CoChat
Sign Up
Cited by 0Retinal Imaging and AnalysisRetinal Diseases and TreatmentsOphthalmology and Visual Health Research

From validation to public health impact: implementing equitable AI-enabled diabetic retinopathy screening

Open access
Chen, Jing, Chen, Bin·2026·Frontiers in Public Health
DOI 10.3389/fpubh.2026.1976572
Open this paper in CoChat — free accountOpen PDF

Abstract

Diabetic retinopathy (DR) remains a preventable cause of vision loss as diabetes prevalence rises worldwide. Artificial intelligence (AI)-enabled retinal screening has progressed from retrospective validation to prospective and real-world deployment, with recent reviews showing high diagnostic performance across fundus photography, smartphone-based systems, and autonomous platforms. Yet diagnostic accuracy alone does not ensure public health benefit. Population impact can be lost when eligible people are not reached, images are ungradable, results are delayed, referrals are incomplete, treatment capacity is insufficient, or disadvantaged groups face financial, geographic, or digital barriers. Unlike reviews centered mainly on diagnostic accuracy or economic performance, this Mini Review reframes AI-enabled DR screening as a population-health intervention spanning detection through completion of effective eye care. We synthesize evidence on diagnostic validity, implementation, economics, referral completion, care cascades, patient-centered outcomes, equity, and governance. We argue that effectiveness and cost-effectiveness are properties of the complete screening-to-treatment pathway, not the algorithm alone, and propose a four-domain framework encompassing technical validity, service integration, population impact, and equity and sustainability. Future research should prioritize comparative implementation studies and outcome sets that capture screening reach, gradability, referral completion, treatment initiation, time to treatment, avoidable vision loss, vision-related quality of life, acceptability, costs, and subgroup equity. The policy question is no longer whether AI can detect DR, but whether health systems can translate automated detection into equitable completion of effective eye care at scale.

AI snapshot

Written from the paper’s full text.

TL;DR

This Mini Review argues that AI-enabled diabetic retinopathy (DR) screening should be evaluated as a population-health intervention spanning the full screening-to-treatment pathway rather than as an isolated diagnostic test, since accuracy alone does not guarantee public health benefit. It proposes a four-domain implementation framework (technical validity, service integration, population impact, equity and sustainability) and calls for comparative implementation studies reporting linked care-cascade outcomes.

Key findings

  • Diagnostic performance is broadly credible: Joseph and colleagues' synthesis of 34 real-world studies reported pooled sensitivity of 94% and specificity of 89% among gradable images, with income level and fundus camera type explaining part of the heterogeneity; Ran and colleagues reviewed 47 prospective validation studies and found high discriminative performance, particularly for vision-threatening DR.
  • Program effectiveness can be diluted at multiple points — before imaging (limited invitation, low attendance, poor access), during imaging (ungradable/non-diagnostic images excluded from accuracy estimates), and after interpretation (delayed result communication, referral loss, insufficient treatment capacity) — so the relevant unit of analysis is the complete care cascade.
  • Economic evidence is pathway evidence, not algorithm evidence: deep learning screening appeared generally favorable in high-income countries, mixed and sensitive to compliance rates, baseline glycemic control, and initial implementation costs in middle-income countries, with sparse low-income country evidence; in urban China an AI-assisted community telemedicine model was not automatically cost-effective versus manual grading but improved when referral compliance increased.
  • Equity is framed as a primary implementation outcome: ungradable images (associated with cataract, small pupils, media opacity, older age, poor imaging conditions) should be reported as an implementation and equity outcome by site, device, age, race/ethnicity, comorbidity, and deprivation/rurality, not treated as a technical exclusion.

Methods

  • Design: structured narrative Mini Review guided by the SANRA framework, explicitly not a systematic review.
  • Data: literature search in PubMed, Web of Science, and Embase for peer-reviewed articles and policy/regulatory documents on AI-enabled DR screening published January 2016–August 2026, with foundational 2016 sources retained; priority given to systematic reviews, meta-analyses, prospective validation studies, real-world implementation studies, randomized or pragmatic evaluations, health economic analyses, patient-centered outcome studies, and policy/regulatory documents, plus reference-list screening of key reviews.
  • Analysis: narrative synthesis; no formal risk-of-bias assessment, exhaustive study inclusion, or quantitative evidence synthesis was performed. The paper also proposes a four-domain framework (technical validity, service integration, population impact, equity and sustainability).

