Systems and methods for predicting the risk of diabetic retinopathy requiring surgical treatment

Abstract

This invention belongs to the field of plasma protein biomarkers and disease risk prediction technology. It discloses a system and method for predicting the risk of surgically treatable diabetic retinopathy (DR). The protein biomarker combination involved includes 13 protein biomarkers positively correlated with the risk of DR and 18 protein biomarkers negatively correlated. This invention focuses on risk prediction for DR as the core clinical outcome. By systematically introducing plasma proteomics information and combining it with traditional clinical risk factors, it achieves early prediction of the risk of this serious outcome. It can quantify the overall risk level of subjects developing DR at the proteomics level, effectively addressing the current lack of risk prediction models for DR as an outcome.

Description

Technical Field

[0001] This invention belongs to the field of plasma protein biomarkers and disease risk prediction technology, and more specifically, relates to a system and method for predicting the risk of diabetic retinopathy requiring surgical treatment. Background Technology

[0002] Diabetic retinopathy is one of the most common and serious microvascular complications of diabetes, affecting approximately one-third of diabetic patients and is a leading cause of vision impairment and blindness among the working-age population worldwide. Epidemiological studies show that the burden of blindness caused by diabetic retinopathy has been steadily increasing over the past few decades, and with the continuous increase in the number of people with diabetes, the public health burden caused by this disease is expected to remain at a high level for a long time to come.

[0003] From a clinical progression perspective, diabetic retinopathy typically undergoes a continuous evolution from early microvascular damage to severe ischemia and neovascularization. Clinically, mild and moderate stages are mainly managed through observation and follow-up. However, when the disease progresses to a severe stage (i.e., severe nonproliferative diabetic retinopathy and proliferative diabetic retinopathy; excluding mild and moderate nonproliferative diabetic retinopathy, such as...),... Figure 1 When diabetic retinopathy (as shown in the image) progresses to a stage where patients often require surgical treatment, including vitrectomy, retinal laser photocoagulation, and intravitreal injection of anti-vascular endothelial growth factor drugs. Such surgically induced diabetic retinopathy not only indicates that the disease has entered a stage threatening visual function but also significantly increases the economic burden on patients and the pressure on the healthcare system. Therefore, identifying high-risk individuals who may develop surgically induced diabetic retinopathy before the disease progresses clinically is crucial for precise prevention and control and the rational allocation of medical resources.

[0004] Plasma proteins, as important molecules actively secreted or passively released into the circulatory system by cells under physiological and pathological conditions, comprehensively reflect various biological processes such as inflammatory responses, angiogenesis, and extracellular matrix remodeling. They are an important source of disease biomarkers and potential intervention targets. Existing studies have found that various circulating proteins are closely related to the risk of diabetic retinopathy, demonstrating a certain risk discrimination ability. In current technologies, the identification of diabetic retinopathy-related biomarkers is mainly carried out through cross-sectional studies, case-control studies, or cohort studies. The types of biomarkers involved are mostly concentrated on traditional clinical indicators, such as glycemic control indicators, diabetes duration, and some inflammatory factors. The diabetic retinopathy risk prediction models built on this basis also mainly rely on the above-mentioned clinical risk factors. These models have problems such as insufficient sensitivity and specificity in the early prediction of diabetic retinopathy.

[0005] Furthermore, the aforementioned technical solutions primarily assess the risk of developing or progressing diabetic retinopathy. They struggle to effectively identify high-risk individuals who may progress to requiring invasive procedures such as vitrectomy, laser photocoagulation, or intravitreal injections, using "needing ophthalmic surgery or invasive treatment" as the core predictive endpoint. Especially in the early stages of the disease, existing predictive models exhibit significant deficiencies in sensitivity and specificity, failing to meet the actual clinical needs for preoperative risk warning and decision support.

[0006] Therefore, there is an urgent need for a risk prediction system centered on surgically treatable diabetic retinopathy as the core clinical outcome, enabling high-risk identification before surgical indications arise, and providing decision support for enhanced follow-up, early intervention, and individualized management. Furthermore, current technologies have not yet sufficiently identified plasma protein biomarkers that reflect early fundus lesions in diabetic patients and can be used to predict the risk of surgically treatable diabetic retinopathy. Summary of the Invention

[0007] Existing predictive models for overall diabetic retinopathy typically rely on clinical indicators such as glycated hemoglobin and diabetes duration as primary inputs, and cannot meet the early prediction needs of surgically treatable diabetic retinopathy phenotypes. The purpose of this invention is to provide a system and method for predicting the risk of surgically treatable diabetic retinopathy. This risk prediction method focuses on surgically treatable diabetic retinopathy as the core clinical outcome. By systematically introducing plasma proteomics information and combining it with traditional clinical risk factors, it achieves early prediction of the risk of this serious outcome. It can quantify the overall risk level of subjects developing surgically treatable diabetic retinopathy at the proteomics level, thereby providing decision support for intensive clinical follow-up, early intervention, and individualized management. This effectively addresses the current lack of risk prediction models for surgically treatable diabetic retinopathy as an outcome.

