Development of machine learning Predictive Model for Type 2 Diabetic Retinopathy Using the Triglyceride-glucose index explained by SHAP method

Development of machine learning Predictive Model for Type 2 Diabetic Retinopathy Using the Triglyceride-glucose index explained by SHAP method | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Development of machine learning Predictive Model for Type 2 Diabetic Retinopathy Using the Triglyceride-glucose index explained by SHAP method XiaoQin Liu, ShuYing Wu, Yue Yang, Yang Li, XinTing Zhang, Ling Qin, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5602589/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Introduction : This study aimed to develop a diabetic retinopathy (DR) Prediction model using various machine learning algorithms incorporating the novel predictor Triglyceride-glucose index (TyG). Furthermore, the model was interpreted using the SHapley Additive exPlanations (SHAP) method. Method : Real-world data were collected from a general hospital in a major city and a county clinic, then divided into the DR Group (1392) and non-DR group (2358). Baseline data were collected, and variables were selected using Recursive Feature Elimination with Cross-Validation (RFECV). The performance of five machine learning algorithms, including Logistic Regression model (LR), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and XGBoost (XGB), was assessed based on accuracy, sensitivity, specificity, and Area Under the Curve (AUC) of the Receiver Operating characteristic Curve (ROC). The optimal model was interpreted using SHAP. Result ：SVM and LR demonstrated superior performance in both the test set and training set (ROC, 0.85 and 0.82, respectively). The top five predictors identified by SHAP analysis included TyG, Insulin therapy, HbA1c, Diabetes Course, HDL. HDL was identified as a protective factor, while the remaining factors were associated with retinopathy. Conclusion ：LR and SVM demonstrated the best performance. This is the first study constructing a DR Prediction model using TyG index. Notably, TyG significantly predicted DR and may serve as a crucial indicator for guiding clinical screening of high DR Risk. Health sciences/Endocrinology Health sciences/Endocrinology/Endocrine system and metabolic diseases Health sciences/Endocrinology/Endocrine system and metabolic diseases/Diabetes Health sciences/Endocrinology/Endocrine system and metabolic diseases/Dyslipidaemias TyG-index Diabetic retinopathy Machine learning Predictive model SHAP Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Research Insights What is currently known about this topic? TyG index is a simple tool for detecting IR. It has been shown to be closely related with macrovascular and microvascular complications in diabetic patients. What is the key research question? Whether TyG can predict the occurrence of DR. What is new? TyG can be used to predict diabetic retinopathy of T2DM. The TyG cut-off value for the prediction of DR is 4. How might this study influence clinical practice? It can be used to screen for high-risk DR to improve early diagnosis and implement timely referral for patients, especially in primary hospitals and poor areas . Introduction Diabetes has a high incidence and mortality rate, thus emerging as a major global health challenge. The global prevalence of diabetes in people aged 20~79 years may increase to 12.2% by 2045[1]. Diabetic retinopathy (DR), the most common and serious ocular complication of type 2 diabetes, can result in visual impairment and even permanent vision loss[2]. To date, DR is the leading cause of blindness among individuals aged 20~74, contributing to new cases of preventable blindness in developing countries[3, 4]. A meta-analysis showed that the worldwide prevalence of DR is about 22.27%[5]. Notably, China has the largest number of diabetic patients in the world[1], with about 27.9% of diabetes cases being DR[6], of which 34% are in rural areas. Most DR patients are asymptomatic or mild in the early stage and are only detected when great damage has occurred or at the late stage[2]. Therefore, early screening of DR is crucial. However, the detection rate of DR remains suboptimal[7], particularly in certain primary hospitals and remote regions due to the constraints in medical conditions and limited social resources. Therefore, more accessible methods are needed to aid healthcare professionals in diagnosis and screening of DR. To date, factors such as diabetes duration, age, BMI, smoking, blood pressure, HbA1c levels, and cholesterol have been identified as risk factors for diabetic retinopathy (DR) [8, 9]. However, only a limited number of studies have incorporated insulin resistance (IR) into DR prediction models, despite substantial evidence[10, 11]demonstrating a strong association between IR and DR. The hyperinsulinemic-euglycemic clamp (HIEC) is the gold standard for detecting IR[12], However, it is expensive and complex, limiting its use. HOMA-IR[13] is the most commonly utilized method for estimating insulin resistance; however, it requires the measurement of fasting insulin levels, which imposes specific technical demands on laboratory capabilities. Additionally, this method is not suitable for patients undergoing insulin therapy and is not widely adopted in primary care settings or resource-limited regions. Therefore, HOMA-IR cannot be widely promoted in primary hospitals and poor areas[14, 15]. A new index has been recently identified for detecting IR: Triglyceride-glucose index (TyG)[16], TyG is calculated by fasting triglyceride (TG) and fasting blood glucose (FBG), providing a simple, reliable, and cheap detection tool[17]. Besides, this index tool has shown better results than HOMA-IR[18], Elsewhere, Srinivasan[19] and Yao[20]have shown that TyG and DR are closely related, while others not. Machine Learning has shown great prospects in disease prediction. Some simple biochemical indicators have been used to predict the occurrence of DR, thus improving the screening rate of disease. Tebeje, Jin, Yang, Li et al.[21-24] have developed various prediction models; however, no comparative analysis has been conducted to determine which model demonstrates the best performance[8]. The newly discovered predictors can improve the predictive value for the prediction model[25, 26], TyG, as a new index for measuring IR, has shown great value in many studies. However, no prediction model uses TyG to predict the DR among the DM2 population. In this study, TyG was incorporated into DR predictor, using various machine learning methods to construct a prediction model. The best method was selected, providing some reference for clinical DR screening. Furthermore, clinical data were collected from primary clinics and advanced medical institutions to increase the generalization of the model. The model was explained using the SHAP method to avoid the limitation of the traditional model "black box". Methods 2. Research design 2.1 Population This is a retrospective study, where patient data were extracted from the real-world database of The First Hospital of Jilin University (a general hospital in a major city) and Meihekou Central Hospital (a primary health care institution in the county) from January 1, 2010 to December 31, 2023. 2.2 Inclusion criteria a. Patients diagnosed with type 2 diabetes(T2DM) following the criteria of the 2024 American Diabetes Association[27]; b. Patients aged⩾18 years; c. Patients with complete indicators (triglyceride and fasting glucose). 2.3 Exclusion criteria a. Patients diagnosed with retinopathy at admission; b. Patients suffering from other retinal diseases, glaucoma, optic neuropathy, or eye diseases caused by systemic diseases; c. Patients with a history of eye surgery; d. Patients with severe systemic diseases (cancer, myocardial infarction, and dialysis history); e. Patients with data loss exceeding 20%. 2.4 Outcome The extraction variable was the first measurement record at first admission. Diagnosis information was extracted from the discharge diagnosis, and the follow-up was conducted until the first DR diagnosis, otherwise the last visit time was selected as the follow-up endpoint. Diagnostic criteria of DR included: A spectrum of retinal microvascular lesions on retinal examination with diabetic patients[28], including mild, moderate, and severe non-proliferative DR and proliferative DR. DR was diagnosed using 45° photos of macular center and indirect ophthalmoscopy when pupils were dilated. DR diagnosis was mainly achieved by endocrinologists and ophthalmologists. Patients diagnosed with retinopathy in other hospitals during the follow-up process were marked as DR, and the follow-up time was based on the earliest diagnosis time. Positive and negative samples were defined as DR and non-DR patients, respectively. 2.5 Ethical approval This study was conducted following the Helsinki Declaration and was approved by the Research Ethics Committee of The First Hospital of Jilin University (approval number: 2024-918). Each participant provided signed written informed consent. This study was reported based on TRIPOD[25]. 2.6 Baseline data collection Data indicators, including sex, age, height, weight, smoking, drinking, course of T2DM, insulin therapy, hypertension history, and laboratory parameters, were mainly obtained from literature reports[2, 29, 30]. Laboratory parameters included glycated hemoglobin (HbA1c, g/dL), total cholesterol (TC, mmol/L), high-density lipoprotein (HDL, mmol/L), low-density lipoprotein (LDL, mmol/L), triglyceride (TG, mg/dL), fasting blood glucose (FPG, mg/dL), fasting C-Peptide (C-PE, ng/ml), fasting insulin (FINS,μU/ml) and C-reactive protein (CRP,μg/L). Body mass index (BMI) was calculated formula as follows: weight (kg)/ height (m) 2 . TyG index: LN [ triglyceride (mg/dl)×plasma glucose (mg/dl)/2]. Three graduate students collated and cross-checked the collected data. A unified training for the data collection and collation personnel was conducted to ensure the accuracy and consistency of the data. 2.7 Sample size calculation The sample size was mainly based on Riley[31]standard for accurate estimation. Four sample sizes (at least 878) (predicted value with small average error, intercept model only, ensuring shrinkage coefficient of 0.9, and ensuring small optimism of apparent model) were calculated. The total sample size was at least 966 (878+10% (878)), considering that some medical record information was incomplete and 10% contingency. 2.8 Model construction Three single model algorithms (LR, DT, and SVM) and two integrated methods (RF and XGBoost) were used to train the model. LR is a machine learning method for solving binary classification problems, and is used to estimate the possibility of something. DT is a simple and easy-to-operate tree-type classification prediction model, providing intuitive and easy-to-understand results. However, DT can easily lead to overfitting during the classification process. SVM is widely used to construct a hyperplane concept to classify the observed values and can be used to deal with classification and regression problems. Compared with the single DT model, the integrated algorithms have higher accuracy but present more complicated and difficult results to explain. 2.9 Statistical analysis 2.9.1 Statistical interpretation： R4.2.1 software was used for all data analysis. The variables missing more than 20%, including fasting C-Peptide, fasting insulin, and CRP were deleted to improve the utilization rate of the data. For individual missing values, the average interpolation method and mode interpolation method were used for counting data and measuring data, respectively. The normally distributed data were expressed as X±S, and compared using two independent samples T-test. The non-normally distributed data were expressed as P50 (P25, P75) and compared using Mann-Whitney U test. The counting data were expressed as frequency (%) and compared using X 2 test. The correlation between the predictive indicators and retinopathy was assessed using univariate and multivariate logistic regression models. Variables with univariate analysis P < .05 were included in multivariate logistic regression analysis. The odds ratio (OR) and corresponding 95% confidence interval (CI) were used to indicate the trend of correlation. 2.9.2 Variable selection The model was trained and verified using Python3.9.15 software and tool kits Sklearn1.0.2, XGBoost1.7.4, and shap0.41.0. The recursive feature elimination with cross-validation (RFECV) was used to screen the optimal index. The least important features were continuously eliminated by training the model. The performance of the model was evaluated until the optimal performance index was reached. In this study, the variables with P < 0.1 in the comparison between groups were included in the RFECV model, and the best indicators were selected for subsequent prediction. 