Limitations

  • Stated: as a Mini Review rather than a systematic review, formal risk-of-bias assessment, exhaustive study inclusion, and quantitative evidence synthesis were not performed.
  • Stated: downstream evidence is thinner than diagnostic evidence — referral uptake has improved in some AI-assisted pathways, but few studies follow patients through treatment and visual outcomes, and care-cascade evidence shows persistent losses across screening, diagnosis, care adherence, and blindness prevention.
  • Stated: evidence from low-income countries remains sparse, and performance and cost-effectiveness estimates from high-income settings should not be extrapolated uncritically to lower-resource settings.
  • Evident from the text: the authors declare that generative AI (ChatGPT, GPT-5.5 Thinking) was used for English language editing, readability improvement, and schematic figure preparation, though not for literature selection, clinical interpretation, data analysis, reference verification, or scientific conclusions.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3389/fpubh.2026.1976572

Cites

31 of 31 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

34 of 62 cited sources checked · 14 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    AI-enabled diabetic retinopathy screening has progressed from retrospective validation to prospective and real-world deployment.

    “Artificial intelligence (AI)-enabled retinal screening has progressed from retrospective validation to prospective and real-world deployment”
    • Also stated in: § Introduction · plain
      “During the past decade, AI-enabled DR screening has moved from landmark retrospective validation studies to prospective evaluations and realworld programmatic deployment (6)(7)(8)(9)(10)(11)(12)(13)(14)(15).”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites [6] Gulshan, 2016, Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs, JAMA. source
    • cites [7] Ting, 2017, Development and validation of a deep learning system for diabetic retinopathy and related eye diseases using retinal images from multiethnic populations with diabetes, JAMA. source

      The study’s aims and Discussion (page unavailable) describe validation using images from an ongoing Singapore screening program and 10 external datasets, extending earlier database-based validation to clinical screening settings. The Discussion examines how the system could be deployed, but our copy does not establish prospective evaluation or actual real-world deployment. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites [8] Abràmoff, 2018, Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices, NPJ Digit Med. source

      The Introduction describes earlier laboratory evaluations of AI algorithms, and the Methods (Study design) report prospective testing at 10 primary care practices. The Abstract says the system was subsequently FDA-authorized for use, but the paper does not establish routine real-world deployment or specify that the earlier evaluations were retrospective. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • Interpretationreasoning

    Autonomous point-of-care, centralized AI or teleophthalmology, and hybrid AI-plus-human grading models cannot be universally ranked because each offers a different balance of immediacy, specialist oversight, infrastructure requirements, and workflow complexity.

    “Autonomous point-of-care, centralized AI or teleophthalmology, and hybrid AI-plus-human grading models should not be ranked universally, because each offers a different balance of immediacy, specialist oversight, infrastructure requirements, and workflow complexity.”
    Quote in the paperNot a substantial claim4 of 6 sources checked
    • cites [8] Abràmoff, 2018, Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices, NPJ Digit Med. source

      The Introduction (page not provided) describes autonomous diagnosis without human-expert review and a real-time decision at the point of care; Methods describes the imaging and operator workflow. These passages support the claim’s description of the autonomous model, but the paper does not compare it with centralized AI, teleophthalmology, or hybrid AI-plus-human grading models, or conclude that the models cannot be universally ranked. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [10] Lee, 2021, Multicenter, head-to-head, real-world validation study of seven automated artificial intelligence diabetic retinopathy screening systems, Diabetes Care. source

      The Conclusions discuss AI as a first screening step with ophthalmologists reviewing flagged cases, and note integration, hardware, and cloud-service considerations for the VA teleretinal system (page not provided). They do not compare autonomous point-of-care, centralized, and hybrid models or conclude that those models should not be ranked universally on the stated criteria. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
    • cites [11] Ruamviboonsuk, 2022, Real-time diabetic retinopathy screening by deep learning in a multisite national screening programme: a prospective interventional cohort study, Lancet Digit Health. source
    • cites [14] Khan, 2025, Diagnostic accuracy of IDx-DR for detecting diabetic retinopathy: a systematic review and meta-analysis, Am J Ophthalmol. source
    • cites [17] Chotcomwongse, 2025, Transforming non-digital, clinical workflows to detect and track vision-threatening diabetic retinopathy via a digital platform integrating artificial intelligence: implementation research, Ophthalmol Ther. source

