Abstract
Objective
This study aims to identify factors associated with diabetic macular edema (DME) presence in working-age (18–60 years) patients with type 2 diabetes mellitus (T2DM) by developing a model that integrates HbA1c, renal function (eGFR), and hemodynamic parameters (SBP). The model addresses critical gaps in current screening strategies by using routinely available biomarkers, thereby enabling non-ophthalmologists to efficiently identify high-risk individuals.
Methods
This cross-sectional study prospectively collected data from 490 patients with type 2 diabetes mellitus (T2DM), aged 18–60 years, who were consecutively enrolled at a single medical center between January 2020 and March 2025. The participants were randomly allocated into two groups: a training cohort (n=343) and a validation cohort (n=147). Predictors were selected via LASSO regression with 10-fold cross-validation from an initial set of 19 variables, encompassing renal function, metabolic parameters, and hemodynamic indices. Subsequently, a multivariate logistic regression model was developed and illustrated through a nomogram. The model’s predictive accuracy was evaluated through receiver operating characteristic (ROC) curves (AUC), calibration curves, and decision curve analysis (DCA).
Results
The overall prevalence of DME in the study cohort was 15.71% (77 of 490). Four predictors independently associated with DME were identified using LASSO regression, namely diabetes duration (OR = 1.460, 95% CI: 1.212–1.457), SBP (OR = 1.066, 95% CI: 1.037–1.095), eGFR (OR = 0.938, 95% CI: 0.916–0.961), and HbA1c (OR = 1.484, 95%CI: 1.189–1.852). The resulting nomogram exhibited robust discriminatory ability (training AUC = 0.905, 95% CI: 0.858–0.951; validation AUC = 0.884, 95% CI: 0.820–0.949) and strong calibration performance (Hosmer-Lemeshow test, P = 0.878). DCA further confirmed substantial clinical applicability within a threshold probability range from 2% to 100%, achieving a maximum net benefit of 0.14, thereby potentially preventing unnecessary intervention in 14 out of every 100 patients.
Conclusion
This nomogram effectively integrates HbA1c, renal function, and hemodynamic parameters to identify key factors associated with diabetic macular edema (DME) risk in working-age patients with type 2 diabetes mellitus (T2DM), demonstrating high accuracy. By utilizing routine clinical measures, it facilitates implementation in primary care settings, offering the potential to reduce vision loss through timely referrals. Future multicenter studies are warranted to verify its generalizability and explore its integration with emerging biomarkers.
Keywords: eGFR, HbA1c, nomogram, systolic blood pressure, type 2 diabetes mellitus
1. Introduction
Diabetic macular edema (DME) is the leading cause of vision impairment in individuals with type 2 diabetes mellitus (T2DM), with prevalence rates ranging from 1.4% to 12.8% (1–3). The global rise in T2DM incidence is expected to increase the number of DME patients to over 30 million by 2045, imposing a substantial economic burden on healthcare systems (1–3). Early diagnosis is crucial for risk stratification, timely intervention, and cost reduction. The working-age population (18–60 years) is particularly vulnerable to DME-related visual impairment, which can lead to central vision loss, distorted vision, and impaired color perception—severely affecting occupational function and quality of life (4, 5). For individuals in visually demanding professions (e.g., drivers, programmers), DME may result in job loss, compounding socioeconomic challenges. Although anti-VEGF therapy remains first-line treatment, poor response rates (30–40%) and high costs limit its effectiveness (6). Frequent injections and follow-ups further reduce compliance among employed patients. Low screening rates in this age group often delay diagnosis, increasing the risk of irreversible vision loss. Thus, a reliable risk prediction tool is urgently needed for early identification of high-risk individuals.
Current screening relies heavily on fundus examination and optical coherence tomography (OCT). However, OCT is seldom available in primary care settings (7). Additionally, a management gap exists between internists overseeing diabetes care and ophthalmologists managing DME, complicating coordinated care. These challenges underscore the need for an accessible, efficient screening tool suitable for non-specialist use. Predictive models offer a promising solution by enhancing screening accessibility, enabling continuous monitoring, and optimizing resource allocation. Yet, no consensus exists on DME risk factors in working-age T2DM patients. The pathophysiology of DME involves chronic hyperglycemia, blood-retinal barrier disruption, inflammation, and oxidative stress (8). While American Diabetes Association (ADA) guidelines emphasize HbA1c and diabetes duration as key predictors, they overlook other metabolic and systemic factors (9). Other studies suggest roles for hypertension, dyslipidemia, nephropathy, and anemia (10). Previous prediction models have been limited by small samples, incomplete variable selection, and poor interpretability.
To address these limitations, we employed least absolute shrinkage and selection operator (LASSO) regression—a robust method for feature selection in chronic disease prediction (11). In this single-center cross-sectional study, we integrated metabolic, renal, and hemodynamic parameters to develop a clinically actionable nomogram for DME risk assessment in working-age T2DM patients. Our objectives were: (1) to establish a non-invasive risk tool for this population; (2) to evaluate the combined predictive value of routine biomarkers; and (3) to provide a visual, interpretable model for individualized intervention. This study aims to improve early detection, reduce unnecessary referrals, and ultimately mitigate vision loss in a high-risk, economically active demographic.