[0008] To achieve the above objectives, according to a first aspect of the present invention, the application of a protein biomarker combination in the preparation of a kit for predicting the risk of surgically treatable diabetic retinopathy is provided, characterized in that the protein biomarker combination simultaneously includes apolipoprotein A-IV, β-1,3-N-acetylglucosamine transferase 7, cocaine esterase, heart-type fatty acid-binding protein, N-acetylgalactosamine transferase 7, proglucagon, growth differentiation factor 15, LRP2-binding protein, neurofilament light chain protein, pancreatin 1α, secretory granulin, and Kazal-type serine protease inhibitor 4, all of which are positively correlated with the risk of surgically treatable diabetic retinopathy. α-2,3-sialyltransferase 1, and detegrin-like metalloproteinase 13 containing platelet-reactive protein motifs that are negatively correlated with the risk of developing diabetic retinopathy requiring surgical treatment, pancreatic α-amylase, α-amylase 2B, type I collagen α1 chain, carboxypeptidase B, CREG1 protein, dentin matrix acidic phosphoprotein 1, desmosome core glycoprotein 4, intercellular adhesion molecule 1, IRRE-like protein 2 homolog, extracellular phosphoglycoproteins of the matrix, B cell and B1 cell-specific proteins in the marginal zone, peptidyl glycine α-amidyl monooxygenase, peptidoglycan recognition protein 1, platelet-activating factor acetylhydrolase, ribokinase, osteopontin, and thiamine pyrophosphate kinase 1.

[0009] According to a second aspect of the present invention, the present invention provides a method for constructing a predictive model based on protein biomarkers for predicting the risk of developing diabetic retinopathy requiring surgical treatment, characterized by comprising the following steps: S1. Prepare a dataset for individuals with abnormal glucose metabolism but without diabetic retinopathy. The dataset includes the Z-values ​​of 31 specific protein biomarkers obtained by each individual with abnormal glucose metabolism, clinical variable data when obtaining the biological samples used to detect the 31 specific protein biomarkers, and whether surgically treated diabetic retinopathy has occurred before a preset subsequent time point. S2. In the dataset, using the Z-valued expression values ​​of 31 specific protein markers and clinical variable data as variables, a predictive model for predicting the risk of developing diabetic retinopathy requiring surgical treatment was obtained by fitting a Cox proportional hazards regression model. Among them, the 31 specific protein biomarkers include apolipoprotein A-IV, β-1,3-N-acetylglucosamine transferase 7, cocaine esterase, heart-type fatty acid-binding protein, N-acetylgalactosamine transferase 7, proglucagon, growth differentiation factor 15, LRP2-binding protein, neurofilament light chain protein, pancreatic stone protein 1α, secretory granulin, Kazal-type serine protease inhibitor 4, α-2,3-sialic acid transferase 1, and are positively correlated with the risk of diabetic retinopathy requiring surgical treatment. The following proteins show a negative correlation with the risk of developing the disease: platelet-reactive protein motif-de-integrin-like metalloproteinase 13, pancreatic α-amylase, α-amylase 2B, type I collagen α1 chain, carboxypeptidase B, CREG1 protein, dentin matrix acidic phosphoprotein 1, desmosome core glycoprotein 4, intercellular adhesion molecule 1, IRRE-like protein 2 homolog, extracellular phosphoglycoproteins in the matrix, B-cell and B1-cell specific proteins in the marginal zone, peptidyl glycine α-amidyl monooxygenase, peptidoglycan recognition protein 1, platelet-activating factor acetylhydrolase, ribokinase, osteopontin, and thiamine pyrophosphate kinase 1.

[0010] As a further preferred embodiment of the present invention, the clinical variables include age, sex, glycated hemoglobin, duration of diabetes, systolic blood pressure, body mass index, total cholesterol, high-density lipoprotein cholesterol, and smoking status.

[0011] As a further preferred embodiment of the present invention, in step S1, the Z-value is obtained by first providing the detection value of a specific protein marker in the plasma of each individual in the group of individuals with abnormal glucose metabolism without diabetic retinopathy in the form of a standardized protein expression value, and calculating the mean and standard deviation of the standardized protein expression value of the specific protein marker; then, for a certain individual, the standardized protein expression value of the specific protein marker in the individual's plasma is first subtracted from the mean, and then divided by the standard deviation to obtain the Z-value of the specific protein marker in the individual's plasma.

[0012] According to a third aspect of the present invention, the present invention provides a predictive model based on protein biomarkers for predicting the risk of developing diabetic retinopathy requiring surgical treatment, constructed by the above-described construction method.

