A risk assessment method, system and storage medium for type 2 diabetes complications

Abstract

The present invention belongs to the technical field of risk assessment of diabetic complications, and in particular, relates to a method, system, and storage medium for risk assessment of type 2 diabetes complications. Risk assessment method: S1, based on the clinical indicator data of patients in the first training set, a two-dimensional clustering tree is constructed using the DDRTree algorithm; the two-dimensional clustering tree contains a number of clustering regions, each clustering region contains a number of nodes, and each node represents a patient; S2, based on the clinical indicator data of the patient to be assessed, a mapping function is used to obtain the position coordinates of the patient to be assessed in the two-dimensional clustering tree; S3, based on the position coordinates of the patient to be assessed, a competing risk model is used to calculate the probability that the patient to be assessed will suffer from various complications in the next r years. The present invention can accurately assess the risk of complications in patients with type 2 diabetes.

Description

Technical Field

[0001] The present invention belongs to the technical field of diabetic complication risk assessment, and in particular relates to a risk assessment method, system and storage medium for type 2 diabetes complications. Background Art

[0002] Type 2 diabetes is a chronic metabolic disease with complications including diabetic retinopathy, heart failure, stroke, myocardial infarction, end-stage renal disease, chronic kidney disease, metabolic-related fatty liver disease, cirrhosis, and diabetic peripheral vascular disease. The complex development of these complications severely impacts patients' quality of life and prognosis. Therefore, risk assessment of type 2 diabetes complications not only allows for preventive interventions in patients' lifestyles but also allows for targeted prognosis and improved outcomes.

[0003] Existing technologies usually use simple statistical methods or single machine learning algorithms based on a single clinical indicator, such as blood sugar or blood pressure, to assess the risk of complications in patients with type 2 diabetes.

[0004] However, complications of type 2 diabetes often affect multiple clinical indicators, and there are potential connections between these indicators that are often difficult for researchers to directly capture. Furthermore, different complications may cause the same clinical indicator to exhibit the same changing trend. For example, long-term follow-up of type 2 diabetes patients with high blood pressure revealed that only a subset of patients developed diabetic foot within a decade. Therefore, existing technologies cannot accurately assess the risk of complications in type 2 diabetes patients using a single clinical indicator.

[0005] Therefore, how to improve the accuracy of risk assessment for complications of type 2 diabetes has become a difficult problem that needs to be solved urgently. Summary of the Invention

[0006] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a method for risk assessment of type 2 diabetes complications, which can accurately assess the risk of complications in type 2 diabetes patients.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A method for risk assessment of type 2 diabetes complications, comprising the following steps:

[0009] S1, based on the clinical indicator data of patients in the first training set, a two-dimensional cluster tree is constructed using the DDRTree algorithm; the two-dimensional cluster tree contains several cluster regions, each cluster region contains several nodes, and each node represents a patient;

[0010] S2, based on the clinical indicator data of the patient to be evaluated, using a mapping function to obtain the position coordinates of the patient to be evaluated in the two-dimensional clustering tree;

[0011] S3, based on the location coordinates of the patient to be evaluated, uses the competing risk model to calculate the probability that the patient to be evaluated will suffer from various complications in the next r years.

[0012] Preferably, after S1 and before S2, S1' is also included: S1', after the technicians mark the common features of the clinical indicators of each clustering area, they use the mapping function to obtain the position coordinates of the patients in the verification set in the two-dimensional clustering tree. If the consistency rate of the common features of the clinical indicators of the patients in the verification set and the corresponding clustering areas is greater than the first threshold, the current two-dimensional clustering tree is judged to be successfully verified; otherwise, the verification fails, and after discarding the current two-dimensional clustering tree, the number of patients in the first training set is increased and the process returns to S1.

[0013] Preferably, in S1, the following sub-steps are also included:

[0014] S11, obtaining a feature matrix X based on a first training set, wherein the first training set contains data of m clinical indicators for n patients; the n rows of the feature matrix X represent n patients, and the m columns represent m clinical indicators; both n and m are positive integers;

[0015] S12, based on the feature matrix X, the optimization objective function F of the DDRTree algorithm is constructed and solved to obtain the key parameters of the two-dimensional clustering tree:

[0016] ;

[0017] Among them, the key parameters are the optimal coordinate matrix and the optimal central node set; Y represents the coordinate matrix, which contains n rows and 2 columns. The n rows of the coordinate matrix Y represent n patients, and are the same as the patients represented by the corresponding rows in the feature matrix X. The first column of the coordinate matrix Y represents the horizontal coordinate, and the second column of the coordinate matrix Y represents the vertical coordinate; Y(f) represents the position coordinate corresponding to the fth row in the coordinate matrix Y; W represents the linear transformation parameter matrix in the DDRTree algorithm; represents the transpose of the linear transformation parameter matrix W; λ represents the first hyperparameter; Indicates the position coordinates of the central node i in the current two-dimensional clustering tree; Indicates the position coordinates of the center node j in the current two-dimensional clustering tree; and are all position coordinates in the coordinate matrix Y; ε represents the edge set of the minimum spanning tree in the current two-dimensional clustering tree; Z represents the set of central node position coordinates; α represents the second hyperparameter; Represents the position coordinates of the central node closest to the position coordinate Y(f); Indicates the coordinate matrix Y and the central node set Z corresponding to the minimum value. At this time, the coordinate matrix Y and the central node set Z are the optimal coordinate matrix and the optimal central node set; represents the square of the 2-norm;

[0018] S13, constructing a two-dimensional clustering tree based on key parameters.

