high sensitivity and specificity. The formal ingestion of CPT 92229 within the HEDIS framework marks a rare instance where quality coding specifications proactively accelerate the adoption of scalable primary care technology.

Socioeconomic Screening Disparities: Advanced diabetic retinopathy rates track low-income status, Medicaid eligibility, and racial/ethnic minority populations due to systemic gaps in baseline preventive care. Because screening adherence correlates heavily with overall income and disability status, the EED measure mandates strict socioeconomic status (SES) stratifications within Medicare Advantage quality reporting.

1) The HEDIS EED Measure (MY 2026)

Formal Nomenclature: Eye Exam for Patients With Diabetes (HEDIS abbreviation: EED). It belongs to the Effectiveness of Care domain. While it tracks identical clinical goals as modern ECDS metrics, it operates via standard administrative and hybrid pathways rather than carrying an "-E" digital designation.

Eligible Population (Denominator): Members aged 18–75 as of December 31 of the measurement year presenting with Type 1 or Type 2 diabetes. Identification requires at least 2 distinct claims containing a diabetes diagnosis code on different dates across the current measurement year or the immediate year prior, OR a single diagnosis code paired alongside a documented insulin or oral hypoglycemic dispensing event within that same 24-month lookback window. Continuous enrollment requires a clean history across the current measurement period with an allowable gap capped at 45 days.

Numerator Requirements: Ingestion of claims or chart documentation confirming one of the following criteria:

• A dilated or retinal eye examination administered by an optometrist or ophthalmologist during the current measurement year.
• A documented negative dilated or retinal eye exam administered by an eye care professional during the year prior to the current measurement year, linked to a complication-free diabetes profile.
• Bilateral retinal imaging interpreted and signed off by a qualified reading center, regardless of the billing provider's primary specialty.
• An autonomous AI screening exam (CPT 92229 or LOINC 105914-6 containing a valid output string) executed by any provider type.
• A negative retinal screening code from the prior year captured via CPT Category II code 3072F.

Required Exclusions (Non-Adjustable):

• Date of death occurring within the current measurement year.
• Active enrollment in hospice programs or use of hospice care services during the measurement year.
• Documentation of ongoing palliative care services (e.g., ICD-10 code Z51.5 or the matching value set).
• Medicare beneficiaries aged 66 and older residing long-term in an Institutional Special Needs Plan (I-SNP) or long-term institutional care settings.
• Members aged 66 and older presenting with a concurrent combination of frailty and advanced illness indicators.
• Anatomical bilateral absence of eyes or history of bilateral enucleation (e.g., SNOMED CT 15665641000119103 or bilateral ICD-10-PCS 08T0XZZ / 08T1XZZ arrays).

Coding Guidance: Workflows must filter out and exclude laboratory encounter entries containing POS 81 codes.

2) Core Code Sets (Representative Sample)

• Retinal Eye Examinations: CPT 92225, 92226, 92227, 92228; CPT 92250 (Fundus Photography); HCPCS G0908 (Diabetic Eye Exam).
• Autonomous AI Systems: CPT 92229 (Autonomous retinal imaging with automated point-of-care interpretative report); LOINC 105914-6 (Autonomous Eye Exam Result or Finding).
• Anatomical & Care Exclusions: SNOMED CT 15665641000119103 (Bilateral absence of eyes); ICD-10-PCS 08T0XZZ (Ocular Enucleation, Left), 08T1XZZ (Ocular Enucleation, Right); Palliative Care Value Set (Z51.5); Hospice Encounter and Hospice Intervention value sets.

3) Clinical Evidence Synthesis

Epidemiological Foundations (Wong et al., Nat Rev Dis Primers): Diabetic retinopathy is an expansive microvascular complication that tracks directly with the duration of the patient's metabolic disease, uncontrolled blood pressure, and chronic hyperglycemia. Legitimate screening structures connected to localized primary care delivery prevent long-term visual impairment by capturing proliferative patterns prior to the onset of macular changes.

The Causal Path of Microvascular Intervention: Because historical trials focused exclusively on treatment efficacy (DRS and ETDRS protocols), contemporary screening intervals are derived inductively. Finding microvascular anomalies early through systematic screening directly unlocks the preventative benefits established by legacy laser and contemporary anti-VEGF treatment trials.

Primary Care AI Ingestion: Point-of-care autonomous AI systems operate as an effective diagnostic triage network. By automatically routing individuals showing more-than-mild background retinopathy directly into specialist networks, these tools bypass traditional scheduling and physical transport barriers.

Teleretinal Network Infrastructure: Systematic reviews confirm that remote digital fundus imaging networks perform with a high level of diagnostic consensus when matched against traditional slit-lamp eye examinations. These systems provide a scalable method to increase access across historically underscreened rural environments.

4) Disparities and Health Equity

• Socioeconomic Disease Concentration: The structural burden of diabetes-related vision loss impacts lower-income groups disproportionately due to compounding challenges with raw food security, clinical medication adherence, and reliable specialty care access.
• The Specialty Care Bottleneck: Legacy quality models that require a patient to schedule and travel to a standalone annual ophthalmology appointment often fail to reach vulnerable, low-income, and Medicaid populations. Moving retinal cameras directly into community-based primary care settings eliminates these access barriers.

