A Simple Scoring Scale for Predicting the Risk of Diabetic Retinopathy over 50 Years Old in China

Purpose: To construct and evaluate a simple scoring scale for predicting the risk of diabetic retinopathy (DR). Methods: Based on the chronic disease management database of Yancheng City, Jiangsu Province, China, 1896 diagnosed patients over the age of 50 with diabetes were randomly selected and subjected to the self-designed epidemiological questionnaire survey and ocular clinical examination. Single-factor and multi-factor logistic regression analysis was used to screen the relevant inuencing factors of DR and then according to the reference value principle, the weights were assigned, and a simple scoring scale was constructed. A receiver operating characteristic curve (ROC) was developed and the accuracy and validity of the simple scale were evaluated. Results: The DR detection rate was 34.8 %. Multivariate analysis showed that family history of diabetes (odds ratio [OR]: 1.322, 95 % condence interval [CI]: 1.012-1.727), diabetes treatment method (OR: 2.074, 95 % CI: 1.696-2.537), diabetes duration (OR: 1.113, 95 % CI: 1.089-1.138), and hemoglobin (Hb)A1c (OR: 1.276, 95 % CI: 1.099-1.482) were independent DR risk factors. After excluding confounding factors and based on the β coecient of the multivariate logistic analysis, the scale had a maximum score of 12 points. The area under the receiver operating characteristic curve was 0.753, the cut-off value was 4, the sensitivity was 66.5 %, and the coincidence rate was 70.9 %. To improve sensitivity, the cut-off value was lowered to 3; the sensitivity was 79.7 %, and the coincidence rate was 65.0 %. Conclusion: The simple scale had good sensitivity and the ability to identify DR patients. 1.322 A association meta-analysis assessed in The showed that expression quantitative trait loci of genes were tested for DR associations. Another study detected an enrichment of the eQTLs from the glucose response genes among small association P-values and identied folliculin a susceptibility gene for retinopathy.


Introduction
Diabetes mellitus (DM) is one of the world's fastest-growing chronic diseases and a leading cause of acquired vision loss. 1 The World Health Organization estimates that the total number of people with diabetes will double, from 171 million in 2000 to 366 million by 2030. 2 Diabetic retinopathy (DR), the primary retinal vascular complication of DM, remains the leading cause of acquired vision loss worldwide in the working-age population. [3][4][5] With the increasing number of people with diabetes, the number of DR and vision-threatening DR cases, including severe non-proliferative DR, proliferative DR, and diabetic macular edema (DME), is estimated to rise to 191.0 million and 56.3 million globally, respectively, by 2030. 6 China, like many countries, has seen a marked increase in the prevalence of DM. As of 2016, the number of DM patients exceeded 140 million, and the proportion of people over age 50 exceeded 90 %. It is estimated that over 60 million people in China will have DM by the year 2030. [7][8][9] About one-third of patients with diabetes suffer from DR 7 ; therefore, the number of DR patients in China will surge to 18 million, seriously affecting the visual function of diabetic patients. 9 In the early course of the disease, DR is generally asymptomatic. If left untreated, DR can seriously impair vision, eventually progressing to blindness. 1,4 Apart from the devastating visual effects that could lead to Loading [MathJax]/jax/output/CommonHTML/jax.js reduced mobility, depression, and lower quality of life, DR is also associated with a higher risk of systemic vascular complications, imposing a noteworthy burden on individuals, households, communities, and societies. [10][11] Therefore, as a progressive disease, the early detection of DR and optimal DM management increases the chance of visual improvement and decreases the DME incidence rate. [12][13] Studies have shown that although more than 80 % of diabetic patients were aware of the effects of diabetes on their eyes, only a small percentage of patients had regular fundus examinations. [14][15][16] In China, patients with diabetes usually see a doctor in the endocrinology department of the hospital rst, but a fundus examination should be performed in the ophthalmology department. Patients' motivation to undergo retinal assessment is poor, hindering early diagnosis and management. Thus, designing a simple DR screening tool suitable for doctors in the clinic is essential to improve the early DR detection rate. A simple scale is practical in the primary public health environment because it is convenient, noninvasive, and low-cost. [17][18] This study combined several conventional clinical indicators for DR patients and assigned them weighted values to simplify the statistical equation model and construct a simple scoring scale for DR screening.
Consequently, the DR risk in diabetes patients can be quanti ed to guide clinical practice.