[0013] According to a fourth aspect of the present invention, the present invention provides a method for using a predictive model based on protein biomarkers constructed by the above-described method for predicting the risk of developing diabetic retinopathy requiring surgical treatment. The method is characterized by using an individual with abnormal glucose metabolism but no diabetic retinopathy as the subject. First, the Z-values ​​of 31 specific protein biomarkers corresponding to these biomarkers in the subject's plasma and the clinical variable data of the subject at the time of obtaining the biological samples used to detect the 31 specific protein biomarkers are obtained. The Z-values ​​of the 31 specific protein biomarkers and the clinical variable data are simultaneously input into the predictive model. The score output by the predictive model represents the subject's future risk of developing diabetic retinopathy requiring surgical treatment; the higher the score, the greater the risk. Among them, the 31 specific protein biomarkers include apolipoprotein A-IV, β-1,3-N-acetylglucosamine transferase 7, cocaine esterase, heart-type fatty acid-binding protein, N-acetylgalactosamine transferase 7, proglucagon, growth differentiation factor 15, LRP2-binding protein, neurofilament light chain protein, pancreatic stone protein 1α, secretory granulin, Kazal-type serine protease inhibitor 4, α-2,3-sialic acid transferase 1, and are positively correlated with the risk of diabetic retinopathy requiring surgical treatment. The following proteins showed a negative correlation with the risk of developing the disease: platelet-reactive protein motif-de-integrin-like metalloproteinase 13, pancreatic α-amylase, α-amylase 2B, type I collagen α1 chain, carboxypeptidase B, CREG1 protein, dentin matrix acidic phosphoprotein 1, desmosome core glycoprotein 4, intercellular adhesion molecule 1, IRRE-like protein 2 homolog, extracellular phosphoglycoproteins of the matrix, marginal zone B cell and B1 cell-specific proteins, peptidyl glycine α-amidyl monooxygenase, peptidoglycan recognition protein 1, platelet-activating factor acetylhydrolase, ribokinase, osteopontin, and thiamine pyrophosphate kinase 1. The Z-values ​​of the 31 specific protein markers in the subject's plasma are obtained by taking into account the mean and standard deviation of the standardized protein expression values ​​of the 31 specific protein markers used in the corresponding construction method of the prediction model. For a specific protein marker, the Z-value of the specific protein marker in the subject's plasma is obtained by subtracting the mean from the standardized protein expression value of the specific protein marker in the subject's plasma and then dividing by the standard deviation.

[0014] As a further preferred embodiment of the present invention, the clinical variable data simultaneously include age, sex, glycated hemoglobin, duration of diabetes, systolic blood pressure, body mass index, total cholesterol, high-density lipoprotein cholesterol, and smoking status.

[0015] According to a fifth aspect of the invention, the present invention provides a predictive system for predicting the risk of developing diabetic retinopathy requiring surgical treatment, characterized in that it comprises: Plasma protein detection and clinical variable data acquisition module: used to obtain the Z-valued expression values ​​of 31 specific protein biomarkers contained in the subject's plasma, as well as the clinical variable data of the subject when obtaining the biological samples used to detect the 31 specific protein biomarkers; The predictive model based on protein biomarkers constructed using the above method for predicting the risk of developing diabetic retinopathy requiring surgical treatment is as follows: It takes the Z-valued expression values ​​of 31 specific protein biomarkers in the subject's plasma and the clinical variable data of the subject when obtaining the biological samples used to detect the 31 specific protein biomarkers as inputs, and the output score is the subject's future risk of developing diabetic retinopathy requiring surgical treatment. The higher the score, the greater the risk.

[0016] As a further preferred embodiment of the present invention, the clinical variable data simultaneously include age, sex, glycated hemoglobin, duration of diabetes, systolic blood pressure, body mass index, total cholesterol, high-density lipoprotein cholesterol, and smoking status.

[0017] Compared with the prior art, the above-described technical solutions conceived in this invention can achieve the following beneficial effects: (1) Plasma proteins are key molecules that maintain and regulate numerous biological activities, enabling a more comprehensive capture of disease progression. However, current technologies have not fully identified plasma protein biomarkers for surgically treatable diabetic retinopathy. The combination of protein biomarkers associated with the risk of developing surgically treatable diabetic retinopathy in this invention includes: i. The following proteins are positively correlated with diabetic retinopathy requiring surgical treatment: apolipoprotein A-IV, β-1,3-N-acetylglucosamine transferase 7, cocaine esterase, heart-type fatty acid-binding protein, N-acetylgalactosamine transferase 7, proglucagon, growth differentiation factor 15, LRP2-binding protein, neurofilament light chain protein, pancreatic stone protein 1α, secretory granulin, Kazal-type serine protease inhibitor 4, and α-2,3-sialyl transferase 1, totaling 13 proteins; ii. Proteins negatively correlated with surgically treatable diabetic retinopathy include: detegrin-like metalloproteinase 13 containing platelet-reactive protein motifs, pancreatic α-amylase, α-amylase 2B, type I collagen α1 chain, carboxypeptidase B, CREG1 protein, dentin matrix acidic phosphoprotein 1, desmosome core glycoprotein 4, intercellular adhesion molecule 1, IRRE-like protein 2 homolog, extracellular phosphoglycoproteins in the matrix, B-cell and B1-cell specific proteins in the marginal zone, peptidyl glycine α-amidyl monooxygenase, peptidoglycan recognition protein 1, platelet-activating factor acetylhydrolase, ribokinase, osteopontin, and thiamine pyrophosphate kinase 1, totaling 18 proteins.

[0018] (2) When diabetic retinopathy progresses to the stage requiring ophthalmic surgery or invasive treatment (e.g., vitrectomy, laser photocoagulation, or intravitreal injection), it is often accompanied by severe and irreversible visual impairment, placing a significant burden on patients' quality of life and the healthcare system. However, existing diabetic retinopathy risk prediction technologies typically rely on clinical indicators such as glycated hemoglobin and the duration of diabetes as the main inputs, and cannot meet the early prediction needs of surgically treatable phenotypes. It is difficult to effectively identify high-risk individuals who may progress to the stage requiring surgical treatment before clinical surgical indications appear.

[0019] This invention introduces 31 specific plasma protein biomarkers associated with the risk of developing diabetic retinopathy requiring surgical intervention. A predictive model is constructed using these biomarkers, and the process is comprehensive, rigorous, and adheres to strict standards. Compared to predictive models that only include traditional clinical risk factors, the predictive model built using these 31 specific protein biomarkers significantly improves predictive efficacy, with marked increases in the C-index, reclassification improvement index, and overall discrimination improvement index. This model can identify high-risk individuals before clinical surgical indications arise. Compared to predictive models based solely on traditional clinical risk factors, this invention's predictive system for the risk of developing diabetic retinopathy requiring surgical intervention, based on 31 specific protein biomarkers, demonstrates higher predictive efficacy in predicting diabetic retinopathy requiring surgical intervention.

