Development of risk assessment models for microvascular complications in type 2 diabetes mellitus patients: a cross-sectional study Authors: Yuhan Lin, Zhenrun Zhan, Ke Zheng, Fengying Zhao, Jiebin Lin, Jiayong Wu, Yalong Tu, Sunjie Yan, Ximei Shen Published: 2026 ABSTRACT This study aimed to develop and validate nomogram risk estimation models for assessing the risk of microvascular complications, specifically diabetic retinopathy (DR) and diabetic kidney disease (DKD), in patients with type 2 diabetes mellitus (T2DM). A retrospective cross-sectional study was conducted involving 6,043 T2DM patients. Clinical and laboratory data were collected. Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to identify significant risk factors from 38 initial variables. These selected variables were then incorporated into multivariate logistic regression analyses to build the risk estimation models. The models were presented as nomograms and were internally validated using bootstrap resampling. Their performance was evaluated by assessing discrimination (using Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC)), calibration (using calibration plots), and clinical utility (using Decision Curve Analysis (DCA)). The LASSO regression identified 8 and 15 independent risk factors for DR and DKD, respectively. The DR model included age, DM duration, insulin use, peripheral neuropathy, heart rate, HbA1c, and total protein. The DKD model included carotid atherosclerosis, diabetic foot, DM duration, α-glucosidase inhibitor use, insulin use, hypertension history, systolic blood pressure, lymphocyte count, total cholesterol, triglycerides, albumin, BMI, uric acid, and creatinine. Multivariate logistic analysis confirmed these associations. The nomograms demonstrated acceptable discrimination, with AUCs of 0.703 (training) and 0.732 (validation) for the DR model, and 0.802 (training) and 0.713 (validation) for the DKD model. Calibration curves showed good agreement between estimated and observed probabilities. This study successfully established and validated effective nomogram models for estimating the risk of DR and DKD in T2DM patients. Utilizing readily available clinical parameters, these models demonstrate moderate discriminative ability and clinical utility, potentially aiding in the early identification of high-risk individuals for targeted screening and improved management of microvascular complications in T2DM.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-57284-w. Introduction Diabetes is a metabolic disease characterized by hyperglycemia resulting from insufficient insulin secretion or impaired insulin function. In China, the incidence of Type 2 Diabetes Mellitus (T2DM) is increasing annually, becoming the third most prevalent disease after cardiovascular diseases and cancer. According to the 2021 International Diabetes Federation (IDF) survey data, there are approximately 571 million diabetic patients globally, and China has the highest number of adult diabetic patients1. This disease has numerous complications that can affect multiple organs, seriously endangering patients’ lives and health2. Microvascular complications are among the common complications of T2DM, primarily related to genetic factors, metabolic abnormalities, increased insulin resistance, etc. Due to insufficient insulin secretion or insulin resistance in T2DM patients, blood glucose levels rise, leading to metabolic disorders. Advanced glycation end products can easily cause vascular endothelial cell damage, triggering inflammatory responses and ultimately leading to the development of peripheral vascular complications3,4. Diabetic Retinopathy (DR) is a highly prevalent microvascular complication in T2DM patients. DR’s main clinical manifestations include neovascularization, retinal hemorrhage and exudation, and macular edema, which can cause mild to moderate retinal damage or even blindness in T2DM patients. Current research confirms that DR is a leading cause of adult blindness5. Simultaneously, Diabetic Kidney Disease (DKD) is also a common and major microvascular complication of diabetes, with a clinical incidence ranging from 9.73 to 47.6%. It is a significant cause of end-stage renal disease and chronic kidney disease, and a major reason for the increased mortality rate among diabetic patients6,7. The typical symptoms of DKD include edema, hypertension, and massive proteinuria. Early-stage DKD is reversible. If diagnosed early and treated promptly, proteinuria can be effectively reduced, further mitigating kidney damage, thereby improving patients’ quality of life and life expectancy8. Therefore, early identification of risk factors associated with microvascular complications in T2DM and implementation of targeted strategies are crucial for preventing these complications and improving prognosis. Currently, clinical research on risk factors for microvascular complications in T2DM is relatively scarce, and there is a lack of scientific, simple risk assessment tools. The nomogram risk estimation model, based on evidence-based medicine and precision treatment, can integrate relevant variables to estimate the probability of disease onset or prognostic outcomes9 and has been widely used to assist clinical diagnosis10,11. This study retrospectively analyzed patient clinical data to establish a quantifiable nomogram risk estimation model for accurately assessing the risk of microvascular complications in T2DM, aiming to identify high-risk populations, enhance risk stratification, and promote personalized medicine. Materials and methods Study subjects A total of 6043 T2DM patients treated in our department from April 2008 to June 2024 were selected as study subjects. The diagnosis of T2DM referred to the diabetes diagnostic criteria12. Inclusion criteria: (1) Clearly diagnosed with T2DM; (2) Complete inpatient medical records. Exclusion criteria: (1) Type 1 diabetes, secondary diabetes, severe liver or kidney dysfunction, cardiac function class III or above; (2) Acute diabetic complications; (3) Various acute or chronic infections; (4) Systemic immune diseases; (5) Hematological diseases; (6) Dialysis or renal transplantation; (7) Malignant tumors; (8) Takayasu’s arteritis; (9) Thromboangiitis obliterans. Diabetic Retinopathy (DR) was diagnosed by ophthalmologists through fundus examination (including fundus photography and/or fluorescein angiography as clinically indicated) according to the diagnostic criteria outlined in the “Expert Consensus on Prevention and Treatment of Diabetic Retinopathy”13. Diagnosis was primarily based on the presence of