Ocular Capillary Ischemia Offers Early Warning of Diabetic Eye Disease Progression

An AI-driven automated optical coherence tomography angiography (OCTA) review system predicted progression of diabetic macular ischemia (DMI) with greater than 90% accuracy, a retrospective cohort study showed.

DMI of the superficial capillary plexus at baseline more than doubled the risk of progression of diabetic retinopathy (DR), and DMI of the deep capillary plexus more than tripled the risk. Evidence of deep capillary plexus DMI at baseline also predicted a significantly increased risk of diabetic macular edema (DME) and deterioration of visual acuity (VA).

The findings suggest OCTA-based evaluation of DMI may improve assessment of a patient's risk of DR progression, DME development, and VA deterioration beyond traditional risk factors and warrant consideration for use in clinical practice, reported Carol Y. Cheung, PhD, of the Chinese University of Hong Kong, and co-authors, in JAMA Ophthalmology.

"Our results showed that the presence of DMI in either one plexus or both plexuses were significantly associated with improved discriminative performance for DR progression," the authors stated. "Previous experimental research has reported that retinal ischemia is an important detrimental factor contributing to DR progression, as progressive retinal ischemia might initiate excessive expression of proangiogenic growth factor to stimulate aberrant formation of retinal neovascularization and ultimately lead to vitreous hemorrhage or tractional retinal detachment."

"As DMI has been found to contribute to progressive visual loss among some patients with DR despite treatment, considering the intricate relations between DMI and DR, our results might advocate the early detection of DMI among a diabetes population to herald early intervention, such as stricter metabolic control, to mitigate both DMI and DR progression and subsequently arrest progressive VA deterioration," they added.

The study has three important implications for evaluation and management of diabetic eye disease, according to the authors of an accompanying editorial.

"First, it provides a foundation for revisiting the classification of early DR stages to incorporate measures of subclinical impairment in capillary perfusion," wrote Amir Kashani, MD, PhD, of John Hopkins Medicine in Baltimore, and co-authors. "Second, it makes a compelling case to adopt OCTA as a standard care platform for assessment of early ischemic retinal disease, at least in patients with diabetes. Third, it is a case example that demonstrates the powerful combination of relatively basic AI methods and highly sensitive noninvasive imaging markers that can provide significant clinically impactful prognostic information about DR."

"Changes to our clinical practice based on these lessons will set the stage for clinical trials aimed at treatment of capillary nonperfusion rather than secondary complications of DR, such as macular edema and neovascularization," they said.

Cheung and colleagues recently proposed binary DMI assessment of superficial capillary plexus and deep capillary plexus OCTA images and developed a deep learning system compatible with two OCTA devices. To continue the research, they conducted a retrospective cohort study to determine whether baseline DMI status provides incremental predictive value for diabetic retinal disease progression and VA deterioration beyond traditional risk factors.

Investigators identified adult patients with a diagnosis of type 2 diabetes in July 2015 and followed for at least 4 years. Each patient underwent the same ophthalmic assessment and interview-based questionnaire.

At baseline and then at each follow-up visit, digital fundus photography was performed, and two experienced readers determined DR status and the presence of other ocular conditions from retinal photographs. DR progression was defined as an increase of at least two steps in severity as compared with baseline.

Data analysis included 321 eyes and 178 patients, who had a median age of 63. All study participants self-identified as Chinese. During a mean follow-up of 50.41 months, 105 (32.71%) eyes had DR progression, 33 (10.28%) developed DME, and 68 (21.18%) had VA deterioration.

After adjustment for traditional risk factors, the presence of DMI of the superficial capillary plexus at baseline was associated with an HR 2.69 for DR progression (95% CI 1.64-4.43, P<0.001). Baseline DMI of the deep capillary plexus increased the hazard for DR to 3.21 (95% CI 1.94-5.30, P<0.001), the hazard for DME development to 4.60 (95% CI 1.15-8.20, P=0.003), and the hazard for VA deterioration to 2.12 (95% CI 1.01-5.22, P=0.04).

The authors acknowledged several limitations of the study: Only a single OCTA system was used in the study; whether the results would be consistent across other devices remains to be determined. Assessment of parafoveal DMI was limited to a single macular image size. Binary DMI classification falls short of quantitative DMI assessment. Future studies of deep learning algorithms have room for improvement on image quality control attained with the algorithm used in the study. The methodology for defining DMI was suboptimal.

Despite the limitations, the system employed in the study achieved greater than 90% area under the receiver operating characteristic curve, sensitivity, and specificity for the study endpoints, Kashani and co-authors noted.

Comment