2. Methods
2.1. Research subjects
This cross-sectional study adopted a prospective data collection approach. We consecutively enrolled patients with type 2 diabetes mellitus (T2DM) aged 18–60 years who visited the hospital between January 2020 and March 2025.The cross-sectional design allows for examination of associations but does not permit inference of temporal relationships. Inclusion criteria were as follows (12): (1) meeting the 2018 diagnostic criteria of the American Diabetes Association; (2) aged 18–60 years; (3) completed standardized ophthalmic examinations; (4) complete clinical data. Exclusion criteria included: (1) other retinal diseases; (2) history of vitrectomy or anti-VEGF therapy; (3) malignant tumors or hematological diseases.
The sample size was calculated using a logistic regression model. This model required that the number of outcome events exceed ten times the number of independent variables. According to reporting guidelines for predictive models and risk-of-bias assessment tools, at least 10 independent variables were necessary (13). In this study, occurrence of DME served as the dependent variable. Based on the literature, 15 independent variables were anticipated in the predictive model, and the prevalence of DME among T2DM patients was approximately 40%. Therefore, the minimum sample size required was 375 (15×10÷40%). To account for potential invalid or incomplete samples, an additional 20% of participants were included, yielding a total enrollment of 490 patients. These participants were randomly allocated at a 7:3 ratio into the training (n=343) and validation cohorts (n=147). The methodological flowchart of the study is detailed in Figure 1. This study was approved by the ethics department of our hospital (Ethics Number: 2020ER035-1), and all subjects gave informed consent to the trial content.
Figure 1.
Research flowchart.
2.2. Data collection
2.2.1. Basic information
Data collection procedures were standardized across the study. Participant age information was verified using official identification documents. BMI values were determined using calibrated electronic scales for height (accurate to 0.1 cm) and weight (accurate to 0.1 kg), with BMI calculated as weight (kg)/[height (m)]2 (14).
2.2.2. Selection of instrumental variables
Fundus examinations were conducted after mydriasis using a Topcon TRC-50DX fundus camera. Seven-field color fundus photographs were captured, including images of the macular center, four quadrants, optic disc, and wide-angle posterior pole. Two ophthalmologists (associate chief physician or higher) independently evaluated images using a double-blind method. Diabetic retinopathy was graded according to the ETDRS standard.
OCT examination was performed using Heidelberg Spectralis OCT and the ETDRS 9-sector grid scanning protocol. The mean thickness of the central 1-mm diameter area was automatically measured as the central subfield thickness (CST). Diagnostic thresholds were: normal (<250 μm), mild edema (250–300 μm), moderate edema (300–400 μm), and severe edema (≥400 μm) (15). Subretinal fluid was defined as a separation between the neuroepithelial layer and the RPE of ≥50 μm, and typical OCT features were recorded if present.
Diagnostic criteria for DME were: OCT-confirmed macular thickening (CST ≥250 μm) accompanied by at least one of the following: (1) low-reflective cystic cavities between retinal layers; (2) dark sub-neuroepithelial fluid spaces; or (3) hard exudate deposits (16).
2.2.3. Clinical indicators
2.2.3.1. Lifestyle and history data
Smoking history, alcohol consumption, and family history of diabetes were collected through standardized questionnaires. Diabetes duration was precisely calculated from the month of first meeting WHO (1999) diabetes diagnostic criteria.
2.2.3.2. Blood pressure measurement
Blood pressure was measured using a validated electronic upper-arm sphygmomanometer after subjects sat quietly for 5 minutes. The average of three consecutive measurements was recorded.
2.2.3.3. Laboratory analyses
HbA1c: For HbA1c analysis, approximately 2–3 mL of venous blood was drawn from the antecubital vein into EDTA-containing tubes. Participants were instructed to avoid strenuous physical activities and dietary extremes prior to blood sampling. Samples were then centrifuged to obtain plasma or serum, and HbA1c levels were subsequently quantified by high-performance liquid chromatography (HPLC), which differentiates glycated from non-glycated hemoglobin based on differences in ionic charge (17).
Fasting plasma glucose (FPG): Subjects fasted for 8–10 hours overnight. Venous blood (2–3 mL) was drawn between 7 and 9 a.m. from the median cubital vein after iodophor disinfection, using vacuum tubes. Samples were centrifuged within 2 hours to separate serum/plasma. Concentrations were determined using the glucose oxidase (GOD-POD) method, which involves enzymatic generation of peroxide quantified colorimetrically (18).
ALT and AST: Subjects fasted 8–12 hours, avoided alcohol, vigorous exercise, and high-fat foods. Venous blood (2–3 mL) was collected without anticoagulants or coagulants, gently inverted 5–8 times, and centrifuged within 2 hours (3,000 rpm, 10 min) to separate serum. Concentrations were measured by the rate method (19).
Estimated glomerular filtration rate (eGFR): Subjects avoided vigorous exercise and high-protein diets for 3 days before sampling. Venous blood (3–5 mL) was collected without anticoagulants, gently inverted 5–8 times, and centrifuged (3,000 rpm, 10 min) within 2 hours to separate serum. Serum creatinine was determined enzymatically (creatinine oxidase method). eGFR was calculated using the Modification of Diet in Renal Disease (MDRD) equation (20).
TC, LDL, and TG: Subjects fasted for 8–12 hours and maintained normal diets for 3 days, avoiding high-fat foods, alcohol, and vigorous exercise. Participants sat quietly for 5 minutes before sampling. Venous blood (3–5 mL) was collected without anticoagulants and centrifuged (3,000 rpm, 10 min) within 30–45 min to separate serum. Concentrations were measured by the cholesterol esterase method (CEH-CHOD-PAP) (21).