2.9.3 Model performance assessment The prediction efficiency of LR, DT, SVM, RF, and XGBoost5 models was evaluated based on accuracy, sensitivity, specificity, and Area Under the Curve (AUC) of receiver operating characteristic curve. Accuracy, precision, recall rate, and F1 score were used as the indicators. SHAP was used to visually explain the machine learning model. The Summary Plot, Shap Heat map, and importance ranking diagram were also drawn. The actual application of a single sample model was visualized via Force Plot. P <.05 was considered a statistically significant difference. 2.9.4 Model interpretation The best final model was determined based on SHAP algorithm[32], which assigns corresponding attribute values (SHAP values) to each variable. These SHAP values quantitatively measure the influence of each feature on the prediction accuracy. A SHAP summary diagram was generated to visualize the contribution of each feature to the model. Results 3.1 Social demographic and clinical characteristics The distribution of study participants is shown in Figure 1. A total of 2014 positive cases were extracted from real-world data, and 5058 negative cases were extracted via the computer random sampling method. Finally, 622 cases were excluded from the positive group (including 402 with missing data, 108 with other diseases affecting vision, 108 with vision surgery, and 94 with critical illness) and 2700 cases were excluded from the negative group (2305 with missing data and 395 with other serious systemic diseases), leaving 3755 cases (1392 cases in the positive group and 2358 cases in the negative group) About 63.09% of the remaining cases were males and 36.91% were females. The subjects were aged 18~91 years, with an average age of 54.8±12.3 (Table 1). Table 1 Comparison of baseline data between the two groups Variables Overall Control Case P ( n=3750 ) ( n=2358 ) ( n=1392 ) Sex Female 1384 (36.91) 800 (33.93) 584 (41.95) <.001 Male 2366 (63.09) 1558 (66.07) 808 (58.05) Age 54.8 ± 12.3 53.3 ± 12.8 57.2 ± 11.1 <.001 BMI 25.9 ± 3.7 26.0 ± 3.8 25.7 ± 3.4 .021 Smoking No 2798 (74.61) 1724 (73.11) 1074 (77.16) .007 Yes 952 (25.39) 634 (26.89) 318 (22.84) Drinking No 2912 (77.65) 1803 (76.46) 1109 (79.67) .025 Yes 838 (22.35) 555 (23.54) 283 (20.33) Hypertension No 1922 (51.25) 1313 (55.68) 609 (43.75) <.001 Yes 1828 (48.75) 1045 (44.32) 783 (56.25) Insulin therapy No 1484 (39.57) 1248 (52.93) 236 (16.95) <.001 Yes 2266 (60.43) 1110 (47.07) 1156 (83.05) Course of diabetes 10.0 (4.0, 15.0) 6.0 (3.0, 10.1) 12.0 (8.0, 20.0) <.001 HbA1c 8.7 (7.3, 10.2) 8.2 (7.0, 9.8) 9.3 (8.2, 10.7) <.001 TC 4.9 (4.2, 5.7) 4.9 (4.3, 5.7) 4.8 (4.1, 5.6) .001 HDL 1.1 (1.1, 1.3) 1.1 (1.1, 1.2) 1.1 (0.9, 1.3) <.001 LDL 3.0 (2.4, 3.5) 3.0 (2.5, 3.6) 2.9 (2.3, 3.4) <.001 TG 2.0 (1.3, 3.4) 2.0 (1.3, 3.2) 2.0 (1.4, 4.0) .072 FBG 8.9 (6.9, 12.0) 8.5 (6.7, 11.6) 9.7 (7.3, 12.6) <.001 TyG 2.2 (1.7, 2.9) 2.2 (1.6, 2.8) 2.3 (1.7, 3.0) <.001 The variables with univariate P<0.05 were included in multivariate logistics regression analysis. The results showed that HDL ( P ＜.001, OR=0.176, 95% CI:0.144~0.215), smoking ( P =.002, OR=0.759, 95% CI:0.620~0.92), LDL ( P =.021, OR=0.811, 95% CI:0.678~0.969), BMI ( P =.001, OR=0.815, 95% CI:0.726~0.915), TyG ( P =.038, OR=1.091, 95% CI:1.005, 1.185), age ( P =.014, OR=1.149, 95% CI:1.028~1.284), hypertension history ( P ＜.001, OR=1.506, 95% CI:1.273~1.781), course of T2DM ( P ＜.001, OR=1.798, 95% CI:1.615~2.002), insulin therapy ( P ＜.001, OR=3.166, 95% CI:2.621~3.824), HbA1c ( P ＜.001, OR=3.443, 95% CI:2.879~4.116), were the independent influencing factors of DR (Table 2). Table 2 Results of multivariate Logistic regression analysis Variable B Wald P OR (95%CI) HDL -1.736 291.731 ＜.001 0.176(0.144, 0.215) Smoking -0.275 7.136 .008 0.759(0.620, 0.929) LDL -0.21 5.3 .021 0.811(0.678, 0.969) BMI -0.205 12.035 .001 0.815(0.726, 0.915) TyG 0.087 4.319 .038 1.091(1.005, 1.185) Age 0.139 6.005 .014 1.149(1.028, 1.284) Hypertension 0.409 22.879 ＜.001 1.506(1.273, 1.781) Course of diabetes 0.587 115.088 ＜.001 1.798(1.615, 2.002) Insulin_therapy 1.152 143.06 ＜.001 3.166(2.621, 3.824) HbA1c 1.236 183.72 ＜.001 3.443(2.879, 4.116) 3.2 RFECV screening The independent variables were selected via RFECV method, and recursive features were eliminated. The results showed that nine indexes, including age, BMI, Diabetes course, Insulin therapy, Hypertension, HbA1c, TC, HDL, and TyG, were retained, and five indexes (sex, smoking, drinking, FBG, and LDL) were excluded (Figure 2). 3.3 Training and verification of the model The data were grouped into the training set and testing set (7:3). The last nine indexes were used as the optimal solution for model training. Moreover, SVM and LR showed the best performance in the test set and training set. The AUC curves of the five models are shown in Figure 3. The comparison of various prediction indexes (Table 3). Table 3 Prediction performance indicators of five models in the training and testing sets Model Train set Test set Precision Recall F1-score Accuracy Precision Recall F1-score Accuracy LR 0.70 0.58 0.64 0.75 0.70 0.69 0.70 0.79 XGBoost 0.69 0.62 0.65 0.75 0.65 0.62 0.64 0.75 SVM 0.70 0.60 0.65 0.75 0.73 0.67 0.7 0.80 RF 0.68 0.61 0.64 0.74 0.64 0.64 0.64 0.75 DT 0.59 0.56 0.58 0.69 0.59 0.64 0.62 0.72 3.4 SHAP model analysis 3.4.1 Ranking of feature importance: The top 5 variables based on SHAP values in the ML model are shown in Figure 4. A summary plot is shown in Figure 4(B), where “red” and “blue” represent higher and lower eigenvalues, respectively. SHAP value0 indicates positive influence. The dispersion of feature distribution (located above the Y axis) was directly related to the importance of the feature. Furthermore, TyG, Insulin therapy, HbA1c, and Diabetes course showed positive effects on retinopathy, while HDL showed negative effects. 3.4.2 Practical application of the model: The force Plot is shown in Figure 5. The blue and red arrows indicate that this factor reduces and increases the risk of retinopathy, respectively. The reference value represents the average SHAP value of all samples. F(x) represents the comprehensive SHAP value of each patient. The model can only predict the patient's retinopathy if the value of f(x) is greater than the base value. The 1045th case (Figure 5) was randomly predicted using the test set. Notably, f(x) was less than the base value, which was accurately predicted as the control group. 3.4.3 The effect of key features on outcomes: A partial dependence diagram of the influence of the first three indicators on infection was drawn, showing the marginal effect relationship between important characteristics and outcome variables (how important influencing factors affect retinopathy). TyG> 4 showed significant impact (Figure 6). Discussion Early screening and identification of people at high risk of developing DR is important for prevention and treatment. However, Li et al.[33]showed that only 17.48% of T2DM undergo routine DR Screening yearly. Doctors in the diabetes department, especially in primary hospitals that lack equipment and professionals, often overlook the early signs of retinopathy in their patients[29, 30]. Wang, Yang, Roşu et al.[34-36]built DR prediction models but did not include IR. Also, Qian[11]used HOMA-IR for prediction, which cannot be easily obtained in primary hospitals. In this study, a simple prediction model was developed using TyG. Results showed that logistic regression and SVM had the best performance among the five models, with area under ROC curve of the testing set and training set of 0.85 and 0.82. respectively. Similarly, Tsao et al.[37]showed that the SVM model has good predictive performance. Jiang et al.[30]also showed that logistic regression has good predictive performance. LR is a linear model suitable for small sample analysis, and it is sensitive to outliers. SVM finds the optimal hyperplane by optimizing the objective function, which is suitable for the analysis of complex data [30].Different algorithms have their own advantages in the modeling process, and no algorithm can be applied to all models. Therefore, the corresponding model should be selected according to the research design. This study was conducted based on the computational minimum sample size. Therefore, a study with a larger sample size is needed to compare the performance between LR and SVM. Herein, the variables included in LR and SVM were consistent. The remaining variables were described in previous studies except for the significant predictive ability of TyG[9, 22, 30, 38, 39].Previous studies showed that the interpretability of ML models is challenging[40]. In the present study, SHAP was used to improve model interpretability. SHAP is used to interpret a "black box" model that calculates a Shapley value for each feature in the prediction model to assess the importance of all feature combinations, reflecting their contribution to the predictive power of the overall model[32]. SHAP results showed that the top 5 variables included TyG, HDL, Insulin therapy, Diabetes course, and HbA1c. In this study, TyG was an important predictor of DR. Notably, increased TyG levels were associated with a high risk of DR due to the excessive production of mitochondrial superoxide in microvascular endothelial cells. This production is caused by pathway-specific insulin resistance, meanwhile, triggering intracellular hyperglycemia and vascular damage[41]. TyG is also a biochemical indicator reflecting the comprehensive effects of glucose and lipid metabolism, which are key causes of DR due to abnormal blood sugar and lipid metabolism[42]. Zhou et al. [43]showed similar results. TyG has never been used as a predictor of DR model before. Mirjalili[44], Zou[45], Yang[46]et al. showed that TyG, as a predictor, has a good predictive effect on coronary heart disease, fatty liver disease, and diabetes, similar to this study. Interestingly, TyG > 4 showed a significant predictive effect on DR. A study[16]reported that the TyG cut-off value for the diagnosis of insulin resistance is 4.65, consistent with this study. Another study[47]showed that the inflection point of TyG for controlling the transformation of diabetes is 8.88. A meta-analysis[48]showed that there is no standardized threshold value for TyG and clear threshold value related to TyG and DR. Therefore, more high-quality studies are needed to determine the optimal cut-off point for TyG in the future. Insulin therapy is another key predictor of DR. Wang, Li et al[34, 39]found that the insulin treatment is associated with a high risk of DR development, possibly because insulin therapy indicates worse islet function. While Ricard found it possibly related to the rapid reduction of blood glucose[49]. Nonetheless, more basic research is needed to confirm the findings. This study, along with others, has demonstrated that the duration of diabetes mellitus is closely associated with the development of DR[21, 35, 38, 50], possibly due to the prolonged exposure of blood vessels to risk factors. Herein, DR was linked to increased HBA1c levels. This may be due to the continuous elevation of blood glucose, which can lead to dysfunction of the retinal vascular endothelium and cause retinal ischemia and increased vascular permeability[2]. Similar findings have been reported previously[11, 21, 39]. In addition, HDL was identified as a key predictor of DR. Low HDL levels may indicate a higher risk of DR, consistent with Roşu, Liu, Li et al[36, 51, 52]. Although the model cannot replace the doctor's diagnosis, this model can assist physicians in identifying high-risk patients with DR, helping in clinical decision-making and patient management. To the best of our knowledge, TyG has never been used to predict diabetic retinopathy in T2DM. In addition, data were collected from multiple medical centers (large urban hospital and regional clinics) in real-world environments, greatly increasing the credibility of the model. Furthermore, the model utilizes routine patient examination data, which are readily accessible and do not impose additional burdens on patients or medical insurance. Besides, the data can be easily popularized in primary hospitals, increasing the screening rate of DR. Limitation: Only internal verification was conducted, thus effective external verification is needed. Conclusion and recommendations In this study, a DR Prediction model was built using TyG and other easily available clinical data. LR and SVM models had the best performance. SHAP showed that the most important predictors of DR were TyG, insulin therapy, diabetes course, HbA1c, and HDL. These findings suggest that baseline characteristics of patients can be used to screen for high-risk DR to improve early diagnosis and implement timely referral for patients. Abbreviations TyG-index: Triglyceride-glucose index DR: Diabetic retinopathy RI: insulin resistance HOMA-IR: Homeostatic Model Assessment for Insulin