[0020] Compared to traditional predictive models based on clinical indicators, the surgically treatable diabetic retinopathy predictive model constructed in this invention incorporates the 31 specific predictive protein biomarkers mentioned above, significantly improving the C-index of the predictive model. Furthermore, the reclassification improvement index and the comprehensive discriminant improvement index indicate that the new model combining "traditional risk factors + 31 specific protein biomarkers" performs better than the old model containing only traditional factors. Therefore, the predictive ability of the surgically treatable diabetic retinopathy predictive model constructed in this invention is significantly improved. Thus, this technology has high value for identifying high-risk groups, early disease prevention, and disease risk prediction.

[0021] (3) This invention uses surgically treatable diabetic retinopathy as the core clinical outcome indicator, directly focusing on the severe stage of disease progression requiring ophthalmic surgical intervention. By predicting the risk of such severe outcomes in advance, this invention can identify potentially high-risk individuals before clinical surgical indications appear, providing a basis for clinicians to implement intensive follow-up, early intervention, and individualized management, thereby helping to delay or avoid disease progression to the stage requiring invasive ophthalmic surgical treatment, reducing surgery-related risks and the resulting visual impairment and medical burden.

[0022] (4) Furthermore, the 31 specific protein biomarkers in this invention can be used to develop kits for predicting the risk of surgically treated diabetic retinopathy, indicating the pathogenesis of surgically treated diabetic retinopathy, and have the potential to serve as potential drug targets to promote the research and development of drugs related to surgically treated diabetic retinopathy. The predictive model constructed in this invention can be used to build an early warning system for surgically treated diabetic retinopathy, and has good application value.

[0023] In summary, this invention provides an effective predictive method for the severe progression stage of diabetic retinopathy (i.e., surgically treatable diabetic retinopathy). It yields a kit, prediction model, and prediction system based on 31 specific protein biomarkers. These can be used to construct risk warning or clinical decision support tools for surgically treatable diabetic retinopathy, and can be used to build a risk warning system for surgically treatable diabetic retinopathy. This provides a basis for clinical implementation of enhanced follow-up, early intervention, and individualized management, thereby helping to delay or avoid invasive ophthalmic surgery, reduce visual impairment, and alleviate medical burden. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the existing clinical screening process for diabetic retinopathy.

[0025] Figure 2 The results are from the multivariate Cox regression analysis in Example 1.

[0026] Figure 3 The graph shows the C-index, reclassification improvement index, and comprehensive discriminant improvement index for a predictive model of diabetic retinopathy requiring surgical treatment; among them... Figure 3 In this context, A corresponds to the C index. Figure 3 In this context, B corresponds to the reclassification improvement indicator within a 10-year time window. Figure 3 The C in the figure corresponds to the comprehensive discriminant improvement index within a 10-year time window. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0028] Example 1: The data used in this invention comes from a large prospective cohort study in the UK, UK Biobank (https: / / www.ukbiobank.ac.uk / ). 9801 participants with abnormal glucose metabolism (6865 prediabetic patients and 2936 diabetic patients) without diabetic retinopathy at baseline provided blood samples at cohort recruitment. Plasma protein levels were quantitatively detected using the Olink Explore 3072 platform, followed by quality control and preprocessing of the plasma proteomics data. During cohort follow-up, newly diagnosed diabetic retinopathy requiring surgical treatment was identified by linking to hospital registration systems, primary care systems, and death registration systems. Cox proportional hazards regression models and minimum absolute contraction and selection operator regression algorithms were used to screen plasma protein biomarkers for diabetic retinopathy requiring surgical treatment, and a predictive model for diabetic retinopathy requiring surgical treatment was constructed based on these biomarkers. The research protocol involved in this invention was approved by the Northwest UK Multicenter Research Ethics Committee, and all participants who participated in the cohort and provided samples signed informed consent forms. The specific steps are as follows: (1) Participant inclusion and questionnaire survey The standardized questionnaires are designed and reviewed by professional researchers. After undergoing standardized training, investigators collect basic demographic, socioeconomic status, lifestyle and medical history information from participants.

[0029] (2) Sample collection, preservation and transportation Participants underwent venous blood collection by uniformly trained nurses, and the blood samples were stored in EDTA vacuum tubes. The blood was then aliquoted into plasma fractions. Each fraction was stored at -80°C. The samples were transported to Olink Analytical Services in Sweden via dry ice for analysis.

[0030] (3) Plasma proteomics detection Plasma proteome analysis was performed using adjacent-site extension analysis. The principle of this technique is as follows: for each target protein, a pair of antibodies is designed, each antibody being coupled with a specific deoxyribonucleotide single strand. When this pair of antibodies binds to the target protein, the two deoxyribonucleotide single strands on the antibody pair complement each other and undergo enzymatic extension to form a double-stranded deoxyribonucleotide. Combined with efficient PCR amplification, the deoxyribonucleotide signal is amplified, allowing for quantitative detection of deoxyribonucleotides and thus inferring the content of the target protein.