TC, LDL, and TG:: Subjects fasted for 12 hours, avoided alcohol and vigorous exercise, and sat quietly for 5 minutes before sampling. Venous blood (2–3 mL) was collected without anticoagulants or using vacuum tubes with separation gel. Samples were centrifuged (4,000 rpm, 10 min) within 30 min to separate serum. Concentrations were measured using the bromocresol green (BCG) method (22).
2.2.4. Quality control system
A three-level quality control system was established for all test indicators. Laboratory indicators were evaluated daily using two levels of quality control materials and regularly participated in CAP external quality assessments (required score ≥90%). Questionnaires underwent double-entry with independent verification, including automatic logical checks. Clinical parameters were extracted automatically from electronic medical records. Key indicators were cross-verified against initial diagnosis records. Additionally, a three-level (red-orange-yellow) early warning system was implemented. A yellow warning (three abnormal indicators) required increased follow-up; an orange warning (five abnormal indicators) indicated the need for specialist consultation; a red warning (≥seven abnormal indicators) triggered immediate multidisciplinary intervention.
2.3. Statistical analysis
Statistical analysis was conducted using R software (version 4.2.0; R Foundation for Statistical Computing, Vienna, Austria). Predictors were identified by applying LASSO regression combined with 10-fold cross-validation, with the optimal λ parameter selected according to the minimum mean square error (MSE). This statistical method effectively reduces coefficients of insignificant variables to zero, thereby improving the accuracy and efficiency of prediction. Variables deemed significant by LASSO regression subsequently entered a multivariate logistic regression analysis, from which odds ratios (OR) and corresponding 95% confidence intervals (95% CI) were determined. A nomogram was then developed utilizing the regression coefficients derived from this model.
The validity and clinical applicability of the model were assessed from three distinct perspectives: (1) Discriminative capacity, evaluated through receiver operating characteristic (ROC) curve analysis, with the area under the ROC curve (AUC) and its associated 95% CI computed; (2) Calibration accuracy, assessed by plotting calibration curves and employing the Hosmer–Lemeshow test to quantify consistency between observed and predicted outcomes; (3) Clinical utility, estimated by decision curve analysis (DCA), which quantified the model’s net benefit across varying threshold probabilities to determine the optimal threshold for intervention.
Continuous variables were summarized as mean ± standard deviation (SD) if normally distributed, or as median and interquartile range [M (P25, P75)] if non-normally distributed. Categorical variables were presented as frequencies with corresponding percentages. Statistical significance was defined as a two-tailed P-value <0.05.
3. Results
3.1. Comparison of clinical data between training and validation sets
In total, 490 T2DM patients participated in the study, with 77 (15.71%) diagnosed with DME. Participants were randomly allocated to either a training cohort (n=343; DME prevalence: 15.74%, 54/343) or a validation cohort (n=147; DME prevalence: 15.65%, 23/147). Comparative analysis revealed no statistically significant differences between the two cohorts in terms of baseline clinical parameters (P>0.05, Table 1).
Table 1.
Baseline characteristics of patients in the training and internal validation sets.
| Total (n=490) | Training set (n=343) | Internal validation set (n=147) | X2/F/H | P | ||||
|---|---|---|---|---|---|---|---|---|
| Non-DME (n=417) | DME (n=73) | Non-DME (n=289) | DME (n=54) | Non-DME (n=124) | DME (n=23) | |||
| Gender (n,%) | 1.801 | 0.180 | ||||||
| Male | 220 (53.27) | 36 (46.75) | 154 (53.29) | 32 (59.26) | 66 (53.23) | 4 (17.39) | ||