5) Operational Pitfalls and Functional Failure Modes

• Over-Restricting Specialty Coding Arrays: Claims ingestion systems frequently limit automated numerator credit exclusively to ophthalmology provider tax IDs. This coding choice accidentally drops valid, compliant screenings performed by licensed optometrists, leading to an artificial drop in the plan's reported rate.
• Prior-Year Negative Exam Omissions: Legacy database queries often fail to look back into the preceding calendar year for negative retinal findings. Missing this historical data forfeits valid 2-year compliance credits for low-risk members.
• Failing to Ingest Specialized Autonomous Codes: Outdated administrative claims infrastructure often overlooks newer codes like CPT 92229 or LOINC 105914-6, which causes modern point-of-care AI screenings to be missed entirely during standard electronic sweeps.
• Incomplete Denominator Scrubbing: Missing historical bilateral enucleation codes or failing to map permanent anatomical eye loss causes clinically ineligible members to remain in the active denominator, suppressing the plan's true compliance score.

6) Quality Improvement Evidence

• Point-of-Care Clinic Integration: Moving from external referral to in-clinic teleretinal or autonomous AI capture resolves specialist-access friction (ADA 2025; FDA 2018).
• Multimodal Outreach: A Cochrane systematic review of 66 RCTs (Lawrenson et al, Cochrane Database Syst Rev [2018]) found QI interventions raised diabetic retinopathy screening attendance by ~12 percentage points versus usual care, with DRS-targeted and general diabetes QI strategies performing similarly, supporting integrated primary-care outreach over siloed reminder systems.
• Clinical-Staff-Led Navigation: Pharmacist-led digital retina scan services embedded in FQHC primary care clinics significantly improve annual diabetic eye exam rates (Mize et al, J Am Pharm Assoc [2024]).
• FQHC Deployment: Ongoing trials are evaluating whether point-of-care AI screening embedded in FQHC workflows improves diabetic retinopathy detection and follow-up rates (Diaz et al, JAMA Netw Open [2025]).

7) Regulatory and Policy Context

The Centers for Medicare & Medicaid Services (CMS) provides coverage for annual dilated eye evaluations for diabetic beneficiaries under Medicare Part B, with matching coverage lines deployed across state-level Medicaid managed care frameworks. Because the EED metric functions as a directly weighted process measure within the Medicare Advantage Star Ratings matrix, plan performance heavily drives federal reimbursement benchmarks.

8) Quality Reporting Implications for Health Plans

Unlike cervical screening metrics, the EED measure impacts plan economics within Part C of the Medicare Advantage Star Ratings framework. Because performance scores are stratified by socioeconomic indicators (such as LIS/DE and disability status), plans are directly accountable for closing care gaps across marginalized subgroups.

Strategic Recommendations

To optimize performance on the HEDIS EED measure, quality teams should implement the following operational strategies:

1. Embed Point-of-Care Imaging Infrastructure: Integrate remote-read digital fundus photography or autonomous AI grading systems directly into high-volume primary care clinics and FQHC networks. Aim to provide point-of-care retinal imaging options at 80% or more of primary care clinics managing diabetic panels.
2. Optimize Prior-Year Lookback Sweeps: Reconfigure technical billing queries to check the preceding year's historical data for negative retinal screens, ensuring these 2-year compliance credits are fully captured.
3. Automate Ingestion of AI and Teleretinal Codes: Ensure core claims engines are configured to recognize and ingest modern teleretinal and autonomous AI codes (CPT 92229 / LOINC 105914-6) to safeguard automated compliance matching.
4. Maintain Precise Denominator Records: Build automated data cleaning sweeps to scan historical files for bilateral eye loss, hospice enrollment, active palliative care codes (Z51.5), and advanced illness modifiers to keep the denominator accurate.
5. Deploy Intelligent Data Extraction (NLP/OCR): Use natural language processing (NLP) to parse unstructured external consultation letters, scan faxed specialist updates, and map unstructured text back into discrete, audit-ready data fields.

Operational Threshold Pivots: If the plan's overall EED rate exceeds 90% but drops significantly within vulnerable LIS/DE or disability cohorts, pivot resources away from broad mailings and invest heavily in targeted point-of-care teleretinal camera deployment within community clinics. If the baseline rate sits below 80%, prioritize broader structural improvements: implement comprehensive teleretinal vendor connections and deploy dedicated patient navigation teams to address fundamental access gaps.

Caveats & Data Realities

• The EED clinical justification rests on a logical deduction from legacy microvascular treatment trials rather than a dedicated screening-cadence RCT. Frame internal and provider communications around this relationship, highlighting that screening serves to connect asymptomatic patients with highly effective, evidence-backed treatment options.
• The 1–2 year screening interval allowed by the ADA is a clinical recommendation for stable, low-risk patients under tight glycemic control. Because HEDIS allows a 2-year credit only following a documented negative exam, the quality metric operates as a structured reflection of the ADA guidelines.
• Autonomous AI systems function primarily as primary care triage mechanisms to identify referable disease. A negative AI screening result does not replace the long-term clinical necessity of standard specialist evaluations for high-risk patients.
• EED remains an administrative measure that permits traditional hybrid reporting pipelines. While paid and denied claims will automatically capture standard screenings, pure cash-pay encounters or unstructured specialist feedback letters will be missed entirely unless captured via structured supplemental data streams or traditional manual chart audits.
• Because EED is directly weighted within the Medicare Advantage Star Ratings framework, data ingestion accuracy and complete lookback tracking must be prioritized to protect plan performance.