Resistance TC: Total cholesterol TG: Triglyceride LDL: Low-density lipoprotein HDL: High-density lipoprotein FBG: Fasting blood glucose BMI: Body mass index LR: Logistic regression DT: Decisiontree RF: Randomforest XCBoost: eXtremegradient boosting SVM: Support vector machine RFECV: Recursive Feature Elimination with Cross-Validation AUC: Area under the curves ROC: Receiver operating characteristic curve SHAP: Shapley additive explanation Declarations Acknowledgement We express our gratitude to all the participants of the study, including patients and researchers. We thank Core Facility of the First Hospital of Jilin University for their help and valuable comments on this manuscript. Contributors XqL and SyW searched the literature, designed the study, analysed the data, interpreted the results, and drafted the manuscript. YY, YL, and XtZ collected and analysed the data. FL conceived, designed, and supervised the study, interpreted the results, and revised the manuscript. LQ collected the data and drafted the manuscript. All authors contributed to the writing of the manuscript. Funding The study was supported by the grants from Jilin Medical and Health Talents Special Project(JLSWSRCZX2023-29). Data Availability The data in this paper comes from The First Hospital of Jilin University Real-World Data Application Platform and Meihekou Central Hospital. The datasets used or analysed during the current study available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate The study was approved by was approved by the Research Ethics Committee of The First Hospital of Jilin University (approval number: 2024-918). Each participant provided signed written informed consent. Consent for publication All the authors gave their consent to publication. Competing interests The authors declare no competing interests. Conflicts of Interest All authors have declared no competing interests. References Sun H, Saeedi P, Karuranga S, Pinkepank M, Ogurtsova K, Duncan BB, Stein C, Basit A, Chan JCN, Mbanya JC et al : IDF Diabetes Atlas: Global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045 . Diabetes Res Clin Pract 2022, 183 :109119. Cheung N, Mitchell P, Wong TY: Diabetic retinopathy . The Lancet 2010, 376 (9735):124-136. Congdon NG, Friedman DS, Lietman T: Important causes of visual impairment in the world today . JAMA-J Am Med Assoc 2003, 290 (15):2057-2060. Maniadakis N, Konstantakopoulou E: Cost Effectiveness of Treatments for Diabetic Retinopathy: A Systematic Literature Review . Pharmacoeconomics 2019, 37 (8). Teo ZL, Tham Y-C, Yu M, Chee ML, Rim TH, Cheung N, Bikbov MM, Wang YX, Tang Y, Lu Y et al : Global Prevalence of Diabetic Retinopathy and Projection of Burden through 2045: Systematic Review and Meta-analysis . Ophthalmology 2021, 128 (11):1580-1591. Zhang G, Chen H, Chen W, Zhang M: Prevalence and risk factors for diabetic retinopathy in China: a multi-hospital-based cross-sectional study . Br J Ophthalmol 2017, 101 (12):1591-1595. Vujosevic S, Aldington SJ, Silva P, Hernández C, Scanlon P, Peto T, Simó R: Screening for diabetic retinopathy: new perspectives and challenges . Lancet Diabetes Endocrinol 2020, 8 (4):337-347. Usman TM, Saheed YK, Nsang A, Ajibesin A, Rakshit S: A systematic literature review of machine learning based risk prediction models for diabetic retinopathy progression . Artif Intell Med 2023, 143 :102617. Guo H, Han F, Qu J-R, Pan C-Q, Sun B, Chen L-M: Scoring and validation of a simple model for predicting diabetic retinopathy in patients with type 2 diabetes based on a meta-analysis approach of 21 cohorts . Annals of Medicine 2024, 56 (1):2413920. Amoaku WM, Ghanchi F, Bailey C, Banerjee S, Banerjee S, Downey L, Gale R, Hamilton R, Khunti K, Posner E et al : Diabetic retinopathy and diabetic macular oedema pathways and management: UK Consensus Working Group . Eye (Lond) 2020, 34 (Suppl 1). Wang Q, Zeng N, Tang H, Yang X, Yao Q, Zhang L, Zhang H, Zhang Y, Nie X, Liao X et al : Diabetic retinopathy risk prediction in patients with type 2 diabetes mellitus using a nomogram model . Front Endocrinol (Lausanne) 2022, 13 :993423. Gastaldelli A: Measuring and estimating insulin resistance in clinical and research settings . Obesity (Silver Spring, Md) 2022, 30 (8):1549-1563. Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC: Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man . Diabetologia 1985, 28 (7):412-419. Zhao Q, Zhang T-Y, Cheng Y-J, Ma Y, Xu Y-K, Yang J-Q, Zhou Y-J: Impacts of triglyceride-glucose index on prognosis of patients with type 2 diabetes mellitus and non-ST-segment elevation acute coronary syndrome: results from an observational cohort study in China . Cardiovascular Diabetology 2020, 19 (1):108. Wang A, Tian X, Zuo Y, Chen S, Meng X, Wu S, Wang Y: Change in triglyceride-glucose index predicts the risk of cardiovascular disease in the general population: a prospective cohort study . Cardiovascular Diabetology 2021, 20 (1):113. Simental-Mendía LE, Rodríguez-Morán M, Guerrero-Romero F: The product of fasting glucose and triglycerides as surrogate for identifying insulin resistance in apparently healthy subjects . Metabolic Syndrome and Related Disorders 2008, 6 (4):299-304. Abbasi F, Reaven GM: Comparison of two methods using plasma triglyceride concentration as a surrogate estimate of insulin action in nondiabetic subjects: triglycerides × glucose versus triglyceride/high-density lipoprotein cholesterol . Metabolism: Clinical and Experimental 2011, 60 (12):1673-1676. Park HM, Lee HS, Lee Y-J, Lee J-H: The triglyceride-glucose index is a more powerful surrogate marker for predicting the prevalence and incidence of type 2 diabetes mellitus than the homeostatic model assessment of insulin resistance . Diabetes Res Clin Pract 2021, 180 :109042. Srinivasan S, Singh P, Kulothungan V, Sharma T, Raman R: Relationship between triglyceride glucose index, retinopathy and nephropathy in Type 2 diabetes . Endocrinology, Diabetes & Metabolism 2021, 4 (1):e00151. Yao L, Wang X, Zhong Y, Wang Y, Wu J, Geng J, Zhou Y, Chen J, Guan P, Xu Y et al : The Triglyceride-Glucose Index is Associated with Diabetic Retinopathy in Chinese Patients with Type 2 Diabetes: A Hospital-Based, Nested, Case-Control Study . Diabetes, Metabolic Syndrome and Obesity : Targets and Therapy 2021, 14 :1547-1555. Mulat Tebeje T, Kindie Yenit M, Gedlu Nigatu S, Bizuneh Mengistu S, Kidie Tesfie T, Byadgie Gelaw N, Moges Chekol Y: Prediction of diabetic retinopathy among type 2 diabetic patients in University of Gondar Comprehensive Specialized Hospital, 2006-2021: A prognostic model . Int J Med Inform 2024, 190 :105536. Jin S, Zhang X, Liu H, Hao J, Cao K, Lin C, Yusufu M, Hu N, Hu A, Wang N: Identification of the Optimal Model for the Prediction of Diabetic Retinopathy in Chinese Rural Population: Handan Eye Study . Journal of Diabetes Research 2022, 2022 :4282953. Yang J, Jiang S: Development and validation of a model that predicts the risk of diabetic retinopathy in type 2 diabetes mellitus patients . Acta Diabetol 2023, 60 (1):43-51. Li H-Y, Dong L, Zhou W-D, Wu H-T, Zhang R-H, Li Y-T, Yu C-Y, Wei W-B: Development and validation of medical record-based logistic regression and machine learning models to diagnose diabetic retinopathy . Graefe's Archive for Clinical and Experimental Ophthalmology 2023, 261 (3):681-689. Collins GS, Reitsma JB, Altman DG, Moons KGM: Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement . BMJ 2015, 350 :g7594. Steyerberg EW, Pencina MJ, Lingsma HF, Kattan MW, Vickers AJ, Van Calster B: Assessing the incremental value of diagnostic and prognostic markers: a review and illustration . European Journal of Clinical Investigation 2012, 42 (2):216-228. 2. Diagnosis and Classification of Diabetes: Standards of Care in Diabetes-2024 . Diabetes Care 2024, 47 (Suppl 1):S20-S42. Wong TY, Cheung CMG, Larsen M, Sharma S, Simó R: Diabetic retinopathy . Nat Rev Dis Primers 2016, 2 :16012. Flaxel CJ, Adelman RA, Bailey ST, Fawzi A, Lim JI, Vemulakonda GA, Ying G-s: Diabetic Retinopathy Preferred Practice Pattern® . Ophthalmology 2020, 127 (1):P66-P145. Jiang W, Li Z: Comparison of Machine Learning Algorithms and Nomogram Construction for Diabetic Retinopathy Prediction in Type 2 Diabetes Mellitus Patients . Ophthalmic Research 2024, 67 (1):537-548. Riley RD, Ensor J, Snell KIE, Harrell FE, Martin GP, Reitsma JB, Moons KGM, Collins G, van Smeden M: Calculating the sample size required for developing a clinical prediction model . BMJ 2020, 368 :m441. Lundberg SM, Lee S-I: A unified approach to interpreting model predictions . In: Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, California, USA: Curran Associates Inc.; 2017: 4768–4777. Li W-Y, Yang M, Song Y-N, Luo L, Nie C, Zhang M-N: An online diabetic retinopathy screening tool for patients with type 2 diabetes . Int J Ophthalmol 2021, 14 (11):1748-1755. Wang G-X, Hu X-Y, Zhao H-X, Li H-L, Chu S-F, Liu D-L: Development and validation of a diabetic retinopathy risk prediction model for middle-aged patients with type 2 diabetes mellitus . Front Endocrinol (Lausanne) 2023, 14 :1132036. Yang Y, Tan J, He Y, Huang H, Wang T, Gong J, Liu Y, Zhang Q, Xu X: Predictive model for diabetic retinopathy under limited medical resources: A multicenter diagnostic study . Front Endocrinol (Lausanne) 2022, 13 :1099302. Roşu CD, Bratu ML, Stoicescu ER, Iacob R, Hațegan OA, Ghenciu LA, Bolintineanu SL: Cardiovascular Risk Factors as Independent Predictors of Diabetic Retinopathy in Type II Diabetes Mellitus: The Development of a Predictive Model . Medicina (Kaunas, Lithuania) 2024, 60 (10). Tsao H-Y, Chan P-Y, Su EC-Y: Predicting diabetic retinopathy and identifying interpretable biomedical features using machine learning algorithms . BMC Bioinformatics 2018, 19 (Suppl 9):283. Chen X, Xie Q, Zhang X, Lv Q, Liu X, Rao H: Nomogram Prediction Model for Diabetic Retinopathy Development in Type 2 Diabetes Mellitus Patients: A Retrospective Cohort Study . Journal of Diabetes Research 2021, 2021 :3825155. Li W, Song Y, Chen K, Ying J, Zheng Z, Qiao S, Yang M, Zhang M, Zhang Y: Predictive model and risk analysis for diabetic retinopathy using machine learning: a retrospective cohort study in China . BMJ Open 2021, 11 (11):e050989. Luo H, Xiang C, Zeng L, Li S, Mei X, Xiong L, Liu Y, Wen C, Cui Y, Du L et al : SHAP based predictive modeling for 1 year all-cause readmission risk in elderly heart failure patients: feature selection and model interpretation . Scientific Reports 2024, 14 (1):17728. Giacco F, Brownlee M: Oxidative stress and diabetic complications . Circ Res 2010, 107 (9):1058-1070. Augustine J, Troendle EP, Barabas P, McAleese CA, Friedel T, Stitt AW, Curtis TM: The Role of Lipoxidation in the Pathogenesis of Diabetic Retinopathy . Front Endocrinol (Lausanne) 2020, 11 :621938. Zhou J, Zhu L, Li Y: Association between the triglyceride glucose index and diabetic retinopathy in type 2 diabetes: a meta-analysis . Front Endocrinol (Lausanne) 2023, 14 :1302127. Mirjalili SR, Soltani S, Heidari Meybodi Z, Marques-Vidal P, Kraemer A, Sarebanhassanabadi M: An innovative model for predicting coronary heart disease using triglyceride-glucose index: a machine learning-based cohort study . Cardiovascular Diabetology 2023, 22 (1):200. Zou H, Zhao F, Lv X, Ma X, Xie Y: Development and validation of a new nomogram to screen for MAFLD . Lipids In Health and Disease 2022, 21 (1):133. Yang H, Yuan L, Wu J, Li X, Long L, Teng Y, Feng W, Lyu L, Xu B, Ma T et al : [Construction of a Predictive Model for Diabetes Mellitus Type 2 in Middle-Aged and Elderly Populations Based on the Medical Checkup Data of National Basic Public Health Service] . Sichuan Da Xue Xue Bao Yi Xue Ban 2024, 55 (3):662-670. Chen X, Liu D, He W, Hu H, Wang W: Predictive performance of triglyceride glucose index (TyG index) to identify glucose status conversion: a 5-year longitudinal cohort study in Chinese pre-diabetes people . J Transl Med 2023, 21 (1):624. Sánchez-García A, Rodríguez-Gutiérrez R, Mancillas-Adame L, González-Nava V, Díaz González-Colmenero A, Solis RC, Álvarez-Villalobos NA, González-González JG: Diagnostic Accuracy of the Triglyceride and Glucose Index for Insulin Resistance: A Systematic Review . Int J Endocrinol 2020, 2020 :4678526. Ricard N, Bailly S, Guignabert C, Simons M: The quiescent endothelium: signalling pathways regulating organ-specific endothelial normalcy . Nat Rev Cardiol 2021, 18 (8):565-580. Semeraro F, Parrinello G, Cancarini A, Pasquini L, Zarra E, Cimino A, Cancarini G, Valentini U, Costagliola C: Predicting the risk of diabetic retinopathy in type 2 diabetic patients . Journal of Diabetes and its Complications 2011, 25 (5):292-297. Liu Y, Zhou Z, Wang Z, Yang H, Zhang F, Wang Q: Construction and clinical validation of a nomogram-based predictive model for diabetic retinopathy in type 2 diabetes . Am J Transl Res 2023, 15 (10):6083-6094. Li Y, Hu B, Lu L, Li Y, Caika S, Song Z, Sen G: Development and external validation of a predictive model for type 2 diabetic retinopathy . Scientific Reports 2024, 14 (1):16741. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5602589","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":391376596,"identity":"d3dbba18-9251-4730-87f4-fafce9334b60","order_by":0,"name":"XiaoQin Liu","email":"","orcid":"","institution":"Jilin University","correspondingAuthor":false,"prefix":"","firstName":"XiaoQin","middleName":"","lastName":"Liu","suffix":""},{"id":391376599,"identity":"6add55d3-e2f6-43f9-a1d0-4f981d5943eb","order_by":1,"name":"ShuYing Wu","email":"","orcid":"","institution":"Jilin University","correspondingAuthor":false,"prefix":"","firstName":"ShuYing","middleName":"","lastName":"Wu","suffix":""},{"id":391376600,"identity":"f8866de1-23ae-4e87-b510-fe71e7fa26b3","order_by":2,"name":"Yue Yang","email":"","orcid":"","institution":"The First Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Yue","middleName":"","lastName":"Yang","suffix":""},{"id":391376601,"identity":"be103c3d-9f5e-48cd-bfdc-b95f71e85452","order_by":3,"name":"Yang Li","email":"","orcid":"","institution":"The First Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"Yang","middleName":"","lastName":"Li","suffix":""},{"id":391376602,"identity":"8da9767d-7068-44b9-8f55-029dcd2ac177","order_by":4,"name":"XinTing Zhang","email":"","orcid":"","institution":"The First Hospital of Jilin University","correspondingAuthor":false,"prefix":"","firstName":"XinTing","middleName":"","lastName":"Zhang","suffix":""},{"id":391376603,"identity":"d011486d-6f53-4692-a72d-a9442dc8f4de","order_by":5,"name":"Ling Qin","email":"","orcid":"","institution":"Meihekou Central Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ling","middleName":"","lastName":"Qin","suffix":""},{"id":391376604,"identity":"aee17d3b-de7f-46dd-b616-189d433dc838","order_by":6,"name":"Fei Li","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAwElEQVRIiWNgGAWjYBACPmYGhg8MBjbMfCAeDzFa2JgZGGcwGKQxsxGvhQGkheEwAwla2BkYGz4UnGdnk0hgfPC2jUHenBiHNc4wuM0M1MJsOLeNwXBnA2Et7I95IFrYpHnbGBIMDhBhS/Mfg3MgLey/idfCYHAAbAsz0VoaewySmdl4HjZLzjknYbiBkBZ+/gOMDT/+2CXzsycf/PCmzEaeoC1ATR9AZDIwfhqAtARB9XBgR7zSUTAKRsEoGHEAAE0sL3E8QLkQAAAAAElFTkSuQmCC","orcid":"","institution":"The First Hospital of Jilin University","correspondingAuthor":true,"prefix":"","firstName":"Fei","middleName":"","lastName":"Li","suffix":""}],"badges":[],"createdAt":"2024-12-08 11:08:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5602589/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5602589/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":71874509,"identity":"0fc9b6bf-0c33-4c45-b35c-1011e42c77f7","added_by":"auto","created_at":"2024-12-19 10:50:12","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":44331,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of participant selection. DR: diabetic retinopathy; NDR: No DR.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-5602589/v1/cfbed98c9a37634735242d4d.png"},{"id":71874510,"identity":"e5fe2657-a97b-4f69-b0d5-ca514bd6882f","added_by":"auto","created_at":"2024-12-19 10:50:12","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":28301,"visible":true,"origin":"","legend":"\u003cp\u003eSelection results of RFECV Method (9 indexes were retained, 5 indexes were excluded)\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-5602589/v1/b30f8bf5177aca3f37c66de8.png"},{"id":71874512,"identity":"efffee46-3a44-4391-b775-78b1971b07d5","added_by":"auto","created_at":"2024-12-19 10:50:12","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":86124,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver Operating Characteristic (ROC) curves based of five machine learning models\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-5602589/v1/b63f7fed49b2b38e392abe89.png"},{"id":71874513,"identity":"24f795c0-3f7b-4ea5-8d10-cb8cbaf90764","added_by":"auto","created_at":"2024-12-19 10:50:12","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":80741,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP interpretation A: The importance ranking of the model prediction features. The horizontal coordinate represents the SHAP values. The larger SHAP values indicate that the variable is more important; B: Each point represents a feature value, and different colors represent the final influence of the feature on the model output results. Red and blue represent larger and smaller values, respectively.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-5602589/v1/1b8c74b4d5b46cb1da3d2135.png"},{"id":71874523,"identity":"89a4c34f-7069-4182-9ca2-4774514337c4","added_by":"auto","created_at":"2024-12-19 10:50:12","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":29725,"visible":true,"origin":"","legend":"\u003cp\u003eForce Plot of SHAP analysis method. NDR patient.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-5602589/v1/5d5c04f2cde8293b0839e130.png"},{"id":71874517,"identity":"7a28f960-9b0f-429d-bddd-828e4aebe373","added_by":"auto","created_at":"2024-12-19 10:50:12","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":198505,"visible":true,"origin":"","legend":"\u003cp\u003ePartial dependence plots of SHAP analysis method (TyG)\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-5602589/v1/1d54552e9d82e87af977b969.png"},{"id":82913885,"identity":"2022ba49-343a-492c-84f3-119bc0d71a94","added_by":"auto","created_at":"2025-05-16 15:46:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3582398,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5602589/v1/3e96f6ee-2eb1-49cc-ac3e-1eedcb1a184e.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development of machine learning Predictive Model for Type 2 Diabetic Retinopathy Using the Triglyceride-glucose index explained by SHAP method","fulltext":[{"header":"Research Insights","content":"\u003cp\u003e\u003cstrong\u003eWhat is currently known about this topic?\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTyG index is a simple tool for detecting IR. It has been shown to be closely related with macrovascular and microvascular complications in diabetic patients.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWhat is the key research question?\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWhether TyG can predict the occurrence of DR.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWhat is new?\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTyG can be used to predict diabetic retinopathy of T2DM. The TyG cut-off value for the prediction of DR is 4.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHow might this study influence clinical practice?\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIt can be used to screen for high-risk DR to improve early diagnosis and implement timely referral for patients, especially in primary hospitals and poor areas .\u003c/p\u003e"},{"header":"Introduction","content":"\u003cp\u003eDiabetes has a high incidence and mortality rate, thus emerging as a major global health challenge. The global prevalence of diabetes in people aged 20~79 years may increase to 12.2% by 2045[1]. Diabetic retinopathy (DR), the most common and serious ocular complication of type 2 diabetes, can result in visual impairment and even permanent vision loss[2]. To date, DR is the leading cause of blindness among individuals aged 20~74, contributing to new cases of preventable blindness in developing countries[3, 4]. A meta-analysis showed that the worldwide prevalence of DR is about 22.27%[5]. Notably, China has the largest number of diabetic patients in the world[1], with about 27.9% of diabetes cases being DR[6], of which 34% are in rural areas. Most DR patients are asymptomatic or mild in the early stage and are only detected when great damage has occurred or at the late stage[2]. Therefore, early screening of DR is crucial. However, the detection rate of DR remains suboptimal[7], particularly in certain primary hospitals and remote regions due to the constraints in medical conditions and limited social resources. Therefore, more accessible methods are needed to aid healthcare professionals in diagnosis and screening of DR.\u003c/p\u003e\n\u003cp\u003eTo date, factors such as diabetes duration, age, BMI, smoking, blood pressure, HbA1c levels, and cholesterol have been identified as risk factors for diabetic retinopathy (DR) [8, 9]. However, only a limited number of studies have incorporated insulin resistance (IR) into DR prediction models, despite substantial evidence[10, 11]demonstrating a strong association between IR and DR. The hyperinsulinemic-euglycemic clamp (HIEC) is the gold standard for detecting IR[12], However, it is expensive and complex, limiting its use. HOMA-IR[13] is the most commonly utilized method for estimating insulin resistance; however, it requires the measurement of fasting insulin levels, which imposes specific technical demands on laboratory capabilities. Additionally, this method is not suitable for patients undergoing insulin therapy and is not widely adopted in primary care settings or resource-limited regions. Therefore, HOMA-IR cannot be widely promoted in primary hospitals and poor areas[14, 15]. A new index has been recently identified for detecting IR:\u0026nbsp;Triglyceride-glucose index (TyG)[16], TyG is calculated by fasting triglyceride (TG) and fasting blood glucose (FBG), providing a simple, reliable, and cheap detection tool[17]. Besides, this index tool has shown better results than HOMA-IR[18], Elsewhere, Srinivasan[19]\u0026nbsp;and Yao[20]have shown that TyG and DR are closely related, while others not.\u003c/p\u003e\n\u003cp\u003eMachine Learning has shown great prospects in disease prediction. Some simple biochemical indicators have been used to predict the occurrence of DR, thus improving the screening rate of disease. Tebeje, Jin, Yang, Li et al.[21-24] have developed various prediction models; however, no comparative analysis has been conducted to determine which model demonstrates the best performance[8]. The newly discovered predictors can improve the predictive value for the prediction model[25, 26], TyG, as a new index for measuring IR, has shown great value in many studies. However, no prediction model uses TyG to predict the DR among the DM2 population.\u003c/p\u003e\n\u003cp\u003eIn this study, TyG was incorporated into DR predictor, using various machine learning methods to construct a prediction model. The best method was selected, providing some reference for clinical DR screening. Furthermore, clinical data were collected from primary clinics and advanced medical institutions to increase the generalization of the model. The model was explained using the SHAP method to avoid the limitation of the traditional model \u0026quot;black box\u0026quot;.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e2. Research design\u003c/p\u003e\n\u003cp\u003e2.1 Population\u003c/p\u003e\n\u003cp\u003eThis is a retrospective study, where patient data were extracted from the real-world database of The First Hospital of Jilin University (a general hospital in a major city) and Meihekou Central Hospital (a primary health care institution in the county) from January 1, 2010 to December 31, 2023.\u003c/p\u003e\n\n\u003cp\u003e2.2 Inclusion criteria\u003c/p\u003e\n\u003cp\u003ea. Patients diagnosed with type 2 diabetes(T2DM) following the criteria of the 2024 American Diabetes Association[27]; b. Patients aged⩾18 years; c. Patients with complete indicators (triglyceride and fasting glucose).\u003c/p\u003e\n\n\u003cp\u003e2.3 Exclusion criteria \u003c/p\u003e\n\u003cp\u003ea. Patients diagnosed with retinopathy at admission; b. Patients suffering from other retinal diseases, glaucoma, optic neuropathy, or eye diseases caused by systemic diseases; c. Patients with a history of eye surgery; d. Patients with severe systemic diseases (cancer, myocardial infarction, and dialysis history); e. Patients with data loss exceeding 20%.\u003c/p\u003e\n\n\u003cp\u003e2.4 Outcome\u003c/p\u003e\n\u003cp\u003eThe extraction variable was the first measurement record at first admission. Diagnosis information was extracted from the discharge diagnosis, and the follow-up was conducted until the first DR diagnosis, otherwise the last visit time was selected as the follow-up endpoint. Diagnostic criteria of DR included: A spectrum of retinal microvascular lesions on retinal examination with diabetic patients[28], including mild, moderate, and severe non-proliferative DR and proliferative DR. DR was diagnosed using 45\u0026deg; photos of macular center and indirect ophthalmoscopy when pupils were dilated. DR diagnosis was mainly achieved by endocrinologists and ophthalmologists. Patients diagnosed with retinopathy in other hospitals during the follow-up process were marked as DR, and the follow-up time was based on the earliest diagnosis time. Positive and negative samples were defined as DR and non-DR patients, respectively.