| Female | 193 (46.73) | 41 (53.25) | 135 (46.71) | 22 (40.74) | 58 (46.77) | 19 (82.61) | ||
| Age (years) | 51.06 ± 11.52 | 55.18 ± 10.6 | 50.77 ± 11.76 | 54.85 ± 11.05 | 51.73 ± 10.97 | 55.96 ± 9.66 | -0.861 | 0.390 |
| BMI (x ± s,kg/m2 ) | 23.81 ± 2.09 | 23.90 ± 2.27 | 23.70 ± 2.08 | 23.91 ± 2.28 | 24.07 ± 2.11 | 23.87 ± 2.31 | -1.498 | 0.135 |
| A long history of smoking (n,%) | 0.090 | 0.764 | ||||||
| No | 287 (69.49) | 55 (71.43) | 200 (69.2) | 38 (70.37) | 87 (70.16) | 17 (73.91) | ||
| Yes | 126 (30.51) | 22 (28.57) | 89 (30.8) | 16 (29.63) | 37 (29.84) | 6 (26.09) | ||
| History of alcohol abuse (n,%) | 0.469 | 0.494 | ||||||
| No | 350 (84.75) | 65 (84.42) | 245 (84.78) | 48 (88.89) | 105 (84.68) | 17 (73.91) | ||
| Yes | 63 (15.25) | 12 (15.58) | 44 (15.22) | 6 (11.11) | 19 (15.32) | 6 (26.09) | ||
| Family history of DM (n,%) | 0.159 | 0.69 | ||||||
| No | 351 (84.99) | 71 (92.21) | 245 (84.78) | 49 (90.74) | 106 (85.48) | 22 (95.65) | ||
| Yes | 62 (15.01) | 6 (7.79) | 44 (15.22) | 5 (9.26) | 18 (14.52) | 1 (4.35) | ||
| DM disease course (years) | 8.3 (7.1,9.6) | 10.2 (8.4,12.3) | 8.2 (7,9.6) | 10.5 (8.4,12.3) | 8.7 (7.5,9.85) | 9.5 (7.1,12.9) | -1.088 | 0.277 |
| SBP (mmHg) | 124.64 ± 14.99 | 138.75 ± 15.42 | 124.11 ± 14.82 | 139.52 ± 16.49 | 125.86 ± 15.39 | 136.96 ± 12.7 | -0.677 | 0.499 |
| DBP (mmHg) | 78.80 ± 7.86 | 82.06 ± 9.40 | 78.84 ± 7.84 | 81.76 ± 9.75 | 78.72 ± 7.96 | 82.78 ± 8.68 | -0.066 | 0.947 |
| HbA1c (%) | 9.20 ± 2.05 | 11.01 ± 1.88 | 9.13 ± 2.01 | 10.96 ± 2.02 | 9.39 ± 2.13 | 11.14 ± 1.53 | -1.183 | 0.238 |
| FPG (mmol/L) | 7.52 ± 1.49 | 8.21 ± 1.70 | 7.53 ± 1.43 | 8.16 ± 1.68 | 7.49 ± 1.63 | 8.31 ± 1.78 | 0.061 | 0.951 |
| ALT (U/L) | 30.3 (28.2,32.5) | 30 (27.9,32.2) | 30.2 (27,33.3) | 29.6 (27.4,32.3) | 30.5 (29.6,31.35) | 30.3 (28.2,31.7) | -0.942 | 0.346 |
| AST (U/L) | 30.5 (27.8,32.5) | 29.9 (26.3,33) | 29.9 (26.5,33.3) | 30.35 (26.3,33.6) | 30.65 (29.65,31.5) | 29.4 (26, 33) | -0.983 | 0.325 |
| eGFR [ml/ (min·1.73 m2 )] |
101.25 ± 19.01 | 79.47 ± 22.95 | 101.52 ± 18.99 | 78.98 ± 22.36 | 100.63 ± 19.13 | 80.63 ± 24.76 | 0.224 | 0.823 |
| TC (mmol/L) | 3.90 ± 1.43 | 4.61 ± 1.48 | 3.95 ± 1.46 | 4.59 ± 1.51 | 3.77 ± 1.38 | 4.66 ± 1.42 | 0.969 | 0.333 |
| LDL (mmol/L) | 2.47 ± 0.98 | 2.77 ± 0.99 | 2.43 ± 0.98 | 2.80 ± 0.94 | 2.57 ± 1.00 | 2.71 ± 1.13 | -1.082 | 0.280 |
| TG | 1.51 ± 0.55 | 1.74 ± 0.44 | 1.52 ± 0.57 | 1.72 ± 0.47 | 1.49 ± 0.50 | 1.79 ± 0.37 | 0.230 | 0.818 |
| ALB (g/L) | 31.98 ± 5.30 | 32.31 ± 4.54 | 31.89 ± 5.16 | 32.23 ± 4.63 | 32.18 ± 5.64 | 32.52 ± 4.43 | -0.556 | 0.579 |
| TP (g/L) | 64.49 ± 7.39 | 63.72 ± 6.15 | 64.40 ± 7.66 | 63.53 ± 6.25 | 64.70 ± 6.74 | 64.17 ± 6.02 | -0.509 | 0.611 |
3.2. Univariate analysis of DME in T2DM patients in the training set
A total of 343 patients with T2DM were included in the training set. LASSO regression screened 19 candidate variables with non-zero coefficients, including gender, age, BMI, smoking, alcohol consumption, family history of DM, diabetes duration, SBP, DBP, ALT, AST, eGFR, FPG, HbA1c, TC, LDL-C, TG, ALB, and TP. Ten-fold cross-validation was conducted to determine the optimal λ parameter, reducing variables while maintaining model fit. After screening, nine predictors with non-zero coefficients remained: age, diabetes duration, SBP, eGFR, FPG, HbA1c, TC, LDL-C, and TG (Figure 2).
Figure 2.
LASSO regression results (A: Variable selection; B: Cross-validation).
3.3. Logistic regression analysis of factors influencing DME in T2DM patients
Logistic regression analysis was performed using DME as the dependent variable. Independent variables evaluated included patient age, diabetes duration, SBP, eGFR, FPG, HbA1c, TC, LDL-C, and TG. Results indicated that diabetes duration, SBP, eGFR, and HbA1c were significantly associated with DME development (Table 2).
Table 2.
Logistic regression analysis of factors influencing DME in T2DM patients.
