Machine Learning-Based Chronic Kidney Disease Progression Risk Prediction System

Abstract

This invention relates to the field of intelligent disease risk prediction technology, specifically to a machine learning-based chronic kidney disease (CKD) progression risk prediction system, comprising: a cumulative impact acquisition module for inputting the current patient's disease information and physiological indicator data into a disease risk propagation network to acquire the cumulative impact; an expected eGFR calculation module for inputting the cumulative impact into an eGFR prediction regression network to obtain the expected eGFR curve; a similarity analysis module for calculating the similarity metric between the current patient's eGFR curve and each target historical patient; a risk value calculation module for obtaining a renal function decline risk value based on the expected eGFR curve, the similarity metric, and the eGFR difference between the current patient and each target historical patient; and a risk determination module for determining whether the current patient is at risk of renal function collapse based on the renal function decline risk value. This invention enables early and accurate prediction of CKD progression risk.

Description

Technical Field

[0001] This invention relates to the field of intelligent disease risk prediction technology, specifically to a machine learning-based system for predicting the progression risk of chronic kidney disease. Background Technology

[0002] Chronic kidney disease (CKD) is a long-term, progressive, and complex systemic disease. The deterioration of kidney function does not occur in isolation but is influenced by the complex interactions of multiple systems and factors, including the cardiovascular system, metabolic status, inflammation levels, and drug exposure. Currently, clinical monitoring and risk assessment of CKD progression primarily rely on time-series analysis of single kidney function indicators such as estimated glomerular filtration rate (eGFR), focusing on the rate of decline, fluctuation amplitude, or acceleration of eGFR. In addition, some studies have attempted to apply traditional machine learning models to classify and predict the risk of CKD occurrence or progression based on patients' static clinical indicators (such as age, blood pressure, blood glucose, and proteinuria) at a specific time point.

[0003] However, the existing methods mentioned above have significant limitations. First, they mostly rely on the trend of a single indicator or a set of static indicators, making it difficult to effectively capture and quantify the dynamic and non-linear interactions and risk transmission relationships between indicators of multiple physiological systems such as the heart, kidneys, and metabolism. Second, these methods often fail to identify early warning signals that may appear during disease progression, especially before approaching a clinically significant deterioration or "collapse" threshold. Studies have shown that when a complex life system (such as the homeostasis of the human body) approaches a critical point of functional instability or state transition, the time series of its key physiological parameters often show changes in statistical characteristics, such as increased variance, enhanced autocorrelation, decreased resilience, and increased fluctuation entropy. These phenomena are considered "early collapse signals" indicating reduced system stability and impending mutations. Currently, the detection and analysis of such signals are mostly limited to single time series (such as a single eGFR sequence) and have not yet been able to organically integrate them with risk transmission networks spanning multiple physiological systems. Therefore, existing technologies lack a method that can predict the risk of CKD progression early and accurately from a holistic system perspective. Summary of the Invention

[0004] To address the technical problem that existing technologies cannot accurately predict the risk of CKD progression from a holistic system perspective, the present invention aims to provide a machine learning-based system for predicting the risk of CKD progression. The specific technical solution adopted is as follows: Cumulative Impact Acquisition Module: This module is used to input the current patient's disease information and physiological indicator data into a preset disease risk propagation network to obtain the cumulative impact of the current patient's existing diseases on the kidneys. The projected eGFR calculation module is used to input the cumulative impact into a preset projected eGFR curve acquisition model to obtain the projected eGFR curve of the current patient. Similarity analysis module: used to filter target historical patients with the same cumulative impact as the current patient, and calculate the similarity measure between the eGFR curves of the current patient and each target historical patient using a preset similarity algorithm; Risk value calculation module: used to obtain the risk value of renal function decline of the current patient based on the predicted eGFR curve, the similarity metric value, and the eGFR difference value between the current patient and each target historical patient; Risk assessment module: used to compare the renal function decline risk value with a preset risk threshold to determine whether the current patient is at risk of renal function collapse.

