FRIDAY, June 12, 2026 (HealthDay News) -- A proteomics- and machine learning (ML)-based precision prediction system enhances early risk stratification for diabetic retinal neurodegeneration (DRN), according to a study published online June 2 in PLOS Medicine.
Huangdong Li, from Sun Yat-Sen University in Guangzhou, China, and colleagues integrated high-throughput plasma proteomics with longitudinal optical coherence tomography (OCT) in two independent populations in a prospective observational study. A total of 1,492 participants were included in the discovery cohort and had baseline plasma proteomics and OCT; 1,218 were followed with repeated OCT over six years. The annualized OCT-derived retinal nerve fiber layer thinning rate was used to quantify DRN.
In multivariable analyses, the researchers identified 71 plasma proteins associated with development and progression of DRN. A proteomics-based DRN model (Pro-DRN) was developed using eight ML algorithms. Pro-DRN achieved a C-index of 0.860 in the independent test set; when integrated with clinical variables, the C-index increased to 0.908. Pro-DRN improved discrimination and reclassification compared with six conventional models. ACTA2, COL6A3, and HSPG2 were among the proteins most consistently driving model performance. Core protein signals and consistent effect directions were seen in cross-ethnic external validation in the U.K. Biobank (502 participants).
"Our study suggests that early retinal nerve damage in diabetes leaves measurable signals in the blood," the authors said in a statement. "By combining plasma proteomics, longitudinal retinal imaging, and explainable AI, Pro-DRN may help move diabetic eye care from detecting established damage toward earlier, molecularly informed risk stratification, so that closer monitoring and future neuroprotective interventions can be directed to the people most likely to benefit."