Participant Recruitment
In March 2019, 2067 patients over 50 years old with con rmed primary diabetes were randomly selected from the chronic disease management system of Yancheng City, Jiangsu Province. In total, 1965 people were chosen, but 69 people declined to participate or dropped out of the investigation or testing, and were excluded; 1896 people were included in the analysis. Patients under 50-years-old were not included because of their high mobility due to work-life factors and the inability to cooperate with epidemiological investigations and ophthalmic examinations.
The investigation followed the principles of research concerning human beings outlined in the Declaration of Helsinki and was approved by the Ethics Committee of the A liated Hospital of Nantong University (Approval No. 2019071). All participants signed written informed consent before the investigation began.

Sample Size
The sample size was calculated using the following formula:

Data Collection
One month before the formal investigation, an ophthalmologist from the A liated Hospital of Nantong University trained the Funing County Center for Disease Control eld workers on the purpose, procedure, and method of our investigation. A trained investigator interviewed each participant using self-designed questionnaires and collected information on demographic data (e.g., sex, age, and education level), living habits (e.g., smoking, drinking, and cell-phone use), diseases history (e.g., hypertension and hyperlipidemia), and diabetes-related data (e.g., family diabetes history, diabetes duration, treatment and control measures, and complications). There was also a physical examination (e.g., weight, height, blood pressure, blood glucose, and hemoglobinA1c [HbA1c]) and an ophthalmologic examination (e.g., the eyelid, conjunctiva, cornea, pupil, and lens) with a slit lamp microscope (YZ5T; Liu Liu, China For univariate analysis, the counting data were expressed as percentages (%), and a chi-square test was used to compare the two groups. Variables with statistical signi cance in the univariate analysis (P < 0.05) were included in the multivariate analysis. The logistic regression stepwise regression method (the forward likelihood ratio [LR] method) was used to adjust for potential confounding factors. The variables were gradually introduced using the partial maximum likelihood estimation to generate the nal regression model until all variables were included. Each variable in the model was statistically signi cant, and the statistical equation model for predicting DR was established. Then, the statistical equation model was simpli ed, and the simple scoring scale was constructed based on the weight of the nal inclusion factors in the multi-factor statistical equation model.

Constructing a Simple Scoring Scale for DR
The simple scoring scale was constructed based on the scoring system method created by Sullivan, Massaro and Agostino 20-21 which uses β coe cients in the nal multi-factor logistic regression equation as the scoring basis, gives different scores based on the weight of each risk factor in the nal regression equation, and takes the total weighted scores as the overall score. These steps simplify the complex mathematical-statistical model, making it clinically practical.

Scoring Scale Evaluation
The constructed scoring scale was comprehensively evaluated from two validity and reliability aspects. A receiver operating characteristic curve (ROC) was used for scale discrimination. All patients were scored according to the constructed scoring scale; the total score of each patient was used as a test variable, and the actual follow-up prognosis of patients was used as a state variable. A ROC curve were drawn to comprehensively evaluate the scoring scale. Combined with the area under the ROC curve (AUC), the sensitivity, speci city, and diagnostic cut-off values of the scoring scale based on the maximum Youden index were calculated according to the coordinates of each point on the ROC curve. The degree of calibration of the scoring scale was assessed using a consistency test (Kappa value).