[0005] Furthermore, the process of constructing the disease risk transmission network includes: Obtain disease information and physiological indicator data of a preset number of historical patients; Based on the disease information and the physiological indicator data, target diseases strongly related to kidney disease are screened out, and the physiological indicator data corresponding to the target diseases are extracted. Analyze the causal relationship between any two of the target diseases, and then determine the strength of the causal relationship; The target disease is used as a disease node in the disease risk transmission network, and the collapse signal of the disease node is analyzed. A learnable propagation matrix is ​​constructed based on the prior matrix corresponding to the strength of the causal relationship and the learnable matrix. Using the collapse signal as input, the cumulative impact is obtained by iteratively transmitting the data through the learnable propagation matrix and accumulating the data.

[0006] Furthermore, the process of obtaining the target disease includes: Calculate the correlation between kidney disease and any other disease. If the absolute value of the correlation is greater than a preset correlation threshold, then any other disease is the target disease.

[0007] Furthermore, the process of obtaining the causal relationship includes: Based on the occurrence count of each of the two target diseases, the number of simultaneous occurrences, and the number of secondary occurrences of the other target disease after the occurrence of one target disease, the conditional probability of the existence of the other target disease after the occurrence of the first target disease is determined as the causal relationship.

[0008] Furthermore, the process of obtaining the strength of the causal relationship includes: Obtain the time difference between the onset of two target diseases in historical patients, wherein the time difference is the onset time of one disease minus the onset time of the other disease; The time differences are converted using a negative exponential function to obtain the corresponding negative exponential conversion values. The negative exponential conversion values ​​are then summed to obtain the first summation result. The strength of the causal relationship is obtained based on the first accumulation result and the causal relationship.

[0009] Furthermore, when the disease node corresponds to a type of physiological indicator data, the process of obtaining the collapse signal includes: Calculate the autocorrelation coefficient and variance of the physiological indicator data corresponding to the disease nodes within the preset time window; The collapse signal corresponding to the disease node is obtained based on the autocorrelation coefficient and variance.

[0010] Furthermore, when the disease node corresponds to at least two types of physiological indicator data, the process of obtaining the collapse signal further includes: Calculate the autocorrelation coefficient and variance of each type of physiological indicator data corresponding to the disease node within the preset time window; For the autocorrelation coefficient and the variance, the autocorrelation coefficient and the variance corresponding to various physiological index data are weighted and summed using preset weights to obtain the aggregated autocorrelation coefficient and aggregated variance corresponding to the disease node; The collapse signal corresponding to the disease node is obtained based on the aggregated autocorrelation coefficient and the aggregated variance.

[0011] Furthermore, the process of obtaining the predicted eGFR curve acquisition model includes: The historical patient's eGFR data and the cumulative impact are input into the eGFR prediction regression network to determine the parameters of the regression equation in the eGFR prediction regression network, wherein the regression equation is a linear equation about the cumulative impact and the eGFR data. Based on the eGFR prediction regression network and the parameters of the regression equation, the model for obtaining the predicted eGFR curve is constructed.

[0012] Furthermore, the process of obtaining the risk value for renal function decline includes: Obtain the slope of the predicted eGFR curve for the current patient; The product result is obtained by multiplying each of the eGFR difference values ​​with the reciprocal of its corresponding similarity measure value, and the product results are accumulated to obtain a second accumulated result; Calculate the negative exponential function value of the slope of the predicted eGFR curve, and obtain the risk value of renal function decline based on the negative exponential function value and the second accumulation result.

[0013] Furthermore, determining whether the current patient is at risk of kidney failure includes: When the risk value of renal function decline is greater than the preset risk threshold, the current patient is at risk of renal function collapse; When the risk value of renal function decline is not greater than the preset risk threshold, the current patient does not have a risk of renal function collapse.