[0031] Antibodies targeting 2,923 proteins are distributed across eight blocks, each focusing on detecting proteins associated with inflammation (first and second inflammation blocks), tumors (first and second tumor blocks), cardiovascular metabolism (first and second cardiovascular metabolism blocks), and nerves (first and second nerve blocks).

[0032] The detection values ​​for each protein are given in the form of standardized protein expression values.

[0033] (4) Quality control and preprocessing of plasma proteomics data During the detection process, a series of internal and external controls were set up to ensure detection quality. Quality control included both measurement quality control and sample quality control. In data preprocessing, proteins with a missing value ratio greater than 30% were first removed; a total of three proteins were removed. Subsequently, missing values ​​in the standardized protein expression values ​​were filled using the mean of each protein. Finally, the standardized protein expression values ​​of each protein were Z-valued, i.e., the mean of the standardized protein expression values ​​of that protein was subtracted, and then divided by the standard deviation of the standardized protein expression values ​​of that protein to obtain the standardized Z-value for each protein.

[0034] (5) Identification of newly diagnosed diabetic retinopathy patients requiring surgical treatment During a median follow-up period of 13.3 years from baseline, newly diagnosed patients requiring surgical treatment of diabetic retinopathy were identified through links to hospital registries, primary care systems, and death registries. The time of initial diagnosis, the International Classification of Disease (10th edition) code for the diagnosed disease, and the Census and Survey Office Surgical Classification Code (4th edition) for the surgical procedure were recorded. A total of 168 newly diagnosed patients requiring surgical treatment of diabetic retinopathy were identified.

[0035] (6) Screening of plasma protein markers for diabetic retinopathy requiring surgical treatment All analyses were performed in R software 4.3.2. The workflow is detailed below: Preliminary screening of plasma protein biomarkers: Multivariate Cox regression analysis was used to screen candidate proteins. The Benjamini-Hochberg false discovery rate (FDR) method was used to correct for multiple tests, and statistically significant results were selected. A statistical significance threshold of 0.05 was set. Specifically, the incidence of surgically treatable diabetic retinopathy (DR) was used as the dependent variable in the multivariate Cox regression analysis, while the standardized Z-scores of each protein were added to the model as independent variables. Other variables that might affect the incidence or protein levels of DR were added as covariates. Covariates included age, sex, race, socioeconomic status, body mass index, dietary habits, exercise level, smoking status, alcohol consumption, glomerular filtration rate, glycated hemoglobin, total cholesterol, high-density lipoprotein cholesterol, systolic blood pressure, history of cardiovascular disease, duration of diabetes, use of antihypertensive medication, and fasting duration. Cox regression analysis yielded the risk ratio of each predictive protein to the incidence of DR requiring surgical treatment. A risk ratio <1 indicated a negative correlation, while a risk ratio >1 indicated a positive correlation. Figure 2 The results of the multivariate Cox regression analysis are shown, where points above the horizontal dashed line represent those corrected by the FDR method. P Proteins with a value <0.05 are significantly associated with the development of diabetic retinopathy requiring surgical treatment. Points below the horizontal dashed line represent proteins not significantly associated with the development of diabetic retinopathy requiring surgical treatment. Points to the left of the vertical dashed line represent proteins downregulated in the diabetic retinopathy case group, and points to the right represent proteins upregulated in the same group. After initial screening, [the following proteins were selected]. Figure 2Above the horizontal dashed line, a total of 39 proteins were significantly associated with the incidence of diabetic retinopathy requiring surgical treatment. Among them, 18 proteins were positively associated with the risk of developing diabetic retinopathy requiring surgical treatment, while 21 proteins were negatively associated with the risk of developing diabetic retinopathy requiring surgical treatment. It is worth noting that some proteins previously reported in the literature as being associated with diabetic retinopathy did not reach the significance threshold after FDR correction in this invention, such as PLXNB2 protein (Plexin-B2), renin, angiopoietin-related protein 4, fatty acid-binding protein 4, tumor necrosis factor receptor superfamily member 1A, interleukin-6, CD276 antigen, T-cell immunoglobulin and mucin domain-containing protein 1, interleukin-1 receptor type 1, and V-set and immunoglobulin domain-containing protein 4. 4) The reason for this may be that: this invention focuses on diabetic retinopathy requiring surgical treatment; moreover, most of the 39 proteins initially screened in this invention are reported for the first time and are significantly associated with diabetic retinopathy requiring surgical treatment, suggesting that these 39 proteins proposed in this invention (including the 31 proteins retained after subsequent screening steps) have certain advantages in severe phenotype identification and differential screening. The 39 proteins initially screened that are associated with the pathogenesis of diabetic retinopathy requiring surgical treatment were selected as candidate proteins and further screened using the minimum absolute contraction and selection operator regression algorithm.