      The Discussion says this study’s AI and manual screening results might not be directly compared because the modalities had different workflows. The Methods and Discussion also describe specialist over-reading, internet and uploading demands, and resource constraints. But the paper does not assess the claimed three-way balance among autonomous point-of-care, centralized AI or teleophthalmology, and hybrid models, or conclude that those models cannot be universally ranked. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [21] Wolf, 2024, Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth: the ACCESS randomized control trial, Nat Commun. source

      The Discussion (page not provided) describes autonomous AI’s immediate point-of-care results and lack of need for ophthalmic oversight, and contrasts its scalability with limitations of teleretinal screening. It calls for future direct comparisons rather than assessing all three models. The paper does not address hybrid AI-plus-human grading or establish a balance of infrastructure and workflow requirements across the three models, so it does not support the full claim about universal ranking. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
  • Premisecitation

    The RAIDERS randomized trial in Rwanda showed that AI assessment could improve referral service uptake in a low-resource setting when results and referral processes were embedded in an intervention rather than treated as a stand-alone test.

    “The RAIDERS randomized trial in Rwanda showed that AI assessment could improve referral service uptake in a low-resource setting when results and referral processes were embedded in an intervention rather than treated as a stand-alone test (18).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [18] Mathenge, 2022, Impact of artificial intelligence assessment of diabetic retinopathy on referral service uptake in a low-resource setting: the RAIDERS randomized trial, Ophthalmol Sci. source
  • Premisecitation

    Independent head-to-head validation has shown that performance can vary materially across algorithms even when systems are applied to the same real-world data.

    “Independent head-to-head validation has shown that performance can vary materially across algorithms even when systems are applied to the same real-world data (10).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [10] Lee, 2021, Multicenter, head-to-head, real-world validation study of seven automated artificial intelligence diabetic retinopathy screening systems, Diabetes Care. source
  • Premisecitation

    A pivotal prospective trial of an autonomous diagnostic system in primary care offices supported the clinical feasibility of point-of-care AI screening without specialist interpretation.

    “Abramoff and colleagues then reported a pivotal prospective trial of an autonomous diagnostic system in primary care offices, supporting the clinical feasibility of point-of-care AI screening without specialist interpretation (8).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [8] Abràmoff, 2018, Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices, NPJ Digit Med. source
  • Classified as probably not a substantial claim. 18
  • May combine two claims. 25

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Diabetic retinopathy (DR) is a common microvascular complication of diabetes and a preventable cause of vision loss when sight-threatening disease is detected and treated early.

    “Diabetic retinopathy (DR) is a common microvascular complication of diabetes and a preventable cause of vision loss when sight-threatening disease is detected and treated early (1).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [1] Ting, 2016, Diabetic retinopathy: global prevalence, major risk factors, screening practices and public health challenges: a review, Clin Exp Ophthalmol. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    This Mini Review was guided by the Scale for the Assessment of Narrative Review Articles (SANRA) framework.

    “This Mini Review was guided by the Scale for the Assessment of Narrative Review Articles (SANRA) framework, with particular attention to the justification of the review topic, clarity of aims, transparent description of literature selection, appropriate referencing, and scientific reasoning (16).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [16] Baethge, 2019, SANRA—a scale for the quality assessment of narrative review articles, Res Integr Peer Rev. source

  • Classified as probably not a substantial claim.

    Premisecitation

    The 2016 review by Ting, Cheung, and Wong framed diabetic retinopathy as a public health problem shaped by inadequate screening coverage, shortages of trained eye-care personnel, cost, uneven access, and limited service capacity in low-resource settings.

    “The 2016 review by Ting, Cheung, and Wong framed DR as a public health problem shaped by inadequate screening coverage, shortages of trained eye-care personnel, cost, uneven access, and limited service capacity in low-resource settings (1).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [1] Ting, 2016, Diabetic retinopathy: global prevalence, major risk factors, screening practices and public health challenges: a review, Clin Exp Ophthalmol. source

  • May combine two claims.

    Premisecitation

    A 2026 meta-analysis of AI-assisted pathways found improved referral uptake, but with high heterogeneity and low certainty, emphasizing that pathway redesign and patient-facing interventions drive much of the observed effect.