[0014] The present invention has the following beneficial effects: The cumulative impact acquisition module can quantify the "impact intensity" of a single disease on the kidneys and the "cumulative impact effect" of multiple diseases on the kidneys along a specific time path, achieving dynamic quantification of the causal cumulative effect between diseases for the first time.

[0015] The predicted eGFR curve generated by the predicted eGFR calculation module can reflect the declining trend of renal function driven by the synergistic effect of multiple system diseases earlier and more sensitively, significantly improving the accuracy and foresight of the prediction.

[0016] The similarity analysis module is performed under the premise of a high degree of consistency in pathophysiological background, so that the output similarity metric value has direct and accurate reference significance for risk assessment.

[0017] The risk value calculation module synthesizes a comprehensive risk value for renal function decline, overcoming the shortcomings of relying solely on its own historical curve or simple comparison with the population mean. The risk assessment module makes decisions based on the risk value of renal function decline, thereby achieving a more reliable early identification of the "risk of renal function collapse".

[0018] In summary, this invention significantly improves the early identification and timely warning of rapidly progressing CKD. Furthermore, by quantifying the cumulative contribution of various diseases and their temporal correlations to renal function, it provides clinical practice with traceable and interpretable risk attribution analysis. This helps to accurately locate high-risk disease targets and key transmission links, thereby guiding targeted early intervention and ultimately improving the evidence-based decision-making and clinical value of the entire management of kidney disease. Attached Figure Description

[0019] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the machine learning-based chronic kidney disease progression risk prediction system provided in the first embodiment of the present invention. Figure 2 A flowchart illustrating the construction process of the disease risk transmission network provided in the second embodiment of the present invention; Figure 3 A flowchart illustrating the process of obtaining the causal relationship strength provided in the third embodiment of the present invention; Figure 4 A flowchart illustrating the process of acquiring a crash signal according to the fourth embodiment of the present invention; Figure 5 Another flowchart illustrating the process of acquiring a crash signal provided in the fifth embodiment of the present invention; Figure 6 This is a flowchart illustrating the process of obtaining the risk value of renal function decline provided in the sixth embodiment of the present invention. Detailed Implementation

[0021] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the machine learning-based chronic kidney disease progression risk prediction system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0023] The specific solution of the machine learning-based chronic kidney disease progression risk prediction system provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Please see Figure 1 The diagram illustrates a machine learning-based chronic kidney disease progression risk prediction system provided in an embodiment of the present invention. The system includes: a cumulative impact acquisition module 101, an estimated eGFR calculation module 102, a similarity analysis module 103, a risk value calculation module 104, and a risk determination module 105.

[0025] Cumulative Impact Acquisition Module 101: This module is used to input the current patient's disease information and physiological indicator data into a preset disease risk propagation network to acquire the cumulative impact of the current patient's existing diseases on the kidneys.

[0026] The disease information can be obtained from the user's electronic medical record, and may include the disease name, disease severity, diagnosis timestamp, etc.

[0027] The physiological indicators can be obtained from the hospital laboratory information system, and include: serum creatinine, eGFR, urine protein, blood glucose, electrolyte concentration, etc.

[0028] In the past, when providing risk warnings for the progression of chronic kidney disease, a moving average algorithm was directly used to perform weighted averaging fusion of eGFR sequences. However, a patient's body is an organic whole composed of multiple organs, and different disease states may have varying degrees of interference and impact on the kidneys. Therefore, when conducting predictions, it is necessary to consider the patient's disease progression to more comprehensively assess its impact on kidney risk.

[0029] When conducting interference analysis on the impact of different disease information on kidney disease, it is necessary to perform statistical analysis on the collected historical patient data to screen out disease information that is truly related to kidney disease. This avoids including irrelevant diseases (such as dermatological or ophthalmic surgical diseases) in the learnable propagation matrix of the disease risk transmission network, thus reducing unnecessary complexity. Relevant pathological information can be screened by analyzing the correlation between disease information and kidney disease. However, it should be noted that simple correlation only indicates an association, not a causal relationship. Therefore, when constructing the learnable propagation matrix, the transmission process of causality must also be considered, and it should be noted that the propagation effect of disease information will not be infinitely amplified; due to the human body's certain self-repair capabilities, the propagation process may stop at some point.

