This Week in Ophthalmology — Aug 8, 2026

Welcome to This Week in Ophthalmology. This week we're covering 10 notable papers spanning screening access and equity, artificial intelligence and standardized language in clinical practice, imaging biomarkers and drug trials in retina, and refinements in cataract and corneal surgery. Let's dive in.

We'll start with who actually gets examined. In the British Journal of Ophthalmology, a multicentre Chinese study bolted opportunistic eye screening onto routine general health examinations across five health examination centres in four provinces, enrolling nearly 64,000 adults with presenting visual acuity and non-mydriatic fundus photography read by two experienced graders [9]. The yield was substantial: non-macular drusen in almost 9 percent, macular degeneration in just over 6 percent, glaucoma suspects in about 5 and a half percent, epiretinal membrane and diabetic retinopathy each around 4 percent. Nearly 6 percent of this relatively young cohort — mean age 45 — already had low vision. The cost worked out to roughly 22 United States dollars per person screened and about 83 dollars per ocular disease suspect identified, which the authors argue makes the model affordable at population scale. Contrast that with the American Journal of Ophthalmology, where a 20 percent sample of United States fee-for-service Medicare claims covering more than a million beneficiaries with diabetes showed annual diabetic retinopathy screening rates that never exceeded about 55 percent, and drifted down to just under 52 percent by 2022 [10]. Adherence fell steadily as neighbourhood Area Deprivation Index rose. Patients in the most deprived neighbourhoods were about a quarter less likely to have an annual exam than those in the least deprived, and after adjusting for demographics, comorbidities, the pandemic, and county-level healthcare resources, they were still roughly 16 percent less likely. So the deprivation gradient is not simply explained by local supply of eye care. Read together, these two papers make a pointed argument: in a well-insured population we are missing nearly half of eligible diabetic patients, disproportionately the poorest, while a low-cost opportunistic model embedded in general health checks can detect multiple fundus diseases cheaply. The actionable step is to think about where your patients already show up — primary care, health screening, endocrinology — rather than waiting for referral.

On the diagnostic technology side, Ophthalmology published a systematic review and meta-analysis of image-based artificial intelligence for thyroid eye disease, pooling 19 studies and nearly 8,750 participants [5]. For diagnosis, pooled sensitivity was 0.93 with specificity 0.84 and an area under the curve of 0.93; for clinical activity score-based activity grading, sensitivity was 0.82 and specificity 0.87. Artificial intelligence had higher sensitivity than the ophthalmologist comparators, but those comparators were selected reader groups working with limited clinical context, so that comparison should be read cautiously. Importantly, the authors do not endorse autonomous deployment: certainty of evidence was low, between roughly one-eighth and one-half of studies were at high risk of bias depending on the task, and while a negative result may be useful as a rule-out aid under supervision, activity grading should not be used to exclude low-clinical-activity-score active disease. Patients with periocular or visual symptoms still need an ophthalmologist. Sitting alongside this is a JAMA Ophthalmology international modified Delphi study that standardized the language of retinal vasculitis [1]. Twenty-two independent experts across ophthalmology, rheumatology, pathology, and imaging reached consensus — 95 to 100 percent agreement in the second round — on definitions for eight terms, including retinal vasculitis, vascular leakage, infectious versus noninfectious vasculitis, perivasculitis, retinal vasculopathy, primary retinal vasculitis, and far-peripheral vascular leakage. The central clinical point is the deliberate separation of angiographic vascular leakage from true vasculitis, a conflation that has driven both overtreatment and uninterpretable literature. And note the link to the previous paper: you cannot train or validate a credible classification algorithm on labels that mean different things in different clinics.

