[Construction of a regional risk prediction model for diabetic retinopathy in patients with type 2 diabetes]

Objective: To construct a risk prediction model for diabetic retinopathy (DR) in patients with type 2 diabetes in Gansu Province. Methods: It was a diagnostic study. Clinical data of patients with type 2 diabetes who were hospitalized in the Department of Endocrinology of the Second Hospital of Lanzhou University and received ophthalmology consultation from November 2023 to October 2024 were collected, including gender, age, duration of diabetes, body mass index, blood pressure, history of hypertension, and laboratory test indicators [fasting blood glucose, glycated hemoglobin, serum creatinine, estimated glomerular filtration rate (eGFR), 24-hour urine protein quantification (24-hUP), and serum uric acid]. The patients were divided into the non-DR group and the DR group based on the results of fundus examination and randomly assigned into the training set and the test set at a ratio of 7∶3 using the computer-generated random sequence method. The univariate binary Logistic regression analysis and least absolute shrinkage and selection operator (LASSO) regression were used to screen variables, and with the multivariate binary Logistic regression analysis, a DR risk prediction nomogram model was constructed. The performance of the prediction model was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curve, precision-recall curve, confusion matrix, and clinical decision curve. The sensitivity was set at 0.8 and 0.9 to compare the corresponding specificity and precision. The statistical analysis was performed using the Mann-Whitney U test and the χ² test. Results: A total of 1 338 patients with type 2 diabetes were enrolled, including 836 males and 502 females. The average age was (61±11) years, with an age range of 18 to 88 years. The total detection rate of DR was 44.4% (594/1 338), with 44.3% (370/836) in males and 44.6% (224/502) in females. Compared with the non-DR group (744 cases), the DR group (594 cases) had a longer duration of diabetes [12.00 (7.00, 20.00) years], higher fasting blood glucose level [8.80 (7.00, 11.00) mmol/L], higher glycated hemoglobin level [8.70% (7.50%, 10.33%)], higher proportion of patients with a history of hypertension [54.5% (324/594)], higher systolic blood pressure [133.00 (120.00, 148.00) mmHg (1 mmHg=0.133 kPa)], and higher proportion of patients with concurrent diabetic nephropathy [56.7% (337/594)], and the differences were statistically significant (all P<0.05). In terms of laboratory tests, the eGFR of the DR group [94.09 (69.67, 103.04) ml/(min·m²)] was lower than that of the non-DR group [97.62 (89.04, 106.50) ml/(min·m²)], and the serum creatinine [68.50 (54.48, 93.18) μmol/L], 24-hUP [0.28 (0.15, 0.81) g], and serum uric acid [334.00 (270.00, 398.00) μmol/L] were higher than those of the non-DR group, and the differences were statistically significant (all P<0.05). The LASSO regression and multivariate binary Logistic regression analysis identified five key variables: duration of diabetes, eGFR, concurrent diabetic nephropathy, 24-hUP, and fasting blood glucose level. The AUC of the DR risk prediction nomogram model constructed using these variables was 0.743 in the training set and 0.736 in the test set. The corresponding specificity for sensitivity of 0.8 and 0.9 was 0.484 and 0.263, and the precision was 0.628 and 0.556, respectively. The comprehensive analysis of calibration curve, precision-recall curve, confusion matrix, and clinical decision curve indicated good discrimination, calibration, and predictive accuracy of the prediction model. Conclusion: The results of the single-center study suggested that the duration of diabetes, blood glucose level, and renal function-related indicators were risk factors for concurrent DR in patients with type 2 diabetes in Gansu Province. The DR risk prediction model established based on these routine clinical indicators had high screening potential and clinical application value, and could better identify high-risk populations.