[0030] The patient's current disease information and physiological indicators are input into a preset disease risk propagation network, which outputs the cumulative impact of the patient's existing diseases on the kidneys.

[0031] The process of constructing the disease risk transmission network will be described in detail in the second embodiment, and will not be repeated here.

[0032] The expected eGFR calculation module 102 is used to input the cumulative impact into a preset expected eGFR curve acquisition model to obtain the expected eGFR curve of the current patient.

[0033] Furthermore, the process of obtaining the predicted eGFR curve acquisition model includes: The historical patient's eGFR data and the cumulative impact are input into the eGFR prediction regression network to determine the parameters of the regression equation in the eGFR prediction regression network, wherein the regression equation is a linear equation about the cumulative impact and the eGFR data. Based on the eGFR prediction regression network and the parameters of the regression equation, the model for obtaining the predicted eGFR curve is constructed.

[0034] Using historical patients' eGFR data as the dependent variable and the cumulative impact of historical patients as the independent variable, the following equation can be constructed: ; Among them, the The cumulative impact is represented by K, which represents the slope and B, which represents the intercept. The values ​​of K and B can be determined by regression.

[0035] Similarity analysis module 103: used to filter target historical patients with the same cumulative impact as the current patient, and calculate the similarity measure between the eGFR curves of the current patient and each target historical patient through a preset similarity algorithm.

[0036] Based on the current patient's cumulative impact, all historical patients with the same cumulative impact are retrieved from historical data as target historical patients. The similarity metric is calculated by comparing the predicted eGFR curve of the current patient with the eGFR curves of these target historical patients.

[0037] Optionally, the similarity between the predicted eGFR curve of the current patient and the eGFR curve of the target historical patient can be quantified by Euclidean distance.

[0038] Risk value calculation module 104: used to obtain the risk value of renal function decline of the current patient based on the predicted eGFR curve, the similarity metric value and the eGFR difference value between the current patient and each target historical patient.

[0039] The process of obtaining the risk value of renal function decline will be described in detail in the sixth embodiment, and will not be repeated here.

[0040] Risk assessment module 105: used to compare the renal function decline risk value with a preset risk threshold to determine whether the current patient is at risk of renal function collapse.

[0041] Specifically, determining whether the current patient is at risk of kidney failure includes: When the risk value of renal function decline is greater than the preset risk threshold, the current patient is at risk of renal function collapse; When the risk value of renal function decline is not greater than the preset risk threshold, the current patient does not have a risk of renal function collapse.

[0042] This invention is applicable to various medical information scenarios, including hospital information systems, electronic medical record systems, regional health management platforms, and chronic disease follow-up monitoring systems. By continuously accessing multidimensional data such as patients' laboratory indicators, vital signs, medication use, and complications across different time points, the system can dynamically construct and update the patient's multi-system physiological risk propagation network, automatically extracting early instability signals at the node level and overall network stability indicators, thus achieving early warning of the risk of rapid decline in kidney function. At the clinical level, it can be widely applied in outpatient follow-up, chronic kidney disease management clinics, chronic disease management centers, predialysis risk assessment, and clinical decision support systems. In terms of application expansion, it can also be used in diverse scenarios such as insurance risk actuarial science, individualized patient treatment pathway design, and clinical trial subject screening, improving the accuracy and foresight of disease management.

[0043] Figure 2 The flowchart illustrates the construction process of a disease risk transmission network provided in the second embodiment of the present invention. The construction process of the disease risk transmission network includes: S201. Obtain disease information and physiological indicator data of a preset number of historical patients.

[0044] The number can be set by the user and is not limited here.

[0045] S202. Based on the disease information and the physiological indicator data, target diseases strongly related to kidney disease are screened out, and the physiological indicator data corresponding to the target diseases are extracted.

