A vascular fistula status prediction method, system, device, and medium

Abstract

The application relates to a vascular fistula state prediction method, system, device and medium. The method comprises the following steps: acquiring continuous dynamic blood glucose monitoring data of a target patient in a preset monitoring period; performing standardization preprocessing on the continuous dynamic blood glucose monitoring data to generate blood glucose concentration time series data; extracting a multi-dimensional dynamic injury dose feature vector of the target patient based on the blood glucose concentration time series data; inputting the dynamic injury dose feature vector and the static clinical feature variable of the target patient into a trained fistula failure risk prediction model to output the risk probability of fistula failure of the target patient in a preset future time window. By using the method, dynamic and quantitative evaluation of the blood glucose fluctuation, an invisible risk source, can be realized, and the blood glucose fluctuation is successfully embedded into a clinical prediction framework, the limitations of traditional dependence on static average indexes are effectively overcome, and the fistula risk early warning can be jumped from reflecting a long-term average state to capturing a real-time and cumulative biological damage process.

Description

Technical Field

[0001] This invention belongs to the field of arteriovenous fistula status prediction technology, and in particular relates to a method, system, device and medium for predicting the status of arteriovenous fistulas. Background Technology

[0002] With the continuous advancement and widespread adoption of blood purification technology, maintenance hemodialysis has become the primary treatment for patients with end-stage renal disease. Establishing and maintaining a functional vascular access is a crucial prerequisite for ensuring adequate dialysis. Among these options, autogenous arteriovenous fistulas (AVFs) are recommended by clinical guidelines as the preferred permanent vascular access due to their long lifespan and relatively few complications. To ensure long-term patency of the fistula, continuous monitoring and risk assessment of the fistula's condition are necessary in clinical practice to detect problems early and intervene accordingly.

[0003] Traditional prediction and risk assessment of arteriovenous fistula (AVF) status primarily relies on the regular monitoring and analysis of a range of macroscopic, static clinical and biochemical indicators. These indicators typically include glycated hemoglobin (HbA1c) reflecting long-term average glycemic control, fasting blood glucose levels, and baseline characteristics such as the patient's age, duration of diabetes, blood pressure, lipid profile, and duration of AVF use. Clinicians or predictive models use these parameters to attempt to determine the patient's risk of AVF failure events such as stenosis or thrombosis.

[0004] However, current prediction methods based on static indicators have significant limitations. The core problem is that they completely fail to capture and quantify the dynamic, cumulative microscopic damage induced by drastic fluctuations in blood glucose at the vascular endothelial cell level. Pathophysiological studies have confirmed that, compared to simple hyperglycemia, sharp fluctuations in blood glucose trigger stronger oxidative stress and inflammatory responses, which are crucial mechanisms leading to endothelial dysfunction and subsequently promoting intimal hyperplasia and stenosis in arteriovenous fistulas. Traditional static indicators only reflect the average blood glucose level over the past 2-3 months, failing to reveal the amplitude and frequency of intraday and interday blood glucose fluctuations. Therefore, they cannot assess the persistent, gradual damage caused by this volatility. This makes it difficult for existing prediction models to distinguish the substantial differences in arteriovenous fistula risk among patients with similar average blood glucose levels but drastically different blood glucose stability, thus affecting the accuracy and timeliness of risk warnings. Summary of the Invention

[0005] Therefore, it is necessary to provide a method for predicting the status of arteriovenous fistulas that can integrate the cumulative effect of microscopic damage caused by dynamic fluctuations in blood glucose levels, thereby achieving a more accurate and forward-looking risk assessment.

[0006] In a first aspect, this application provides a method for predicting the status of arteriovenous fistulas, including: Acquire continuous dynamic blood glucose monitoring data of the target patient within a preset monitoring period; perform standardized preprocessing on the continuous dynamic blood glucose monitoring data to generate blood glucose concentration time series data; Based on blood glucose concentration time series data, multidimensional dynamic injury dose characteristics of target patients are obtained; among which, multidimensional dynamic injury dose characteristics include baseline total characteristics, injury variability characteristics, injury rhythm characteristics, and injury peak characteristics. Combine multi-dimensional dynamic damage dose features to generate a dynamic damage dose feature vector; The dynamic injury dose feature vector and the static clinical feature variables of the target patient are input into the trained arteriovenous fistula failure risk prediction model, and the output is the risk probability of the target patient experiencing arteriovenous fistula failure within a preset future time window; where the static clinical feature variables are stable patient baseline features.