[0019] Preferably, S1' further includes the following contents:

[0020] S11′, the technicians mark the common features of the clinical indicators of each cluster area according to the clinical indicators of the patients in each cluster area in the current two-dimensional cluster tree;

[0021] S12´, use the mapping function to obtain the position coordinates of each patient in the validation set in the two-dimensional clustering tree;

[0022] S13′, determining the clustering region where each patient in the validation set is located according to the position coordinates of each patient in the validation set in the two-dimensional cluster tree;

[0023] S14′, after determining whether the clinical indicator characteristics of each patient in the validation set are consistent with the common characteristics of the corresponding cluster region, calculate the consistency rate CR = NUM(C) / NUM(All); where NUM (·) represents the number; NUM(C) represents the number of patients in the validation set whose clinical indicator characteristics are consistent with the common characteristics of the corresponding cluster region, and NUM(All) represents the total number of patients in the validation set;

[0024] If the consistency rate CR is greater than the first threshold, the current two-dimensional clustering tree is determined to be successfully verified; if the consistency rate CR is below the first threshold, the current two-dimensional clustering tree is determined to have failed verification. After discarding the current two-dimensional clustering tree, a new first training set is constructed or the number of patients in the first training set is increased, and then the process returns to S1.

[0025] Preferably, a machine learning model is used to obtain the position coordinates of the patient G in the two-dimensional cluster tree based on a mapping function. ;

[0026] The mapping function is:

[0027] ;

[0028] ;

[0029] in, The vertical coordinate represents the position coordinate of patient G in the two-dimensional cluster tree; The horizontal coordinate represents the position coordinate of patient G in the two-dimensional cluster tree; represents the first global bias parameter; represents the second global bias parameter; represents the age regression coefficient; represents the gender regression coefficient; represents the age of patient G; represents the gender parameter of patient G; represents the data of the pth clinical indicator of patient G; express kth power of represents the intermediate value of the pth clinical index of patient G, max(·) represents the maximum value, Indicates that the node d in the two-dimensional clustering tree corresponds to the data of the p-th clinical indicator of the patient, and the two-dimensional clustering tree includes n nodes in total.

[0030] Preferably, in S3, undiagnosed complications of the patient to be evaluated are evaluated, where u is a positive integer and includes the following:

[0031] The competing risk model is based on the location coordinates of the patient R to be evaluated, and calculates the probability that the patient will suffer from the Vth complication in the next r years. Probability , where r ≥ 0 and V is a positive integer less than or equal to u:

[0032] ;

[0033] ;

[0034] Where t represents the future t-year of the patient R to be evaluated starting from the current moment; Indicates that the patient R to be evaluated will have complications in the next t years The instantaneous probability of represents the first regression coefficient; represents the second regression coefficient; The ordinate represents the position coordinate of the patient R to be evaluated in the two-dimensional cluster tree; The horizontal coordinate represents the position coordinate of the patient R to be evaluated in the two-dimensional cluster tree; Indicates complications The benchmark subdistribution risk function that varies with future time t.

[0035] Preferably, before using the competing risk model to evaluate the probability of a patient suffering from various complications in the next r years, the competing risk model is trained using the second training set; the second training set includes e training data, each training data corresponds to a patient who has suffered from more than one complication, and each training data includes the time when the patient was diagnosed with type 2 diabetes, the time when various complications were diagnosed, and the clinical indicator data when the complication was most recently diagnosed; e is a positive integer.

[0036] Preferably, the clinical indicator data of patients in the first training set, the validation set, and the second training set are all cleaned clinical indicator data, and the cleaned clinical indicator data further includes the following:

[0037] Step 1: Extract clinical indicator data from several patients. If a patient has more than one type of dirty data, all clinical indicator data for the current patient will be removed. Dirty data refers to clinical indicator data that is outside the 5 standard deviations of the same clinical indicator data of several patients.

[0038] Step 2: The remaining clinical indicator data in step 1 are converted into clinical indicator data within the range of 0 to 1 through the rank normalization method. At this time, the clinical indicator data cleaning is completed.

[0039] The present invention also provides a risk assessment system for type 2 diabetes complications, comprising: a two-dimensional clustering tree construction module, a position coordinate mapping module and a risk assessment module; the two-dimensional clustering tree construction module includes a DDRTree algorithm, the DDRTree algorithm constructs a two-dimensional clustering tree based on the clinical indicator data of patients in a first training set, and sends the constructed two-dimensional clustering tree to the position coordinate mapping module; the position coordinate mapping module is used to convert the clinical indicator data of the patient to be assessed into position coordinates on the two-dimensional clustering tree, and send the position coordinates of the patient to be assessed to the risk assessment module; the risk assessment module includes a competing risk model, the competing risk model calculates the probability of the patient to be assessed suffering from various complications in the next r years based on the position coordinates of the patient to be assessed and outputs the calculated probability; each module is configured to execute the steps of the above-mentioned method for risk assessment of type 2 diabetes complications.