characteristic microvascular lesions such as microaneurysms, retinal hemorrhages, hard exudates, cotton-wool spots, intraretinal microvascular abnormalities, venous beading, and neovascularization. Cases of hypertensive retinopathy and retinal arteriosclerosis were carefully excluded. Diabetic Kidney Disease (DKD) was diagnosed with reference to the “Chinese Expert Consensus on Clinical Diagnosis of Diabetic Kidney Disease in Adults”14. The diagnosis was based on the presence of persistent albuminuria (urinary albumin-to-creatinine ratio (UACR) ≥ 30 mg/g on at least two occasions within 3–6 months) and/or a reduced estimated glomerular filtration rate (eGFR < 60 mL/min/1.73 m²) in patients with T2DM, in the absence of other primary kidney diseases or significant non-diabetic renal pathology. Based on the above criteria, the study subjects were divided into a Diabetic Retinopathy group (n = 1306) and a No Diabetic Retinopathy group (n = 4737), and a Diabetic Kidney Disease group (n = 1280) and a No Diabetic Kidney Disease group (n = 4763). For external validation, an independent cohort of 664 T2DM patients treated at Heping Hospital Affiliated to Changzhi Medical College from January 2021 to June 2024 was enrolled. The same inclusion and exclusion criteria were applied to ensure comparability between the development and validation cohorts. DR and DKD were diagnosed using identical criteria and procedures as described above. This study was approved by the Ethics Committee of The First Affiliated Hospital of Fujian Medical University (Approval No. MRCTA, ECFAH of FMU [2017]131), and written informed consent was obtained from all participants. Data collection for the validation cohort was approved by the Ethics Committee of Heping Hospital Affiliated to Changzhi Medical College (Approval No. [2024]057). All experiments were performed in accordance with relevant guidelines and regulations. Methods This study used a cross-sectional research method. General data of the subjects were collected, including gender, age, Body Mass Index (BMI), diabetes duration, smoking history, alcohol history, Systolic Blood Pressure (SBP), Diastolic Blood Pressure (DBP), statin use (patients taking medication for ≥ 1 month before hospitalization were included in the statin group15, and comorbidities. Laboratory data of the subjects were collected, including Ankle-Brachial Index (ABI), Neutrophil count (NEUT), Lymphocyte count (LYM), Platelet count (PLT), Glycated Hemoglobin (HbA1c), Total Cholesterol (TC), Triglycerides (TG), High-Density Lipoprotein Cholesterol (HDL-C), Low-Density Lipoprotein Cholesterol (LDL-C), prealbumin, total protein, albumin, globulin, Aspartate Aminotransferase (AST), Alanine Aminotransferase (ALT), uric acid, creatinine, etc. As this is a cross-sectional study, the developed models are intended to assess the association between clinical parameters and the presence of microvascular complications at the time of evaluation, serving as risk stratification tools to identify patients who may have undiagnosed complications, rather than as predictive models for future events. Statistical analysis All data statistical analyses were performed using R software (version 4.3.0). Normally distributed measurement data were expressed as mean ± standard deviation (x̄±s), and intergroup comparisons were made using t-tests. Non-normally distributed measurement data were expressed as median (interquartile range) [M (P25, P75)], and intergroup comparisons were made using the Mann-Whitney U test. Count data were expressed as rates, proportions, etc., and intergroup comparisons were made using the chi-square test or Fisher’s exact test. Variables with statistical significance (P < 0.05) were identified through univariate analysis. Using the Matrix and glmnet packages in R, Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis was employed for optimized screening. Multivariate logistic regression analysis was performed on the selected variables to finally identify independent risk factors for constructing a model estimating the probability of microvascular complications in T2DM patients. Using the rms package in R, the Nomogram function was used to integrate the predictors and transform them into a visual Nomogram risk estimation model for microvascular complications in T2DM patients. Using the car, tcltk, pROC, and glm packages, the Bootstrap method was used for internal validation with 1000 resamples. Using the rms package in R, the calibrate function was used to construct calibration curves for the nomogram predicting the risk of microvascular complications to assess its accuracy. The test level was α = 0.05, and P < 0.05 was considered statistically significant. Results Comparison of clinical data between T2DM patients with and without retinopathy and with and without nephropathy This study included 6043 T2DM patients, comprising 1306 patients with retinopathy (21.6%) and 4737 patients without retinopathy (78.4%). Among them, 1280 patients (21.2%) had kidney disease, and 4763 patients (78.8%) did not. In the comparison between groups with and without retinopathy, characteristics such as lower extremity arterial disease, sulfonylurea use, non-sulfonylurea secretagogues, thiazolidinedione use, smoking history, alcohol history, platelet count, neutrophil count, triglycerides, HDL-C, and BMI showed no statistical difference. Compared to T2DM patients without retinopathy, those with retinopathy were older, had longer diabetes duration, higher SBP, DBP, heart rate, HbA1c, TC, ALT, AST, globulin, albumin, total protein, and LDL-C levels, lower lymphocyte count, creatinine, and uric acid levels, and higher proportions of carotid atherosclerosis, kidney disease, peripheral neuropathy, diabetic foot, use of biguanides, glycosidase inhibitors, insulin, statins, and history of hypertension. The differences were statistically significant (P < 0.05). In the comparison between groups with and without kidney disease, characteristics such as gender, lower extremity arterial disease, biguanide use history, sulfonylurea use history, non-sulfonylurea secretagogue use history, thiazolidinedione use history, statin use history, smoking history, alcohol history, platelet count, and LDL-C showed no statistical difference. Compared to T2DM patients without kidney disease, those with kidney disease were older, had longer diabetes duration, higher SBP, DBP, BMI, heart rate, neutrophil count, HbA1c, TC, TG, globulin, uric acid, and creatinine levels, lower HDL-C, ALT, AST, albumin, and total protein levels, and higher proportions of carotid atherosclerosis, retinopathy, peripheral