    “A 2026 meta-analysis of AI-assisted pathways found improved referral uptake, but with high heterogeneity and low certainty, emphasizing that pathway redesign and patient-facing interventions drive much of the observed effect (22).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [22] Leigh, 2026, Referral uptake after diabetic retinopathy screening with artificial intelligence-assisted care pathways: a systematic review and meta-analysis, NPJ Digit Med. source

      The Results report improved referral uptake with substantial heterogeneity (I² = 91.9%), and the Discussion describes the evidence as low certainty. The Discussion emphasizes pathway redesign and patient-facing co-interventions, but says their relative contributions remain uncertain; it does not establish that they drive much of the observed effect. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    The ACCESS randomized trial showed that autonomous AI increased diabetic eye exam completion among youth, but also highlighted the need to measure follow-through after abnormal results.

    “The ACCESS randomized trial showed that autonomous AI increased diabetic eye exam completion among youth, but also highlighted the need to measure follow-through after abnormal results (21).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [21] Wolf, 2024, Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth: the ACCESS randomized control trial, Nat Commun. source

  • May combine two claims.

    Premisecitation

    In Thailand, a digital platform integrating AI had to transform non-digital workflows, establish tracking, and coordinate follow-up for vision-threatening DR.

    “In Thailand, a digital platform integrating AI had to transform non-digital workflows, establish tracking, and coordinate follow-up for vision-threatening DR (17).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [17] Chotcomwongse, 2025, Transforming non-digital, clinical workflows to detect and track vision-threatening diabetic retinopathy via a digital platform integrating artificial intelligence: implementation research, Ophthalmol Ther. source

  • May combine two claims.

    Premisecitation

    In urban China, an AI-assisted community telemedicine model was not automatically cost-effective compared with manual grading, but its cost-effectiveness improved when referral compliance increased.

    “In urban China, Lin and colleagues found that an AI-assisted community telemedicine model was not automatically cost-effective compared with manual grading, but its value improved when referral compliance increased (24).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [24] Lin, 2023, Artificial intelligence in community-based diabetic retinopathy telemedicine screening in urban China: cost-effectiveness and cost-utility analyses with real-world data, JMIR Public Health Surveill. source

  • May combine two claims.

    Premisecitation

    Autonomous AI systems can provide same-visit results and reduce dependence on asynchronous grading; centralized AI/teleophthalmology models can support multisite quality assurance and specialist oversight; hybrid AI-plus-human models retain human review for selected or uncertain cases.

    “Autonomous systems can provide same-visit results and reduce dependence on asynchronous grading, whereas centralized models may support multisite quality assurance and specialist oversight; hybrid models retain human review for selected or uncertain cases (8,10,11,14,17,21).”
    Quote in the paperSubstantial claim4 of 6 sources checked
    • cites [8] Abràmoff, 2018, Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices, NPJ Digit Med. source

      Split into 3 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [10] Lee, 2021, Multicenter, head-to-head, real-world validation study of seven automated artificial intelligence diabetic retinopathy screening systems, Diabetes Care. source

      Split into 3 atoms — 1 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [11] Ruamviboonsuk, 2022, Real-time diabetic retinopathy screening by deep learning in a multisite national screening programme: a prospective interventional cohort study, Lancet Digit Health. source
    • cites [14] Khan, 2025, Diagnostic accuracy of IDx-DR for detecting diabetic retinopathy: a systematic review and meta-analysis, Am J Ophthalmol. source
    • cites [17] Chotcomwongse, 2025, Transforming non-digital, clinical workflows to detect and track vision-threatening diabetic retinopathy via a digital platform integrating artificial intelligence: implementation research, Ophthalmol Ther. source

      Split into 3 atoms — 1 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [21] Wolf, 2024, Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth: the ACCESS randomized control trial, Nat Commun. source

      Split into 3 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • May combine two claims.

    Interpretationcitation

    After AI interpretation, delayed result communication, unclear referral pathways, insufficient patient-facing communication, referral loss, or inadequate treatment capacity can prevent automated detection from becoming completed care.