[0007] Furthermore, based on blood glucose concentration time series data, multidimensional dynamic damage dose characteristics of the target patient are obtained, including: Based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations and blood glucose concentration time series data, the instantaneous endothelial injury rate data were calculated. Within the time range of the preset monitoring period, the instantaneous endothelial injury rate data are numerically integrated to generate the cumulative injury dose; wherein, the cumulative injury dose is used to characterize the total load borne by intravascular cells within the preset monitoring period. Based on transient endothelial injury rate data and cumulative injury dose, multidimensional dynamic injury dose characteristics of the target patient are generated.

[0008] Furthermore, based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations and blood glucose concentration time-series data, transient endothelial injury rate data were calculated, including: Based on blood glucose concentration time series data, the rate of change of blood glucose and the degree of deviation of blood glucose at each sampling time point are calculated, and based on each rate of change of blood glucose and each degree of deviation of blood glucose, a blood glucose change rate sequence and a blood glucose deviation sequence are generated. Based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations, a damage integral kernel function is constructed. The specific expression of the damage integral kernel function is as follows: ; in, For the damage integral kernel function, For time delay variables; This is an individualized damage amplitude coefficient; Weights for fast response components; The fast response time constant; The slow response time constant; It is a unit step function; Based on the blood glucose change rate sequence, blood glucose deviation sequence, and damage activation function, the damage activation intensity at each sampling time point is calculated, and a damage activation intensity sequence is generated based on each damage activation intensity; wherein, the damage activation intensity is proportional to the power of the absolute value of the blood glucose change rate. The instantaneous endothelial damage rate data are obtained by convolving the damage excitation intensity sequence and the damage integral kernel function.

[0009] Furthermore, based on transient endothelial injury rate data and cumulative injury dose, multidimensional dynamic injury dose characteristics of the target patient are generated, including: The cumulative damage dose is determined as the baseline total dose characteristic; Calculate the standard deviation and coefficient of variation of instantaneous endothelial injury rate data within a preset monitoring period, and determine the standard deviation and coefficient of variation as the characteristics of injury variability; The day is divided into multiple preset time periods. The proportion of damage dose in each preset time period to the total daily damage dose is calculated, and the proportion is determined as the damage rhythm characteristic. The number of peaks, average peak height, and total duration of instantaneous endothelial injury rate exceeding a preset injury rate threshold are statistically analyzed, and the number of peaks, average peak height, and total duration are determined as injury peak characteristics.

[0010] Furthermore, based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations, after constructing the damage integral kernel function, it also includes: Obtain blood samples from target patients and detect the concentrations of biomarkers associated with endothelial injury and oxidative stress in the blood samples; Based on a pre-defined calibration mapping relationship, the individualized damage amplitude coefficient in the damage integral kernel function is calibrated according to the concentration of biomarkers, resulting in calibrated individualized damage amplitude coefficients. These calibrated individualized damage amplitude coefficients are used to calculate the damage integral kernel function.

[0011] Furthermore, the trained risk prediction model is obtained through the following method: Longitudinal data from historical patient cohorts were collected, and multiple observation records were constructed for each patient based on the longitudinal data. The observation records included dynamic injury dose feature vectors, static clinical feature variables, monitoring period time intervals, fistula failure event status, and fistula failure event time information for the corresponding monitoring period. The time-varying Cox proportional hazards model is used as the basic framework for the arteriovenous fistula (AVF) failure risk prediction model; the hazard function of the AVF failure risk prediction model is: ; in, In time The risk of arteriovenous fistula failure. For time, For the baseline risk function, In time The corresponding dynamic damage dose feature vector, This is a vector of static clinical characteristic variables. Let be the first coefficient vector to be estimated. Let the second coefficient vector be the one to be estimated. Based on observation records, the arteriovenous fistula failure risk prediction model was trained to obtain a well-trained arteriovenous fistula failure risk prediction model.