| B | S.E. | Wald | P | OR | 95%CI | ||
|---|---|---|---|---|---|---|---|
| Age | 0.035 | 0.021 | 2.824 | 0.093 | 1.035 | 0.994 | 1.078 |
| DM disease course (years) | 0.379 | 0.094 | 16.145 | <0.001 | 1.460 | 1.214 | 1.757 |
| SBP(mmHg) | 0.064 | 0.014 | 20.84 | <0.001 | 1.066 | 1.037 | 1.095 |
| HbA1c | 0.395 | 0.113 | 12.167 | <0.001 | 1.484 | 1.189 | 1.852 |
| FPG | 0.199 | 0.142 | 1.980 | 0.159 | 1.221 | 0.925 | 1.611 |
| eGFR | -0.064 | 0.012 | 26.719 | <0.001 | 0.938 | 0.916 | 0.961 |
| TC | 0.168 | 0.150 | 1.251 | 0.263 | 1.183 | 0.881 | 1.589 |
| LDL | 0.362 | 0.214 | 2.861 | 0.091 | 1.436 | 0.944 | 2.183 |
| TG | 0.712 | 0.405 | 3.089 | 0.079 | 2.038 | 0.921 | 4.509 |
| Quantity | -17.936 | 3.221 | 31.01 | 0.000 | 0.000 | ||
3.4. Construction and evaluation of nomogram model for T2DM patients with DME
Based on multivariate logistic regression analysis, a nomogram predicting DME risk in T2DM patients was constructed using diabetes duration, SBP, eGFR, and HbA1c (Figure 3). This nomogram provided a visual tool for individualized risk assessment. The usage method involved locating each patient’s values on corresponding variable axes, summing scores on the “Total Score” axis, and projecting the total onto the “Risk” axis to determine predicted probability.
Figure 3.
Nomogram model for DME in T2DM patients.
3.5. Model validation
The four variables identified from logistic regression were included in the nomogram (Figure 3). ROC curve analysis was conducted to evaluate model discrimination. Results showed excellent discrimination between T2DM patients with and without DME. The training set AUC was 0.905 (95% CI: 0.858–0.951; Figure 4A), and the validation set AUC was 0.884 (95% CI: 0.820–0.949; Figure 4B), outperforming individual variables.
Figure 4.
ROC curves of the model in training (A) and internal validation sets (B).
A systematic comparison was performed between single-variable models and the comprehensive nomogram. Delong test results showed statistically superior predictive ability of the comprehensive model compared to single-variable models (Figure 5, Table 3).
Figure 5.
ROC curve analysis of individual factors.
Table 3.
Comparison of individual factors and nomogram predictive values.
| AUC | 95%CI | P | Sensitivity | Specificity | Cut-off | Delong Z | P | |
|---|---|---|---|---|---|---|---|---|
| Course of DM-only model | 0.713 | 0.627-0.799 | <0.001 | 51.9 | 86.5 | 10.45 | 4.452 | <0.001 |
| SBP-only model | 0.748 | 0.676-0.821 | <0.001 | 57.4 | 81.7 | 137.5 | 4.263 | <0.001 |
| HbA1c-only model | 0.740 | 0.669-0.812 | <0.001 | 74.1 | 65.7 | 9.95 | 4.261 | <0.001 |
| eGFR-only model | 0.777 | 0.703-0.851 | <0.001 | 68.5 | 78.2 | -86.5 | 3.805 | <0.001 |
| Nomogram model | 0.905 | 0.858-0.951 | <0.001 | 72.2 | 95.5 |
Delong test results indicated that predictions from the nomogram were significantly different from those of the other four models.
Calibration curve analysis showed high consistency between predicted and actual probabilities (Figures 6A, B). The average calibration error was 0.020 (training set, Hosmer-Lemeshow test: χ²=3.748, P = 0.879) and 0.032 (validation set, Hosmer-Lemeshow test: χ²=5.024, P = 0.755). DCA evaluated net clinical benefit (Figures 6C, D). In the training cohort, the nomogram’s net benefit surpassed those of the “all-intervention” or “no-intervention” strategies across threshold probabilities ranging from 2% to 100%, achieving a peak net benefit of 0.14, implying the avoidance of 14 unnecessary interventions per 100 patients. Similarly, in the validation cohort, the net benefit remained consistently superior to the reference strategies at thresholds between 1.8% and 100%, supporting its effectiveness within moderate-to-high-risk groups.
Figure 6.
Calibration curves (A, B) and decision curves (C, D) for the training and internal validation sets.
4. Discussion
In this study, we developed and validated a nomogram for predicting diabetic macular edema (DME) risk specifically in working-age adults (18–60 years) with type 2 diabetes mellitus (T2DM). Using LASSO regression, we identified four key predictors—diabetes duration, systolic blood pressure (SBP), estimated glomerular filtration rate (eGFR), and HbA1c—and integrated them into a clinically actionable model. The nomogram demonstrated excellent discriminatory power, with AUC values of 0.905 and 0.884 in the training and validation sets, respectively, and well-calibrated prediction accuracy.
4.1. Novel contributions and clinical implications
This study addresses a critical gap in DME risk prediction by developing the first nomogram specifically tailored for working-age T2DM patients using exclusively routine clinical parameters. Unlike previous models that required specialized ophthalmologic equipment or focused on older populations, our tool leverages readily available biomarkers (HbA1c, eGFR, SBP) already collected in primary care settings. This approach makes DME risk assessment accessible to non-ophthalmologists, potentially transforming screening workflows in resource-limited environments. By enabling early identification of high-risk individuals in primary care, our model facilitates timely referrals to ophthalmology, optimizes resource allocation, and may significantly reduce vision loss in a demographic where visual impairment carries substantial socioeconomic consequences. The integration of renal function (eGFR) with metabolic and hemodynamic factors also provides new insights into the multifactorial pathophysiology of DME, highlighting the interconnectedness of microvascular complications in diabetes.
4.2. Interpretation of key predictors
The duration of diabetes emerged as a strong predictor (OR = 1.460), consistent with established literature linking longer disease duration to microvascular damage (23, 24). The slightly lower odds ratio compared to some previous studies may reflect better overall diabetes management in our younger cohort or their preserved metabolic compensatory mechanisms. Pathophysiologically, prolonged hyperglycemia drives DME through advanced glycation end-product accumulation, oxidative stress, and hemodynamic abnormalities that disrupt the blood-retinal barrier (25, 26). Our findings support intensified screening (e.g., every 3–6 months) for patients with diabetes duration ≥10 years.