    “After AI interpretation, delayed result communication, unclear referral pathways, insufficient patientfacing communication, referral loss, or inadequate treatment capacity can prevent automated detection from becoming completed care (17)(18)(19)(20)(21)(22)(23).”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites [17] Chotcomwongse, 2025, Transforming non-digital, clinical workflows to detect and track vision-threatening diabetic retinopathy via a digital platform integrating artificial intelligence: implementation research, Ophthalmol Ther. source

      The Results (page unavailable) show that 115 of 129 AI-screened true-positive referrals attended the tertiary center, and that nurses could not contact some patients about changed referral decisions. The Discussion (page unavailable) identifies appointments, reminders, and timely information as benefits of a digital platform. These passages support the importance of follow-up and communication after detection, but do not establish delayed result communication, unclear referral pathways, or inadequate treatment capacity as barriers in this study. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [18] Mathenge, 2022, Impact of artificial intelligence assessment of diabetic retinopathy on referral service uptake in a low-resource setting: the RAIDERS randomized trial, Ophthalmol Sci. source

      The Abstract Conclusions and Discussion support the roles of delayed results, unclear referral processes, patient counselling, and failure to attend referral care; the Discussion reports that nearly half of AI-screened patients did not attend referral care within 30 days. The paper does not establish inadequate treatment capacity as a barrier, and it measures referral attendance rather than completed treatment. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [19] Bresnick, 2020, Adherence to ophthalmology referral, treatment and follow-up after diabetic retinopathy screening in the primary care setting, BMJ Open Diabetes Res Care. source

      The Discussion (page not supplied) documents referral loss: some patients who missed an ophthalmology appointment had no recorded formal referral, and the authors call for better referral tracking. The Introduction says patients need education about timely specialist care. But this study used human retinal graders, not AI, and the cited passages do not establish delayed result communication or inadequate treatment capacity as barriers. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    The most diagnostically accurate DR screening model is not necessarily the most cost-effective option; the preferred trade-off between sensitivity and specificity depends on prevalence, willingness-to-pay thresholds, costs of false positives, costs of missed disease, and downstream treatment capacity.

    “Wang and colleagues showed that the most diagnostically accurate model is not necessarily the most cost-effective option; the preferred trade-off between sensitivity and specificity depends on prevalence, willingness-to-pay thresholds, costs of false positives, costs of missed disease, and downstream treatment capacity (25).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [25] Wang, 2024, Economic evaluation for medical artificial intelligence: accuracy vs. cost-effectiveness in a diabetic retinopathy screening case, NPJ Digit Med. source

  • May combine two claims.

    Premisecitation

    Dow and colleagues reported improved follow-up in an AI-based workflow, but the result was linked to workflow differences, result timing, and patient communication rather than the algorithm alone.

    “Dow and colleagues reported improved follow-up in an AI-based workflow, but the result was linked to workflow differences, result timing, and patient communication rather than the algorithm alone (20).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [20] Dow, 2023, Artificial intelligence improves patient follow-up in a diabetic retinopathy screening program, Clin Ophthalmol. source

  • May combine two claims.

    Premisecitation

    AI-enabled screening can reproduce or worsen inequity if devices are unavailable, internet connectivity is weak, staff are inadequately trained, costs are shifted to patients, or algorithms perform differently across populations and camera systems.

    “However, the same technology can reproduce or worsen inequity if devices are unavailable, internet connectivity is weak, staff are inadequately trained, costs are shifted to patients, or algorithms perform differently across populations and camera systems (10,(31)(32)(33).”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites [10] Lee, 2021, Multicenter, head-to-head, real-world validation study of seven automated artificial intelligence diabetic retinopathy screening systems, Diabetes Care. source

      The Conclusions report performance differences between sites and warn that algorithms may not generalize across populations; they also link image quality to technician training. The Methods identify camera-system specifications as part of the acquisition protocol. The paper does not establish that unavailable devices, weak internet connectivity, or costs shifted to patients worsen inequity; its systems were tested without an internet connection. (LLM, conf=0.87)

      Experimental — accuracy not yet measured
    • cites [31] Abràmoff, 2023, Considerations for addressing bias in artificial intelligence for health equity, NPJ Digit Med. source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [32] Tsai, 2022, Cross-camera external validation for artificial intelligence software in diagnosis of diabetic retinopathy, J Diabetes Res. source

      The Introduction says real-world performance may be less satisfactory because images differ across racial groups and fundus cameras; the Discussion calls for cross-camera validation. The paper does not address weak internet connectivity, inadequate staff training, device unavailability, or costs shifted to patients, so it supports only the population-and-camera portion of the inequity claim. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    AI-enabled screening may be particularly valuable for rural and remote populations, low-income patients, older adults, and people who face long travel distances or limited specialist availability.