[0012] Secondly, this application also provides a system for predicting the status of arteriovenous fistulas, comprising: The data acquisition module is used to acquire continuous dynamic blood glucose monitoring data of the target patient within a preset monitoring period; and to perform standardized preprocessing on the continuous dynamic blood glucose monitoring data to generate blood glucose concentration time series data. The feature extraction module is used to obtain multi-dimensional dynamic injury dose characteristics of the target patient based on blood glucose concentration time series data; among which, the multi-dimensional dynamic injury dose characteristics include baseline total characteristics, injury variability characteristics, injury rhythm characteristics, and injury peak characteristics; The vector generation module is used to combine multi-dimensional dynamic damage dose features to generate dynamic damage dose feature vectors. The state prediction module is used to input the dynamic damage dose feature vector and the static clinical feature variables of the target patient into the trained arteriovenous fistula failure risk prediction model, and output the risk probability of the target patient experiencing arteriovenous fistula failure within a preset future time window; wherein, the static clinical feature variables are stable patient baseline features.

[0013] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement any of the arteriovenous fistula state prediction methods described in the embodiments of this application.

[0014] Fourthly, this application also provides a computer-readable storage medium storing at least one piece of program code, which is loaded and executed by a processor to implement a method for predicting the state of arteriovenous fistulas as described in any of the embodiments of this application.

[0015] The aforementioned method, system, device, and medium for predicting arteriovenous fistula (AVF) status acquires continuous dynamic blood glucose monitoring data of a target patient within a preset monitoring period and performs standardized preprocessing to generate time-series blood glucose concentration data. Multidimensional dynamic damage dose characteristics of the target patient are extracted from this data. These dynamic damage dose characteristics and the target patient's static clinical characteristics are then input into a trained AVF failure risk prediction model, outputting the probability of AVF failure within a preset future time window. This enables dynamic and quantitative assessment of the hidden risk source of blood glucose fluctuations and successfully embeds it into a clinical prediction framework, effectively overcoming the limitations of traditional methods relying on static average indicators. This allows AVF risk warning to leap from reflecting long-term average states to capturing real-time, cumulative biological damage processes. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a method for predicting the state of an arteriovenous fistula in one embodiment. Figure 2 This is a flowchart illustrating the steps of obtaining multi-dimensional dynamic damage dose characteristics of a target patient based on blood glucose concentration time series data in one embodiment. Figure 3 This is a schematic diagram of the structure of an arteriovenous fistula status prediction system in one embodiment. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] In one embodiment, a method for predicting the state of arteriovenous fistulas is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. Figure 1 As shown, in this embodiment, the method includes the following steps: Step S101: Obtain continuous dynamic blood glucose monitoring data for the target patient within a preset monitoring period; perform standardized preprocessing on the continuous dynamic blood glucose monitoring data to generate blood glucose concentration time series data.

[0020] Among them, the preset monitoring period refers to a complete monitoring period with clinical assessment significance set in advance, such as 14 or 28 consecutive days, to ensure that diverse blood glucose fluctuation patterns, including dialysis days and non-dialysis days, can be captured; standardized preprocessing refers to eliminating interference caused by differences in dimensions / scales of data, so that different features or variables are in a unified dimension that can be fairly compared and calculated; the target patient is the patient whose arteriovenous fistula status is to be assessed.

[0021] For example, raw monitoring data collected by a continuous glucose monitor (CGM) worn by the target patient is acquired. This data covers a preset monitoring period, and the acquired raw monitoring data is a discrete point sequence containing timestamps and corresponding blood glucose concentration values. Data cleaning of the raw monitoring data can be performed to remove obvious sensor anomalies based on physiologically reasonable thresholds (e.g., 2.0-30.0 mmol / L); transient spike noise can be identified and filtered out using a sliding window-based standard deviation method to obtain cleaned data. Linear interpolation can then be used to resample the cleaned data into equally spaced sequences at fixed time intervals (e.g., 5 minutes), ultimately yielding clean, continuous blood glucose concentration time series data, denoted as . ,in, Equal time intervals; CGM is a medical device for continuous, real-time monitoring of blood glucose levels. Its core advantage lies in overcoming the limitations of traditional finger-prick blood testing with single-point sampling, providing more comprehensive blood glucose fluctuation data for diabetic patients or those requiring blood glucose management, thus assisting in precise blood glucose control; Data cleaning is the core step in data preprocessing, aiming to address issues in the raw data and improve data quality; Physiologically reasonable threshold ranges refer to the upper and lower limits of the numerical range of various physiological indicators such as blood pressure, blood glucose, body temperature, and heart rate under normal physiological conditions to maintain homeostasis and ensure normal organ function. Exceeding or falling below these ranges may indicate abnormal physiological function or even lead to health risks; The standard deviation method based on sliding windows identifies anomalies through local data statistical characteristics and then replaces the anomalies with local reasonable values; Sliding window is an efficient algorithm commonly used for linear data structures such as arrays and strings, through... Maintaining a window (i.e., a continuous sub-interval in the data) and sliding this window across the data while dynamically adjusting its size or position; the standard deviation method is a core method in statistics used to measure the dispersion of data. It reflects the dispersion or central tendency of data by calculating the average deviation of data from the mean; transient spike noise refers to abnormal fluctuations in a signal that appear suddenly, last for a very short time (such as milliseconds), and have an amplitude much higher than the normal signal. Its characteristics are suddenness, isolation, and abnormal amplitude; linear interpolation is a numerical estimation method based on linear relationships. When the precise coordinates or data of two points are known, it assumes that the change of variables between these two points follows a linear trend, thereby estimating the unknown value at any position between the two points; resampling refers to adjusting the sampling frequency or sample density of data. Simply put, it makes the intervals of data in the time / space dimension denser (upsampling) or sparser (downsampling).