neuropathy, diabetic foot, use of glycosidase inhibitors, insulin, and history of hypertension. The differences were statistically significant (P < 0.05). As shown in Table 1. Table 1Baseline characteristics [x ± s, n( %), M( P25, P75) ]ParameterTotal (6043 cases)Retinopathy patients (1306 cases)Non-retinopathy patients (4737 cases)P valueNephropathy patients (1280 cases)Non-nephropathy patients (4763 cases)P valueCarotid atherosclerosis (cases, %)3809(63)871(66.7)2938(62.0)0.002876(68.4)2933(61.6)0.001Retinopathy (cases, %)1306(21.6)–––553(43.2)753(15.8)0.001Lower extremity arterial vascular lesions (cases, %)1893(31.3)415(31.8)1478(31.2)0.717408(31.9)1485(31.2)0.657Nephropathy (cases, %)1280(21.2)553(42.3)727(15.3)0.001–––Diabetic foot (cases, %)274(4.5)90(6.9)184(3.9)0.001119(9.3)155(3.3)0.001Peripheral neuropathy (cases, %)4480(74.1)1050(80.4)3430(72.4)0.0011009(78.8)3471(72.9)0.001Medication history Biguanides (cases, %)3716(61.5)863(66.1)2853(60.2)0.001786(61.4)2930(61.5)0.969 Sulfonylureas (cases, %)2540(42.0)569(43.6)1971(41.6)0.216549(42.9)1991(41.8)0.503 Non-sulfonylurea secretagogues (cases, %)996(16.5)205(15.7)791(16.7)0.411199(15.5)797(16.7)0.330 Glycosidase inhibitors (cases, %)2538(42.0)593(45.4)1945(41.1)0.005628(49.1)1910(40.1)0.001 Thiazolidinediones (cases, %)520(8.6)113(8.7)407(8.6)0.956108(8.4)412(8.7)0.854 Insulin (cases, %)2220(36.7)686(52.5)1534(32.4)0.001713(55.7)1507(31.6)0.001 Statins (cases, %)1487(24.6)365(27.9)1122(23.7)0.002341(26.6)1146(24.1)0.062 Smoking history (cases, %)1790(29.6)395(30.2)1395(29.4)0.601407(31.8)1383(29.0)0.059 Hypertension history (cases, %)3265(54.0)775(59.3)2490(52.6)0.001925(72.3)2340(49.1)0.001 Alcohol history (cases, %)911(15.1)190(14.5)721(15.2)0.577192(15.0)719(15.1)0.967Male (cases, %)3282(54.3)686(52.5)2596(54.8)0.153704(55.0)2578(54.1)0.599Age (years)62(53,70)62(54,68)62(53,70)0.04565(57,72)61(53,69)0.001DM duration (years)8(3,13)10(7,17)7(2,11)0.00110(7,18)7(2,11)0.001Systolic blood pressure (mmHg)134(122,148)138(126,154)133(122,146)0.001142(130,160)132(120,145)0.001Diastolic blood pressure (mmHg)78(71,85)78(72,86)78(71,85)0.03779(71,87)78(71,85)0.011Heart rate (bpm)78(71,88)80(73,89)78(70,87)0.00180(72,89)78(70,87)0.001Lymphocyte count (×10⁹/L)1.82(1.39,2.20)1.77(1.33,2.15)1.84(1.41,2.22)0.0011.62(1.21,2.01)1.86(1.45,2.25)0.001Platelet count (×10⁹/L)230(186,270)230(188,273)230(186,269)0.346235(186,276)229(186,268)0.140Neutrophil count (×10⁹/L)3.89(3.00,4.96)3.96(3.07,5.04)3.87(2.98,4.94)0.0744.32(3.32,5.44)3.80(2.92,4.80)0.001Glycated hemoglobin (%)9.07(7.30,10.40)9.07(7.80,10.60)8.9(7.1,10.3)0.0019.07(7.50,10.28)9(7.20,10.40)0.035LDL-C (mmol/L)2.84(2.16,3.42)2.86(2.19,3.49)2.81(2.14,3.39)0.0092.86(2.07,3.54)2.84(2.17,3.38)0.481Total cholesterol (mmol/L)4.55(3.75,5.21)4.58(3.85,5.37)4.53(3.73,5.16)0.0014.58(3.72,5.52)4.53(3.76,5.16)0.017Triglycerides (mmol/L)1.50(1.01,2.10)1.60(1.02,2.12)1.48(1.01,2.10)0.0841.61(1.08,2.18)1.47(0.99,2.09)0.001HDL-C (mmol/L)1.08(0.89,1.26)1.07(0.89,1.25)1.08(0.89,1.26)0.5191.06(0.86,1.23)1.09(0.90,1.27)0.001ALT (U/L)21.00(15.00,29.00)20(14,28.48)22(15,31)0.00119(13,28.48)22(16,31)0.001AST (U/L)21.00(17.00,24.66)21(16,24.66)22(17,25)0.00121(16,24.66)21(17,25)0.037Globulin (g/L)27.04(24.80,28.60)27.04(24.40,28.40)27.04(24.80,28.70)0.02627.04(24.90,29.30)27.04(24.70,28.50)0.020Albumin (g/L)39.40(37.10,42.30)39.40(35.90,41.30)39.70(37.40,42.50)0.00137.90(33.83,40.20)39.90(37.90,42.70)0.001Total protein (g/L)66.44(63.70,69.40)66.44(62.50,68.53)66.44(64.00,69.70)0.00166.44(61.30,68.20)66.44(64.20,69.70)0.001BMI (kg/m²)24.62(22.37,26.45)24.56(22.20,26.30)24.62(22.46,26.47)0.08324.62(22.81,27.12)24.62(22.26,26.27)0.001Uric acid (µmol/L)325(259.70,382.00)327.91(270.08,389.47)320.60(256.40,380.25)0.001345.90(290.78,425.98)314.40(252.50,368.50)0.001Creatinine (µmol/L)63.00(50.20,75.00)66.00(51.00,81.20)62.20(50.10,74.50)0.00174.50(60.63,109.18)60(49.00,74.10)0.001 Table 1Baseline characteristics [x ± s, n( %), M( P25, P75) ]ParameterTotal (6043 cases)Retinopathy patients (1306 cases)Non-retinopathy patients (4737 cases)P valueNephropathy patients (1280 cases)Non-nephropathy patients (4763 cases)P valueCarotid atherosclerosis (cases, %)3809(63)871(66.7)2938(62.0)0.002876(68.4)2933(61.6)0.001Retinopathy (cases, %)1306(21.6)–––553(43.2)753(15.8)0.001Lower extremity arterial vascular lesions (cases, %)1893(31.3)415(31.8)1478(31.2)0.717408(31.9)1485(31.2)0.657Nephropathy (cases, %)1280(21.2)553(42.3)727(15.3)0.001–––Diabetic foot (cases, %)274(4.5)90(6.9)184(3.9)0.001119(9.3)155(3.3)0.001Peripheral neuropathy (cases, %)4480(74.1)1050(80.4)3430(72.4)0.0011009(78.8)3471(72.9)0.001Medication history Biguanides (cases, %)3716(61.5)863(66.1)2853(60.2)0.001786(61.4)2930(61.5)0.969 Sulfonylureas (cases, %)2540(42.0)569(43.6)1971(41.6)0.216549(42.9)1991(41.8)0.503 Non-sulfonylurea secretagogues (cases, %)996(16.5)205(15.7)791(16.7)0.411199(15.5)797(16.7)0.330 Glycosidase inhibitors (cases, %)2538(42.0)593(45.4)1945(41.1)0.005628(49.1)1910(40.1)0.001 Thiazolidinediones (cases, %)520(8.6)113(8.7)407(8.6)0.956108(8.4)412(8.7)0.854 Insulin (cases, %)2220(36.7)686(52.5)1534(32.4)0.001713(55.7)1507(31.6)0.001 Statins (cases, %)1487(24.6)365(27.9)1122(23.7)0.002341(26.6)1146(24.1)0.062 Smoking history (cases, %)1790(29.6)395(30.2)1395(29.4)0.601407(31.8)1383(29.0)0.059 Hypertension history (cases, %)3265(54.0)775(59.3)2490(52.6)0.001925(72.3)2340(49.1)0.001 Alcohol history (cases, %)911(15.1)190(14.5)721(15.2)0.577192(15.0)719(15.1)0.967Male (cases, %)3282(54.3)686(52.5)2596(54.8)0.153704(55.0)2578(54.1)0.599Age (years)62(53,70)62(54,68)62(53,70)0.04565(57,72)61(53,69)0.001DM duration (years)8(3,13)10(7,17)7(2,11)0.00110(7,18)7(2,11)0.001Systolic blood pressure (mmHg)134(122,148)138(126,154)133(122,146)0.001142(130,160)132(120,145)0.001Diastolic blood pressure (mmHg)78(71,85)78(72,86)78(71,85)0.03779(71,87)78(71,85)0.011Heart rate (bpm)78(71,88)80(73,89)78(70,87)0.00180(72,89)78(70,87)0.001Lymphocyte count (×10⁹/L)1.82(1.39,2.20)1.77(1.33,2.15)1.84(1.41,2.22)0.0011.62(1.21,2.01)1.86(1.45,2.25)0.001Platelet count (×10⁹/L)230(186,270)230(188,273)230(186,269)0.346235(186,276)229(186,268)0.140Neutrophil count (×10⁹/L)3.89(3.00,4.96)3.96(3.07,5.04)3.87(2.98,4.94)0.0744.32(3.32,5.44)3.80(2.92,4.80)0.001Glycated hemoglobin (%)9.07(7.30,10.40)9.07(7.80,10.60)8.9(7.1,10.3)0.0019.07(7.50,10.28)9(7.20,10.40)0.035LDL-C (mmol/L)2.84(2.16,3.42)2.86(2.19,3.49)2.81(2.14,3.39)0.0092.86(2.07,3.54)2.84(2.17,3.38)0.481Total cholesterol (mmol/L)4.55(3.75,5.21)4.58(3.85,5.37)4.53(3.73,5.16)0.0014.58(3.72,5.52)4.53(3.76,5.16)0.017Triglycerides (mmol/L)1.50(1.01,2.10)1.60(1.02,2.12)1.48(1.01,2.10)0.0841.61(1.08,2.18)1.47(0.99,2.09)0.001HDL-C (mmol/L)1.08(0.89,1.26)1.07(0.89,1.25)1.08(0.89,1.26)0.5191.06(0.86,1.23)1.09(0.90,1.27)0.001ALT (U/L)21.00(15.00,29.00)20(14,28.48)22(15,31)0.00119(13,28.48)22(16,31)0.001AST (U/L)21.00(17.00,24.66)21(16,24.66)22(17,25)0.00121(16,24.66)21(17,25)0.037Globulin (g/L)27.04(24.80,28.60)27.04(24.40,28.40)27.04(24.80,28.70)0.02627.04(24.90,29.30)27.04(24.70,28.50)0.020Albumin (g/L)39.40(37.10,42.30)39.40(35.90,41.30)39.70(37.40,42.50)0.00137.90(33.83,40.20)39.90(37.90,42.70)0.001Total protein (g/L)66.44(63.70,69.40)66.44(62.50,68.53)66.44(64.00,69.70)0.00166.44(61.30,68.20)66.44(64.20,69.70)0.001BMI (kg/m²)24.62(22.37,26.45)24.56(22.20,26.30)24.62(22.46,26.47)0.08324.62(22.81,27.12)24.62(22.26,26.27)0.001Uric acid (µmol/L)325(259.70,382.00)327.91(270.08,389.47)320.60(256.40,380.25)0.001345.90(290.78,425.98)314.40(252.50,368.50)0.001Creatinine (µmol/L)63.00(50.20,75.00)66.00(51.00,81.20)62.20(50.10,74.50)0.00174.50(60.63,109.18)60(49.00,74.10)0.001 Baseline characteristics [x ± s, n( %), M( P25, P75) ] LASSO regression screening of high-risk factor indicators for assessing the risk of kidney disease and retinopathy in T2DM patients As shown in Fig. 1A–B, as the parameter logλ increased, the regression coefficients continuously converged, eventually converging to 0. As shown in Fig. 1C–D, the dashed line on the right represents λ1.se, meaning one standard error to the right of λmin, corresponding to 8 and 16 variables, respectively. At this λ value, the model’s fit is good, and fewer variables are included, making the model simpler. Ultimately, 7 high-risk factors related to retinopathy in T2DM patients were identified: Age, DM Duration, Insulin Use, Concurrent Peripheral Neuropathy, Heart Rate, HbA1c, Total Protein; and 15 high-risk factors related to kidney disease in T2DM patients: Age, Concurrent Carotid Atherosclerosis, Concurrent Diabetic Foot, DM Duration, Use of Glycosidase Inhibitors, Use of Insulin, History of Hypertension, SBP, Lymphocyte Count, TC, TG, Albumin, BMI, Uric Acid, and Creatinine. Fig. 1Coefficient Path Plots for T2DM with retinopathy (A) and nephropathy (B): This study included 38 variables. Each curve represents the change trajectory of one independent variable’s coefficient. The vertical axis is the coefficient value, the lower horizontal axis is log λ, and the upper horizontal axis is the number of non-zero coefficients in the model at that point. Cross-Validation Curves for LASSO Regression of T2DM with retinopathy (C) and nephropathy (D): The X-axis is the logarithm of the penalty coefficient, log(λ); the Y-axis is the likelihood deviation. The two dashed lines on the plot represent two specific log(λ) values: the left dashed line is λ min (the λ value at which the deviation is minimized), and the right dashed line is λ 1se (the λ value within 1 standard error of the minimum deviation, often used to select a more parsimonious model). Coefficient Path Plots for T2DM with retinopathy (A) and nephropathy (B): This study included 38 variables. Each curve represents the change trajectory of one independent variable’s coefficient. The vertical axis is the coefficient value, the lower horizontal axis is log λ, and the upper horizontal axis is the number of non-zero coefficients in the model at that point. Cross-Validation Curves for LASSO Regression of T2DM with retinopathy (C) and nephropathy (D): The X-axis is the logarithm of the penalty coefficient, log(λ); the Y-axis is the likelihood deviation. The two dashed lines on the plot represent two specific log(λ) values: the left dashed line is λ min (the λ value at which the deviation is minimized), and the right dashed line is λ 1se (the λ value within 1 standard error of the minimum deviation, often used to select a more parsimonious model). Multivariate logistic regression analysis Multivariate logistic regression analysis was performed on the 8 high-risk factors for retinopathy screened by LASSO regression. As shown in Table 2 below, Age (OR = 0.976, 95%CI 0.970–0.982, P = 0.001), Kidney Disease (OR = 3.197, 95%CI 2.761–3.701, P = 0.001), Diabetes Duration (OR = 1.071, 95%CI 1.060–1.082, P = 0.001), Insulin (OR = 1.386, 95%CI 1.205–1.594, P = 0.001), Peripheral Neuropathy (OR = 1.420, 95%CI 1.208–1.669, P = 0.001), Heart Rate (OR = 1.010, 95%CI 1.004–1.015, P = 0.001), HbA1c (OR = 1.062, 95%CI 1.032–1.093, P = 0.001), Total Protein (OR = 0.979, 95%CI 0.968–0.990, P = 0.001) were independent risk factors for retinopathy in T2DM patients. Simultaneously, multivariate logistic regression analysis was performed on the 16 high-risk factors for kidney disease screened by LASSO regression. As shown in Table 3 below, the Age factor was not statistically significant for model construction and was excluded; while Carotid Atherosclerosis (OR = 1.27, 95%CI 1.090–1.481, P = 0.002), Retinopathy (OR = 3.066, 95%CI 2.617–3.592, P = 0.001), Diabetic Foot (OR = 1.635, 95%CI 1.220–2.193, P = 0.001), Diabetes Duration (OR = 1.027, 95%CI 1.016–1.038, P = 0.001), Use of Glycosidase Inhibitors (OR = 1.265, 95%CI 1.091–1.467, P = 0.002), Insulin (OR = 1.573, 95%CI 1.349–1.834, P = 0.001), History of Hypertension (OR = 1.564, 95%CI 1.320–1.852, P = 0.001), SBP (OR = 1.013, 95%CI 1.009–1.017, P = 0.001), Lymphocyte Count (OR = 0.792, 95%CI 0.701–0.894, P = 0.001), TC (OR = 1.121, 95%CI 1.053–1.193, P = 0.001), TG (OR = 1.068, 95%CI 1.018–1.122, P = 0.007), Albumin (OR = 0.906, 95%CI 0.891–0.922, P = 0.001), BMI (OR = 1.024, 95%CI 1.004–1.043, P = 0.016), Uric Acid (OR = 1.002, 95%CI 1.001–1.003, P = 0.001), Creatinine (OR = 1.007, 95%CI 1.006–1.009, P = 0.001) were independent risk factors for kidney disease in T2DM patients. Variance inflation factor (VIF) analysis was performed to assess multicollinearity, with results presented in Tables S1 and S2, confirming that no concerning collinearity existed among the predictors (all VIF < 1.3). Table 2Multivariate logistic regression analysis of the risk of retinopathy in patients with T2DM.ParameterBStandard ErrorWaldP valueOR95%CIAge− 0.0250.00366.4570.0010.9760.970–0.982DM duration0.0690.005181.4080.0011.0711.060–1.082Insulin use0.3260.07120.8990.0011.3861.205–1.594Peripheral neuropathy0.3500.08218.0790.0011.4201.208–1.669Heart rate0.0090.00313.9170.0011.0101.004–1.015HbA1c0.0600.01517.0850.0011.0621.032–1.093Total protein− 0.0210.00614.1120.0010.9790.968–0.990 Table 2Multivariate logistic regression analysis of the risk of retinopathy in patients with T2DM.ParameterBStandard ErrorWaldP valueOR95%CIAge− 0.0250.00366.4570.0010.9760.970–0.982DM