[0046] Before screening for target diseases strongly associated with kidney disease, some diseases that are clearly unrelated to kidney disease, such as fractures and burns, can be eliminated.

[0047] The process of obtaining the target disease includes: Calculate the correlation between kidney disease and any other disease. If the absolute value of the correlation is greater than a preset correlation threshold, then any other disease is the target disease.

[0048] Optionally, the correlation between kidney disease and any other disease can be calculated using the existing Spearman algorithm. It should be noted that the correlation calculation between two diseases refers to the correlation between the physiological indicator data sequences corresponding to the two diseases. Each disease has its corresponding physiological indicator data; for example, the physiological indicator data corresponding to chronic kidney disease is eGFR time-series data, and the physiological indicator data corresponding to hypertension is blood pressure time-series data. When a disease corresponds to at least two types of physiological indicator data, the Spearman algorithm can be used to calculate the correlation components between the physiological indicator data corresponding to kidney disease and any physiological indicator data of that disease. A corresponding weight is assigned to each correlation component, and a weighted summation method is used to obtain the correlation.

[0049] The correlation threshold can be set by the user, preferably 0.3.

[0050] S203. Analyze the causal relationship between any two of the target diseases, and then determine the strength of the causal relationship.

[0051] Furthermore, the process of obtaining the causal relationship includes: Based on the occurrence count of each of the two target diseases, the number of simultaneous occurrences, and the number of secondary occurrences of the other target disease after the occurrence of one target disease, the conditional probability of the existence of the other target disease after the occurrence of the first target disease is determined as the causal relationship.

[0052] The causal relationship can be expressed by the following formula: ; Among them, the Indicates the first The first time after the occurrence of the target disease The number of secondary infections of the target disease, the Indicates the first Target disease and the first The number of times the target disease occurs simultaneously, the Indicates the first The number of occurrences of the target disease, the Indicates the first The number of occurrences of the target disease, the Indicates the first The target disease in the first The conditional probability of the existence of a target disease is the causal relationship.

[0053] The As a weight in the analysis, the closer the value is to 1, the more significant the temporal relationship between the two target diseases. It should be noted that, to ensure the calculation results are meaningful, in this embodiment of the invention, when performing fractional operations, if the denominator is 0, a parameter adjustment factor greater than 0 needs to be added to the denominator to prevent it from being 0. The value of the parameter adjustment factor is set by the implementer according to the actual situation; in this application, it is set to 0.1.

[0054] The strength of the causal relationship between two target diseases can be determined by the conditional probability and the time difference of onset between them.

[0055] The process of obtaining the strength of the causal relationship will be described in detail in the third embodiment, and will not be repeated here.

[0056] S204. Using the target disease as a disease node in the disease risk transmission network, analyze the collapse signal of the disease node.

[0057] When conducting risk propagation, it's insufficient to simply multiply the edge weights (causal strength) between disease nodes with a causal relationship to calculate the cumulative risk along the path. For example, in the causal network A → B → C, the overall impact of A on C is not equal to the causal strength of A → B multiplied by the causal strength of B → C. This is because each node (disease state) within the patient possesses a certain risk tolerance and adjustment capacity. Therefore, during propagation, collapse signals at each disease node should be introduced to dynamically reflect the state changes and buffering effects of disease nodes during risk transmission, thereby more realistically simulating the transmission and attenuation of risk.

[0058] The process of obtaining the crash signal will be described in detail in the fourth and fifth embodiments, and will not be repeated here.

[0059] S205. Construct a learnable propagation matrix based on the prior matrix corresponding to the strength of the causal relationship and the learnable matrix.

[0060] The learnable propagation matrix can be represented as: ; Among them, the This represents a normalization function, such as a max-min normalization function. The prior matrix representing the strength of the causal relationship, the This refers to an activation function, such as the Sigmoid activation function. Represents the balancing hyperparameters, the The Represents a learnable matrix, the This represents the learnable propagation matrix.