\u003c/p\u003e\n\n\u003cp\u003e2.5 Ethical approval\u003c/p\u003e\n\u003cp\u003eThis study was conducted following the Helsinki Declaration and was approved by the Research Ethics Committee of The First Hospital of Jilin University (approval number: 2024-918). Each participant provided signed written informed consent. This study was reported based on TRIPOD[25].\u003c/p\u003e\n\n\u003cp\u003e2.6 Baseline data collection\u003c/p\u003e\n\u003cp\u003eData indicators, including sex, age, height, weight, smoking, drinking, course of T2DM, insulin therapy, hypertension history, and laboratory parameters, were mainly obtained from literature reports[2, 29, 30].\u003c/p\u003e\n\u003cp\u003eLaboratory parameters included glycated hemoglobin (HbA1c, g/dL), total cholesterol (TC, mmol/L), high-density lipoprotein (HDL, mmol/L), low-density lipoprotein (LDL, mmol/L), triglyceride (TG, mg/dL), fasting blood glucose (FPG, mg/dL), fasting C-Peptide (C-PE, ng/ml), fasting insulin (FINS,\u0026mu;U/ml) and C-reactive protein (CRP,\u0026mu;g/L).\u003c/p\u003e\n\u003cp\u003eBody mass index (BMI) was calculated formula as follows: weight (kg)/ height (m)\u003csup\u003e2\u003c/sup\u003e. TyG index: LN [ triglyceride (mg/dl)\u0026times;plasma glucose (mg/dl)/2]. Three graduate students collated and cross-checked the collected data. A unified training for the data collection and collation personnel was conducted to ensure the accuracy and consistency of the data.\u003c/p\u003e\n\n\u003cp\u003e2.7 Sample size calculation\u003c/p\u003e\n\u003cp\u003eThe sample size was mainly based on Riley[31]standard for accurate estimation. Four sample sizes (at least 878) (predicted value with small average error, intercept model only, ensuring shrinkage coefficient of 0.9, and ensuring small optimism of apparent model) were calculated. The total sample size was at least 966 (878+10% (878)), considering that some medical record information was incomplete and 10% contingency.\u003c/p\u003e\n\n\u003cp\u003e2.8 Model construction\u003c/p\u003e\n\u003cp\u003eThree single model algorithms (LR, DT, and SVM) and two integrated methods (RF and XGBoost) were used to train the model. LR is a machine learning method for solving binary classification problems, and is used to estimate the possibility of something. DT is a simple and easy-to-operate tree-type classification prediction model, providing intuitive and easy-to-understand results. However, DT can easily lead to overfitting during the classification process. SVM is widely used to construct a hyperplane concept to classify the observed values and can be used to deal with classification and regression problems. Compared with the single DT model, the integrated algorithms have higher accuracy but present more complicated and difficult results to explain.\u003c/p\u003e\n\n\u003cp\u003e2.9 Statistical analysis\u003c/p\u003e\n\u003cp\u003e2.9.1 Statistical interpretation：\u003c/p\u003e\n\u003cp\u003eR4.2.1 software was used for all data analysis. The variables missing more than 20%, including fasting C-Peptide, fasting insulin, and CRP were deleted to improve the utilization rate of the data. For individual missing values, the average interpolation method and mode interpolation method were used for counting data and measuring data, respectively. The normally distributed data were expressed as X\u0026plusmn;S, and compared using two independent samples T-test. The non-normally distributed data were expressed as P50 (P25, P75) and compared using Mann-Whitney U test. The counting data were expressed as frequency (%) and compared using X\u003csup\u003e2\u003c/sup\u003e test. The correlation between the predictive indicators and retinopathy was assessed using univariate and multivariate logistic regression models. Variables with univariate analysis \u003cem\u003eP\u003c/em\u003e \u0026lt; .05 were included in multivariate logistic regression analysis. The odds ratio (OR) and corresponding 95% confidence interval (CI) were used to indicate the trend of correlation.\u003c/p\u003e\n\n\u003cp\u003e2.9.2 Variable selection\u003c/p\u003e\n\u003cp\u003eThe model was trained and verified using Python3.9.15 software and tool kits Sklearn1.0.2, XGBoost1.7.4, and shap0.41.0. The recursive feature elimination with cross-validation (RFECV) was used to screen the optimal index. The least important features were continuously eliminated by training the model. The performance of the model was evaluated until the optimal performance index was reached. In this study, the variables with P \u0026lt; 0.1 in the comparison between groups were included in the RFECV model, and the best indicators were selected for subsequent prediction.\u003c/p\u003e\n\n\u003cp\u003e2.9.3 Model performance assessment \u003c/p\u003e\n\u003cp\u003eThe prediction efficiency of LR, DT, SVM, RF, and XGBoost5 models was evaluated based on accuracy, sensitivity, specificity, and Area Under the Curve (AUC) of receiver operating characteristic curve. Accuracy, precision, recall rate, and F1 score were used as the indicators. SHAP was used to visually explain the machine learning model. The Summary Plot, Shap Heat map, and importance ranking diagram were also drawn. The actual application of a single sample model was visualized via Force Plot. \u003cem\u003eP\u003c/em\u003e\u0026lt;.05 was considered a statistically significant difference.\u003c/p\u003e\n\n\u003cp\u003e2.9.4 Model interpretation\u003c/p\u003e\n\u003cp\u003eThe best final model was determined based on SHAP algorithm[32], which assigns corresponding attribute values (SHAP values) to each variable. These SHAP values quantitatively measure the influence of each feature on the prediction accuracy. A SHAP summary diagram was generated to visualize the contribution of each feature to the model.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e3.1 Social demographic and clinical characteristics\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;The distribution of study participants is shown in Figure 1. A total of 2014 positive cases were extracted from real-world data, and 5058 negative cases were extracted via the computer random sampling method. Finally, 622 cases were excluded from the positive group (including 402 with missing data, 108 with other diseases affecting vision, 108 with vision surgery, and 94 with critical illness) and 2700 cases were excluded from the negative group (2305 with missing data and 395 with other serious systemic diseases), leaving 3755 cases (1392 cases in the positive group and 2358 cases in the negative group) About 63.09% of the remaining cases were males and 36.91% were females. The subjects were aged 18~91 years, with an average age of 54.8\u0026plusmn;12.3 (Table 1).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 1 Comparison of baseline data between the two groups\u003c/p\u003e\n\u003cdiv align=\"center\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"117%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" rowspan=\"2\" style=\"width: 32px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVariables\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOverall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eControl\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCase\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width: 12px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eP\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e(\u003c/strong\u003e\u003cstrong\u003en=3750\u003c/strong\u003e\u003cstrong\u003e)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e(\u003c/strong\u003e\u003cstrong\u003en=2358\u003c/strong\u003e\u003cstrong\u003e)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e(\u003c/strong\u003e\u003cstrong\u003en=1392\u003c/strong\u003e\u003cstrong\u003e)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eSex\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e1384 (36.91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e800 (33.93)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e584 (41.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e2366 (63.09)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1558 (66.07)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e808 (58.05)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e54.8 \u0026plusmn; 12.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e53.3 \u0026plusmn; 12.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e57.2 \u0026plusmn; 11.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e25.9 \u0026plusmn; 3.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e26.0 \u0026plusmn; 3.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e25.7 \u0026plusmn; 3.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e.021\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eSmoking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e2798 (74.61)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1724 (73.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1074 (77.16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e.007\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e952 (25.39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e634 (26.89)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e318 (22.84)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eDrinking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e2912 (77.65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1803 (76.46)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1109 (79.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e.025\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e838 (22.35)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e555 (23.54)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e283 (20.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eHypertension\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e1922 (51.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1313 (55.68)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e609 (43.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e1828 (48.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1045 (44.32)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e783 (56.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eInsulin therapy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e1484 (39.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1248 (52.93)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e236 (16.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e2266 (60.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1110 (47.07)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1156 (83.05)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eCourse of diabetes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e10.0 (4.0, 15.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e6.0 (3.0, 10.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e12.0 (8.0, 20.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eHbA1c\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e8.7 (7.3, 10.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e8.2 (7.0, 9.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e9.3 (8.2, 10.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e4.9 (4.2, 5.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e4.9 (4.3, 5.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e4.8 (4.1, 5.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eHDL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e1.1 (1.1, 1.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1.1 (1.1, 1.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e1.1 (0.9, 1.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eLDL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e3.0 (2.4, 3.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e3.0 (2.5, 3.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e2.9 (2.3, 3.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eTG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e2.0 (1.3, 3.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e2.0 (1.3, 3.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e2.0 (1.4, 4.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e.072\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eFBG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e8.9 (6.9, 12.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e8.5 (6.7, 11.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e9.7 (7.3, 12.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22px;\"\u003e\n \u003cp\u003eTyG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20px;\"\u003e\n \u003cp\u003e2.2 (1.7, 2.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e2.2 (1.6, 2.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e2.3 (1.7, 3.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u0026lt;.