[0040] The present invention also provides a computer-readable storage medium: the computer-readable storage medium stores a computer program programmed or configured to execute the above-mentioned method for risk assessment of type 2 diabetes complications.

[0041] The beneficial effects of the present invention are:

[0042] (1) The risk assessment method for type 2 diabetes complications of the present invention is based on multiple clinical characteristic data of the patient to be assessed to comprehensively assess the probability of the patient suffering from certain complications in the future, thereby improving the accuracy of the assessment.

[0043] (2) The two-dimensional clustering tree constructed by the present invention greatly simplifies the presentation of the patient's clinical indicator data without losing the information represented by the patient's clinical indicator data, thereby improving the computational efficiency in the evaluation process; and the node position coordinates in the two-dimensional clustering tree also include the potential connections between the patient's own multiple clinical indicator data mined; and the distribution of nodes in the two-dimensional clustering tree also includes the potential connections between the various clinical indicator data of patients.

[0044] (3) After constructing the two-dimensional clustering tree, the present invention also uses a validation set to verify the current two-dimensional clustering tree, ensuring that only two-dimensional clustering trees with sufficient generalization and accuracy will be used in subsequent complication probability assessment, further improving the accuracy of complication probability assessment.

[0045] (4) The present invention uses a competing risk model to calculate the probability of the patient to be evaluated suffering from various complications in the next r years based on a two-dimensional clustering tree. The two-dimensional clustering tree constructed by the present invention not only ensures that the patient's clinical indicator information is complete and covers potential contact information, but also displays this information in the form of simple two-dimensional position coordinates. For technical personnel: technical personnel do not care about the two-dimensional position coordinates of the two-dimensional clustering tree and the massive information represented by the clustering tree structure, but only want to know the probability of the final complication. Therefore, whether the technical personnel understand all the information covered by the two-dimensional clustering tree has no effect on the final competing risk model's evaluation of the probability of the complication. For the competing risk model: the input of the competing risk model is simplified. Only the two-dimensional position coordinates of the patient to be evaluated need to be used as the input of the model, so that the competing risk model can receive the various clinical indicator information and potential contact information of the patient to be evaluated represented by the two-dimensional position coordinates. While improving the evaluation accuracy of the competing risk model, the computational overhead of the competing risk model is greatly reduced, and the evaluation efficiency is improved.

[0046] (5) When evaluating the probability of a patient suffering from a certain complication within the next r years, the present invention takes into account that the probability of suffering from the disease will also increase with the passage of time. Therefore, the time is integrated in the evaluation process, making the evaluation result of the probability of suffering from the disease more scientific.

[0047] (6) The present invention can assess the probability of a patient developing various complications within the next r years. Therefore, patients can assess the probability of developing different complications at different times in the future based on their current clinical indicators, which is very flexible and convenient. Regular complication risk assessment is also beneficial for providing feedback to doctors and patients on the intervention and prognosis effects of type 2 diabetes. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a method for risk assessment of type 2 diabetes complications according to the present invention;

[0049] Figure 2 Schematic diagram of the two-dimensional clustering tree constructed for the present invention. DETAILED DESCRIPTION

[0050] In order to make the technical solution of the present invention clearer and more specific, the present invention is clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Ordinary technicians in this field, without making any creative work, make equivalent substitutions for the technical features of the technical solution of the present invention and solutions derived from conventional reasoning all fall within the scope of protection of the present invention.

[0051] The "patients" mentioned in the present invention are all patients who have been diagnosed with type 2 diabetes, and no further explanation will be given below.

[0052] like Figure 1 FIG. 1 is a flow chart of a method for risk assessment of type 2 diabetes complications according to the present invention, comprising the following steps:

[0053] S1, based on the clinical indicator data of patients in the first training set, a two-dimensional cluster tree is constructed using the DDRTree algorithm; the two-dimensional cluster tree contains several cluster regions, each cluster region contains several nodes, and each node represents a patient;

[0054] S2, based on the clinical indicator data of the patient to be evaluated, using a mapping function to obtain the position coordinates of the patient to be evaluated in the two-dimensional clustering tree;

[0055] S3, based on the location coordinates of the patient to be evaluated, uses the competing risk model to calculate the probability that the patient to be evaluated will suffer from various complications in the next r years.

[0056] In S1, the following sub-steps are also included:

[0057] S11. Obtain a feature matrix X based on the first training set, where the first training set contains data on m clinical indicators of n patients; each row of the feature matrix X represents m clinical indicators of one patient, and each column represents one clinical indicator; both n and m are positive integers.

[0058] The feature matrix X is a matrix with n rows and m columns, and the elements in the matrix represent the data of a certain clinical indicator of a certain patient.