[0061] The learnable propagation matrix is ​​not a fixed matrix, but a learnable, personalized disease impact network.

[0062] The learnable propagation matrix satisfies the constraint that the spectral radius is less than 1, thereby ensuring the stability of the network and preventing the risk from being excessively amplified during disease transmission.

[0063] S206. Using the collapse signal as input, the cumulative impact is obtained by iteratively transmitting the data through the learnable propagation matrix and accumulating the data.

[0064] The crash signal This represents the strength of an early warning signal for a disease node at time t, used to quantify the current instability or activity level of the disease.

[0065] Initial values ​​for iteration: , representing the initial characteristics of the node itself; 1-step dissemination: ; k+1 step propagation: ; Among them, the These are the coefficients for the residual connection, used to prevent information loss during propagation.

[0066] After K-step iterative propagation, the impact of all steps needs to be summarized to obtain the risk impact vector: ; Among them, the As the weight of each step, the This represents the risk impact vector experienced by a node.

[0067] With learnable node sensitivity coefficients Transform the risk impact vector into node-level risk characteristics (scalars): The aforementioned This indicates the cumulative impact situation.

[0068] Figure 3 The flowchart below shows the process for obtaining the causal relationship strength according to the third embodiment of the present invention. The process for obtaining the causal relationship strength includes: S301. Obtain the time difference between the onset of two target diseases in historical patients, wherein the time difference is the onset time of one disease minus the onset time of the other disease.

[0069] The time difference can be expressed as: That is, the first The onset time of the target disease minus the first The onset time of the target disease.

[0070] S302. Convert each of the time differences into a negative exponential function value to obtain the corresponding negative exponential conversion value, and accumulate each of the negative exponential conversion values ​​to obtain a first accumulation result.

[0071] The first accumulated result can be expressed as: , wherein Represented by natural constant An exponential function with base 0, the Indicates the first The first time after the occurrence of the target disease The number of secondary infections of the target disease.

[0072] S303. Obtain the strength of the causal relationship based on the first accumulation result and the causal relationship.

[0073] The strength of the causal relationship can be expressed as: ; Among them, the Indicates the first The target disease in the first The strength of the causal relationship that exists after a target disease.

[0074] Figure 4 The flowchart below illustrates the process of acquiring a collapse signal according to the fourth embodiment of the present invention. When the disease node corresponds to a type of physiological indicator data, the process of acquiring the collapse signal includes: S401. Calculate the autocorrelation coefficient and variance of the physiological indicator data corresponding to the disease nodes within the preset time window.

[0075] The autocorrelation coefficient of the physiological indicator data corresponding to the disease node within the preset time window can be expressed as: .

[0076] The variance of the physiological indicator data corresponding to the disease node within the preset time window can be expressed as: .

[0077] The time window can be set independently according to user needs, preferably in a size of 4×1, and the physiological index data within the time window is the latest physiological index data of historical patients.

[0078] S402. Obtain the collapse signal corresponding to the disease node based on the autocorrelation coefficient and variance.

[0079] The collapse signal can be represented as: ; Among them, the This indicates the crash signal.

[0080] Figure 5 Another flowchart of the process for obtaining the collapse signal provided in the fifth embodiment of the present invention, wherein when the disease node corresponds to at least two types of physiological indicator data, the process for obtaining the collapse signal further includes: S501. Calculate the autocorrelation coefficient and variance of each type of physiological indicator data corresponding to the disease node within the preset time window.

[0081] The autocorrelation coefficient of the physiological indicator data corresponding to the disease node within the preset time window can be expressed as: , wherein This indicates the number of types of physiological indicator data.

[0082] The variance of the physiological indicator data corresponding to the disease node within the preset time window can be expressed as: .

[0083] The time window can be set independently according to user needs, preferably in a size of 4×1, and the physiological index data within the time window is the latest physiological index data of historical patients.