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eThe variables with univariate P\u0026lt;0.05 were included in multivariate logistics regression analysis. The results showed that HDL (\u003cem\u003eP\u003c/em\u003e＜.001, OR=0.176, 95% CI:0.144~0.215), smoking (\u003cem\u003eP\u003c/em\u003e=.002, OR=0.759, 95% CI:0.620~0.92), LDL (\u003cem\u003eP\u003c/em\u003e=.021, OR=0.811, 95% CI:0.678~0.969), BMI (\u003cem\u003eP\u003c/em\u003e=.001, OR=0.815, 95% CI:0.726~0.915), TyG (\u003cem\u003eP\u003c/em\u003e=.038, OR=1.091, 95% CI:1.005, 1.185), age (\u003cem\u003eP\u003c/em\u003e=.014, OR=1.149, 95% CI:1.028~1.284), hypertension history (\u003cem\u003eP\u003c/em\u003e＜.001, OR=1.506, 95% CI:1.273~1.781), course of T2DM (\u003cem\u003eP\u003c/em\u003e＜.001, OR=1.798, 95% CI:1.615~2.002), insulin therapy (\u003cem\u003eP\u003c/em\u003e＜.001, OR=3.166, 95% CI:2.621~3.824), HbA1c (\u003cem\u003eP\u003c/em\u003e＜.001, OR=3.443, 95% CI:2.879~4.116), were the independent influencing factors of DR (Table 2).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 2 Results of multivariate Logistic regression analysis\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"99%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVariable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eB\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eWald\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eP\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cem\u003eOR\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003eHDL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e-1.736\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e291.731\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e＜.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e0.176(0.144, 0.215)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003eSmoking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e-0.275\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e7.136\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e.008\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e0.759(0.620, 0.929)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003eLDL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e-0.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e5.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e.021\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e0.811(0.678, 0.969)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e-0.205\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e12.035\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e0.815(0.726, 0.915)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003eTyG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e0.087\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e4.319\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e.038\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e1.091(1.005, 1.185)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e0.139\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e6.005\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e.014\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e1.149(1.028, 1.284)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003eHypertension\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e0.409\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e22.879\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e＜.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e1.506(1.273, 1.781)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003eCourse of diabetes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e0.587\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e115.088\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e＜.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e1.798(1.615, 2.002)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003eInsulin_therapy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e1.152\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e143.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e＜.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e3.166(2.621, 3.824)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 28px;\"\u003e\n \u003cp\u003eHbA1c\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e1.236\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e183.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e＜.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 30px;\"\u003e\n \u003cp\u003e3.443(2.879, 4.116)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e3.2 RFECV screening\u003c/p\u003e\n\u003cp\u003eThe independent variables were selected via RFECV method, and recursive features were eliminated. The results showed that nine indexes, including age, BMI, Diabetes course, Insulin therapy, Hypertension, HbA1c, TC, HDL, and TyG, were retained, and five indexes (sex, smoking, drinking, FBG, and LDL) were excluded (Figure 2).\u003c/p\u003e\n\u003cp\u003e3.3 Training and verification of the model\u003c/p\u003e\n\u003cp\u003eThe data were grouped into the training set and testing set (7:3). The last nine indexes were used as the optimal solution for model training. Moreover, SVM and LR showed the best performance in the test set and training set. The AUC curves of the five models are shown in Figure 3. The comparison of various prediction indexes (Table 3).\u003c/p\u003e\n\u003cp\u003eTable 3 Prediction performance indicators of five models in the training and testing sets\u003c/p\u003e\n\u003cdiv align=\"center\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"99%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width: 12px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" style=\"width: 44px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTrain set\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width: 2px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" style=\"width: 40px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTest set\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1-score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1-score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.80\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12px;\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e3.4 SHAP model analysis\u003c/p\u003e\n\u003cp\u003e3.4.1 Ranking of feature importance: The top 5 variables based on SHAP values in the ML model are shown in Figure 4. A summary plot is shown in Figure 4(B), where \u0026ldquo;red\u0026rdquo; and \u0026ldquo;blue\u0026rdquo; represent higher and lower eigenvalues, respectively. SHAP value\u0026lt;0 indicates negative influence, SHAP value\u0026gt;0 indicates positive influence. The dispersion of feature distribution (located above the Y axis) was directly related to the importance of the feature. Furthermore, TyG, Insulin therapy, HbA1c, and Diabetes course showed positive effects on retinopathy, while HDL showed negative effects.\u003c/p\u003e\n\u003cp\u003e3.4.2 Practical application of the model: The force Plot is shown in Figure 5. The blue and red arrows indicate that this factor reduces and increases the risk of retinopathy, respectively. The reference value represents the average SHAP value of all samples. F(x) represents the comprehensive SHAP value of each patient. The model can only predict the patient\u0026apos;s retinopathy if the value of f(x) is greater than the base value. The 1045th case (Figure 5) was randomly predicted using the test set. Notably, f(x) was less than the base value, which was accurately predicted as the control group.\u003c/p\u003e\n\u003cp\u003e3.4.3 The effect of key features on outcomes: A partial dependence diagram of the influence of the first three indicators on infection was drawn, showing the marginal effect relationship between important characteristics and outcome variables (how important influencing factors affect retinopathy). TyG\u0026gt; 4 showed significant impact (Figure 6).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eEarly screening and identification of people at high risk of developing DR is important for prevention and treatment. However, Li et al.[33]showed that only 17.48% of T2DM undergo routine DR Screening yearly. Doctors in the diabetes department, especially in primary hospitals that lack equipment and professionals, often overlook the early signs of retinopathy in their patients[29, 30]. Wang, Yang, Roşu et al.[34-36]built DR prediction models but did not include IR. Also, Qian[11]used HOMA-IR for prediction, which cannot be easily obtained in primary hospitals. In this study, a simple prediction model was developed using TyG.\u003c/p\u003e\n\u003cp\u003eResults showed that logistic regression and SVM had the best performance among the five models, with area under ROC curve of the testing set and training set of 0.85 and 0.82. respectively. Similarly, Tsao et al.[37]showed that the SVM model has good predictive performance. Jiang et al.[30]also showed that logistic regression has good predictive performance. LR is a linear model suitable for small sample analysis, and it is sensitive to outliers. SVM finds the optimal hyperplane by optimizing the objective function, which is suitable for the analysis of complex data [30].Different algorithms have their own advantages in the modeling process, and no algorithm can be applied to all models. Therefore, the corresponding model should be selected according to the research design. This study was conducted based on the computational minimum sample size. Therefore, a study with a larger sample size is needed to compare the performance between LR and SVM.\u003c/p\u003e\n\u003cp\u003eHerein, the variables included in LR and SVM were consistent. The remaining variables were described in previous studies except for the significant predictive ability of TyG[9, 22, 30, 38, 39].Previous studies showed that the interpretability of ML models is challenging[40]. In the present study, SHAP was used to improve model interpretability. SHAP is used to interpret a \u0026quot;black box\u0026quot; model that calculates a Shapley value for each feature in the prediction model to assess the importance of all feature combinations, reflecting their contribution to the predictive power of the overall model[32]. SHAP results showed that the top 5 variables included TyG, HDL, Insulin therapy, Diabetes course, and HbA1c.\u003c/p\u003e\n\u003cp\u003eIn this study, TyG was an important predictor of DR. Notably, increased TyG levels were associated with a high risk of DR due to the excessive production of mitochondrial superoxide in microvascular endothelial cells. This production is caused by pathway-specific insulin resistance, meanwhile, triggering intracellular hyperglycemia and vascular damage[41]. TyG is also a biochemical indicator reflecting the comprehensive effects of glucose and lipid metabolism, which are key causes of DR due to abnormal blood sugar and lipid metabolism[42]. Zhou et al. [43]showed similar results. TyG has never been used as a predictor of DR model before. Mirjalili[44], Zou[45], Yang[46]et al. showed that TyG, as a predictor, has a good predictive effect on coronary heart disease, fatty liver disease, and diabetes, similar to this study. Interestingly, TyG \u0026gt; 4 showed a significant predictive effect on DR. A study[16]reported that the TyG cut-off value for the diagnosis of insulin resistance is 4.65, consistent with this study. Another study[47]showed that the inflection point of TyG for controlling the transformation of diabetes is 8.88. A meta-analysis[48]showed that there is no standardized threshold value for TyG and clear threshold value related to TyG and DR. Therefore, more high-quality studies are needed to determine the optimal cut-off point for TyG in the future.\u003c/p\u003e\n\u003cp\u003eInsulin therapy is another key predictor of DR. Wang,\u0026nbsp;Li et al[34, 39]found\u0026nbsp;that the insulin treatment is associated with a high risk of DR development,\u0026nbsp;possibly because insulin therapy indicates worse islet function. While\u0026nbsp;Ricard\u0026nbsp;found it\u0026nbsp;possibly related to the rapid reduction of blood glucose[49]. Nonetheless, more basic research is needed to confirm the findings. This study, along with others, has demonstrated that the duration of diabetes mellitus is closely associated with the development of DR[21, 35, 38, 50],\u0026nbsp;possibly due to the prolonged exposure of blood vessels to risk factors.\u003c/p\u003e\n\u003cp\u003eHerein, DR was linked to increased HBA1c levels. This may be due to the continuous elevation of blood glucose, which can lead to dysfunction of the retinal vascular endothelium and cause retinal ischemia and increased vascular permeability[2]. Similar findings have been reported previously[11, 21, 39].\u003c/p\u003e\n\u003cp\u003eIn addition, HDL was identified as a key predictor of DR. Low HDL levels may indicate a higher risk of DR, consistent with Roşu, Liu, Li et al[36, 51, 52].\u003c/p\u003e\n\u003cp\u003eAlthough the model cannot replace the doctor\u0026apos;s diagnosis, this model can assist physicians in identifying high-risk patients with DR, helping in clinical decision-making and patient management.