[0059] The clinical indicator data of patients in the first training set came from the China Kidney Disease Database, and were the clinical indicator data of patients in the first year after being diagnosed with type 2 diabetes.

[0060] In this embodiment, n=26000; m=10, and they are high-density lipoprotein cholesterol, triglycerides, systolic blood pressure, alanine aminotransferase, glycosylated hemoglobin, low-density lipoprotein cholesterol, creatinine, heart rate, body mass index, and diastolic blood pressure, respectively.

[0061] The patient's age and age are known data, but we do not need to use them when constructing the training set and feature matrix X.

[0062] S12, based on the feature matrix X, the optimization objective function F of the DDRTree algorithm is constructed and solved to obtain the key parameters of the two-dimensional clustering tree:

[0063] ;

[0064] Among them, the key parameters are the optimal coordinate matrix and the optimal central node set; Y represents the coordinate matrix, which contains n rows and 2 columns. Each row of the coordinate matrix Y represents a patient in the first training set and is the same as the patient represented by the corresponding row in the feature matrix X. The first column of the coordinate matrix Y represents the horizontal coordinate, and the second column of the coordinate matrix Y represents the vertical coordinate; Y(f) represents the fth row in the coordinate matrix Y, which is a position coordinate; W represents the linear transformation parameter matrix in the DDRTree algorithm; represents the transpose of the linear transformation parameter matrix W; λ represents the first hyperparameter, which is set by technicians based on experience and is used to control the compactness between nodes in the two-dimensional clustering tree; Represents the position coordinates of the center node i in the current two-dimensional clustering tree, which is also a position coordinate in the coordinate matrix Y; represents the position coordinates of the central node j in the current two-dimensional clustering tree, which is also a position coordinate in the coordinate matrix Y; ε represents the edge set of the minimum spanning tree in the current two-dimensional clustering tree, which is generated by the DDRTree algorithm in the process of learning the potential relationship between the clinical indicators of patients in the first dataset; Z represents the set of central node position coordinates; α represents the second hyperparameter, which is set by technicians based on experience and is used to control the degree to which non-central nodes in the two-dimensional clustering tree move closer to the central node; Represents the position coordinates of the central node closest to the position coordinate Y(f); Indicates the coordinate matrix Y and the central node set Z corresponding to the minimum value. At this time, the coordinate matrix Y and the central node set Z are the optimal coordinate matrix and the optimal central node set; Represents the square of the 2-norm.

[0065] S13, constructing a two-dimensional clustering tree based on key parameters.

[0066] The schematic diagram of the two-dimensional clustering tree is as follows Figure 2 As shown, Figure 2 Each gray point in the graph represents a patient in the first training set and is also a node in the two-dimensional clustering tree.

[0067] In the DDRTree algorithm, a two-dimensional clustering tree is constructed according to key parameters. This is an existing technology and will not be described in detail here.

[0068] Optionally, after S1 and before S2, also include S1´:

[0069] S1´, after the technicians mark the common features of the clinical indicators of each cluster area, they use the mapping function to obtain the position coordinates of the patients in the verification set in the two-dimensional cluster tree. If the consistency rate of the common features of the clinical indicators of the patients in the verification set and the corresponding cluster areas is greater than the first threshold, the current two-dimensional cluster tree is judged to be successfully verified; otherwise, the verification fails, and after discarding the current two-dimensional cluster tree, the number of patients in the first training set is increased and the process returns to S1.

[0070] The set of m clinical indicators of patient G is , Represents the data of the pth clinical indicator of patient G, 1≤p≤m, and p is a positive integer.

[0071] Use machine learning model based on mapping function to obtain the position coordinates of patient G in the two-dimensional clustering tree ;

[0072] The mapping function is:

[0073] ;

[0074] ;

[0075] in, The vertical coordinate represents the position coordinate of patient G in the two-dimensional cluster tree; The horizontal coordinate represents the position coordinate of patient G in the two-dimensional cluster tree; represents the first global bias parameter; represents the second global bias parameter, and It is a known quantity within the machine learning model and is generated by the machine learning model itself; represents the age regression coefficient; represents the gender regression coefficient; represents the age of patient G; Indicates the gender parameter of patient G. Male and female correspond to different gender parameters, and the gender parameter is set by the technician; represents the data of the pth clinical indicator of patient G; express kth power of represents the intermediate value of the pth clinical index of patient G, max(·) represents the maximum value, Indicates that the node d in the two-dimensional clustering tree corresponds to the data of the p-th clinical indicator of the patient, and the two-dimensional clustering tree includes n nodes in total.

[0076] In S1´, the following are also included:

[0077] S11´, the technicians mark the common features of the clinical indicators of each cluster area based on the clinical indicators of the patients in each cluster area in the current two-dimensional cluster tree.

[0078] For example, the common characteristics of clinical indicators in the first cluster area are high body mass index (BMI greater than 30) and high blood sugar (fasting blood sugar greater than 9.8mmol / L), and the common characteristics of clinical indicators in the second cluster area are high high-density lipoprotein cholesterol (greater than 2.6mmol / L) and high heart rate (resting heart rate greater than 100 beats / minute).