[0036] Further screening of predictive proteins: Proteomics data are characterized by high dimensionality and high collinearity. Minimum absolute shrinkage and selection operator regression is a regularization method used to avoid overfitting and achieve dimensionality reduction. The entire analysis population was divided into training and validation sets in a 7:3 ratio. In the training set, the aforementioned covariates and 39 proteins associated with the pathogenesis of surgically treatable diabetic retinopathy were added to the minimum absolute shrinkage and selection operator regression model, but no penalty was imposed on the covariates, only on the proteins. The penalty term parameter lambda was determined through 5-fold cross-validation using the C-exponent of cross-validation as the loss function. To obtain a predictive model with the fewest possible predictive variables, we chose the lambda value at one standard error of the minimum lambda value as the penalty term parameter. Under this lambda, a total of 31 predictive proteins were screened (as shown in Table 1). The other 8 proteins that were removed were: amphiregulin, protein FAM3D, hepatocyte growth factor, osteomodulin, urokinase-type plasminogen activator, lithostathine-1-beta, regenerating islet-derived protein 4, and serine protease inhibitor Kazal-type 1. Compared to previous cross-sectional or case-control studies, this invention can screen for proteins associated with the pathogenesis of surgically treatable diabetic retinopathy, rather than proteins associated with the development of surgically treatable diabetic retinopathy. The causal sequence between proteins and surgically treatable diabetic retinopathy is more clearly defined in this invention.

[0037] Table 1: 31 predictive proteins associated with surgically treatable diabetic retinopathy

[0038] In Table 1, "...E..." in the "FDR-P value" column indicates the exponent of 10 (e.g., 1.49E-02 means 1.49 × 10). -2 ).

[0039] (7) Construction and evaluation of a predictive model for diabetic retinopathy requiring surgical treatment In the training set, a predictive model for diabetic retinopathy requiring surgical treatment was constructed using 31 selected predictive proteins.

[0040] Based on the risk ratio of 31 predictive proteins to the incidence of diabetic retinopathy requiring surgical treatment, the following 31 predictive proteins were identified: apolipoprotein A-IV, UDP-GlcNAc:betaGal beta-1,3-N-acetylglucosaminyltransferase 7, cocaine esterase, heart-type fatty acid-binding protein, N-acetylgalactosaminyltransferase 7, pro-glucagon, growth / differentiation factor 15, LRP2-binding protein, neurofilament light polypeptide, lithostathine-1-alpha, secretagogin, and Kazal-type serine protease inhibitor 4. The risk ratio of inhibitor Kazal-type 4 and α-2,3-sialyltransferase 1 (CMP-N-acetylneuraminate-beta-galactosamide-alpha-2,3-sialyltransferase 1) to the incidence of diabetic retinopathy requiring surgical treatment is greater than 1, and they are positively correlated with the risk of diabetic retinopathy requiring surgical treatment.The following proteins contain platelet-reactive protein motifs: disintegrin and metalloproteinase with thrombospondin motifs 13, pancreatic alpha-amylase, alpha-amylase 2B, collagen alpha-1(I) chain, carboxypeptidase B, protein CREG1, dentin matrix acidic phosphoprotein 1, desmoglein-4, intercellular adhesion molecule 1, IRRE-like protein 2 homolog, matrix extracellular phosphoglycoprotein, and marginal zone B- and B1-cell-specific proteins. The risk ratios of the following proteins in patients with diabetic retinopathy requiring surgical treatment were less than 1, indicating a negative correlation with the risk of developing diabetic retinopathy requiring surgical treatment.

[0041] Considering the existing predictive models for diabetic retinopathy in clinical practice, we also constructed these existing models and compared their predictive performance with that of the model incorporating the 31 predictive proteins obtained in this invention. Specifically, the following two models were constructed in the training set: ① Model 1: Includes 9 variables: age, gender, glycated hemoglobin, duration of diabetes, systolic blood pressure, body mass index, total cholesterol, high-density lipoprotein cholesterol, and smoking status. ② Model 2: Includes the 9 variables from Model 1 plus the 31 predictive proteins. By fitting a Cox proportional hazards regression model, the participants' risk values ​​for developing diabetic retinopathy were output; the higher the risk value, the greater the future risk of developing diabetic retinopathy requiring surgical treatment.

[0042] In the validation set, we evaluated the predictive performance of each model using three indicators: the C-index, the reclassification improvement index, and the comprehensive discriminant improvement index. A higher C-index indicates better predictive performance. The reclassification improvement index quantifies the difference between the new and old models in the number of correctly classified subjects. The comprehensive discriminant improvement index quantifies the difference between the new and old models in predictive probabilities (increased predictive probability for patients and decreased predictive probability for non-patients). Both the reclassification improvement index and the comprehensive discriminant improvement index are greater than 0, indicating that the new model predicts better than the old model; equal to 0, indicating no difference in predictive performance; and less than 0, indicating that the new model predicts worse than the old model. Figure 3 The results show that the C-index of the traditional Model 1 is 0.833 (95% confidence interval: 0.781, 0.877). After adding 31 predictive proteins to the traditional model, the C-index of Model 2 increased to 0.882 (95% confidence interval: 0.842, 0.917). The reclassification improvement index and the comprehensive discrimination improvement index within the 10-year time window both indicate that Model 2 has a significant improvement in predictive performance compared to Model 1.