duration0.0690.005181.4080.0011.0711.060–1.082Insulin use0.3260.07120.8990.0011.3861.205–1.594Peripheral neuropathy0.3500.08218.0790.0011.4201.208–1.669Heart rate0.0090.00313.9170.0011.0101.004–1.015HbA1c0.0600.01517.0850.0011.0621.032–1.093Total protein− 0.0210.00614.1120.0010.9790.968–0.990 Multivariate logistic regression analysis of the risk of retinopathy in patients with T2DM. Table 3Multivariate logistic regression analysis of the risk of nephropathy in patients with T2DM.ParameterBStandard ErrorWaldP valueOR95%CIAge0.0060.0042.8850.0891.0060.999–1.014Carotid atherosclerosis0.2390.0789.3610.0021.271.090–1.481Diabetic foot0.4920.1510.8070.0011.6351.220–2.193DM duration0.0260.00622.6050.0011.0271.016–1.038Glycosidase inhibitors0.2350.0769.660.0021.2651.091–1.467Insulin use0.4530.07833.5120.0011.5731.349–1.834History of hypertension0.4470.08626.8240.0011.5641.320–1.852Systolic blood pressure0.0130.00247.7810.0011.0131.009–1.017Lymphocyte count− 0.2330.06214.1220.0010.7920.701–0.894Total cholesterol0.1140.03212.9160.0011.1211.053–1.193Triglycerides0.0660.0257.1520.0071.0681.018–1.122Albumin− 0.0980.009123.0490.0010.9060.891–0.922BMI0.0230.015.7540.0161.0241.004–1.043Uric acid0.002020.650.0011.0021.001–1.003Creatinine0.0070.00162.070.0011.0071.006–1.009 Table 3Multivariate logistic regression analysis of the risk of nephropathy in patients with T2DM.ParameterBStandard ErrorWaldP valueOR95%CIAge0.0060.0042.8850.0891.0060.999–1.014Carotid atherosclerosis0.2390.0789.3610.0021.271.090–1.481Diabetic foot0.4920.1510.8070.0011.6351.220–2.193DM duration0.0260.00622.6050.0011.0271.016–1.038Glycosidase inhibitors0.2350.0769.660.0021.2651.091–1.467Insulin use0.4530.07833.5120.0011.5731.349–1.834History of hypertension0.4470.08626.8240.0011.5641.320–1.852Systolic blood pressure0.0130.00247.7810.0011.0131.009–1.017Lymphocyte count− 0.2330.06214.1220.0010.7920.701–0.894Total cholesterol0.1140.03212.9160.0011.1211.053–1.193Triglycerides0.0660.0257.1520.0071.0681.018–1.122Albumin− 0.0980.009123.0490.0010.9060.891–0.922BMI0.0230.015.7540.0161.0241.004–1.043Uric acid0.002020.650.0011.0021.001–1.003Creatinine0.0070.00162.070.0011.0071.006–1.009 Multivariate logistic regression analysis of the risk of nephropathy in patients with T2DM. Construction of nomogram risk estimation models for the risk of retinopathy and kidney disease in T2DM patients Through multivariate logistic regression, Age, Kidney Disease, Diabetes Duration, Insulin, Peripheral Neuropathy, Heart Rate, HbA1c, and Total Protein were identified as factors associated with the risk of retinopathy in T2DM patients (as shown in Fig. 2A). Simultaneously, through multivariate logistic regression, Carotid Atherosclerosis, Diabetic Foot, Diabetes Duration, Glycosidase Inhibitors, Insulin, History of Hypertension, SBP, Lymphocyte Count, TC, TG, Serum Albumin, BMI, Serum Creatinine, and Serum Uric Acid were identified as factors associated with the risk of kidney disease in T2DM patients (as shown in Fig. 2B). Each variable has a scale on its corresponding line representing its value range. Each value corresponds to a single score. The total score obtained by summing the individual scores corresponds to the probability of developing the complication. Fig. 2(A) Nomogram risk estimation model for retinopathy in patients with T2DM; (B) nomogram risk estimation model for nephropathy in patients with T2DM. (A) Nomogram risk estimation model for retinopathy in patients with T2DM; (B) nomogram risk estimation model for nephropathy in patients with T2DM. Evaluation of the discrimination of the nomogram risk estimation models for retinopathy and kidney disease risk in T2DM patients As shown in Fig. 3, Receiver Operating Characteristic (ROC) curve analysis for the diabetic retinopathy risk assessment model yielded an Area Under the Curve (AUC) = 0.703 (95% CI 0.687–0.719) for the training set and an AUC = 0.732 (95% CI 0.683–0.781) for the external validation set (Fig. 3A). Similarly, ROC analysis for the diabetic kidney disease prediction model demonstrated an AUC = 0.802 (95% CI 0.788–0.816) for the training set and an AUC = 0.713 (95% CI 0.665–0.760) for the external validation set (Fig. 3B). The model fit statistics, including AIC and BIC values for both prediction models, are presented in Table 4, with the diabetic retinopathy model showing AIC = 5803.472 and BIC = 5857.125, and the diabetic kidney disease model showing AIC = 4995.475 and BIC = 5096.075. These results indicate that both risk assessment models exhibit good discrimination ability and satisfactory model fit based on comprehensive evaluation metrics. Fig. 3Receiver operating characteristic (ROC) curves for the risk assessment models of retinopathy (A) and nephropathy (B) in patients with T2DM. Receiver operating characteristic (ROC) curves for the risk assessment models of retinopathy (A) and nephropathy (B) in patients with T2DM. Table 4Model fit statistics for diabetic retinopathy and diabetic kidney disease risk assessment models.ModelDatasetAUCAICBIC1DRTraining0.703 (0.687–0.719)5803.4725857.1252DRValidation0.732 (0.683–0.781)––3DKDTraining0.802 (0.788–0.816)4995.4755096.0754DKDValidation0.713(0.665–0.760)–– Table 4Model fit statistics for diabetic retinopathy and diabetic kidney disease risk assessment models.ModelDatasetAUCAICBIC1DRTraining0.703 (0.687–0.719)5803.4725857.1252DRValidation0.732 (0.683–0.781)––3DKDTraining0.802 (0.788–0.816)4995.4755096.0754DKDValidation0.713(0.665–0.760)–– Model fit statistics for diabetic retinopathy and diabetic kidney disease risk assessment models. Evaluation of the accuracy and clinical net benefit analysis of the nomogram risk estimation models for retinopathy and kidney disease risk in T2DM patients As shown in Fig. 4, calibration curves were established to evaluate the models’ accuracy. For the DR model, the calibration slope was 0.96, the intercept was − 0.03, and the Brier score was 0.148. For the DKD model, the calibration slope was 0.94, the intercept was 0.01, and the Brier score was 0.132. These quantitative metrics confirm good calibration and overall model performance. The DCA curves demonstrate that the nomogram models provide a positive net benefit compared to the ‘treat-all’ or ‘treat-none’ strategies across a wide range of clinically relevant threshold probabilities. For the DR model (Fig. 5A), net benefit is observed for threshold probabilities between 0% and approximately 72%. For the DKD model (Fig. 5B), net benefit is observed for threshold probabilities between 0% and approximately 83%. This indicates that within these ranges, using the model to guide