[0079] After the two-dimensional cluster tree is generated, the range of each cluster area in the current two-dimensional cluster tree is determined.

[0080] S12´, use the mapping function to obtain the position coordinates of each patient in the validation set in the two-dimensional clustering tree.

[0081] S13´, determine the clustering region where each patient in the validation set is located according to the position coordinates of each patient in the validation set in the two-dimensional clustering tree.

[0082] As long as the position coordinates of a patient fall into a clustering area of ​​the current two-dimensional clustering tree, the area is the clustering area where the current patient is located.

[0083] S14′, after determining whether the clinical indicator characteristics of each patient in the validation set are consistent with the common characteristics of the corresponding cluster region, calculate the consistency rate CR = NUM(C) / NUM(All), where NUM (·) represents the number, NUM(C) represents the number of patients in the validation set whose clinical indicator characteristics are consistent with the common characteristics of the corresponding cluster region, and NUM(All) represents the total number of patients in the validation set;

[0084] If the consistency rate CR is greater than the first threshold, the current two-dimensional clustering tree is determined to be successfully verified; if the consistency rate CR is below the first threshold, the current two-dimensional clustering tree is determined to have failed verification. After discarding the current two-dimensional clustering tree, a new first training set is constructed or the number of patients in the first training set is increased, and then the process returns to S1.

[0085] The clinical indicator data of the patients in the validation set also come from the China Kidney Disease Database and are also the clinical indicator data of the patients in the first year after being diagnosed with type 2 diabetes. However, there is no overlap between the patients in the validation set and the patients in the first training set. In this embodiment, the number of patients in the validation set is 6501.

[0086] S1 and its sub-steps of the present invention construct a two-dimensional cluster tree; and S1' and its sub-steps are used to verify whether the structure of the current two-dimensional cluster tree has sufficient generalization and accuracy. If the structure of the two-dimensional cluster tree has sufficient generalization and accuracy, then the clustering area into which the patients in the verification set fall through the mapping function must have clinical indicator characteristics consistent with those of the patients in the verification set. Even if there is inconsistency, it is extremely small. Therefore, in S14', the consistency rate is finally calculated to determine whether the structure of the current two-dimensional cluster tree has sufficient generalization and accuracy. If the consistency rate is too low, it means that the generalization and accuracy of the current two-dimensional cluster tree are insufficient and cannot be used for subsequent calculations. Therefore, the present invention adopts the method of reconstructing the first training set or expanding the number of patients in the original first training set (i.e., increasing the number of samples) to address this situation, and then reconstructing the two-dimensional cluster tree based on the new first training set until the consistency rate is greater than the first threshold.

[0087] In S1', S12' and S2, the mapping function is used to obtain the position coordinates of the patient in the two-dimensional clustering tree, which has been described above and will not be repeated here.

[0088] Figure 2 The red dots in the figure represent the positions of the patients to be evaluated in the two-dimensional cluster tree.

[0089] Before using the competing risk model to assess the probability of a patient suffering from various complications in the next r years, the competing risk model is trained using the second training set, which also includes the following:

[0090] The second training set includes e training data, each of which corresponds to a patient who has suffered from one or more complications. Each training data includes the time when the patient was diagnosed with type 2 diabetes, the time when various complications were diagnosed, and the clinical indicator data when the complications were most recently diagnosed; e is a positive integer.

[0091] In this embodiment, e=48000.

[0092] The clinical indicator data and complication data of patients in the second training set also came from the Chinese Kidney Disease Database.

[0093] The main complications of type 2 diabetes include: myocardial infarction (MI), stroke (including ischemic and hemorrhagic), heart failure (HF), metabolic-associated fatty liver disease (MAFLD), cirrhosis, diabetic retinopathy (DR), chronic kidney disease (CKD), end-stage renal disease (ESRD) and diabetic peripheral vascular disease (DPD).

[0094] The clinical indicator data of patients in the first training set, validation set, and second training set are all cleaned clinical indicator data. The cleaned clinical indicator data also includes the following:

[0095] Step 1: Extract clinical indicator data from several patients. If a patient has more than one type of dirty data, all clinical indicator data for the current patient will be removed. Dirty data refers to clinical indicator data that is outside the 5 standard deviations of the same clinical indicator data of several patients.

[0096] Step 2: The remaining clinical indicator data in step 1 are converted into clinical indicator data within the range of 0 to 1 through the rank normalization method. At this time, the clinical indicator data cleaning is completed.

[0097] Cleaning clinical indicator data can eliminate obviously abnormal clinical indicator data. Even if a very small amount of clinical indicator data with less abnormality is retained, the subsequent impact on other normal clinical indicator data will be reduced due to rank normalization.

[0098] In this embodiment, the clinical indicator data of the patient to be evaluated are clinical indicator data within the past three months.