Systolic blood pressure (SBP) was independently associated with DME risk (OR = 1.066), aligning with large-scale studies emphasizing blood pressure control in diabetic retinopathy (27). Notably, diastolic pressure showed no significant association, suggesting pulse pressure may be more relevant in DME pathogenesis. Hypertension likely contributes to DME by damaging endothelial tight junctions and promoting VEGF expression (28, 29).
The inverse relationship between eGFR and DME risk (OR = 0.938) underscores the role of renal dysfunction in DME development. This association may be explained by uremic toxin accumulation, chronic inflammation, and renal anemia-induced retinal hypoxia (30–33). Our results support integrating renal function assessment into DME risk stratification.
HbA1c levels significantly predicted DME risk (OR = 1.484), consistent with the known role of chronic hyperglycemia in microvascular complications (34, 35). The persistence of elevated risk even in patients with currently controlled HbA1c highlights the “metabolic memory” phenomenon (35–38), emphasizing the importance of early and sustained glycemic control.
5. Limitations
While this study successfully developed an effective DME prediction model, several limitations warrant careful consideration. The single-center, cross-sectional design, despite employing prospective data collection, may limit the generalizability of the findings, particularly as the sample was drawn from a single institution and external validation was not performed. Furthermore, the observed DME prevalence (15.7%) differed substantially from the initial estimate used for sample size calculation (40%), and the relatively small number of DME events compared to candidate predictors raises concerns about potential overfitting, despite the use of LASSO regression. It is also critical to acknowledge that the cross-sectional nature of the study means that temporal causality cannot be inferred from the observed associations. Clinically, the model’s scope is limited by the deliberate exclusion of specialized imaging parameters, such as OCT-measured retinal thickness, which are known to correlate with DME severity, and by not accounting for the impact of interventions like anti-VEGF therapy that can alter DME progression. Finally, while focusing on working-age adults (18–60 years) enhances relevance for that demographic, it restricts applicability to older populations who bear a higher disease burden.
These limitations highlight clear directions for future research. Priority should be given to conducting multi-center prospective studies with larger, more diverse cohorts to enable external validation and refine the model’s general applicability. Subsequent model iterations could significantly enhance predictive power by incorporating multi-modal imaging data, such as OCT and color fundus photography, and by considering the modifying effects of standard treatments. Further research should also explore optimal intervention thresholds and validate the model’s performance specifically in older age groups to provide more comprehensive clinical guidance.
6. Conclusion
This nomogram effectively integrates HbA1c, renal function, and hemodynamic parameters to identify key factors associated with diabetic macular edema (DME) risk in working-age patients with type 2 diabetes mellitus (T2DM), demonstrating high accuracy. By utilizing routine clinical measures, it facilitates implementation in primary care settings, offering the potential to reduce vision loss through timely referrals. Future multicenter studies are warranted to verify its generalizability and explore its integration with emerging biomarkers.
Acknowledgments
The authors sincerely appreciate all the subjects and their families who took part in this research.
Funding Statement
The author(s) declared that financial support was not received for this work and/or its publication.
Footnotes
Edited by: Raba Thapa, Tilganga Institute of Ophthalmology, Nepal
Reviewed by: Barbara Ciastek, University of Opole, Poland
Laura De Luca, University of Messina, Italy
Data availability statement
The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding author.
Ethics statement
The Ethics Committee of the Affiliated Hospital of North Sichuan Medical College approved the study protocol, and all enrolled participants provided informed consent.
Author contributions
QP: Conceptualization, Methodology, Writing – original draft, Writing – review & editing. LW: Data curation, Formal analysis, Validation, Visualization, Writing – review & editing. RH: Investigation, Resources, Software, Supervision, Writing – review & editing. XL: Funding acquisition, Project administration, Writing – review & editing. BL: Writing – original draft, Writing – review & editing, Validation.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that Generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
- 1. Sorour OA, Levine ES, Baumal CR, Elnahry AG, Braun P, Girgis J, et al. Persistent diabetic macular edema: definition, incidence, biomarkers, and treatment methods. Surv Ophthalmol. (2023) 68:147–74. doi: 10.1016/j.survophthal.2022.11.008, PMID: [DOI] [PubMed] [Google Scholar]
- 2. Gurreri A, Pazzaglia A. Diabetic macular edema: state of art and intraocular pharmacological approaches. Adv Exp Med Biol. (2021) 1307:375–89. doi: 10.1007/5584_2020_535, PMID: [DOI] [PubMed] [Google Scholar]