[0022] Step S102: Based on blood glucose concentration time series data, obtain multi-dimensional dynamic injury dose characteristics of the target patient; wherein, the multi-dimensional dynamic injury dose characteristics include baseline total characteristics, injury variability characteristics, injury rhythm characteristics, and injury peak characteristics.

[0023] For example, transient endothelial injury rate data is calculated by combining time-series blood glucose concentration data and the biological response characteristics of vascular endothelial cells to blood glucose fluctuations. Within a preset monitoring period, the transient endothelial injury rate data is numerically integrated to generate a cumulative injury dose characterizing the total load borne by vascular cells within the preset monitoring period. Based on the transient endothelial injury rate data and the cumulative injury dose, the baseline total injury characteristics, injury variability characteristics, injury rhythm characteristics, and injury peak characteristics of the target patient are generated.

[0024] Step S103: Combine multi-dimensional dynamic damage dose features to generate a dynamic damage dose feature vector.

[0025] For example, the basic total characteristic representing the total damage, the damage fluctuation characteristic describing the stability of the damage process, the damage rhythm characteristic reflecting the temporal regularity of damage occurrence, and the damage peak characteristic characterizing the intensity of acute injury events are arranged and spliced ​​in a preset order. This transforms the complex, time-dependent dynamic process of damage within a monitoring cycle into a structured, fixed-dimensional mathematical expression, namely, a dynamic damage dose feature vector, denoted as... For example, a feature vector can be constructed whose elements are, in order: cumulative damage dose value, damage rate standard deviation, damage rate coefficient of variation, nighttime damage percentage, morning damage percentage, afternoon damage percentage, evening damage percentage, peak count, and average peak height. The preset order refers to the pre-defined arrangement of the features when combined into a vector.

[0026] Step S104: Input the dynamic damage dose feature vector and the static clinical feature variables of the target patient into the trained arteriovenous fistula failure risk prediction model, and output the risk probability of the target patient experiencing arteriovenous fistula failure within a preset future time window; wherein, the static clinical feature variables are stable patient baseline features.

[0027] For example, a pre-trained arteriovenous fistula failure risk prediction model is invoked. This model is built on a time-varying Cox proportional hazards survival analysis framework and can handle covariates updated over time. The generated dynamic injury dose feature vector is then used. As the core time-varying covariate, a set of static clinical characteristic variables that do not change or change slowly over time are also extracted from the patient's electronic medical record system. For example, factors such as age, duration of diabetes, baseline glycated hemoglobin, fistula location, and smoking history are input into the prediction model. The prediction model calculates the cumulative risk function value for the target patient within a given future time window (e.g., the next 12 months) based on its learned parameters (i.e., the weighting of each feature's contribution to the risk rate). This calculated value is directly converted into an intuitive risk probability, such as "the probability of fistula failure in the next year is X%". The electronic medical record system is a core information system used by medical institutions for the digital storage, management, retrieval, and sharing of patient medical information. It replaces traditional paper medical records, achieving standardized and efficient management of medical data.

[0028] In this embodiment, the raw blood glucose fluctuation signal is transformed into dynamic endothelial injury quantitative features based on physiological mechanisms. These patterned features representing microscopic cumulative damage are fused with macroscopic static clinical indicators, and a pre-trained, fully-developed arteriovenous fistula (AVF) failure risk prediction model is used to calculate the individualized future failure risk. This enables dynamic and quantitative assessment of blood glucose variability, a hidden risk source, and successfully embeds it into the clinical prediction framework. It effectively overcomes the limitations of traditional methods that rely on static average indicators, allowing AVF risk warning to leap from reflecting long-term average states to capturing real-time, cumulative biological damage processes. This enables earlier and more accurate identification of high-risk patients, providing a quantitative basis for proactive intervention and optimized patient management.