[0099] In S3, only the undiagnosed complications of the patient to be evaluated are evaluated, where u is a positive integer, and the following are also included:

[0100] The competing risk model is based on the location coordinates of the patient R to be evaluated, and calculates the probability that the patient will suffer from the Vth complication in the next r years. Probability , where r ≥ 0 and V is a positive integer less than or equal to u:

[0101] ;

[0102] ;

[0103] Where t represents the future t-year of the patient R to be evaluated starting from the current moment; Indicates that the patient R to be evaluated will have complications in the next t years The instantaneous probability of represents the first regression coefficient; represents the second regression coefficient; The ordinate represents the position coordinate of the patient R to be evaluated in the two-dimensional cluster tree; The horizontal coordinate represents the position coordinate of the patient R to be evaluated in the two-dimensional cluster tree; Indicates complications The benchmark sub-distribution risk function that changes with future time t; 、 as well as It is obtained by the competing risk model during the training and optimization process.

[0104] In this embodiment, r=10.

[0105] The present invention provides a method for risk assessment of type 2 diabetes complications, which is based on multiple clinical characteristic data of the patient to be assessed to comprehensively assess the probability of the patient suffering from certain complications in the future, thereby improving the accuracy of the assessment.

[0106] The present invention first constructs a two-dimensional cluster tree based on the various clinical indicator data of patients in the first training set using the DDRTree algorithm, converting patients carrying multiple clinical indicator data into nodes in the two-dimensional cluster tree. Subsequently, we only need to process the position coordinates of the nodes, achieving dimensionality reduction of the patient's multidimensional information. This dimensionality reduction of information does not directly delete information of certain dimensions, but rather integrates and compresses the multidimensional information. During the training process, the DDRTree algorithm not only mines the potential connections between the multiple clinical indicator data of each patient, but also learns the potential connections between the various clinical indicator data between patients. After this, the two-dimensional node clustering method is used to reflect the potential connections between the multiple clinical indicators between patients through the perspective of node distribution on a two-dimensional plane. That is, the two-dimensional cluster tree constructed by the present invention greatly simplifies the representation of the patient's clinical indicator data without losing the information represented by the patient's clinical indicator data, thereby improving the computational efficiency during the evaluation process. Moreover, the node position coordinates in the two-dimensional cluster tree also include the potential connections between the patient's own multiple clinical indicator data that have been mined. Moreover, the distribution of nodes in the two-dimensional cluster tree also includes the potential connections between the various clinical indicator data between patients.

[0107] After constructing the two-dimensional clustering tree, the present invention also uses a validation set to verify the current two-dimensional clustering tree, ensuring that only two-dimensional clustering trees with sufficient generalization and accuracy will be used in subsequent complication probability assessment, further improving the accuracy of complication probability assessment.

[0108] The present invention uses a competing risk model to calculate the probability of a patient to be evaluated suffering from various complications within the next r years based on a two-dimensional clustering tree. The two-dimensional clustering tree constructed by the present invention not only ensures that the patient's clinical indicator information is complete and covers potential contact information, but also displays this information in the form of simple two-dimensional position coordinates. For technical personnel: technical personnel do not care about the two-dimensional position coordinates of the two-dimensional clustering tree and the massive information represented by the clustering tree structure, but only hope to know the probability of the final complication. Therefore, whether the technical personnel understand all the information covered by the two-dimensional clustering tree has no effect on the final competing risk model's assessment of the probability of the complication. For the competing risk model: the input of the competing risk model is simplified. Only the two-dimensional position coordinates of the patient to be evaluated need to be used as the input of the model, so that the competing risk model can receive the various clinical indicator information and potential contact information of the patient to be evaluated represented by the two-dimensional position coordinates. While improving the evaluation accuracy of the competing risk model, the computational overhead of the competing risk model is greatly reduced, and the evaluation efficiency is improved.

[0109] When evaluating the probability of a patient suffering from a certain complication in the next r years, the present invention takes into account that the probability of suffering from the disease will also increase cumulatively with the passage of time, so time is integrated in the evaluation process, making the evaluation result of the probability of suffering from the disease more scientific.

[0110] This invention can estimate the probability of a patient developing various complications within the next r years. Therefore, patients can assess the probability of developing different complications at different times in the future based on their current clinical indicators, providing a flexible and convenient method. Based on the assessment results, doctors can implement preventive interventions in patients' lifestyles and provide targeted prognoses, thereby improving treatment outcomes. Regular complication risk assessments also facilitate feedback to doctors and patients on the effectiveness of interventions and prognoses, facilitating timely adjustments.

[0111] Using clinical data from 5,000 patients diagnosed with complications at the time of their type 2 diabetes diagnosis, researchers used the complication risk assessment method of the present invention to estimate the probability of developing various complications within three years of their type 2 diabetes diagnosis. For these 5,000 patients, the complication with the highest estimated probability, greater than 85%, had an accuracy rate of 73.2% in predicting that the patient would actually develop the disease within three years of their type 2 diabetes diagnosis. This compares to the average accuracy of 46.9% for existing risk assessments of type 2 diabetes complications.