- 3. Teo ZL, Tham YC, Yu M, Chee ML, Rim TH, Cheung N, et al. Global prevalence of diabetic retinopathy and projection of burden through 2045: systematic review and meta-analysis. Ophthalmology. (2021) 128:1580–91. doi: 10.1016/j.ophtha.2021.04.027, PMID: [DOI] [PubMed] [Google Scholar]
- 4. GBD 2019 Blindness and Vision Impairment Collaborators, & Vision Loss Expert Group of the Global Burden of Disease Study Causes of blindness and vision impairment in 2020 and trends over 30 years, and prevalence of avoidable blindness in relation to VISION 2020: the Right to Sight: an analysis for the Global Burden of Disease Study. The Lancet. Global health. (2021) 9:e144–e60. doi: 10.1016/S2214-109X(20)30489-7, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Kohli P, Tripathy K, Patel BC. Macular edema. In: Statpearls. StatPearls Publishing LLC; (2025). Treasure Island (FL) ineligible companies. Disclosure: Koushik Tripathy declares no relevant financial relationships with ineligible companies. Disclosure: Bhupendra Patel declares no relevant financial relationships with ineligible companies. (Treasure Island (FL): StatPearls Publishing Copyright © 2025). [Google Scholar]
- 6. Uludag G, Hassan M, Matsumiya W, Pham BH, Chea S, Trong Tuong Than N, et al. Efficacy and safety of intravitreal anti-vegf therapy in diabetic retinopathy: what we have learned and what should we learn further? Expert Opin Biol Ther. (2022) 22:1275–91. doi: 10.1080/14712598.2022.2100694, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Kalra G, Zarranz-Ventura J, Chahal R, Bernal-Morales C, Lupidi M, Chhablani J. Optical coherence tomography (Oct) angiolytics: A review of oct angiography quantitative biomarkers. Surv Ophthalmol. (2022) 67:1118–34. doi: 10.1016/j.survophthal.2021.11.002, PMID: [DOI] [PubMed] [Google Scholar]
- 8. Forrester JV, Kuffova L, Delibegovic M. The role of inflammation in diabetic retinopathy. Front Immunol. (2020) 11:583687. doi: 10.3389/fimmu.2020.583687, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. De Boer IH, Khunti K, Sadusky T, Tuttle KR, Neumiller JJ, Rhee CM, et al. Diabetes management in chronic kidney disease: A consensus report by the american diabetes association (Ada) and kidney disease: improving global outcomes (Kdigo). Diabetes Care. (2022) 45:3075–90. doi: 10.2337/dci22-0027, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Lin KY, Hsih WH, Lin YB, Wen CY, Chang TJ. Update in the epidemiology, risk factors, screening, and treatment of diabetic retinopathy. J Diabetes Investig. (2021) 12:1322–5. doi: 10.1111/jdi.13480, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Kim IM, Radgoudarzi N, Chen EM, Liu J, Moussa K, Shantha JG, et al. Time to progression to proliferative diabetic retinopathy in patients with type 2 diabetes. JAMA Netw Open. (2025) 8:e2521150. doi: 10.1001/jamanetworkopen.2025.21150, PMID: [DOI] [PubMed] [Google Scholar]
- 12. Li Y, Teng D, Shi X, Qin G, Qin Y, Quan H, et al. Prevalence of diabetes recorded in mainland China using 2018 diagnostic criteria from the american diabetes association: national cross sectional study. Bmj. (2020) 369:m997. doi: 10.1136/bmj.m997, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Harhay MO, Au DH, Dell SD, Gould MK, Redline S, Ryerson CJ, et al. Methodologic guidance and expectations for the development and reporting of prediction models and causal inference studies. Ann Am Thorac Soc. (2020) 17:679–82. doi: 10.1513/AnnalsATS.202002-141ED, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Gutin I. In bmi we trust: reframing the body mass index as a measure of health. Soc Theory Health. (2018) 16:256–71. doi: 10.1057/s41285-017-0055-0, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Group CMAOSFD . Guidelines for clinical diagnosis and treatment of diabetic retinopathy in China (2014). Chin J Ophthalmol. (2014) 50:851–65. [Google Scholar]
- 16. Chinese Elderly Type 2 Diabetes Prevention and Treatment of Clinical Guidelines Writing Group, Geriatric Endocrinology and Metabolism Branch of Chinese Geriatric Society, Geriatric Endocrinology and Metabolism Branch of Chinese Geriatric Health Care Society, Geriatric Professional Committee of Beijing Medical Award Foundation, & National Clinical Medical Research Center for Geriatric Diseases (PLA General Hospital). Zhonghua nei ke za zhi. (2022) 61:12–50. doi: 10.3760/cma.j.cn112138-20211027-00751, PMID: [DOI] [PubMed] [Google Scholar]
- 17. Casadei G, Filippini M, Brognara L. Glycated hemoglobin (Hba1c) as a biomarker for diabetic foot peripheral neuropathy. Diseases. (2021) 9. doi: 10.3390/diseases9010016, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Bergman M, Manco M, Satman I, Chan J, Schmidt MI, Sesti G, et al. International diabetes federation position statement on the 1-hour post-load plasma glucose for the diagnosis of intermediate hyperglycaemia and type 2 diabetes. Diabetes Res Clin Pract. (2024) 209:111589. doi: 10.1016/j.diabres.2024.111589, PMID: [DOI] [PubMed] [Google Scholar]
- 19. Kwo PY, Cohen SM, Lim JK. Acg clinical guideline: evaluation of abnormal liver chemistries. Am J Gastroenterol. (2017) 112:18–35. doi: 10.1038/ajg.2016.517, PMID: [DOI] [PubMed] [Google Scholar]