clinical decision-making would result in a better clinical outcome than referring all patients or referring none. Fig. 4Calibration curves for the risk assessment models of retinopathy (A) and nephropathy (B) in patients with T2DM. Calibration curves for the risk assessment models of retinopathy (A) and nephropathy (B) in patients with T2DM. Fig. 5Decision curve analysis (DCA) for risk estimation models of retinopathy (A, C) and Nephropathy (B, D) in Patients with T2DM. The y-axis represents net benefit, calculated as (true positives / N) - (false positives / N) × (pt/(1-pt)), where pt is the threshold probability. The x-axis shows threshold probabilities—the minimum probability of complications at which a clinician or patient would opt for intervention (e.g., referral for ophthalmologic examination or nephrology consultation). The horizontal lines represent two extreme strategies: the “treat-all” line (assuming all patients have the complication and receive intervention) and the “treat-none” line (assuming no patients have the complication and receive no intervention, resulting in a net benefit of zero). The red line (“Nomogram model”) shows the net benefit of using our model to guide intervention decisions across different threshold probabilities. Decision curve analysis (DCA) for risk estimation models of retinopathy (A, C) and Nephropathy (B, D) in Patients with T2DM. The y-axis represents net benefit, calculated as (true positives / N) - (false positives / N) × (pt/(1-pt)), where pt is the threshold probability. The x-axis shows threshold probabilities—the minimum probability of complications at which a clinician or patient would opt for intervention (e.g., referral for ophthalmologic examination or nephrology consultation). The horizontal lines represent two extreme strategies: the “treat-all” line (assuming all patients have the complication and receive intervention) and the “treat-none” line (assuming no patients have the complication and receive no intervention, resulting in a net benefit of zero). The red line (“Nomogram model”) shows the net benefit of using our model to guide intervention decisions across different threshold probabilities. Discussion The Nomogram risk estimation model is a graphical mathematical model in statistics, a visual representation based on regression analysis, widely used for the diagnosis and risk assessment of various diseases16. Currently, domestic and international research on risk estimation models for complications in T2DM patients has been conducted17. The application of these risk estimation models has, to some extent, facilitated the early identification of high-risk populations for T2DM complications among healthcare workers, possessing certain clinical value. However, the predictive indicators of some models are not easily obtainable, and their applicability needs further evaluation. In comparison to other risk assessment models reported in the literature, our nomogram models offer distinct advantages. For diabetic retinopathy, previous models such as those by Xu et al.18. primarily utilized genetic markers or advanced imaging parameters, which may not be routinely available in primary care settings. In contrast, our model relies on readily available clinical and laboratory variables. Similarly, for diabetic kidney disease, while models like the one from the CRIC study19 incorporate specialized biomarkers like TNF receptors, our model achieves competitive discriminative performance (AUC = 0.802 in training, 0.713 in validation) using standard clinical data, enhancing its practical applicability for early risk stratification in diverse clinical environments. Meanwhile, compared to established risk assessment tools such as the UKPDS risk engine, our nomogram models offer several distinct advantages. First, the UKPDS models were developed primarily from European cohorts diagnosed between 1977 and 199720, potentially limiting their applicability to contemporary Chinese populations with different ethnic backgrounds, lifestyle factors, and treatment paradigms. Second, our models incorporate a broader range of routinely available clinical variables—including medication use, heart rate, lymphocyte count, and uric acid—that reflect current clinical practice, whereas UKPDS models rely on more limited predictors. Third, while UKPDS provides long-term risk prediction (e.g., 10-year risk), our cross-sectional models are specifically designed for point-of-care risk stratification to identify patients who may already have undiagnosed complications, complementing rather than competing with prognostic tools. Fourth, to our knowledge, this is one of the few studies in a Chinese population simultaneously developing and internally validating models for both DR and DKD using LASSO regression to address multicollinearity, followed by a comprehensive evaluation including decision curve analysis. The primary clinical utility of these nomogram models lies in their application as targeted screening tools in routine diabetes care. In outpatient or primary care settings where resources for universal screening may be limited, these models can assist clinicians in identifying individuals at elevated risk for undiagnosed diabetic retinopathy or kidney disease at the time of evaluation. A patient identified as high-risk—for example, with an estimated probability exceeding a clinically determined threshold—could be prioritized for confirmatory diagnostic evaluation, such as dilated fundus examination for DR or urine albumin-to-creatinine ratio testing for DKD. This targeted approach may improve early detection rates and facilitate timely intervention, particularly in resource-constrained healthcare environments. Importantly, these models are not intended to replace clinical judgment or guideline-recommended screening protocols, but rather to complement them by providing a quantitative, individualized risk estimate based on routinely collected clinical variables. Diabetic microvascular complications are common in T2DM, primarily consisting of diabetic kidney disease and diabetic retinopathy. Diabetic kidney disease is a major microvascular complication, affecting about 40% of diabetic patients21, and is a leading cause of chronic kidney disease and end-stage renal failure globally, imposing a severe economic burden. Simultaneously, diabetic retinopathy has become a leading cause of blindness in T2DM patients. The increasing number of T2DM patients exacerbates the prevalence of diabetic retinopathy. Early identification of diabetic retinopathy is crucial for delaying T2DM progression, reducing patient