[0029] In one embodiment, such as Figure 2 As shown, based on blood glucose concentration time series data, multidimensional dynamic damage dose characteristics of the target patient are obtained, including: Step S201: Based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations and blood glucose concentration time series data, the instantaneous endothelial injury rate data is calculated.

[0030] Among them, vascular endothelial cells are important sensing and response units for blood glucose fluctuations. Their biological response to blood glucose fluctuations has three core characteristics: immediacy, bidirectionality, and cumulative damage. Immediate metabolic response means that they can quickly adapt to changes in glucose supply. Bidirectional functional regulation means that they can compensate for and protect against normal fluctuations (physiological fluctuations, such as slight postprandial hyperglycemia) and activate against abnormal fluctuations (pathological blood glucose fluctuations, such as alternating postprandial hyperglycemia and nocturnal hypoglycemia in diabetic patients). The cumulative nature of long-term response means that endothelial cells have a certain ability to repair damage from a single blood glucose fluctuation (such as clearing damaged organelles through autophagy), but long-term repeated blood glucose fluctuations will lead to an imbalance between damage and repair. The essence of endothelial cell response to blood glucose fluctuations is a dynamic balance between adapting to physiological needs and resisting pathological damage. The frequency, amplitude, and duration of abnormal fluctuations are the key factors that determine whether they shift from compensatory protection to damage and disability.

[0031] For example, receiving preprocessed, time-series blood glucose concentration data at equal time intervals. Based on the understanding of the pathophysiological response of vascular endothelial cells under blood glucose fluctuations, the transient endothelial injury rate data were calculated.

[0032] Step S202: Within the time range of the preset monitoring period, the instantaneous endothelial injury rate data is numerically integrated to generate the cumulative injury dose; wherein, the cumulative injury dose is used to characterize the total load borne by intravascular cells within the preset monitoring period.

[0033] Numerical integration refers to the process of accumulating the "rate-time" surface over multiple small time intervals to approximate the total result, since instantaneous rate is discrete data that changes with time and is not a continuous mathematical formula.

[0034] For example, based on transient endothelial injury rate data that fully cover a preset monitoring period T (e.g., 14 consecutive days). This data is a continuous function or high-density discrete sequence that corresponds one-to-one with the time points of the input blood glucose sequence. Its values ​​directly reflect the immediate intensity of endothelial biological damage induced by blood glucose fluctuations at each time point. Numerical integration is performed over the entire monitoring period T in the time domain: an approximate calculation can be performed using a discrete summation method, i.e., according to a fixed time step. All time points within period T Corresponding damage rate value The formula is to multiply the time step by the time step and then sum them up. Through this integral operation, the damage rate curve, which fluctuates over time, is condensed into a single scalar value, namely the cumulative damage dose. This dose, in a pathophysiological sense, is equivalent to the total accumulated biological damage suffered by vascular endothelial cells due to oxidative stress and inflammatory responses caused by blood glucose fluctuations during the monitoring period T; that is, the total load. Discrete summation is a mathematical method for calculating the sum of all or some terms of discrete data or discrete functions, where variables take the value of isolated points, such as integers or countable sequences. It is used to solve the problem of accumulating non-continuous data and is distinct from integral operations for continuous functions. Approximate calculation is a calculation method that obtains an approximate result close to the true value through simplification, estimation, or truncation in scenarios where precise results are not needed or cannot be obtained.

[0035] Step S203: Based on the transient endothelial injury rate data and cumulative injury dose, generate multidimensional dynamic injury dose characteristics of the target patient.

[0036] For example, by analyzing transient endothelial injury rate data and cumulative damage dose Multi-dimensional feature extraction was performed to obtain basic total features, injury fluctuation features, injury rhythm features, and injury peak features. Among them, multi-dimensional feature extraction refers to the process of mining multiple key information (i.e., features) from different angles and types from the raw data. The core is to break through the limitations of single-dimensional information and more comprehensively characterize the essence of the data. Injury fluctuation features are used to describe the stability or degree of violent fluctuations in the occurrence of injury; injury rhythm features are used to reveal whether the injury occurs in a concentrated period of specific physiological or treatment time; and injury peak features are used to quantify the acute onset pattern of injury.