    “It may be particularly valuable for rural and remote populations, lowincome patients, older adults, and people who face long travel distances or limited specialist availability (28)(29)(30).”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites [28] Whitestone, 2024, Feasibility and acceptance of artificial intelligence-based diabetic retinopathy screening in Rwanda, Br J Ophthalmol. source

      The available Abstract reports AI-based diabetic retinopathy screening at four diabetes clinics in Rwanda, but does not say it is particularly valuable for the listed populations or people facing travel and specialist-access barriers. Searches for rural and remote populations, low income, older adults, travel distance, and specialist availability found no passage addressing that claim; the available abstract text ends mid-sentence. (LLM, conf=0.68)

      Experimental — accuracy not yet measured
    • cites [29] Bhambhwani, 2025, Feasibility and patient experience of a pilot artificial intelligence-based diabetic retinopathy screening program in Northern Ontario, Ophthalmic Epidemiol. source

      The Introduction describes rural and remote areas of Northwestern Ontario as facing long travel distances and physician shortages, and the Discussion says AI screening may improve access where eye-care services are limited. The Results report a mean participant age of 70.8 years, but the paper does not establish that screening is particularly valuable for older adults or address low-income patients as a group. It also calls for further study in more remote and rural settings. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [30] Leong, 2026, Autonomous AI-assisted diabetic retinopathy screening at primary care is associated with increased presentation to eye care by at risk patients, NPJ Digit Med. source

      The Discussion (page not provided) suggests AI-assisted diabetic retinopathy screening may improve access for African-American patients and reports that a previous study found improved screening adherence among Medicaid-covered patients. It does not establish particular value for rural or remote populations, older adults, or people facing long travel distances or limited specialist availability; the study served a predominantly urban population, and its current Medicaid analysis found no significant difference. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The public health challenge of diabetic retinopathy is expanding as diabetes prevalence rises.

    “The public health challenge is expanding as diabetes prevalence rises: the International Diabetes Federation estimates that 589 million adults were living with diabetes in 2024, with projections approaching 853 million by 2050 (2).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [2] International, 2025, IDF

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning

    Because this was a Mini Review rather than a systematic review, formal risk-of-bias assessment, exhaustive study inclusion, and quantitative evidence synthesis were not performed.

    “Because this was a Mini Review rather than a systematic review, formal risk-of-bias assessment, exhaustive study inclusion, and quantitative evidence synthesis were not performed.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    More than 100 million adults already have diabetic retinopathy.

    “Updated global estimates suggest that more than 100 million adults already have DR”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [3] Teo, 2021, Global prevalence of diabetic retinopathy and projection of burden through 2045: systematic review and meta-analysis, Ophthalmology. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The International Diabetes Federation projects that diabetes prevalence will approach 853 million by 2050.

    “with projections approaching 853 million by 2050 (2)”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [2] International, 2025, IDF

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The International Diabetes Federation estimates that 589 million adults were living with diabetes in 2024.

    “the International Diabetes Federation estimates that 589 million adults were living with diabetes in 2024”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [2] International, 2025, IDF

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationnone given

    Population impact can be lost when eligible people are not reached, images are ungradable, results are delayed, referrals are incomplete, treatment capacity is insufficient, or disadvantaged groups face financial, geographic, or digital barriers.

    “Population impact can be lost when eligible people are not reached, images are ungradable, results are delayed, referrals are incomplete, treatment capacity is insufficient, or disadvantaged groups face financial, geographic, or digital barriers.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    A structured narrative literature search was conducted in PubMed, Web of Science, and Embase.

    “A structured narrative literature search was conducted in PubMed, Web of Science, and Embase to identify peer-reviewed articles and relevant policy or regulatory documents addressing AI-enabled diabetic retinopathy screening.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Digital interventions do not substitute for functioning health systems and require governance, training, infrastructure, and equity safeguards.

    “the World Health Organization has cautioned that digital interventions do not substitute for functioning health systems and require governance, training, infrastructure, and equity safeguards (5).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5] 2019, WHO

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Premisecitation

    Screening is not simply an ophthalmic service; it is a population-level strategy that links chronic disease management, primary care, imaging infrastructure, referral pathways, and timely treatment.