[0043] In addition, the above steps can correspond to the construction of a predictive system based on protein biomarkers for predicting the risk of developing diabetic retinopathy requiring surgical treatment, which may include: Information collection module: Collects information on age, gender, race, socioeconomic status, body mass index, dietary habits, exercise volume, smoking status, alcohol consumption status, glomerular filtration rate, glycated hemoglobin, total cholesterol, high-density lipoprotein cholesterol, systolic blood pressure, history of cardiovascular disease, duration of diabetes, use of antihypertensive drugs, and fasting duration, and collects plasma samples from participants; Plasma protein level detection module: Ortho-extended analysis technology is used for the detection and quantification of plasma proteins; Plasma protein data quality control and preprocessing module: used for quality control and preprocessing of proteomics data; removes proteins with missing values ​​greater than 30%, and fills in the missing values ​​of standardized protein expression values ​​using the mean of each protein; performs Z-score normalization on the standardized protein expression values ​​of each protein. The protein biomarker screening module for the risk of surgically treated diabetic retinopathy (DR) included: ① Preliminary screening: Multivariate Cox regression analysis was used to screen proteins in the entire population. Specifically, the presence and duration of DR were used as the dependent variable in the multivariate Cox regression analysis, while the standardized Z-scores of each protein were added to the model as independent variables. Other variables that might affect the incidence or protein levels of DR were added as covariates, including age, sex, race, socioeconomic status, body mass index, dietary habits, exercise level, smoking status, alcohol consumption, glomerular filtration rate, glycated hemoglobin, total cholesterol, high-density lipoprotein cholesterol, systolic blood pressure, history of cardiovascular disease, duration of diabetes, use of antihypertensive medication, and fasting duration. After preliminary screening, a total of 39 proteins were associated with the incidence of DR, of which 18 proteins were positively correlated and 21 proteins were negatively correlated. ② Further screening: The 39 proteins initially identified as being associated with the development of surgically treatable diabetic retinopathy (DR) were selected as candidate proteins. The entire analysis population was divided into a training set and a validation set at a ratio of 7:3. In the training set, predictive proteins for DR were further screened using a minimum absolute contraction and selection operator regression algorithm. Ultimately, 31 predictive proteins were identified, of which 13 were positively correlated with the risk of developing DR. These are apolipoprotein A-IV (APOA4), UDP-GlcNAc:betaGalbeta-1,3-N-acetylglucosaminyltransferase 7 (B3GNT7), cocaine esterase (CES2), and heart-shaped fatty acid-binding protein (FLC).The following proteins are listed: heart, FABP3, N-acetylgalactosaminyltransferase 7 (GALNT7), proglucagon (GCG), growth / differentiation factor 15 (GDF15), LRP2-binding protein (LRP2BP), neurofilament light polypeptide (NEFL), lithostathine-1-alpha (REG1A), secretagogin (SCGN), Kazal-type serine protease inhibitor 4 (SPINK4), and α-2,3-sialic acid transferase 1 (CMP-N-acetylneuraminate-beta-galactosamide-alpha-2).3-Sialyltransferase 1 (ST3GAL1); while 18 proteins were negatively associated with the risk of developing diabetic retinopathy requiring surgical treatment. These are: a disintegrin and metalloproteinase with thrombospondin motifs 13 (ADAMTS13), pancreatic alpha-amylase (AMY2A), alpha-amylase 2B (AMY2B), collagen alpha-1(I) chain (COL1A1), carboxypeptidase B (CPB1), protein CREG1 (CREG1), and dentin matrix acidic phosphoprotein 1. 1. DMP1), Desmoglein-4 (DSG4), Intercellular adhesion molecule 1 (ICAM1), Kin of IRRE-like protein 2 (KIRREL2), Matrix extracellular phosphoglycoprotein (MEPE), Marginal zone B- and B1-cell-specific protein (MZB1), Peptidyl-glycine alpha-amidating monooxygenase (PAM), Peptidoglycan recognition protein 1 (PGLYRP1), Platelet-activating factor acetylhydrolase Acetylhydrolase (PLA2G7), ribokinase (RBKS), osteopontin (SPP1), and thiamin pyrophosphokinase 1 (TPK1); Risk Prediction Module for Surgically Treatable Diabetic Retinopathy: Considering the existing clinical models for predicting surgically treatable diabetic retinopathy, we constructed these models and added the aforementioned 31 predictive proteins. Specifically, the following two models were built in the training set: ① Model 1: Includes 9 variables: age, gender, glycated hemoglobin, duration of diabetes, systolic blood pressure, body mass index, total cholesterol, high-density lipoprotein cholesterol, and smoking status. ② Model 2: Includes the 9 variables from Model 1 plus the 31 predictive proteins. By fitting a Cox proportional hazards regression model, the risk value of each participant is output; a higher risk value indicates a greater risk of developing surgically treatable diabetic retinopathy in the future.

[0044] Evaluation module for predicting the incidence risk of surgically treated diabetic retinopathy: In the validation set, three indicators were used to compare the predictive performance of existing surgically treated diabetic retinopathy prediction models in clinical practice with those of prediction models that incorporate 31 predictive proteins.

[0045] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. Use of a reagent for detecting a combination of protein markers in the preparation of a kit for predicting the risk of developing a surgical treatment type diabetic retinopathy, characterized in that, The protein biomarker combination includes apolipoprotein A-IV, β-1,3-N-acetylglucosamine transferase 7, cocaine esterase, heart-type fatty acid-binding protein, N-acetylgalactosamine transferase 7, proglucagon, growth differentiation factor 15, LRP2-binding protein, neurofilament light chain protein, pancreatic stone protein 1α, secretory granulin, Kazal-type serine protease inhibitor 4, and α-2,3-sialyl transferase 1, which are positively correlated with the risk of diabetic retinopathy requiring surgical treatment. Synthinase-like metalloproteinase 13, pancreatic α-amylase, α-amylase 2B, type I collagen α1 chain, carboxypeptidase B, CREG1 protein, dentin matrix acidic phosphoprotein 1, desmosome core glycoprotein 4, intercellular adhesion molecule 1, IRRE-like protein 2 homolog, extracellular phosphoglycoprotein of the matrix, marginal zone B cell and B1 cell specific proteins, peptidyl glycine α-amidyl monooxygenase, peptidoglycan recognition protein 1, platelet-activating factor acetylhydrolase, ribokinase, osteopontin, thiamine pyrophosphate kinase 1; the combination of these protein biomarkers is used in a kit to predict the risk of developing diabetic retinopathy requiring surgical treatment.