[0037] In this embodiment, the instantaneous damage rate is calculated based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations. This rate is then quantified into a total load through time integration, and multi-dimensional pattern features are extracted from the rate curve and total load. This innovative approach constructs a complete computational chain from blood glucose fluctuation signals to the quantification of cumulative endothelial damage and the characterization of dynamic damage patterns. It enables dynamic, multi-dimensional, and quantifiable assessment of previously unobservable microscopic damage caused by blood glucose fluctuations. This provides subsequent risk prediction models with an innovative predictive factor that far surpasses traditional static blood glucose indicators, precisely reflecting the true biological pressure state and change patterns of individual arteriovenous fistulas.

[0038] In one embodiment, based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations and blood glucose concentration time-series data, transient endothelial injury rate data are calculated, including: Step S301: Based on the blood glucose concentration time series data, calculate the blood glucose change rate and blood glucose deviation at each sampling time point, and generate a blood glucose change rate sequence and a blood glucose deviation sequence based on each blood glucose change rate and each blood glucose deviation.

[0039] For example, receiving blood glucose concentration time series data , Represents a fixed time interval Distributed discrete sampling time points: for each time point except the starting point Through formula Calculate its instantaneous rate of change. This value quantifies the rate of increase or decrease of blood glucose within a unit of time; a positive value represents an increase in blood glucose, and a negative value represents a decrease. By iterating through all time points, a blood glucose change rate sequence corresponding to the time points of the original sequence is generated. Simultaneously, it calculates blood glucose deviation: a personalized blood glucose target reference value is preset or calculated based on the target patient's data. For example, the median blood glucose level during the monitoring period of the target patient or a clinically set personalized target can be used, through a formula. For each time point, calculate the difference between its blood glucose concentration and the target reference value. By iterating through all time points, a difference sequence corresponding to the time points of the original sequence is generated. The blood glucose deviation sequence is used to characterize the magnitude and direction of the deviation of blood glucose levels from the ideal steady state at each time point. Positive deviation indicates hyperglycemia, and negative deviation indicates hypoglycemia.

[0040] Step S302: Based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations, a damage integral kernel function is constructed. The specific expression of the damage integral kernel function is as follows:

[0041] in, For the damage integral kernel function, For time delay variables; This is an individualized damage amplitude coefficient; Weights for fast response components; The fast response time constant; The slow response time constant; It is a unit step function.

[0042] Among them, the fast response time constant and slow response time constant These are preset parameters, used to characterize the characteristic timescales of rapid activation of oxidative stress and the characteristic timescales of sustained release of inflammatory mediators, respectively; weighting coefficients. The relative contribution of the fast and slow response components can be adjusted based on pathophysiological knowledge or fitting of previous data; unit step function. It is a piece of program logic used to ensure causality, that is, the response only occurs after the triggering event. Individualized damage amplitude coefficient It is an adjustable scaling factor used to calibrate the overall difference in the intensity of the response of endothelial cells in different patients to the same blood glucose fluctuations. Its value can be personalized based on the patient's blood biomarker levels.

[0043] For example, based on the specific pathophysiological response patterns exhibited by vascular endothelial cells when encountering blood glucose fluctuations, a mathematical model, namely the damage integral kernel function, is constructed and stored. The construction of this function is based on the biological understanding that the stress response of endothelial cells to a blood glucose fluctuation event is not instantaneous and disappears immediately, but rather includes a rapid initial burst phase and a subsequent slow decay phase. Therefore, a weighted sum of a double exponential decay function can be used to simulate this two-phase response characteristic. The specific expression of the function is as follows: This kernel function The physical meaning is to describe a unit-intensity blood glucose perturbation occurring at time zero, in the future time... The distribution profile of the intensity of the endothelial injury response.

[0044] Step S303: Based on the blood glucose change rate sequence, blood glucose deviation sequence, and damage activation function, calculate the damage activation intensity at each sampling time point, and generate a damage activation intensity sequence based on each damage activation intensity; wherein, the damage activation intensity is proportional to the power of the absolute value of the blood glucose change rate.