    “In this context, screening is not simply an ophthalmic service; it is a population-level strategy that links chronic disease management, primary care, imaging infrastructure, referral pathways, and timely treatment (4).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [4] Wong, 2018, Guidelines on diabetic eye care: the International Council of Ophthalmology recommendations for screening, follow-up, referral, and treatment based on resource settings, Ophthalmology. source

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning

    The policy question is no longer whether AI can detect diabetic retinopathy, but whether health systems can translate automated detection into equitable completion of effective eye care at scale.

    “The policy question is no longer whether AI can detect DR, but whether health systems can translate automated detection into equitable completion of effective eye care at scale.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The key question has shifted from whether algorithms can classify retinal images accurately to whether health systems can use automated detection to deliver completed, timely, and equitable eye care.

    “As this evidence has matured, the key question has shifted from whether algorithms can classify retinal images accurately to whether health systems can use automated detection to deliver completed, timely, and equitable eye care.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The number of adults with diabetic retinopathy is expected to increase substantially by 2045.

    “and the number is expected to increase substantially by 2045 (3)”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [3] Teo, 2021, Global prevalence of diabetic retinopathy and projection of burden through 2045: systematic review and meta-analysis, Ophthalmology. source

  • May combine two claims.

    Premisecitation

    Deep learning-based DR screening appeared generally favorable in high-income countries, while results in middle-income countries were mixed and sensitive to compliance rates, baseline glycemic control, and initial implementation costs; evidence from low-income countries remains sparse.

    “Ran and colleagues found that deep learning-based DR screening appeared generally favorable in high-income countries, whereas results in middle-income countries were mixed and sensitive to compliance rates, baseline glycemic control, and initial implementation costs; evidence from low-income countries remains sparse (15).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15] Ran, 2026, Real-world prospective validation and economic evaluation of deep learning-based diabetic retinopathy detection from fundus photographs: a systematic review and meta-analysis, Diabetes Care. source

  • May combine two claims.

    Premisecitation

    AI DR screening has high discriminative performance, particularly for vision-threatening DR, but performance is affected by region, clinical pathway, mydriasis, image quality control, sample size, grading criteria, reference standard, and model architecture.

    “Ran and colleagues reviewed 47 prospective validation studies and found high discriminative performance, particularly for vision-threatening DR, while also showing that region, clinical pathway, mydriasis, image quality control, sample size, grading criteria, reference standard, and model architecture affected performance (15).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15] Ran, 2026, Real-world prospective validation and economic evaluation of deep learning-based diabetic retinopathy detection from fundus photographs: a systematic review and meta-analysis, Diabetes Care. source

  • May combine two claims.

    Premisecitation

    Product-specific evidence for IDx-DR suggests strong pooled accuracy, but performance is tied to the device, protocol, target disease definition, and intended-use population.

    “Product-specific evidence for IDx-DR also suggests strong pooled accuracy, but performance is tied to the device, protocol, target disease definition, and intended-use population (14).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [14] Khan, 2025, Diagnostic accuracy of IDx-DR for detecting diabetic retinopathy: a systematic review and meta-analysis, Am J Ophthalmol. source

  • May combine two claims.

    Premisecitation

    Real-time national deployment in Thailand and African clinical validation studies show that AI performance depends on workflow, image acquisition, and the surrounding screening pathway.

    “Real-time national deployment in Thailand and African clinical validation studies further show that AI performance depends on workflow, image acquisition, and the surrounding screening pathway (9,11).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [9] Bellemo, 2019, Artificial intelligence using deep learning to screen for referable and vision-threatening diabetic retinopathy in Africa: a clinical validation study, Lancet Digit Health. source
    • cites [11] Ruamviboonsuk, 2022, Real-time diabetic retinopathy screening by deep learning in a multisite national screening programme: a prospective interventional cohort study, Lancet Digit Health. source

  • May combine two claims.

    Premisecitation

    Pooled sensitivity of 94% and specificity of 89% among gradable images for AI DR screening, with income level and fundus camera type explaining part of the heterogeneity.