\u003c/p\u003e\n\u003cp\u003eTo the best of our knowledge, TyG has never been used to predict diabetic retinopathy in T2DM. In addition, data were collected from multiple medical centers (large urban hospital and regional clinics) in real-world environments, greatly increasing the credibility of the model. Furthermore, the model utilizes routine patient examination data, which are readily accessible and do not impose additional burdens on patients or medical insurance. Besides, the data can be easily popularized in primary hospitals, increasing the screening rate of DR.\u003c/p\u003e\n\u003cp\u003eLimitation: Only internal verification was conducted, thus effective external verification is needed.\u003c/p\u003e"},{"header":"Conclusion and recommendations","content":"\u003cp\u003eIn this study, a DR Prediction model was built using TyG and other easily available clinical data. LR and SVM models had the best performance. SHAP showed that the most important predictors of DR were TyG, insulin therapy, diabetes course, HbA1c, and HDL. These findings suggest that baseline characteristics of patients can be used to screen for high-risk DR to improve early diagnosis and implement timely referral for patients.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eTyG-index: Triglyceride-glucose index\u003c/p\u003e\n\u003cp\u003eDR: Diabetic retinopathy\u003c/p\u003e\n\u003cp\u003eRI: insulin resistance\u003c/p\u003e\n\u003cp\u003eHOMA-IR: Homeostatic Model Assessment for Insulin Resistance\u003c/p\u003e\n\u003cp\u003eTC: Total cholesterol\u003c/p\u003e\n\u003cp\u003eTG: Triglyceride\u003c/p\u003e\n\u003cp\u003eLDL: Low-density lipoprotein\u003c/p\u003e\n\u003cp\u003eHDL: High-density lipoprotein\u003c/p\u003e\n\u003cp\u003eFBG: Fasting blood glucose\u003c/p\u003e\n\u003cp\u003eBMI: Body mass index\u003c/p\u003e\n\u003cp\u003eLR: Logistic regression\u003c/p\u003e\n\u003cp\u003eDT: Decisiontree\u003c/p\u003e\n\u003cp\u003eRF: Randomforest\u003c/p\u003e\n\u003cp\u003eXCBoost: eXtremegradient boosting\u003c/p\u003e\n\u003cp\u003eSVM: Support vector machine\u003c/p\u003e\n\u003cp\u003eRFECV: Recursive Feature Elimination with Cross-Validation\u003c/p\u003e\n\u003cp\u003eAUC: Area under the curves\u003c/p\u003e\n\u003cp\u003eROC: Receiver operating characteristic curve\u003c/p\u003e\n\u003cp\u003eSHAP: Shapley additive explanation\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eAcknowledgement\u003c/p\u003e\n\u003cp\u003eWe express our gratitude to all the participants of the study, including patients and researchers. We thank Core Facility of the First Hospital of Jilin University for their help and valuable comments on this manuscript. \u003c/p\u003e\n\n\u003cp\u003eContributors\u003c/p\u003e\n\u003cp\u003eXqL and SyW searched the literature, designed the study, analysed the data, interpreted the results, and drafted the manuscript. YY, YL, and XtZ collected and analysed the data. FL conceived, designed, and supervised the study, interpreted the results, and revised the manuscript. LQ collected the data and drafted the manuscript. All authors contributed to the writing of the manuscript.\u003c/p\u003e\n\n\u003cp\u003eFunding\u003c/p\u003e\n\u003cp\u003eThe study was supported by the grants from Jilin Medical and Health Talents Special Project(JLSWSRCZX2023-29).\u003c/p\u003e\n\n\u003cp\u003eData Availability\u003c/p\u003e\n\u003cp\u003eThe data in this paper comes from The First Hospital of Jilin University Real-World Data Application Platform and Meihekou Central Hospital. The datasets used or analysed during the current study available from the corresponding author on reasonable request.\u003c/p\u003e\n\n\u003cp\u003eDeclarations\u003c/p\u003e\n\n\u003cp\u003eEthics approval and consent to participate\u003c/p\u003e\n\u003cp\u003eThe study was approved by was approved by the Research Ethics Committee of The First Hospital of Jilin University (approval number: 2024-918). Each participant provided signed written informed consent.\u003c/p\u003e\n\n\u003cp\u003eConsent for publication\u003c/p\u003e\n\u003cp\u003eAll the authors gave their consent to publication.\u003c/p\u003e\n\n\u003cp\u003eCompeting interests\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\n\u003cp\u003eConflicts of Interest\u003c/p\u003e\n\u003cp\u003eAll authors have declared no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSun H, Saeedi P, Karuranga S, Pinkepank M, Ogurtsova K, Duncan BB, Stein C, Basit A, Chan JCN, Mbanya JC\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eIDF Diabetes Atlas: Global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045\u003c/strong\u003e. \u003cem\u003eDiabetes Res Clin Pract \u003c/em\u003e2022, \u003cstrong\u003e183\u003c/strong\u003e:109119.\u003c/li\u003e\n\u003cli\u003eCheung N, Mitchell P, Wong TY: \u003cstrong\u003eDiabetic retinopathy\u003c/strong\u003e. \u003cem\u003eThe Lancet \u003c/em\u003e2010, \u003cstrong\u003e376\u003c/strong\u003e(9735):124-136.\u003c/li\u003e\n\u003cli\u003eCongdon NG, Friedman DS, Lietman T: \u003cstrong\u003eImportant causes of visual impairment in the world today\u003c/strong\u003e. \u003cem\u003eJAMA-J Am Med Assoc \u003c/em\u003e2003, \u003cstrong\u003e290\u003c/strong\u003e(15):2057-2060.\u003c/li\u003e\n\u003cli\u003eManiadakis N, Konstantakopoulou E: \u003cstrong\u003eCost Effectiveness of Treatments for Diabetic Retinopathy: A Systematic Literature Review\u003c/strong\u003e. \u003cem\u003ePharmacoeconomics \u003c/em\u003e2019, \u003cstrong\u003e37\u003c/strong\u003e(8).\u003c/li\u003e\n\u003cli\u003eTeo ZL, Tham Y-C, Yu M, Chee ML, Rim TH, Cheung N, Bikbov MM, Wang YX, Tang Y, Lu Y\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eGlobal Prevalence of Diabetic Retinopathy and Projection of Burden through 2045: Systematic Review and Meta-analysis\u003c/strong\u003e. \u003cem\u003eOphthalmology \u003c/em\u003e2021, \u003cstrong\u003e128\u003c/strong\u003e(11):1580-1591.\u003c/li\u003e\n\u003cli\u003eZhang G, Chen H, Chen W, Zhang M: \u003cstrong\u003ePrevalence and risk factors for diabetic retinopathy in China: a multi-hospital-based cross-sectional study\u003c/strong\u003e. \u003cem\u003eBr J Ophthalmol \u003c/em\u003e2017, \u003cstrong\u003e101\u003c/strong\u003e(12):1591-1595.\u003c/li\u003e\n\u003cli\u003eVujosevic S, Aldington SJ, Silva P, Hern\u0026aacute;ndez C, Scanlon P, Peto T, Sim\u0026oacute; R: \u003cstrong\u003eScreening for diabetic retinopathy: new perspectives and challenges\u003c/strong\u003e. \u003cem\u003eLancet Diabetes Endocrinol \u003c/em\u003e2020, \u003cstrong\u003e8\u003c/strong\u003e(4):337-347.\u003c/li\u003e\n\u003cli\u003eUsman TM, Saheed YK, Nsang A, Ajibesin A, Rakshit S: \u003cstrong\u003eA systematic literature review of machine learning based risk prediction models for diabetic retinopathy progression\u003c/strong\u003e. \u003cem\u003eArtif Intell Med \u003c/em\u003e2023, \u003cstrong\u003e143\u003c/strong\u003e:102617.\u003c/li\u003e\n\u003cli\u003eGuo H, Han F, Qu J-R, Pan C-Q, Sun B, Chen L-M: \u003cstrong\u003eScoring and validation of a simple model for predicting diabetic retinopathy in patients with type 2 diabetes based on a meta-analysis approach of 21 cohorts\u003c/strong\u003e. \u003cem\u003eAnnals of Medicine \u003c/em\u003e2024, \u003cstrong\u003e56\u003c/strong\u003e(1):2413920.\u003c/li\u003e\n\u003cli\u003eAmoaku WM, Ghanchi F, Bailey C, Banerjee S, Banerjee S, Downey L, Gale R, Hamilton R, Khunti K, Posner E\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eDiabetic retinopathy and diabetic macular oedema pathways and management: UK Consensus Working Group\u003c/strong\u003e. \u003cem\u003eEye (Lond) \u003c/em\u003e2020, \u003cstrong\u003e34\u003c/strong\u003e(Suppl 1).\u003c/li\u003e\n\u003cli\u003eWang Q, Zeng N, Tang H, Yang X, Yao Q, Zhang L, Zhang H, Zhang Y, Nie X, Liao X\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eDiabetic retinopathy risk prediction in patients with type 2 diabetes mellitus using a nomogram model\u003c/strong\u003e. \u003cem\u003eFront Endocrinol (Lausanne) \u003c/em\u003e2022, \u003cstrong\u003e13\u003c/strong\u003e:993423.\u003c/li\u003e\n\u003cli\u003eGastaldelli A: \u003cstrong\u003eMeasuring and estimating insulin resistance in clinical and research settings\u003c/strong\u003e. \u003cem\u003eObesity (Silver Spring, Md) \u003c/em\u003e2022, \u003cstrong\u003e30\u003c/strong\u003e(8):1549-1563.\u003c/li\u003e\n\u003cli\u003eMatthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC: \u003cstrong\u003eHomeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man\u003c/strong\u003e. \u003cem\u003eDiabetologia \u003c/em\u003e1985, \u003cstrong\u003e28\u003c/strong\u003e(7):412-419.\u003c/li\u003e\n\u003cli\u003eZhao Q, Zhang T-Y, Cheng Y-J, Ma Y, Xu Y-K, Yang J-Q, Zhou Y-J: \u003cstrong\u003eImpacts of triglyceride-glucose index on prognosis of patients with type 2 diabetes mellitus and non-ST-segment elevation acute coronary syndrome: results from an observational cohort study in China\u003c/strong\u003e. \u003cem\u003eCardiovascular Diabetology \u003c/em\u003e2020, \u003cstrong\u003e19\u003c/strong\u003e(1):108.\u003c/li\u003e\n\u003cli\u003eWang A, Tian X, Zuo Y, Chen S, Meng X, Wu S, Wang Y: \u003cstrong\u003eChange in triglyceride-glucose index predicts the risk of cardiovascular disease in the general population: a prospective cohort study\u003c/strong\u003e. \u003cem\u003eCardiovascular Diabetology \u003c/em\u003e2021, \u003cstrong\u003e20\u003c/strong\u003e(1):113.\u003c/li\u003e\n\u003cli\u003eSimental-Mend\u0026iacute;a LE, Rodr\u0026iacute;guez-Mor\u0026aacute;n M, Guerrero-Romero F: \u003cstrong\u003eThe product of fasting glucose and triglycerides as surrogate for identifying insulin resistance in apparently healthy subjects\u003c/strong\u003e. \u003cem\u003eMetabolic Syndrome and Related Disorders \u003c/em\u003e2008, \u003cstrong\u003e6\u003c/strong\u003e(4):299-304.\u003c/li\u003e\n\u003cli\u003eAbbasi F, Reaven GM: \u003cstrong\u003eComparison of two methods using plasma triglyceride concentration as a surrogate estimate of insulin action in nondiabetic subjects: triglycerides \u0026times; glucose versus triglyceride/high-density lipoprotein cholesterol\u003c/strong\u003e. \u003cem\u003eMetabolism: Clinical and Experimental \u003c/em\u003e2011, \u003cstrong\u003e60\u003c/strong\u003e(12):1673-1676.\u003c/li\u003e\n\u003cli\u003ePark HM, Lee HS, Lee Y-J, Lee J-H: \u003cstrong\u003eThe triglyceride-glucose index is a more powerful surrogate marker for predicting the prevalence and incidence of type 2 diabetes mellitus than the homeostatic model assessment of insulin resistance\u003c/strong\u003e. \u003cem\u003eDiabetes Res Clin Pract \u003c/em\u003e2021, \u003cstrong\u003e180\u003c/strong\u003e:109042.\u003c/li\u003e\n\u003cli\u003eSrinivasan S, Singh P, Kulothungan V, Sharma T, Raman R: \u003cstrong\u003eRelationship between triglyceride glucose index, retinopathy and nephropathy in Type 2 diabetes\u003c/strong\u003e. \u003cem\u003eEndocrinology, Diabetes \u0026amp; Metabolism \u003c/em\u003e2021, \u003cstrong\u003e4\u003c/strong\u003e(1):e00151.\u003c/li\u003e\n\u003cli\u003eYao L, Wang X, Zhong Y, Wang Y, Wu J, Geng J, Zhou Y, Chen J, Guan P, Xu Y\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eThe Triglyceride-Glucose Index is Associated with Diabetic Retinopathy in Chinese Patients with Type 2 Diabetes: A Hospital-Based, Nested, Case-Control Study\u003c/strong\u003e. \u003cem\u003eDiabetes, Metabolic Syndrome and Obesity : Targets and Therapy \u003c/em\u003e2021, \u003cstrong\u003e14\u003c/strong\u003e:1547-1555.\u003c/li\u003e\n\u003cli\u003eMulat Tebeje T, Kindie Yenit M, Gedlu Nigatu S, Bizuneh Mengistu S, Kidie Tesfie T, Byadgie Gelaw N, Moges Chekol Y: \u003cstrong\u003ePrediction of diabetic retinopathy among type 2 diabetic patients in University of Gondar Comprehensive Specialized Hospital, 2006-2021: A prognostic model\u003c/strong\u003e. \u003cem\u003eInt J Med Inform \u003c/em\u003e2024, \u003cstrong\u003e190\u003c/strong\u003e:105536.\u003c/li\u003e\n\u003cli\u003eJin S, Zhang X, Liu H, Hao J, Cao K, Lin C, Yusufu M, Hu N, Hu A, Wang N: \u003cstrong\u003eIdentification of the Optimal Model for the Prediction of Diabetic Retinopathy in Chinese Rural Population: Handan Eye Study\u003c/strong\u003e. \u003cem\u003eJournal of Diabetes Research \u003c/em\u003e2022, \u003cstrong\u003e2022\u003c/strong\u003e:4282953.\u003c/li\u003e\n\u003cli\u003eYang J, Jiang S: \u003cstrong\u003eDevelopment and validation of a model that predicts the risk of diabetic retinopathy in type 2 diabetes mellitus patients\u003c/strong\u003e. \u003cem\u003eActa Diabetol \u003c/em\u003e2023, \u003cstrong\u003e60\u003c/strong\u003e(1):43-51.\u003c/li\u003e\n\u003cli\u003eLi H-Y, Dong L, Zhou W-D, Wu H-T, Zhang R-H, Li Y-T, Yu C-Y, Wei W-B: \u003cstrong\u003eDevelopment and validation of medical record-based logistic regression and machine learning models to diagnose diabetic retinopathy\u003c/strong\u003e. \u003cem\u003eGraefe\u0026apos;s Archive for Clinical and Experimental Ophthalmology \u003c/em\u003e2023, \u003cstrong\u003e261\u003c/strong\u003e(3):681-689.