Results
The average age was 68.28 ± 8.73 years, the gender ratio was 1: The eight statistically signi cant factors identi ed in the univariate analysis were taken as independent variables in the multivariate logistic regression analysis (the forward LR method), which determined that the family diabetes history, the treatment method, diabetes duration, and glycosylated hemoglobin should be included in the nal regression model. After excluding other confounding factors (e.g., sex, age, education, hypertension, hyperlipidemia, diabetic nephropathy, diabetic foot, and BMI), the above factors in the nal regression model were associated with and independent predictors of DR. Table 2 presents the partial regression coe cient (β) , odds ratio (OR), and P-value for each factor. The statistical equation model (Eq. 1) for predicting DR was: The complexity of the statistical equation models requires reliance on a calculator or computer, which limits the clinical applicability. Therefore, it was simpli ed to construct a simple scoring scale.
Based on the practical clinical signi cance, the diabetes duration was divided into four categories (< 5, 5-9, 10-14, and ≥ 15), and the HbA1c level was divided into three categories (< 7.0, 7.0-9.9, ≥ 10 %). Based Loading [MathJax]/jax/output/CommonHTML/jax.js on the scoring system and the regression Eq. 2, a simple scoring scale for DR was constructed and had a maximum score of 12 points. Table 3 presents the score assignments of various risk factors.
calculation notes: Classify the risk factors and determine the reference value of each category as: C(i=1, ... ,n;j=1, ... ,Ci; Ci is the total number of categories of risk factors). Based on the original data distribution, the lower limit of hemoglobinA1c (HbA1c; %) was 3, the upper limit was 15, the lower limit of the diabetes duration was 1, and the upper limit was 40.
W iREF : One category for each risk is a reference and assigned 0 points, recorded as: W iREF (i=1, ...n). The HbA1c reference was <7.0 % and the diabetes duration reference was <5 years. The simple scoring scale was comprehensively evaluated by the ROC curve (Fig. 1); the AUC was 0.753 (95 % con dence interval: 0.731-0.776). According to the coordinates of each point on the ROC curve, the diagnostic cut-off value at the maximum Youden Index was 3.5 points. According to the actual scoring, the cut-off value was 4 points. The sensitivity value was 0.665, the speci city value was 0.732, the positive predictive value was 0.570, the negative predictive value was 0.804, and the coincidence rate was 0.709 using a 4-point cut-off value (Table 4). For early detection and to improve the sensitivity, a 3-point cut-off value was evaluated. The sensitivity value was 0.797, speci city value was 0.571, positive predictive value was 0.498, negative predictive value was 0.840, and the coincidence rate was 0.650.