[0084] S502. For the autocorrelation coefficient and the variance, respectively, the autocorrelation coefficient and the variance corresponding to various physiological index data are weighted and summed using preset weights to obtain the aggregated autocorrelation coefficient and aggregated variance corresponding to the disease node.

[0085] The aggregation autocorrelation coefficient can be expressed as: ,in, The The above , The above All are non-negative numbers.

[0086] The aggregation variance can be expressed as: ,in, The The above , The above All are non-negative numbers.

[0087] S503. Obtain the collapse signal corresponding to the disease node based on the aggregated autocorrelation coefficient and the aggregated variance.

[0088] The collapse signal can be represented as: ; Among them, the This indicates the crash signal.

[0089] Figure 6 This is a flowchart of the process for obtaining the risk value of renal function decline provided in the sixth embodiment of the present invention. The process for obtaining the risk value of renal function decline includes: S601. Obtain the slope of the predicted eGFR curve for the current patient.

[0090] The slope can be used To express.

[0091] S602. Multiply each of the eGFR difference values ​​by the reciprocal of its corresponding similarity measure value to obtain a product result, and sum the product results to obtain a second accumulated result.

[0092] The second accumulated result can be expressed as: ; Among them, the The eGFR difference value represents the arithmetic mean of the differences between the current patient's predicted eGFR curve and the target historical patient's eGFR curve at corresponding points in time. The similarity metric is represented by M, where M represents the number of target historical patients.

[0093] S603. Calculate the negative exponential function value of the slope of the predicted eGFR curve, and obtain the risk value of renal function decline based on the negative exponential function value and the second accumulation result.

[0094] ; Among them, the The risk value for renal function decline is indicated by the value of the renal function decline. Represents the value of the negative exponential function, the Represented by natural constant An exponential function with base 0.

[0095] The present invention has the following beneficial effects: The cumulative impact acquisition module can quantify the "impact intensity" of a single disease on the kidneys and the "cumulative impact effect" of multiple diseases on the kidneys along a specific time path, achieving dynamic quantification of the causal cumulative effect between diseases for the first time.

[0096] The predicted eGFR curve generated by the predicted eGFR calculation module can reflect the declining trend of renal function driven by the synergistic effect of multiple system diseases earlier and more sensitively, significantly improving the accuracy and foresight of the prediction.

[0097] The similarity analysis module is performed under the premise of a high degree of consistency in pathophysiological background, so that the output similarity metric value has direct and accurate reference significance for risk assessment.

[0098] The risk value calculation module synthesizes a comprehensive risk value for renal function decline, overcoming the shortcomings of relying solely on its own historical curve or simple comparison with the population mean.

[0099] The risk assessment module makes decisions based on the risk value of renal function decline, thereby achieving a more reliable early identification of the "risk of renal function collapse".

[0100] In summary, this invention significantly improves the early identification and timely warning of rapidly progressing CKD. Furthermore, by quantifying the cumulative contribution of various diseases and their temporal correlations to renal function, it provides clinical practice with traceable and interpretable risk attribution analysis. This helps to accurately locate high-risk disease targets and key transmission links, thereby guiding targeted early intervention and ultimately improving the evidence-based decision-making and clinical value of the entire management of kidney disease.