    “Joseph and colleagues synthesized 34 real-world studies and reported pooled sensitivity of 94% and specificity of 89% among gradable images, with income level and fundus camera type explaining part of the heterogeneity (12).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [12] Joseph, 2024, Diagnostic accuracy of artificial intelligence-based automated diabetic retinopathy screening in real-world settings: a systematic review and meta-analysis, Am J Ophthalmol. source

  • May combine two claims.

    Premisecitation

    The effectiveness of digital health depends on infrastructure, governance, workforce, and integration into existing services.

    “This pattern is consistent with the broader digital health principle that effectiveness depends on infrastructure, governance, workforce, and integration into existing services (5).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [5] 2019, WHO

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    Before imaging, limited invitation, low attendance, or poor access can reduce screening reach.

    “Before imaging, limited invitation, low attendance, or poor access can reduce screening reach.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    DR severity is associated with a graded decline in vision-related quality of life, and even non-vision-threatening DR may be associated with worse VRQoL.

    “Zayed and colleagues' systematic review and meta-analysis found that DR severity is associated with a graded decline in vision-related quality of life (VRQoL), and that even non-vision-threatening DR may be associated with worse VRQoL (27).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [27] Zayed, 2024, Diabetic retinopathy and quality of life: a systematic review and meta-analysis, JAMA Ophthalmol. source

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Patient-centered outcomes are central to evaluating the public health value of AI-enabled DR screening.

    “Beyond diagnostic and economic endpoints, patient-centered outcomes are central to evaluating the public health value of AI-enabled DR screening.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Same-day feedback may support follow-up only when patients understand what a positive, negative, or non-diagnostic result means and receive clear counseling, referral explanation, and navigation.

    “Same-day feedback may support followup only when patients understand what a positive, negative, or non-diagnostic result means and receive clear counseling, referral explanation, and navigation.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationcitation

    The purpose of earlier detection in AI-enabled DR screening is not merely to identify retinal lesions but to prevent disability and protect daily functioning.

    “These findings matter for AI screening because the purpose of earlier detection is not merely to identify retinal lesions, but to prevent disability and protect daily functioning.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Patient-centered outcomes are particularly important when AI provides same-visit results, because screening shifts from passive image acquisition to immediate risk communication.

    “These outcomes are particularly important when AI provides same-visit results, shifting screening from passive image acquisition to immediate risk communication.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Generic health-related quality-of-life instruments may be less sensitive than vision-specific measures for DR, and visual acuity alone does not fully explain patient experience.

    “The review also suggested that generic health-related quality-of-life instruments may be less sensitive than vision-specific measures, and that visual acuity alone does not fully explain patient experience.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [27] Zayed, 2024, Diabetic retinopathy and quality of life: a systematic review and meta-analysis, JAMA Ophthalmol. source

  • May combine two claims.

    Premisecitation

    Referral uptake has improved in some AI-assisted pathways, yet few studies follow patients through treatment and visual outcomes.

    “Referral uptake has improved in some AI-assisted pathways, yet few studies follow patients through treatment and visual outcomes (19)(20)(21)(22)(23).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites (19)(20)(21)(22)(23)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

39 of 39

  • The paper relies on this without support.

    Diagnostic studies often evaluate gradable images, whereas screening programs are accountable for the eligible population with diabetes.

    “The central issue is a shift in denominator: diagnostic studies often evaluate gradable images, whereas screening programs are accountable for the eligible population with diabetes.”
1 of 1

Cites

Flagged first, then by citations

Authors
Chen, J., & Chen, B.
Affiliations
Leshan Normal University (CN)
Corresponding author
Chen, Jing
Venue
Frontiers in Public Health
Publisher
Frontiers Media SA
Published
2026-10-05
Volume
14
ISSN
2296-2565 (ISSN-L)
DOI
10.3389/fpubh.2026.1976572
OpenAlex
W7220439764
Semantic Scholar
cb26a8aac8e0cf64616dec84e32421d899cf2f5c
Cited by
0
References
31
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
  • Frontiers in Public Health (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Topics
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Retinal Imaging and Analysis
  • Health Sciences › Medicine › Ophthalmology › Retinal Diseases and Treatments
  • Health Sciences › Medicine › Ophthalmology › Ophthalmology and Visual Health Research
Keywords
diabetic retinopathy, population health impact, health equity, implementation, referral completion, cascade of care, cost-effectiveness, fundus photography, smartphone-based screening, blindness prevention, AI governance

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-13.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-13.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review