Discussion
In this study, we combined several conventional clinical indicators to create a simple scoring scale for calculating DR risk in diabetes patients. The simpli ed scoring scale consisted of only the diabetes family history, treatment method, and duration, and the HbA1c level had a maximum score of 12 points. The ROC curve demonstrated that the optimum cut-off for DR screening was 4 points, with 0.753 AUC, 0.665 sensitivity, and a 0.709 coincidence rate, indicating that the simple scale had a good ability to screen out diabetic patients with DR. However, as a microvascular complication of progressive diabetes, the goal of the simple scale is to identify as many DR patients as early as possible. Therefore, to improve the sensitivity, the cut-off value was set to 3 points, which improved the sensitivity to 79.7 %. Practically, clinicians can quickly score patients using the 3-or 4-point cut-off to predict the possibility of DR in diabetes patients. The simpli ed scale can not only solve the prognosis prediction inaccuracy by a single index but also avoid the complexity of the prognosis statistical equation model. This allows doctors to quickly diagnose and provide treatment suggestions and gives patients an early warning. However, implementing a new clinical prediction tool for screening DR needs time and clinical practice. For patients with complicated conditions, the scoring scale constructed in this study only complemented the clinical understanding and could not replace clinical evaluation and judgment.
The univariate and multivariate logistic analyses showed that the HbA1c level was an independent highrisk factor for DR and played an important role in the overall scoring scale. The 2020 edition of the Chinese Guidelines for the Prevention and Treatment method of diabetes stipulated that the Loading [MathJax]/jax/output/CommonHTML/jax.js individualized HbA1c level should be kept below 7 %. 22 Therefore, in this study, an HbA1c level of < 7 % was used as the reference standard, which was assigned as 0 points on the scoring scale; 7 to 9.9 % was considered generally under control and received 1 point, and ≥ 10 % was under poor control and received 3 points. The worse the glycemic control, the higher the HbA1c level and the greater the risk of DR in diabetes patients, which was consistent with the results of many studies in China and abroad. In two clinical trials, the United Kingdom Prospective Diabetes Study and the Diabetes Control and Complications Trial reported that the strict glycemic control (i.e., HbA1c, 7 %) decreased the DR incidence rate in type 1 and 2 DM. [23][24] Further, the risk of retinopathy was reduced by 30 to 40 % for every percent that the HbA1c level was lowered (e.g., from 8 to 7 %); the effect was called metabolic memory. [24][25] Recently, two cross-sectional studies from China also showed that the HbA1c level was an independent risk factor for DR, with ORs of 1.237 and 1.66, respectively. Thus, stringent glucose control is very important to reduce the occurrence and progression of DR. [26][27] The diabetes duration was also an important DR risk factor. A retrospective cohort analysis from an academic outpatient department of endocrinology and metabolic diseases in Germany analyzed 17 461 consultations of 4513 patients with DM between 1987 and 2014. This study demonstrated that the retinopathy prevalence increased depending on diabetes duration, starting at 1.1 % at diagnosis and ending at 63 % after 30 years or more. 28 Based on our study, the disease duration was divided into < 5, 5-9, 10-14, ≥ 15 years. According to the German results, the DR prevalence was lower if the patient's diabetes duration was less than ve years. Therefore, diabetes diagnosed for < 5 years was the reference group, and the DR risk score increased by 1 point for each additional ve years. The maximum score was 12 points, and the highest score awarded based on the diabetes duration was 5 points, indicating that the diabetes duration was crucial for DR occurrence.
DM family history was an independent predictor of DR and increased the risk by 1.322 times. A genomewide association meta-analysis assessed genes exhibiting the largest difference in glucose response for their association with DR. 29 The results showed that expression quantitative trait loci (eQTLs) of the glucose response genes were tested for DR associations. Another study detected an enrichment of the eQTLs from the glucose response genes among small association P-values and identi ed folliculin as a susceptibility gene for diabetic retinopathy. 30 Insulin, as a treatment for diabetes, also played an important role in the simple scale of this study. Although the mechanism of insulin affecting DR remains unclear, there are three possibilities. First, insulin-dependent diabetes mellitus is more prone to microvascular and nervous system complications. 31 Second, a rapid improvement of blood glucose in diabetes patients due to insulin use caused the early worsening of diabetic retinopathy. 32 Third, an acute abundance of (exogenously administered) insulin enhanced ischemia-induced vascular endothelial growth factor expression, worsening DR. 33 For rather than a substitute for clinical evaluation and judgment. The present study has several limitations. First, although 93 % of patients with diabetes in China were over 50 years old, only individuals over 50 were included in this study because of the eld investigation response rate. Thus, extrapolating the results should be done with caution. Second, the value of some predicted variables was a cross-sectional measurement, which observes a point value during the patient investigation, and the timing sequence of outcomes and exposure could not be determined.

Conclusions
In this study, diabetic patients were randomly selected from the local chronic disease management system to comprise the investigation population. Thus, a group case-control study was designed to screen for DR risk factors and construct a simpli ed predictive scale for identifying DR. The research subjects were well-represented, and the constructed scale had good effectiveness and accuracy.
Therefore, we recommend that general practitioners use our simpli ed predictive scale for the quick and non-invasive early detection of DR.

Declarations
Ethics approval and consent to participate The investigation followed the principles of research concerning human beings outlined in the Declaration of Helsinki and was approved by the Ethics Committee of the A liated Hospital of Nantong University (Approval No. 2019071). All participants signed written informed consent before the investigation began.

Consent for publication
Not applicable.

Availability of data and materials
The data, analytic methods, and study materials will not be made available to other researchers for purposes of reproducing the results or replicating the procedure.

Competing interests
No potential con icts of interest relevant to this article were reported.

Funding/Support
The research received funding through grants MS22019005 (Gu YJ) from Nantong science and  Caption not included with this version Loading [MathJax]/jax/output/CommonHTML/jax.js