2. A method for constructing a predictive model based on protein biomarkers for predicting the risk of developing diabetic retinopathy requiring surgical treatment, characterized in that, Includes the following steps: S1. Prepare a dataset for individuals with abnormal glucose metabolism but without diabetic retinopathy. The dataset includes the Z-values ​​of 31 specific protein biomarkers obtained by each individual with abnormal glucose metabolism, clinical variable data when obtaining the biological samples used to detect the 31 specific protein biomarkers, and whether surgically treated diabetic retinopathy has occurred before a preset subsequent time point. S2. In the dataset, using the Z-valued expression values ​​of 31 specific protein markers and clinical variable data as variables, a predictive model for predicting the risk of developing diabetic retinopathy requiring surgical treatment was obtained by fitting a Cox proportional hazards regression model. Among them, the 31 specific protein biomarkers include apolipoprotein A-IV, β-1,3-N-acetylglucosamine transferase 7, cocaine esterase, heart-type fatty acid-binding protein, N-acetylgalactosamine transferase 7, proglucagon, growth differentiation factor 15, LRP2-binding protein, neurofilament light chain protein, pancreatic stone protein 1α, secretory granulin, Kazal-type serine protease inhibitor 4, α-2,3-sialic acid transferase 1, and are positively correlated with the risk of diabetic retinopathy requiring surgical treatment. The following proteins showed a negative correlation with the risk of developing the disease: platelet-reactive protein motif-de-integrin-like metalloproteinase 13, pancreatic α-amylase, α-amylase 2B, type I collagen α1 chain, carboxypeptidase B, CREG1 protein, dentin matrix acidic phosphoprotein 1, desmosome core glycoprotein 4, intercellular adhesion molecule 1, IRRE-like protein 2 homolog, extracellular phosphoglycoproteins of the matrix, marginal zone B cell and B1 cell-specific proteins, peptidyl glycine α-amidyl monooxygenase, peptidoglycan recognition protein 1, platelet-activating factor acetylhydrolase, ribokinase, osteopontin, and thiamine pyrophosphate kinase 1. The clinical variables include age, sex, glycated hemoglobin, duration of diabetes, systolic blood pressure, body mass index, total cholesterol, high-density lipoprotein cholesterol, and smoking status.

3. The method for constructing a predictive model based on protein biomarkers for predicting the risk of surgically treated diabetic retinopathy as described in claim 2, characterized in that, In step S1, the Z-value is obtained by first providing the detection value of a specific protein marker in the plasma of each individual in the group of individuals with abnormal glucose metabolism without diabetic retinopathy in the form of a standardized protein expression value, and calculating the mean and standard deviation of the standardized protein expression value of the specific protein marker; then, for a certain individual, the standardized protein expression value of the specific protein marker in the individual's plasma is first subtracted from the mean, and then divided by the standard deviation to obtain the Z-value of the specific protein marker in the individual's plasma.

4. A predictive model for the risk of diabetic retinopathy requiring surgical treatment, constructed using the method described in any one of claims 2-3, which is based on protein biomarkers and used to predict the risk of diabetic retinopathy requiring surgical treatment.

5. A method for using the predictive model for predicting the risk of surgically treated diabetic retinopathy, constructed using the method described in any one of claims 2-3, characterized in that... Using individuals with abnormal glucose metabolism but no diabetic retinopathy as subjects, the Z-values ​​of 31 specific protein biomarkers in the subject's plasma and the clinical variable data of the subject at the time of obtaining the biological samples used to detect the 31 specific protein biomarkers were first obtained. The Z-values ​​of the 31 specific protein biomarkers and the clinical variable data were simultaneously input into the prediction model. The score output by the prediction model is the subject's risk of developing diabetic retinopathy requiring surgical treatment in the future; the higher the score, the greater the risk. The Z-values ​​of the 31 specific protein markers in the subject's plasma are obtained by taking into account the mean and standard deviation of the standardized protein expression values ​​of the 31 specific protein markers used in the corresponding construction method of the prediction model. For a specific protein marker, the Z-value of the specific protein marker in the subject's plasma is obtained by subtracting the mean from the standardized protein expression value of the specific protein marker in the subject's plasma and then dividing by the standard deviation.

6. A predictive system for predicting the risk of developing diabetic retinopathy requiring surgical treatment, characterized in that, include: Plasma protein detection and clinical variable data acquisition module: used to obtain the Z-valued expression values ​​of 31 specific protein biomarkers contained in the subject's plasma, as well as the clinical variable data of the subject when obtaining the biological samples used to detect the 31 specific protein biomarkers; The predictive model for predicting the risk of surgically treatable diabetic retinopathy, constructed using the method described in any one of claims 2-3, is as follows: It takes as input the Z-valued expression values ​​of 31 specific protein markers present in the subject's plasma and the clinical variable data of the subject when obtaining the biological samples used to detect the 31 specific protein markers. The output score represents the subject's future risk of developing surgically treatable diabetic retinopathy; the higher the score, the greater the risk.

Images