disease burden, and improving outcomes22. Therefore, this study established Nomogram risk estimation models for the risk of retinopathy and kidney disease in T2DM patients. The final retinopathy risk estimation model included 7 variables: Age, Diabetes Duration, Insulin, Peripheral Neuropathy, Heart Rate, HbA1c, and Total Protein. The kidney disease risk estimation model included 14 variables: Carotid Atherosclerosis, Diabetic Foot, Diabetes Duration, Glycosidase Inhibitors, Insulin, History of Hypertension, SBP, Lymphocyte Count, TC, TG, Serum Albumin, BMI, Serum Creatinine, and Serum Uric Acid. The model’s predictors are easily obtainable clinically, facilitating early identification of high-risk populations for T2DM complications by medical staff, holding certain clinical significance. The models also demonstrated good discrimination and accuracy. This study differs from previous statistical methods and studies that used only logistic regression analysis. We attempted to use new statistical methods and models based on previous research to study risk factors for diabetic retinopathy and diabetic kidney disease, explaining the issues from different perspectives. Previous studies used univariate analysis to screen for multivariate analysis or performed stepwise regression and cyclic logistic regression analysis to obtain results23,24. However, in this process, various confounding factors must be considered alongside multicollinearity issues. In this study, LASSO regression analysis provided a better solution and more accurate results for this problem. LASSO regression analysis was used to screen variables, followed by traditional logistic regression analysis. ROC, calibration, and DCA curves were constructed to verify the accuracy and stability of the two models25. Although our final predictive models do not include cross-complication variables to avoid circularity, the strong association between kidney disease and retinopathy observed in our univariate analysis (Table 1) is consistent with previous research. The coincidence of retinal and renal lesions in diabetic patients is widely recognized, with some literature proposing the definition of “reno-retinal syndrome”26, where microvascular endothelial dysfunction is an essential common pathological mechanism in the early development stages of both diseases. Hyperglycemia is associated with retinal ischemia and hypoxia, oxidative stress, accumulation of inflammatory cells/factors, and activation of various metabolic pathways, leading to the destruction of the endothelial barrier structure and increased microvascular permeability. An important observation from our validation analysis was the decline in discriminative performance for the DKD model from the training set (AUC = 0.802) to the temporal validation set (AUC = 0.713). This decline suggests some degree of model overfitting or temporal instability, which may be attributable to several factors. First, changes in clinical practice and diagnostic criteria over the extended study period (2008–2024) may have introduced heterogeneity between the training and validation cohorts. Second, the temporal validation set was smaller (n = 1,000) and may not fully represent the spectrum of disease severity captured in the larger training set. Third, the exclusion of patients with severe kidney dysfunction may have truncated the outcome spectrum, potentially amplifying the impact of case-mix differences between cohorts. This performance decline underscores the importance of further validation in larger, multi-center cohorts before widespread clinical implementation, and highlights that the current models are best suited for risk stratification in populations similar to our development cohort. In our models, younger age emerged as a factor associated with an increased risk of diabetic retinopathy. We interpret this finding not as a direct effect of chronological youth but as a clinical marker likely identifying a subgroup with early-onset T2DM, leading to longer lifetime exposure to hyperglycemia—a primary driver of microvascular damage—and potentially reflecting a more aggressive contemporary metabolic phenotype. While this may initially appear counterintuitive, it underscores a critical epidemiological shift and reflects complex underlying pathways rather than a direct biological effect of chronological youth. We interpret this finding through several plausible, interlinked explanations. Firstly, a younger age at presentation often signifies an earlier onset of T2DM. These patients accumulate a longer duration of exposure to hyperglycemia over their lifetime, which is a primary driver of microvascular damage. Secondly, more pronounced metabolic disorders are often observed in patients with early-onset diabetes, which may be a key reason for the earlier appearance of complications27,28. Thirdly, younger age serves as a proxy for contemporary lifestyle and environmental risk factors. Individuals developing T2DM at a younger age in the current era may be exposed to distinct dietary patterns, reduced physical activity, and other obesogenic factors that contribute to a more aggressive metabolic phenotype29,30. Therefore, in the context of our predictive model, “younger age” likely acts as a composite clinical marker identifying a subgroup of patients with early-onset disease, prolonged metabolic insult, and a high-risk profile, warranting vigilant monitoring and early intervention. Hyperglycemia is a well-known factor associated with the development of diabetic retinopathy. Our study also showed that higher HbA1c is associated with an increased risk of diabetic retinopathy. One study reported that by categorizing HbA1c into deciles and calculating the prevalence of diabetic retinopathy for each decile, the prevalence of DR increased sharply when HbA1c exceeded 6.4%. The threshold of HbA1c for detecting DR was almost consistent with the WHO diagnostic criteria for diabetes31, meaning that once T2DM is diagnosed, diabetic retinopathy already exists or the possibility is high. Insulin is crucial for biology. It is involved not only in metabolism but also in the regulation of growth factors like IGF-1. Acute replenishment of exogenously administered insulin and IGF-1 enhances the expression of vascular endothelial growth factor (VEGF) induced by ischemia. A critical ratio of tissue sensitivity to VEGF: the availability of VEGF can trigger vascular proliferation (i.e., microaneurysms and/or abnormal blood vessels)32. Furthermore, it is important to consider that the association between insulin use and an increased risk of diabetic