\u003c/li\u003e\n\u003cli\u003eCollins GS, Reitsma JB, Altman DG, Moons KGM: \u003cstrong\u003eTransparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement\u003c/strong\u003e. \u003cem\u003eBMJ \u003c/em\u003e2015, \u003cstrong\u003e350\u003c/strong\u003e:g7594.\u003c/li\u003e\n\u003cli\u003eSteyerberg EW, Pencina MJ, Lingsma HF, Kattan MW, Vickers AJ, Van Calster B: \u003cstrong\u003eAssessing the incremental value of diagnostic and prognostic markers: a review and illustration\u003c/strong\u003e. \u003cem\u003eEuropean Journal of Clinical Investigation \u003c/em\u003e2012, \u003cstrong\u003e42\u003c/strong\u003e(2):216-228.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e2. Diagnosis and Classification of Diabetes: Standards of Care in Diabetes-2024\u003c/strong\u003e. \u003cem\u003eDiabetes Care \u003c/em\u003e2024, \u003cstrong\u003e47\u003c/strong\u003e(Suppl 1):S20-S42.\u003c/li\u003e\n\u003cli\u003eWong TY, Cheung CMG, Larsen M, Sharma S, Sim\u0026oacute; R: \u003cstrong\u003eDiabetic retinopathy\u003c/strong\u003e. \u003cem\u003eNat Rev Dis Primers \u003c/em\u003e2016, \u003cstrong\u003e2\u003c/strong\u003e:16012.\u003c/li\u003e\n\u003cli\u003eFlaxel CJ, Adelman RA, Bailey ST, Fawzi A, Lim JI, Vemulakonda GA, Ying G-s: \u003cstrong\u003eDiabetic Retinopathy Preferred Practice Pattern\u0026reg;\u003c/strong\u003e. \u003cem\u003eOphthalmology \u003c/em\u003e2020, \u003cstrong\u003e127\u003c/strong\u003e(1):P66-P145.\u003c/li\u003e\n\u003cli\u003eJiang W, Li Z: \u003cstrong\u003eComparison of Machine Learning Algorithms and Nomogram Construction for Diabetic Retinopathy Prediction in Type 2 Diabetes Mellitus Patients\u003c/strong\u003e. \u003cem\u003eOphthalmic Research \u003c/em\u003e2024, \u003cstrong\u003e67\u003c/strong\u003e(1):537-548.\u003c/li\u003e\n\u003cli\u003eRiley RD, Ensor J, Snell KIE, Harrell FE, Martin GP, Reitsma JB, Moons KGM, Collins G, van Smeden M: \u003cstrong\u003eCalculating the sample size required for developing a clinical prediction model\u003c/strong\u003e. \u003cem\u003eBMJ \u003c/em\u003e2020, \u003cstrong\u003e368\u003c/strong\u003e:m441.\u003c/li\u003e\n\u003cli\u003eLundberg SM, Lee S-I: \u003cstrong\u003eA unified approach to interpreting model predictions\u003c/strong\u003e. In: \u003cem\u003eProceedings of the 31st International Conference on Neural Information Processing Systems.\u003c/em\u003e Long Beach, California, USA: Curran Associates Inc.; 2017: 4768\u0026ndash;4777.\u003c/li\u003e\n\u003cli\u003eLi W-Y, Yang M, Song Y-N, Luo L, Nie C, Zhang M-N: \u003cstrong\u003eAn online diabetic retinopathy screening tool for patients with type 2 diabetes\u003c/strong\u003e. \u003cem\u003eInt J Ophthalmol \u003c/em\u003e2021, \u003cstrong\u003e14\u003c/strong\u003e(11):1748-1755.\u003c/li\u003e\n\u003cli\u003eWang G-X, Hu X-Y, Zhao H-X, Li H-L, Chu S-F, Liu D-L: \u003cstrong\u003eDevelopment and validation of a diabetic retinopathy risk prediction model for middle-aged patients with type 2 diabetes mellitus\u003c/strong\u003e. \u003cem\u003eFront Endocrinol (Lausanne) \u003c/em\u003e2023, \u003cstrong\u003e14\u003c/strong\u003e:1132036.\u003c/li\u003e\n\u003cli\u003eYang Y, Tan J, He Y, Huang H, Wang T, Gong J, Liu Y, Zhang Q, Xu X: \u003cstrong\u003ePredictive model for diabetic retinopathy under limited medical resources: A multicenter diagnostic study\u003c/strong\u003e. \u003cem\u003eFront Endocrinol (Lausanne) \u003c/em\u003e2022, \u003cstrong\u003e13\u003c/strong\u003e:1099302.\u003c/li\u003e\n\u003cli\u003eRoşu CD, Bratu ML, Stoicescu ER, Iacob R, Hațegan OA, Ghenciu LA, Bolintineanu SL: \u003cstrong\u003eCardiovascular Risk Factors as Independent Predictors of Diabetic Retinopathy in Type II Diabetes Mellitus: The Development of a Predictive Model\u003c/strong\u003e. \u003cem\u003eMedicina (Kaunas, Lithuania) \u003c/em\u003e2024, \u003cstrong\u003e60\u003c/strong\u003e(10).\u003c/li\u003e\n\u003cli\u003eTsao H-Y, Chan P-Y, Su EC-Y: \u003cstrong\u003ePredicting diabetic retinopathy and identifying interpretable biomedical features using machine learning algorithms\u003c/strong\u003e. \u003cem\u003eBMC Bioinformatics \u003c/em\u003e2018, \u003cstrong\u003e19\u003c/strong\u003e(Suppl 9):283.\u003c/li\u003e\n\u003cli\u003eChen X, Xie Q, Zhang X, Lv Q, Liu X, Rao H: \u003cstrong\u003eNomogram Prediction Model for Diabetic Retinopathy Development in Type 2 Diabetes Mellitus Patients: A Retrospective Cohort Study\u003c/strong\u003e. \u003cem\u003eJournal of Diabetes Research \u003c/em\u003e2021, \u003cstrong\u003e2021\u003c/strong\u003e:3825155.\u003c/li\u003e\n\u003cli\u003eLi W, Song Y, Chen K, Ying J, Zheng Z, Qiao S, Yang M, Zhang M, Zhang Y: \u003cstrong\u003ePredictive model and risk analysis for diabetic retinopathy using machine learning: a retrospective cohort study in China\u003c/strong\u003e. \u003cem\u003eBMJ Open \u003c/em\u003e2021, \u003cstrong\u003e11\u003c/strong\u003e(11):e050989.\u003c/li\u003e\n\u003cli\u003eLuo H, Xiang C, Zeng L, Li S, Mei X, Xiong L, Liu Y, Wen C, Cui Y, Du L\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eSHAP based predictive modeling for 1 year all-cause readmission risk in elderly heart failure patients: feature selection and model interpretation\u003c/strong\u003e. \u003cem\u003eScientific Reports \u003c/em\u003e2024, \u003cstrong\u003e14\u003c/strong\u003e(1):17728.\u003c/li\u003e\n\u003cli\u003eGiacco F, Brownlee M: \u003cstrong\u003eOxidative stress and diabetic complications\u003c/strong\u003e. \u003cem\u003eCirc Res \u003c/em\u003e2010, \u003cstrong\u003e107\u003c/strong\u003e(9):1058-1070.\u003c/li\u003e\n\u003cli\u003eAugustine J, Troendle EP, Barabas P, McAleese CA, Friedel T, Stitt AW, Curtis TM: \u003cstrong\u003eThe Role of Lipoxidation in the Pathogenesis of Diabetic Retinopathy\u003c/strong\u003e. \u003cem\u003eFront Endocrinol (Lausanne) \u003c/em\u003e2020, \u003cstrong\u003e11\u003c/strong\u003e:621938.\u003c/li\u003e\n\u003cli\u003eZhou J, Zhu L, Li Y: \u003cstrong\u003eAssociation between the triglyceride glucose index and diabetic retinopathy in type 2 diabetes: a meta-analysis\u003c/strong\u003e. \u003cem\u003eFront Endocrinol (Lausanne) \u003c/em\u003e2023, \u003cstrong\u003e14\u003c/strong\u003e:1302127.\u003c/li\u003e\n\u003cli\u003eMirjalili SR, Soltani S, Heidari Meybodi Z, Marques-Vidal P, Kraemer A, Sarebanhassanabadi M: \u003cstrong\u003eAn innovative model for predicting coronary heart disease using triglyceride-glucose index: a machine learning-based cohort study\u003c/strong\u003e. \u003cem\u003eCardiovascular Diabetology \u003c/em\u003e2023, \u003cstrong\u003e22\u003c/strong\u003e(1):200.\u003c/li\u003e\n\u003cli\u003eZou H, Zhao F, Lv X, Ma X, Xie Y: \u003cstrong\u003eDevelopment and validation of a new nomogram to screen for MAFLD\u003c/strong\u003e. \u003cem\u003eLipids In Health and Disease \u003c/em\u003e2022, \u003cstrong\u003e21\u003c/strong\u003e(1):133.\u003c/li\u003e\n\u003cli\u003eYang H, Yuan L, Wu J, Li X, Long L, Teng Y, Feng W, Lyu L, Xu B, Ma T\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003e[Construction of a Predictive Model for Diabetes Mellitus Type 2 in Middle-Aged and Elderly Populations Based on the Medical Checkup Data of National Basic Public Health Service]\u003c/strong\u003e. \u003cem\u003eSichuan Da Xue Xue Bao Yi Xue Ban \u003c/em\u003e2024, \u003cstrong\u003e55\u003c/strong\u003e(3):662-670.\u003c/li\u003e\n\u003cli\u003eChen X, Liu D, He W, Hu H, Wang W: \u003cstrong\u003ePredictive performance of triglyceride glucose index (TyG index) to identify glucose status conversion: a 5-year longitudinal cohort study in Chinese pre-diabetes people\u003c/strong\u003e. \u003cem\u003eJ Transl Med \u003c/em\u003e2023, \u003cstrong\u003e21\u003c/strong\u003e(1):624.\u003c/li\u003e\n\u003cli\u003eS\u0026aacute;nchez-Garc\u0026iacute;a A, Rodr\u0026iacute;guez-Guti\u0026eacute;rrez R, Mancillas-Adame L, Gonz\u0026aacute;lez-Nava V, D\u0026iacute;az Gonz\u0026aacute;lez-Colmenero A, Solis RC, \u0026Aacute;lvarez-Villalobos NA, Gonz\u0026aacute;lez-Gonz\u0026aacute;lez JG: \u003cstrong\u003eDiagnostic Accuracy of the Triglyceride and Glucose Index for Insulin Resistance: A Systematic Review\u003c/strong\u003e. \u003cem\u003eInt J Endocrinol \u003c/em\u003e2020, \u003cstrong\u003e2020\u003c/strong\u003e:4678526.\u003c/li\u003e\n\u003cli\u003eRicard N, Bailly S, Guignabert C, Simons M: \u003cstrong\u003eThe quiescent endothelium: signalling pathways regulating organ-specific endothelial normalcy\u003c/strong\u003e. \u003cem\u003eNat Rev Cardiol \u003c/em\u003e2021, \u003cstrong\u003e18\u003c/strong\u003e(8):565-580.\u003c/li\u003e\n\u003cli\u003eSemeraro F, Parrinello G, Cancarini A, Pasquini L, Zarra E, Cimino A, Cancarini G, Valentini U, Costagliola C: \u003cstrong\u003ePredicting the risk of diabetic retinopathy in type 2 diabetic patients\u003c/strong\u003e. \u003cem\u003eJournal of Diabetes and its Complications \u003c/em\u003e2011, \u003cstrong\u003e25\u003c/strong\u003e(5):292-297.\u003c/li\u003e\n\u003cli\u003eLiu Y, Zhou Z, Wang Z, Yang H, Zhang F, Wang Q: \u003cstrong\u003eConstruction and clinical validation of a nomogram-based predictive model for diabetic retinopathy in type 2 diabetes\u003c/strong\u003e. \u003cem\u003eAm J Transl Res \u003c/em\u003e2023, \u003cstrong\u003e15\u003c/strong\u003e(10):6083-6094.\u003c/li\u003e\n\u003cli\u003eLi Y, Hu B, Lu L, Li Y, Caika S, Song Z, Sen G: \u003cstrong\u003eDevelopment and external validation of a predictive model for type 2 diabetic retinopathy\u003c/strong\u003e. \u003cem\u003eScientific Reports \u003c/em\u003e2024, \u003cstrong\u003e14\u003c/strong\u003e(1):16741.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"TyG-index, Diabetic retinopathy, Machine learning, Predictive model, SHAP","lastPublishedDoi":"10.21203/rs.3.rs-5602589/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5602589/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eIntroduction\u003c/strong\u003e: This study aimed to develop a diabetic retinopathy (DR) Prediction model using various machine learning algorithms incorporating the novel predictor Triglyceride-glucose index (TyG). Furthermore, the model was interpreted using the SHapley Additive exPlanations (SHAP) method.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethod\u003c/strong\u003e: Real-world data were collected from a general hospital in a major city and a county clinic, then divided into the DR Group (1392) and non-DR group (2358). Baseline data were collected, and variables were selected using Recursive Feature Elimination with Cross-Validation (RFECV). The performance of five machine learning algorithms, including Logistic Regression model (LR), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and XGBoost (XGB), was assessed based on accuracy, sensitivity, specificity, and Area Under the Curve (AUC) of the Receiver Operating characteristic Curve (ROC). The optimal model was interpreted using SHAP.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResult\u003c/strong\u003e：SVM and LR demonstrated superior performance in both the test set and training set (ROC, 0.85 and 0.82, respectively). The top five predictors identified by SHAP analysis included TyG, Insulin therapy, HbA1c, Diabetes Course, HDL. HDL was identified as a protective factor, while the remaining factors were associated with retinopathy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion\u003c/strong\u003e：LR and SVM demonstrated the best performance. This is the first study constructing a DR Prediction model using TyG index. Notably, TyG significantly predicted DR and may serve as a crucial indicator for guiding clinical screening of high DR Risk.\u003c/p\u003e","manuscriptTitle":"Development of machine learning Predictive Model for Type 2 Diabetic Retinopathy Using the Triglyceride-glucose index explained by SHAP method","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-19 10:50:07","doi":"10.21203/rs.3.rs-5602589/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f51f1a95-8a7c-4729-9fc5-6204eeeec2ca","owner":[],"postedDate":"December 19th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":41713511,"name":"Health sciences/Endocrinology"},{"id":41713512,"name":"Health sciences/Endocrinology/Endocrine system and metabolic diseases"},{"id":41713513,"name":"Health sciences/Endocrinology/Endocrine system and metabolic diseases/Diabetes"},{"id":41713514,"name":"Health sciences/Endocrinology/Endocrine system and metabolic diseases/Dyslipidaemias"}],"tags":[],"updatedAt":"2025-05-16T15:38:37+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-19 10:50:07","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5602589","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5602589","identity":"rs-5602589","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}