Objective:To explore the relationship between the levels of serum midkine (Midkine), a disintegrin and metalloproteinase domain-containing protein 10 (ADAM10), and angiopoietin-like protein 3 (ANGPTL3) and the occurrence and progression of diabetic retinopathy (DR), and to construct a DR risk stratification model incorporating the aforementioned serum markers.Methods:A cross-sectional study. A total of 152 patients with type 2 diabetes (T2DM) (including 300 eyes) who were hospitalized in Department of Ophthalmology and/or Endocrinology of Taizhou Traditional Chinese Medicine Hospital from June 2024 to June 2025 were included in the study. According to the severity of DR, the patients were divided into the no DR (NDR) group, the non-proliferative DR (NPDR) group and the proliferative DR (PDR) group. Each group consisted of 50 cases (100 eyes), 61 cases (120 eyes) and 41 cases (80 eyes), respectively. The demographic characteristics and laboratory test data of patients, including blood glucose, blood lipids, and serum levels of Midkine, ADAM10, and ANGPTL3 were systematically collected. The generalized linear mixed-effects model (GLMM, Logit link function) was used to screen the independent risk factors for the occurrence and progression of DR, and calculate the odds ratio and its 95% confidence interval. Three risk-stratification models were developed via scoring based on GLMM regression coefficients (
β-coefficients): Model 1 included only conventional clinical indicators [T2DM duration, fasting blood glucose (FBG), glycated hemoglobin (HbA1c), low-density lipoprotein-cholesterol (LDL-C), high-density lipoprotein-cholesterol (HDL-C)]. Model 2 incorporated solely serum biomarkers (Midkine, ADAM10, ANGPTL3); Model 3 was the full combined model integrating both conventional indicators and serum biomarkers. Receiver-operating characteristic curves and the area under the curve (AUC) were applied to evaluate predictive performance of each model, and 10-fold cross-validation was performed to assess model prediction accuracy.
Results:Compared with the NDR group, the NPDR and PDR groups exhibited elevated levels of T2DM duration (
F=89.21), FBG (
F=19.87), HbA1c (
F=74.35), LDL-C (
F=16.92), Midkine (
F=162.18), ADAM10 (
F=69.85), and ANGPTL3 (
F=4.47), accompanied by decreased HDL-C level (
F=18.95). All differences were statistically significant (
P<0.05). GLMM analysis demonstrated that increased levels of T2DM duration, FBG, HbA1c, LDL-C, Midkine, ADAM10, and ANGPTL3 significantly raised the risk of DR onset and progression (
β = 0.285, 0.256, 0.658, 0.486, 1.135, 0.003, 0.016;
P<0.05), whereas elevated HDL-C level reduced such risk (
β=?0.732,
P<0.05). GLMM analysis revealed that elevated levels of T2DM duration, FBG, HbA1c, LDL-C, Midkine, ADAM10, and ANGPTL3 significantly increased the risk of DR onset and progression (
β=0.285, 0.256, 0.658, 0.486, 1.135, 0.003, 0.016;
P<0.05), whereas increased HDL-C level reduced this risk (
β = ?0.732,
P<0.05). The AUCs of Model 1, 2 and 3 for predicting DR onset and progression were 0.759, 0.847 and 0.927, with 10-fold cross-validation accuracies of 72.10%, 80.23% and 88.20%, respectively. Model 3 had statistically superior AUC and cross-validation accuracy versus Model 1 and Model 2 (
P<0.05). Model 3 produced risk scores ranging from ?12.3 to 18.7. Eyes were stratified into low-risk (≤2.1), medium-risk (2.2-8.5), and high-risk (≥8.6) subgroups according to score tertiles, which correspond to NDR, mild-to-moderate NPDR, and severe NPDR-PDR respectively.
Conclusions:Serum levels of Midkine, ADAM10 and ANGPTL3 are significantly positively correlated with the onset and progression of DR. The full-combined model exhibits better predictive performance for DR onset than single-model approaches using either conventional indicators or serum biomarkers alone.