[0112] Technicians used data from patients with type 2 diabetes who had confirmed complications to validate the model's performance using the complication risk assessment method of the present invention. Based on a two-dimensional clustering tree structure and a competing risk model, the present invention demonstrated good predictive performance in predicting diabetic complications within 10 years. Specifically:

[0113] (1) Model discrimination ability: The ROC AUC (area under the receiver operating characteristic curve) of the method of the present invention reached 0.82-0.91 when predicting major complications (including myocardial infarction, stroke, heart failure, metabolic-related fatty liver, cirrhosis, diabetic retinopathy, chronic kidney disease, end-stage renal disease and diabetic peripheral vascular disease), indicating that the model has excellent discrimination.

[0114] (2) External validation performance: In the external validation of an independent dataset, the ROC AUC remained between 0.79 and 0.86, demonstrating the strong generalization ability of the model.

[0115] Compared with traditional methods that only use a single clinical indicator or simple statistical methods, this invention integrates multidimensional clinical data and utilizes dimensionality reduction technology to significantly improve prediction efficiency and accuracy while retaining complete information, providing an efficient and reliable technical solution for personalized risk assessment of type 2 diabetes complications.

[0116] The present invention also provides a risk assessment system for type 2 diabetes complications, comprising:

[0117] Two-dimensional clustering tree construction module, location coordinate mapping module and risk assessment module,

[0118] The two-dimensional clustering tree construction module includes the DDRTree algorithm, which constructs a two-dimensional clustering tree based on the clinical indicator data of the patients in the first training set, and sends the constructed two-dimensional clustering tree to the position coordinate mapping module;

[0119] The position coordinate mapping module is used to convert the clinical indicator data of the patient to be evaluated into position coordinates on a two-dimensional clustering tree, and send the position coordinates of the patient to be evaluated to the risk assessment module;

[0120] The risk assessment module includes a competing risk model, which calculates the probability of the patient suffering from various complications in the next r years based on the patient's location coordinates and outputs it;

[0121] Each module is configured to execute the steps of the above-mentioned method for risk assessment of complications of type 2 diabetes.

[0122] The present invention also provides a computer-readable storage medium: the computer-readable storage medium stores a computer program programmed or configured to execute the above-mentioned method for risk assessment of type 2 diabetes complications.

[0123] The technologies, shapes, and structures not described in detail in the present invention are all well-known technologies. It should also be pointed out that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. The various components or steps in the embodiments of the present invention can be decomposed and / or recombined, and such decompositions and / or recombinations should be regarded as equivalent solutions of the present application and should fall within the scope of protection of the present invention.