[0045] The damage excitation function takes the following form: The design principle of this function is based on a pathophysiological mechanism: the rate of change in blood glucose is a key driving force for triggering oxidative stress response, and the function takes the absolute value of the rate of change. And apply a power greater than 1. This model is used to reflect the nonlinearly amplified damaging effect of rapid blood glucose changes (whether rising or falling) compared to slow changes; the degree to which blood glucose deviates from the ideal state determines the baseline level of damage or background enhancement, mediated by a sigmoid function. To handle blood sugar deviation This function maps the deviation to a weight value between 0 and 1, thereby simulating the different contribution characteristics of hyperglycemia and hypoglycemia to injury activation. For example, the activation weight of extreme hyperglycemia approaches 1, while extreme hypoglycemia may contribute through other mechanisms, and its weight is not zero in this function. The S-shaped function is essentially a type of monotonically increasing S-shaped continuous function with an output range between (0,1), used to map any real number input to the interval [0,1].

[0046] For example, based on the generated blood glucose rate of change sequence and blood glucose deviation sequence The potential damage intensity that may be triggered by blood glucose fluctuations at each sampling time point is calculated using a damage activation function. This is achieved by iterating through all time points. , will the corresponding and Substituting this damage excitation function into the calculation, a new time series is generated, namely the damage excitation intensity series. This sequence is used to characterize the transient intensity of events that occur at each discrete time point, as revealed by raw blood glucose data, and that may trigger an endothelial cell stress response.

[0047] Step S304: Convolve the damage excitation intensity sequence and the damage integral kernel function to obtain instantaneous endothelial damage rate data.

[0048] Convolution is essentially a local weighted summation mathematical operation used to extract local features of data; the mathematical model of convolution is: The pathophysiological explanation is: at any observation time The rate of damage to endothelial cells is not solely triggered by fluctuations in blood glucose at a given moment, but rather by the cumulative effect of the residual influence of all triggering events over a past period; each past triggering event... Its impact is determined according to the kernel function. The described pattern decays over time, and the convolution integral is a weighted sum of all these residual effects.

[0049] For example, discrete damage excitation intensity sequences and continuous damage integral kernel function Convolution operations are performed to simulate the continuous damage output generated after a sequence of blood glucose fluctuation events passes through the biological response system of endothelial cells. The continuous kernel function... At the same fixed time interval Discrete sampling is performed on the discrete excitation intensity sequence and the discrete kernel sequence to obtain a discrete kernel sequence. Following the mathematical model of convolution, discrete convolution is performed on the discrete excitation intensity sequence and the discrete kernel sequence to obtain instantaneous endothelial injury rate data. The instantaneous endothelial injury rate data is a sum of input time series data. Time-axis aligned, continuous, or high-density sampled sequences. This data is used to characterize the instantaneous rate of change of biological damage stress actually experienced by vascular endothelial cells under the continuous driving force of blood glucose fluctuation events. Discrete sampling refers to extracting a continuously changing signal or physical quantity into a series of discrete, quantifiable numerical points at fixed time intervals (or spatial intervals), thereby converting the continuous signal into discrete data that can be processed by a computer. A biological response system is a complex regulatory network within an organism that can sense external or internal signals, such as chemical substances, physical stimuli, and changes in physiological state, and converts these signals and triggers targeted responses through specific mechanisms to maintain homeostasis or adapt to the environment. Discrete convolution is essentially a mathematical operation that generates a new discrete sequence from two discrete sequences through four core steps: flipping, shifting, multiplying, and summing. It is used to describe the weighting effect of one sequence on another.

[0050] In this embodiment, the dynamic characteristics of blood glucose at each sampling time point are calculated using blood glucose concentration time series data; a biological response kernel function is constructed; based on the dynamic characteristics of blood glucose and the damage activation function, the damage activation intensity at each sampling time point is calculated, resulting in a damage activation intensity sequence; this sequence is then convolved with the damage integral kernel function to obtain instantaneous endothelial damage rate data. This approach can accurately simulate how blood glucose fluctuations are transformed into persistent cumulative damage through endothelial cells at the data processing level. The output instantaneous endothelial damage rate data is an interpretable model result that integrates pathophysiological mechanisms, fundamentally solving the core problem that traditional methods cannot dynamically and mechanistically quantify the hidden risk of blood glucose fluctuations.

[0051] In one embodiment, based on transient endothelial injury rate data and cumulative injury dose, a multidimensional dynamic injury dose profile of the target patient is generated, including: Step S401: The cumulative damage dose is determined as a basic total characteristic.

[0052] For example, the cumulative damage dose This serves as the baseline total dose characteristic. It acts as the total baseline for all subsequent derived characteristics, reflecting the total dose of injury, similar to the total dose administered in drug therapy or the total radiation dose in radiation exposure.

