AAO 2019: Single session devoted to diabetes

October 30, 2019

News

Article

At this year’s Retina Subspecialty Day during the American Academy of Ophthalmology (AAO), presentations as varied as artificial intelligence (AI), news from the Diabetic Retinopathy Clinical Research Retina Network (DRCR.net), using optical coherence tomography (OCT) to predict visual outcomes, and how to manage patients with high-risk nonproliferative diabetic retinopathy (DR) without diabetic macular edema (DME) were among the highlights.

Here is a brief overview:

Tien Yin Wong, MBBS, MMed(Ophth) provided an overview of current AI-deep learning systems for DR screening and noted a full automated system has “significant potential for increasing efficiency, reproducibility, and coverage of DR screening programs, reducing cost and access barriers, and improving patient outcomes by providing accurate early detection of DR and, thus, referral for appropriate treatment.”

For example, the FDA has approved IDx’s IDx-DR (a software program that uses an AI algorithm to analyze images of the eye taken with the Topcon NW400). In 2018, the IDx-DR’s first approval was for detecting greater than a mild level of DR in adults who have diabetes without an assisted interpretation by a clinician.

Google Health developed a deep learning system for detecting referable DR using more than 128,000 images that were graded 3 to 7 times for DR, diabetic macular edema (DME), and image gradeability; the company’s algorithm achieved a sensitivity of >87% and specificity of >90%. The Singapore Eye Lesion Analyzer (SELENA) screens for not only DR, but also glaucoma and age-related macular degeneration.

These systems are not without potential biases that may limit commercial acceptance, including medicolegal and operational issues, he said. Other challenges include that the “AI black box” remains unacceptable to physicians and likely to patients as well; AI-deep learning models require large amount of high-quality data with “gold-standard labelling,” he said.

So training that uses a single clinical dataset is limited by potential biases, and data sharing across countries and centers is complex and requires regulatory approval. Finally, most current AI-deep learning algorithms are designed in research settings and may not be applicable in real world settings.

Neil M. Bressler, MD, noted visual acuity responses at 12 weeks following three monthly injections of anti-vascular endothelial growth factor (VEGF) agents is associated with 2-year outcomes, regardless of which anti-VEGF is used (at the time of this study, brolucizumab had not been approved).

Dr. Bressler presented findings from a deeper analysis of Protocol T on behalf of the DRCR.net. (Protocol T compared bevacizumab, aflibercept, and ranibizumab over the course of 2 years; conclusions favored aflibercept after year 1, but there were no differences between aflibercept and ranibizumab by year 2.) One question the group hoped to address is whether or not the findings from Protocol I were supported by Protocol T.

Dr. Bressler said that “when continuing to follow the DRCR.net treatment regimen for DME beyond 12 weeks, a suboptimal response (defined as <5 letter gain) from baseline to week 12 often was followed by a subsequent meaningful vision improvement (defined as a minimum 2-line gain) from baseline to 2 years. This vision improvement occurred even when patients remained on the same intravitreal therapy.

“There is little evidence to suggest that switching from DRCR.net anti-VEGF treatment regimens for DME will result in better vision results,” he said.

As an example, Protocol U showed that mean VA improvement by 6 months was no better in the combined dexamethasone/ranibizumab group than in the ranibizumab group, “even though, on average, there was a greater reduction in retinal thickness in the combination group.”

Managing patients with high-risk nonproliferative DR but who do not have DME is the underlying purpose of the PANORAMA study, and Rishi P. Singh, MD, presented the 1 year results. PANORAMA compared the efficacy and safety of intravitreal aflibercept to sham; the study enrolled 402 patients whose Diabetic Retinopathy Severity Scale (DRSS) were 47 or 53, who had an absence of center-involved DME, and who had a baseline VA score of ≥ 69 letters.

At week 52, 65% and 80% of aflibercept every 16 weeks and every 8 weeks, respectively, had a ≥2-step improvement in DRSS score, compared with only 15% in the sham group. Between 3% and 4% of patients in the aflibercept arms developed vision-threatening complications, compared with 20% in the sham group (p < 0.0001 for both).

John Wells, MD, presented results on behalf of the DRCR.net’s Protocol T extension study. Participants who had completed the original 2-year Protocol T were asked to complete a follow-up visit about 5 years after they were randomized. (At the end of that 2-year study, eyes with better baseline vision of 20/32 to 20/40, there was no difference among eyes treated with aflibercept, bevacizumab, or ranibizumab, with a mean gain of about 8 letters in all three groups.)

Of the original 660 patients, 317 (about 67% of the available patients) returned for the 5-year follow-up visit.

Generally, Dr. Wells said 95% of those patients had undergone at least one retinal exam over the 3 years, there was a mean of 14 visits over those additional 3 years, and 70% had at least one treatment for DME or DR over the 3 years.

Aflibercept was given most often (38% of patients); only 10% of eyes received any steroid treatment, and only 8% received laser treatment, Dr. Wells said. Overall, an average of four treatments were given over the 3 years.

In terms of vision gains, Dr. Wells said at 2 years there was about a 12-letter gain, but this had dropped to 7 letters by 5 years, regardless of where patients started on the vision scale.

About half the eyes still had 20/25 vision, however. “Interestingly, the OCT central subfield thickness did not change from year 2 to year 5, regardless of treatment group,” he said.