Claims

1. A method for risk assessment of type 2 diabetes complications, characterized in that: The following steps are involved: S1, based on the clinical indicator data of patients in the first training set, a two-dimensional cluster tree is constructed using the DDRTree algorithm; the two-dimensional cluster tree contains several cluster regions, each cluster region contains several nodes, and each node represents a patient; S2, based on the clinical indicator data of the patient to be evaluated, using a mapping function to obtain the position coordinates of the patient to be evaluated in the two-dimensional clustering tree; S3, based on the location coordinates of the patient to be evaluated, uses the competing risk model to calculate the probability that the patient to be evaluated will suffer from various complications in the next r years; After S1 and before S2, S1' is also included: S1', after the technicians mark the common features of the clinical indicators of each cluster area, they use the mapping function to obtain the position coordinates of the patients in the verification set in the two-dimensional cluster tree. If the consistency rate of the common features of the clinical indicators of the patients in the verification set and the corresponding cluster areas is greater than the first threshold, the current two-dimensional cluster tree is judged to be successfully verified; otherwise, the verification fails, the current two-dimensional cluster tree is discarded, the number of patients in the first training set is increased, and the process returns to S1; In S1, the following sub-steps are also included: S11, obtaining a feature matrix X based on a first training set, wherein the first training set contains data of m clinical indicators for n patients; the n rows of the feature matrix X represent n patients, and the m columns represent m clinical indicators; both n and m are positive integers; S12, based on the feature matrix X, the optimization objective function F of the DDRTree algorithm is constructed and solved to obtain the key parameters of the two-dimensional clustering tree: ； Among them, the key parameters are the optimal coordinate matrix and the optimal central node set; Y represents the coordinate matrix, which contains n rows and 2 columns. The n rows of the coordinate matrix Y represent n patients, and are the same as the patients represented by the corresponding rows in the feature matrix X. The first column of the coordinate matrix Y represents the horizontal coordinate, and the second column of the coordinate matrix Y represents the vertical coordinate; Y(f) represents the position coordinate corresponding to the fth row in the coordinate matrix Y; W represents the linear transformation parameter matrix in the DDRTree algorithm; represents the transpose of the linear transformation parameter matrix W; λ represents the first hyperparameter; Indicates the position coordinates of the central node i in the current two-dimensional clustering tree; Indicates the position coordinates of the center node j in the current two-dimensional clustering tree; and are all position coordinates in the coordinate matrix Y; ε represents the edge set of the minimum spanning tree in the current two-dimensional clustering tree; Z represents the set of central node position coordinates; α represents the second hyperparameter; Represents the position coordinates of the central node closest to the position coordinate Y(f); · Indicates the coordinate matrix Y and the central node set Z corresponding to the minimum value. At this time, the coordinate matrix Y and the central node set Z are the optimal coordinate matrix and the optimal central node set; represents the square of the 2-norm; S13, constructing a two-dimensional clustering tree based on key parameters; In S1´, the following are also included: S11′, the technicians mark the common features of the clinical indicators of each cluster area according to the clinical indicators of the patients in each cluster area in the current two-dimensional cluster tree; S12´, use the mapping function to obtain the position coordinates of each patient in the validation set in the two-dimensional clustering tree; S13′, determining the clustering region where each patient in the validation set is located according to the position coordinates of each patient in the validation set in the two-dimensional cluster tree; S14′, after determining whether the clinical indicator characteristics of each patient in the validation set are consistent with the common characteristics of the corresponding cluster region, calculate the consistency rate CR = NUM(C) / NUM(All); where NUM (·) represents the number; NUM(C) represents the number of patients in the validation set whose clinical indicator characteristics are consistent with the common characteristics of the corresponding cluster region, and NUM(All) represents the total number of patients in the validation set; If the consistency rate CR is greater than the first threshold, the current two-dimensional clustering tree is determined to be successfully verified; if the consistency rate CR is below the first threshold, the current two-dimensional clustering tree is determined to have failed verification, and the current two-dimensional clustering tree is discarded. A new first training set is constructed or the number of patients in the first training set is increased, and then the process returns to S1. Use machine learning model based on mapping function to obtain the position coordinates of patient G in the two-dimensional clustering tree , ; The mapping function is: ； ； in, The vertical coordinate represents the position coordinate of patient G in the two-dimensional cluster tree; The horizontal coordinate represents the position coordinate of patient G in the two-dimensional cluster tree; represents the first global bias parameter; represents the second global bias parameter; represents the age regression coefficient; represents the gender regression coefficient; represents the age of patient G; represents the gender parameter of patient G; represents the data of the pth clinical indicator of patient G; express kth power of represents the intermediate value of the pth clinical index of patient G, max(·) represents the maximum value, Indicates that the node d in the two-dimensional cluster tree corresponds to the data of the patient's p-th clinical indicator, and the two-dimensional cluster tree includes n nodes in total; In S3, the undiagnosed complications of the patient to be evaluated are evaluated, where u is a positive integer and includes the following: The competing risk model is based on the location coordinates of the patient R to be evaluated, and calculates the probability that the patient will suffer from the Vth complication in the next r years. Probability , where r ≥ 0 and V is a positive integer less than or equal to u: ； ； Where t represents the future t-year of the patient R to be evaluated starting from the current moment; Indicates that the patient R to be evaluated will have complications in the next t years The instantaneous probability of represents the first regression coefficient; represents the second regression coefficient; The ordinate represents the position coordinate of the patient R to be evaluated in the two-dimensional cluster tree; The horizontal coordinate represents the position coordinate of the patient R to be evaluated in the two-dimensional cluster tree; Indicates complications The benchmark sub-distribution risk function that changes with future time t; Before using the competing risk model to assess the probability of a patient developing various complications within the next r years, the competing risk model is trained using the second training set; the second training set includes e training data, each of which corresponds to a patient who has developed more than one complication. Each training data includes the time when the patient was diagnosed with type 2 diabetes, the time when various complications were diagnosed, and the clinical indicator data when the complication was most recently diagnosed; e is a positive integer; The clinical indicator data of patients in the first training set, validation set, and second training set are all cleaned clinical indicator data. The cleaned clinical indicator data also includes the following: Step 1: Extract clinical indicator data from several patients. If a patient has more than one type of dirty data, all clinical indicator data for the current patient will be removed. Dirty data refers to clinical indicator data that is outside the 5 standard deviations of the same clinical indicator data of several patients. Step 2: The remaining clinical indicator data in step 1 are converted into clinical indicator data within the range of 0 to 1 through the rank normalization method. At this time, the clinical indicator data cleaning is completed.

2. A risk assessment system for type 2 diabetes complications, characterized in that: include: Two-dimensional clustering tree construction module, location coordinate mapping module and risk assessment module, The two-dimensional clustering tree construction module includes the DDRTree algorithm, which constructs a two-dimensional clustering tree based on the clinical indicator data of the patients in the first training set, and sends the constructed two-dimensional clustering tree to the position coordinate mapping module; The position coordinate mapping module is used to convert the clinical indicator data of the patient to be evaluated into position coordinates on a two-dimensional clustering tree, and send the position coordinates of the patient to be evaluated to the risk assessment module; The risk assessment module includes a competing risk model, which calculates the probability of the patient suffering from various complications in the next r years based on the patient's location coordinates and outputs it; Each module is configured to execute the steps of a method for risk assessment of type 2 diabetes complications according to claim 1.

3. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program programmed or configured to execute the method for risk assessment of type 2 diabetes complications according to claim 1 .

Images