- 20. Levey AS, Coresh J, Tighiouart H, Greene T, Inker LA. Measured and estimated glomerular filtration rate: current status and future directions. Nat Rev Nephrol. (2020) 16:51–64. doi: 10.1038/s41581-019-0191-y, PMID: [DOI] [PubMed] [Google Scholar]
- 21. Jung E, Kong SY, Ro YS, Ryu HH, Shin SD. Serum cholesterol levels and risk of cardiovascular death: A systematic review and a dose-response meta-analysis of prospective cohort studies. Int J Environ Res Public Health. (2022) 19. doi: 10.3390/ijerph19148272, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Seidu S, Kunutsor SK, Khunti K. Serum albumin, cardiometabolic and other adverse outcomes: systematic review and meta-analyses of 48 published observational cohort studies involving 1,492,237 participants. Scand Cardiovasc J. (2020) 54:280–93. doi: 10.1080/14017431.2020.1762918, PMID: [DOI] [PubMed] [Google Scholar]
- 23. Wong TY, Cheung CM, Larsen M, Sharma S, Simó R. Diabetic retinopathy. Nat Rev Dis Primers. (2016) 2:16012. doi: 10.1038/nrdp.2016.12, PMID: [DOI] [PubMed] [Google Scholar]
- 24. Poráčová J, Nagy M, Mydlárová Blaščáková M, Konečná M, Sedlák V, Zahatňanská M, et al. Diabetes mellitus and its influence on the incidence and process of diabetic retinopathy. Cent Eur J Public Health. (2023) 31:S4–s9. doi: 10.21101/cejph.a7838, PMID: [DOI] [PubMed] [Google Scholar]
- 25. Zhang J, Zhang J, Zhang C, Zhang J, Gu L, Luo D, et al. Diabetic macular edema: current understanding, molecular mechanisms and therapeutic implications. Cells. (2022) 11. doi: 10.3390/cells11213362, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Stana D, Potop V, Istrate SL, Eniceicu C, Mihalcea AR, Paşca IG, et al. Variability of diabetic macular edema in correlation with hypertension retinopathy in patients with diabetes mellitus and essential hypertension. Rom J Ophthalmol. (2019) 63:327–38. doi: 10.22336/rjo.2019.52, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Do DV, Han G, Abariga SA, Sleilati G, Vedula SS, Hawkins BS. Blood pressure control for diabetic retinopathy. Cochrane Database Syst Rev. (2023) 3:Cd006127. doi: 10.1002/14651858.CD006127.pub3, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Haydinger CD, Ferreira LB, Williams KA, Smith JR. Mechanisms of macular edema. Front Med (Lausanne). (2023) 10:1128811. doi: 10.3389/fmed.2023.1128811, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Chou YB, Chang JY, Chou YJ, Pu C. Association between renal function and the treatment of diabetic macular edema in long-term cohort study. Sci Rep. (2024) 14:26098. doi: 10.1038/s41598-024-77530-3, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Wautier JL, Wautier MP. Vascular permeability in diseases. Int J Mol Sci. (2022) 23. doi: 10.3390/ijms23073645, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Ou SH, Chang WC, Wu LY, Wang SI, Wei JC, Lee PT. Diabetic macular edema is predictive of renal failure in patients with diabetes mellitus and chronic kidney disease. J Clin Endocrinol Metab. (2024) 109:761–70. doi: 10.1210/clinem/dgad581, PMID: [DOI] [PubMed] [Google Scholar]
- 32. Zhang X, Hao X, Wang L, Xie L. Association of abnormal renal profiles with subretinal fluid in diabetic macular edema. J Ophthalmol. (2022) 2022:5581679. doi: 10.1155/2022/5581679, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Wang L, Jin L, Wang W, Gong X, Li Y, Li W, et al. Association of renal function with diabetic retinopathy and macular oedema among chinese patients with type 2 diabetes mellitus. Eye (Lond). (2023) 37:1538–44. doi: 10.1038/s41433-022-02173-5, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Eleftheriadou A, Riley D, Zhao SS, Austin P, Hernández G, Lip GYH, et al. Risk of diabetic retinopathy and diabetic macular oedema with sodium-glucose cotransporter 2 inhibitors and glucagon-like peptide 1 receptor agonists in type 2 diabetes: A real-world data study from a global federated database. Diabetologia. (2024) 67:1271–82. doi: 10.1007/s00125-024-06132-5, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Suzuki Y, Kiyosawa M. Relationship between diabetic nephropathy and development of diabetic macular edema in addition to diabetic retinopathy. Biomedicines. (2023) 11. doi: 10.3390/biomedicines11051502, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Bhutia CU, Kaur P, Singh K, Kaur S. Evaluating peripheral blood inflammatory and metabolic biomarkers as predictors in diabetic retinopathy and diabetic macular edema. Indian J Ophthalmol. (2023) 71:2521–5. doi: 10.4103/ijo.Ijo_345_23, PMID: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Meduri A, De Luca L, Oliverio GW, Mancini M, Minutoli L, Silvagno F, et al. DEXAMETHASONE INTRAVITREAL INJECTION IN DIABETIC PATIENTS UNDERGOING CATARACT SURGERY: an updated literature review. Retina (Philadelphia Pa.). (2025) 45:1030–42. doi: 10.1097/IAE.0000000000004381, PMID: [DOI] [PubMed] [Google Scholar]
- 38. Oliverio GW, Meduri A, Brancati VU, Ingrande I, De Luca L, Raimondo ED, et al. Clinical and optical coherence tomography biomarkers as prognostic factors in dexamethasone intravitreal implant for diabetic macular edema. Eur J Ophthalmol. (2024) 34:1810–8. doi: 10.1177/11206721241235242, PMID: [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding author.
The Ethics Committee of the Affiliated Hospital of North Sichuan Medical College approved the study protocol, and all enrolled participants provided informed consent.