[0101] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0102] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A machine learning-based system for predicting the risk of progression of chronic kidney disease, characterized in that, The system includes: Cumulative Impact Acquisition Module: This module is used to input the current patient's disease information and physiological indicator data into a preset disease risk propagation network to obtain the cumulative impact of the current patient's existing diseases on the kidneys. The projected eGFR calculation module is used to input the cumulative impact information into a preset projected eGFR curve acquisition model to obtain the projected eGFR curve for the current patient. Similarity analysis module: used to screen target historical patients with the same cumulative impact as the current patient, and calculate the similarity measure between the eGFR curves of the current patient and each target historical patient using a preset similarity algorithm; Risk value calculation module: used to obtain the risk value of renal function decline of the current patient based on the predicted eGFR curve, the similarity metric value, and the difference value of eGFR between the current patient and each target historical patient; Risk assessment module: Used to compare the risk value of renal function decline with a preset risk threshold to determine whether the current patient is at risk of renal function collapse; The process of constructing a disease risk transmission network includes: Obtain disease information and physiological indicator data of a preset number of historical patients; Based on disease information and the physiological indicator data, target diseases strongly related to kidney disease are screened out, and physiological indicator data corresponding to the target diseases are extracted. Analyze the causal relationship between any two target diseases, and then determine the strength of the causal relationship; The target disease is used as a disease node in the disease risk transmission network, and the collapse signal of the disease node is analyzed. A learnable propagation matrix is ​​constructed by combining the prior matrix corresponding to the strength of causal relationship with the learnable matrix. Using the collapse signal as input, the data is iteratively transmitted through a learnable propagation matrix and accumulated to obtain the cumulative impact situation. The process of obtaining causal relationships includes: Based on the occurrence number of each of the two target diseases, the simultaneous occurrence number, and the secondary occurrence number of the other target disease after the occurrence of one target disease, the conditional probability of the existence of the other target disease after the occurrence of one target disease is determined as the causal relationship. The process of obtaining the strength of causal relationship includes: Obtain the time difference between the onset of two target diseases in historical patients, where the time difference is the onset time of one disease minus the onset time of the other disease; The time differences are converted using a negative exponential function to obtain the corresponding negative exponential conversion values. The negative exponential conversion values ​​are then summed to obtain the first summation result. The strength of the causal relationship is obtained based on the first accumulation result and the causal relationship. When a disease node corresponds to a type of physiological indicator data, the process of obtaining the collapse signal includes: calculating the autocorrelation coefficient and variance of the physiological indicator data corresponding to the disease node within a preset time window. The collapse signal corresponding to the disease node is obtained based on the autocorrelation coefficient and variance; When a disease node corresponds to at least two types of physiological indicator data, the process of obtaining the collapse signal also includes: Calculate the autocorrelation coefficient and variance of each type of physiological indicator data corresponding to disease nodes within the preset time window; For the autocorrelation coefficient and variance, the autocorrelation coefficient and variance of various physiological index data are weighted and summed using preset weights to obtain the aggregated autocorrelation coefficient and aggregated variance of the disease node. Collapse signals corresponding to disease nodes are obtained based on aggregated autocorrelation coefficients and aggregated variances; The process of obtaining the risk value for renal function decline includes: Obtain the slope of the predicted eGFR curve for the current patient; The product of each eGFR difference value and the inverse of its corresponding similarity measure value is obtained, and the product results are summed to obtain the second summed result. Calculate the negative exponential function value of the slope of the predicted eGFR curve, and obtain the risk value of renal function decline based on the negative exponential function value and the second summation result; The learnable propagation matrix satisfies the constraint that the spectral radius is less than 1.

2. The machine learning-based chronic kidney disease progression risk prediction system as described in claim 1, characterized in that, The process of obtaining the target disease includes: Calculate the correlation between kidney disease and any other disease. If the absolute value of the correlation is greater than a preset correlation threshold, then any other disease is the target disease.

3. The machine learning-based chronic kidney disease progression risk prediction system as described in claim 1, characterized in that, The process of obtaining the model for predicting the eGFR curve includes: The historical patient's eGFR data and the cumulative impact are input into the eGFR prediction regression network to determine the parameters of the regression equation in the eGFR prediction regression network, wherein the regression equation is a linear equation about the cumulative impact and the eGFR data. Based on the eGFR prediction regression network and the parameters of the regression equation, the model for obtaining the predicted eGFR curve is constructed.

4. The machine learning-based chronic kidney disease progression risk prediction system as described in claim 1, characterized in that, The determination of whether a patient is at risk of kidney failure includes: When the risk value of renal function decline is greater than the preset risk threshold, the current patient is at risk of renal function collapse; When the risk value of renal function decline is not greater than the preset risk threshold, the current patient does not have a risk of renal function collapse.

Images