retinopathy in our model likely reflects indication bias. Patients with more advanced diabetes, longer disease duration, or poorer glycemic control on oral agents—all factors that heighten the risk for microvascular complications—are more frequently prescribed insulin. Thus, insulin therapy may serve as a clinical surrogate marker for a high-risk diabetes phenotype rather than a direct promoter of retinopathy pathogenesis32. This underscores that our model identifies associated risk indicators for concurrent complication assessment, with insulin use contributing to risk stratification by encapsulating greater disease severity. Nerve damage is also a complication of T2DM. Research by Weimin Wang et al. showed that diabetic peripheral neuropathy is related to concurrently existing diabetic retinopathy, diabetic kidney disease, and liver damage, and is positively correlated33. Additionally, another study, after multivariate logistic analysis, concluded that diabetic retinopathy is associated with a high heart rate. An elevated heart rate may be a predictor of advanced retinopathy34, consistent with the results of this study. The results also found that total protein is a protective factor against the occurrence of diabetic retinopathy. In clinical practice, targeted early intervention for diabetic patients can reduce the occurrence of diabetic retinopathy. On the other hand, regarding the research results for diabetic kidney disease. We found that patients with a longer duration of diabetes have a higher risk of developing kidney disease35. Advanced age also increases the risk of kidney disease in diabetic patients35,36. A history of hypertension and elevated SBP is another important independent risk factor for kidney disease35,37. The coincidence of retinal and renal lesions in diabetic patients is widely recognized, with some literature proposing the definition of “reno-retinal syndrome”26; they are mutual risk factors, consistent with the results of this study. Obesity is also associated with an increased risk of diabetic kidney disease, and BMI is a risk factor for concurrent kidney disease35,38,39. Dyslipidemia also seems to play a role in the pathogenesis of diabetic kidney disease. In previous studies, lower Triglyceride (TG) levels were associated with a reduced risk of the degree of kidney lesions40. In T2DM patients, elevated Total Cholesterol (TC) levels were also associated with an increased risk of developing moderate and severely elevated kidney lesions41. The occurrence of diabetic foot is positively correlated with concurrent kidney disease in diabetes, consistent with the research of Gauthier Borderie42. Several observational studies have indicated that serum creatinine is a risk factor for the development and progression of diabetic kidney disease43. Higher serum albumin levels are significantly associated with a lower risk of diabetic kidney disease44, all consistent with the results of this study. In summary, this study retrospectively analyzed relevant medical record data of T2DM patients, constructed and validated risk estimation models for microvascular complications. The included indicators are simple and easily obtainable in clinical work, helping clinicians early assess the risk of microvascular complications in T2DM patients. Currently, there are no specific interventions for microvascular complications clinically. Efforts often focus on active regular exercise, controlling diet, BMI, blood pressure, total cholesterol, total protein, HbA1c, LDL-C, lymphocyte count, triglycerides, uric acid, etc., to reduce the risk of microvascular complications in T2DM patients. Conclusion This study developed nomogram models incorporating readily available clinical variables to assess the risk of diabetic retinopathy and nephropathy in T2DM patients based on cross-sectional data. Utilizing LASSO regression for feature selection enhanced model accuracy and addressed multicollinearity. The models demonstrated moderate discriminatory power (validation AUCs: 0.732 for DR, 0.713 for DKD) and satisfactory calibration, facilitating early identification of individuals at high risk. Findings reaffirmed known risk factors (e.g., hyperglycemia, hypertension, renal-retinal syndrome) and highlighted modifiable predictors. These tools support personalized risk stratification and targeted screening, potentially reducing the burden of microvascular complications through early detection and management. Further multi-center validation is recommended to reinforce generalizability. Limitation Furthermore, this risk estimation model has certain limitations that warrant consideration. Firstly, as a cross-sectional study, the models should be interpreted as tools for assessing concurrent risk associations and stratifying patients based on their current clinical profile, rather than for predicting future complication development. Their primary clinical utility lies in helping clinicians identify individuals who may already have undiagnosed complications or are at high risk at the time of evaluation, thereby facilitating timely screening and management. Secondly, the model was developed using retrospective, single-center data. Its generalizability to other populations or healthcare settings requires further verification through external validation and multi-center studies. Future research should aim to include larger, more diverse cohorts and incorporate additional disease-related factors to enhance the model’s accuracy and universal applicability. Thirdly, the exclusion of patients with severe liver or kidney dysfunction at enrollment may have introduced selection bias, particularly affecting the representativeness of the DKD group. While this exclusion criterion was implemented to minimize confounding from non-diabetic causes of kidney disease, it likely resulted in the underrepresentation of patients with more advanced DKD in our cohort. Consequently, our findings may not fully capture the risk factor profile associated with severe diabetic kidney disease, and the models’ performance in patients with advanced renal impairment remains unknown. This selection bias may also partially explain the observed decline in DKD model performance upon temporal validation, as the training cohort may not fully represent the full spectrum of DKD severity encountered in broader clinical practice. Future studies should consider including patients across all stages of kidney dysfunction with careful etiological classification to better characterize risk factors across the entire disease spectrum. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1