A blood test using artificial intelligence has identified 71 proteins linked to early retinal neurodegeneration in diabetes—potentially allowing doctors to predict and prevent vision loss years before symptoms appear.
Researchers at the Guangdong Provincial Clinical Research Center for Ocular Diseases in Guangzhou, China, published their findings June 2 in PLOS Medicine, revealing how machine learning could transform diabetic eye care by detecting subtle biochemical signals in blood long before retinal damage becomes visible. The model, called Pro-DRN, outperforms existing methods by 26% and has already been deployed as an interactive risk assessment tool for clinicians.
What the Blood Test Detects—and Why It Matters
Diabetic retinal neurodegeneration (DRN) is a silent but devastating complication of diabetes, affecting more than half a billion people worldwide. Unlike traditional diabetic retinopathy—which causes bleeding or swelling in the retina—DRN involves the gradual breakdown of retinal neurons, the cells responsible for converting light into neural signals. The damage often goes undetected until severe visual impairment sets in, by which point treatments are far less effective.
What makes this study groundbreaking is its focus on early molecular signals in blood plasma. By analyzing samples from 1,492 type 2 diabetes patients in the Guangzhou Diabetic Eye Study—none of whom had visible DRN at enrollment—the researchers identified 71 proteins associated with the condition. These proteins are involved in inflammation, cellular maintenance, and structural integrity of nerve and muscle tissue, suggesting that DRN may leave a measurable biochemical footprint in the bloodstream years before symptoms appear.
“Our study suggests that early retinal nerve damage in diabetes leaves measurable signals in the blood.”
The team then validated these findings in an independent cohort of 502 diabetic patients from the UK Biobank, confirming that the same protein signatures appeared in a geographically distinct population. This cross-continental replication strengthens confidence in the model’s reliability, according to Inside Precision Medicine, which noted that the three most predictive proteins—ACTA2, COL6A3, and HSPG2—are structural components critical for maintaining nerve and muscle integrity in the eye.
The 26% Advantage: How Pro-DRN Outperforms Existing Methods
The researchers developed Pro-DRN, a machine learning model that integrates proteomic data with longitudinal retinal imaging data collected over six years. Unlike current diagnostic methods—which rely on retinal scans that only detect damage after it’s irreversible—Pro-DRN uses blood protein levels to predict DRN risk with 26% greater accuracy than the best-performing existing model.
This improvement is not incremental; it represents a potential paradigm shift in early detection. According to the study published in PLOS Medicine, the model was trained on data from 1,218 participants in the Guangzhou study and tested on the UK Biobank cohort. The consistency of results across two diverse populations suggests the model could be broadly applicable, though real-world deployment will require further validation in clinical settings.
The implications extend beyond vision. DRN is often one of the first signs of broader diabetic neurodegeneration, which can include cognitive decline, dementia, and peripheral neuropathy. By identifying high-risk patients early, Pro-DRN could enable targeted interventions—not just for the eyes, but for the nervous system as a whole.
A Tool Already in Use: How Clinicians Can Access Pro-DRN
The researchers have made Pro-DRN available as an interactive, web-based risk assessment tool, allowing doctors to input patient blood protein data and receive a risk stratification score. This could enable earlier, more frequent monitoring for patients identified as high-risk, potentially catching DRN before it causes irreversible damage.
While the model is not yet widely adopted, its deployment marks a significant step toward personalized diabetes care. As News-Medical reported, the tool is designed to complement—not replace—existing diagnostic methods. Clinicians would still rely on retinal imaging for confirmation, but Pro-DRN could flag patients who need closer surveillance.
What Comes Next: Challenges and Opportunities
The road to clinical integration is not without hurdles. First, the model’s accuracy must be validated in larger, more diverse populations. The current study included patients primarily from China and the UK, and its generalizability to other ethnic groups or diabetes subtypes remains untested. Additionally, the cost and accessibility of proteomic blood testing could limit its immediate adoption in resource-constrained settings.
Regulatory approval will also be critical. While the study was published in a peer-reviewed journal, Pro-DRN has not yet received clearance from health authorities like the FDA or EMA. If approved, it could become the first AI-driven diagnostic tool for early DRN detection, paving the way for similar applications in other neurodegenerative conditions linked to diabetes.
Beyond DRN, the findings raise broader questions about diabetes as a systemic neurodegenerative disease. If blood protein signatures can predict retinal damage, could they also forecast cognitive decline or peripheral neuropathy? Early evidence suggests they might, but more research is needed to explore these connections.
“These findings suggest that a simple blood test analyzed with artificial intelligence may help identify people with diabetes who are at highest risk of early retinal nerve damage, well before visible damage appears on the retina.”
The potential for such a test to transform diabetes care is enormous. Currently, nearly 50% of diabetic patients develop some form of retinal damage, often without early warning signs. Pro-DRN could shift the focus from reactive treatment to preventive care, allowing clinicians to intervene before irreversible damage occurs.
Yet, as with any AI-driven diagnostic tool, ethical and practical questions remain. Who will have access to this technology? How will insurance systems cover the associated costs? And perhaps most importantly, how will clinicians integrate these predictions into their decision-making without overburdening patients with unnecessary anxiety?
The Bigger Picture: Diabetes as a Nervous System Disorder
DRN is not just an eye disease—it’s a window into the broader neurodegenerative effects of diabetes. The retina is one of the most metabolically active tissues in the body, making it particularly vulnerable to metabolic stress. When retinal neurons begin to degenerate, it often signals that similar damage is occurring in other parts of the nervous system, including the brain and peripheral nerves.
This study underscores the idea that diabetes is not merely a metabolic disorder but a neurodegenerative condition with far-reaching implications. The proteins identified in the blood—many of which are involved in inflammation and cellular maintenance—are not unique to the retina. They appear in other tissues affected by diabetes, suggesting that a single blood test could one day provide a comprehensive snapshot of a patient’s neurological risk profile.
If further research confirms these links, Pro-DRN could evolve into a broader screening tool for diabetic neurodegeneration. Imagine a future where a routine blood test not only checks for retinal risk but also flags patients at higher risk for cognitive decline or peripheral neuropathy. That future may be closer than we think.
For now, the focus remains on the eyes—but the implications stretch far beyond.