Our third theme is retinal imaging biomarkers and the disappointing state of retinal pharmacotherapy. In Investigative Ophthalmology and Visual Science, an analysis of 102 untreated fellow eyes from the MAHALO trial tested the long-standing assumption that hyperautofluorescent perilesional rings mark the direction of geographic atrophy growth [8]. Using normalized autofluorescence intensity in perilesional zones out to 250 and 500 microns and clock-hour sector mapping of border expansion, higher hyperautofluorescence density was not associated with greater expansion — in fact the regression slopes were consistently negative — while higher hypoautofluorescence density was significantly associated with greater expansion across every ring size and threshold tested, including in unifocal lesions. If replicated, that reverses a common interpretive habit at the viewing box. Complementing this, a Survey of Ophthalmology review synthesizes what we know about macular atrophy in neovascular age-related macular degeneration, which becomes progressively more common in treated eyes and is now a principal limit on durable visual gain despite good exudative control [7]. Risk factors include older age, poor baseline acuity, type 3 macular neovascularization, intraretinal fluid, subretinal hyperreflective material, and early atrophic signs at baseline, whereas type 1 neovascularization may progress more slowly. The authors highlight the unresolved controversy over whether cumulative anti-VEGF exposure independently drives atrophy or simply marks more aggressive disease — worth remembering when counselling patients who ask whether injections are damaging their retina. On the therapeutic front, Ophthalmology Retina published a systematic review and meta-analysis of pharmacologic prevention and treatment of proliferative vitreoretinopathy, including 69 studies with 31 in the meta-analysis [4]. Five-fluorouracil with low-molecular-weight heparin for prevention, methotrexate for established disease, corticosteroids, and anti-VEGF therapy each showed no significant improvement in anatomical outcomes individually. Pooled across agents there was a reduction in redetachment risk of roughly half, driven mainly by the methotrexate and anti-VEGF subgroups, and retinoids showed a possible signal for reattachment — but certainty of evidence was low to very low. The conclusion is plain: current evidence does not support routine pharmacologic adjuvants, and management remains surgical pending adequately powered randomized trials. Also in Ophthalmology Retina, a Korean nationwide cohort of over 462,000 newly diagnosed cancer patients found modestly increased risks of retinal artery and vein occlusion versus the general population — about 27 percent higher for arterial and 15 percent higher for venous occlusion [6]. The signal concentrated in hematologic malignancies, where arterial occlusion risk was roughly two and a half times that of the general population, and in eye and adnexal cancers, where venous occlusion risk was about three times higher. Absolute five-year incidence remained low, at 0.13 percent for arterial and 0.91 percent for venous occlusion, so this is a reason to lower your threshold for hematologic workup in a cancer survivor with an occlusion, not a reason for routine surveillance.

Finally, two papers for the surgeons. In the American Journal of Ophthalmology, a multicentre retrospective database study of 3,738 eyes with prior myopic laser vision correction, operated by 145 surgeons across United States centres, compared 14 no-history intraocular lens power formulas using IOLMaster 700 biometry [2]. Modern formulas beat early-generation Haigis-L and Shammas-PL, and formulas incorporating posterior corneal curvature beat their own non-posterior versions. Among posterior-curvature formulas, Cooke K6 led at a mean absolute error of 0.37 dioptres, with PEARL-DGS, EVO 2.0, and Barrett True K all at 0.38. The clinically meaningful message is where the differences matter: in steep corneas, K6 put an additional 8 to 10 percent of eyes within half a dioptre compared with Barrett True K, and in short eyes under 23.5 millimetres it gained more than 13 percent within half a dioptre. Those extreme eyes were 22 percent of the cohort; in typical post-laser-correction eyes the differences were negligible. When full biometry is unavailable, Shammas-Cooke is a reasonable fallback. And in Survey of Ophthalmology, a systematic review of preloaded Descemet membrane endothelial keratoplasty pooled 39 studies, 19 clinical and 20 laboratory [3]. Across 1,420 clinical eyes the weighted rebubbling rate was 25.5 percent overall — 30.4 percent for endothelium-in versus 19.3 percent for endothelium-out — while visual improvement and twelve-month endothelial cell loss of 38 versus 32 percent did not differ significantly between configurations. Laboratory work showed only about 10 percent cell loss after up to five days stored in the delivery device, supporting the shipping model. The endothelium-out rebubbling advantage is a trend, not a proven difference, given considerable heterogeneity.

If you only have time for one paper this week, make it the American Journal of Ophthalmology study of no-history intraocular lens formulas after myopic laser vision correction [2]. It tells you precisely which subgroup of post-refractive eyes — short or steep — justifies switching formulas, and where chasing the newest calculator buys you nothing.

Here are the key takeaways from this week in Ophthalmology. First, roughly half of United States Medicare beneficiaries with diabetes are missing annual retinopathy screening, and the deficit worsens with neighbourhood deprivation even after adjusting for local resources — while opportunistic screening embedded in general health examinations detects substantial disease for about 22 dollars a head. Second, artificial intelligence for thyroid eye disease performs well enough to serve as a supervised rule-out aid but not for autonomous activity grading, and consensus definitions for retinal vasculitis now give us the vocabulary such tools depend on. Third, stop treating hyperautofluorescent rings as the direction of geographic atrophy growth; in this analysis, hypoautofluorescence-dominant sectors predicted expansion. Fourth, no pharmacologic adjuvant is yet established for proliferative vitreoretinopathy — surgery remains the answer. Fifth, in cancer survivors, especially with hematologic or ocular adnexal malignancy, retinal vascular occlusion risk is modestly elevated, so investigate rather than screen. And sixth, for post-laser-correction cataract eyes, use a modern posterior-corneal-curvature formula, and pay particular attention in short or steep eyes.

That's your roundup for This Week in Ophthalmology. The full transcript and references are available on the episode page in your AudioScholar library. This is an AI-curated summary — for clinical decisions, always consult primary sources and current guidelines. See you next week.

If this weekly briefing is useful, follow the show in your podcast app so new episodes arrive automatically. And if you have a paper you have been meaning to read, upload the PDF, or paste any link, at audioscholar dot C C. We will turn it into audio like this one, in any of thirty-one languages.

This is an automated summary generated by artificial intelligence, which can make mistakes. Always review the original source materials.