[0053] Step S402: Calculate the standard deviation and coefficient of variation of the instantaneous endothelial injury rate data within a preset monitoring period, and determine the standard deviation and coefficient of variation as the injury variability characteristics.

[0054] For example, calling instantaneous endothelial injury rate data Perform standard statistical analysis on the data sequence. Calculate the standard deviation of the sequence using the following formula: ,in, Standard deviation is used to reflect the absolute magnitude of the fluctuation of the damage rate value around its average level. Let T be the mean of the sequence over the period T. Let be the total number of data points. For example, the coefficient of variation is further calculated using the following formula: Coefficient of variation It is a dimensionless relative dispersion index used to eliminate the influence of mean size, allowing for comparison of the variability of the injury process among patients with different average injury levels. The calculated standard deviation... and coefficient of variation They were collectively identified as having a damage fluctuation characteristic.

[0055] Step S403: Divide the day into multiple preset time periods, calculate the proportion of damage dose in each preset time period to the total damage dose of the day, and determine the proportion as the damage rhythm characteristic.

[0056] For example, based on the same instantaneous endothelial injury rate data This study introduces chronobiological analysis to explore whether injury occurrence exhibits specific circadian rhythms or patterns associated with life / treatment events. The 24-hour day is divided into several non-overlapping consecutive time periods according to pre-defined rules, such as nighttime (00:00-06:00), morning (06:00-12:00), afternoon (12:00-18:00), and evening (18:00-24:00). For each day within the monitoring period, the injury dose occurring within each pre-defined time period is calculated: this is done by analyzing all time points within that time period. The values ​​are discretely integrated (summed). Simultaneously, the total damage dose for the day is calculated. Then, for each preset time period of each day, the percentage of damage dose during that time period relative to the total damage dose for that day is calculated. The average percentage of the same time period across all days within the monitoring period is taken to obtain the final average damage dose percentage for each preset time period. This series of percentage values, for example, nighttime average percentage X%, morning average percentage Y%, etc., is defined as the damage rhythm characteristic. This set of characteristics is used to reveal the distribution pattern of damage load throughout the day, helping to identify whether damage is concentrated in specific high-risk windows such as nighttime hypoglycemia periods, postprandial hyperglycemia periods, or before and after dialysis treatment. Among them, the preset rules refer to the time-segmented logic based on the inherent physiological laws of the human body (such as circadian rhythms, metabolic cycles, and fluctuations in hormone secretion) or clinical needs (such as medication intervals and the frequency of disease monitoring); discrete integrals (also known as summation) are cumulative calculations for discrete data sequences. By adding up the discrete small parts, the total amount or cumulative effect of the whole is obtained, which corresponds to the integral of a continuous function. Continuous integrals are the sum of infinitely many infinitesimal quantities, while discrete integrals are the sum of a finite number (or countably infinite number) of discrete quantities; chronobiological analysis is the core research method based on the discipline of chronobiology. Its core is to explore the relationship between time rhythms and life activities in organisms. Through systematic observation, quantification, and interpretation of the laws of biological rhythms, it reveals the influence of time factors on physiological functions and pathological processes.

[0057] Step S404: Count the number of peaks, average peak height, and total duration of instantaneous endothelial injury rate data exceeding a preset injury rate threshold, and determine the number of peaks, average peak height, and total duration as injury peak characteristics.

[0058] The preset damage rate threshold refers to a critical value predefined based on historical data distribution, such as the median or 75th percentile of damage rate values ​​for all patients, or a pathophysiological significance, used to distinguish between background damage and significant high-intensity damage events.

[0059] For example, traversing the entire monitoring period T The sequence identifies continuous intervals formed by all data points where the instantaneous endothelial injury rate exceeds a preset injury threshold, defining each such interval as a peak injury event. The total number of these peak injury events is counted. For each identified peak injury event, its highest point is recorded. The value is used as the height of the peak. The average peak height is obtained by averaging all peak heights. Simultaneously, the total time covered by all peak events is summed, i.e., the total time from when each event exceeds a preset damage threshold to when it falls back below the preset damage threshold, to obtain the total peak duration. The number of peaks, the average peak height, and the total duration are collectively used as the damage peak characteristics.

[0060] In this embodiment, the cumulative damage dose is determined as the basic total characteristic; the dispersion of instantaneous endothelial damage rate data is analyzed to obtain the damage fluctuation characteristics; the damage load distribution is calculated by dividing the daytime period to obtain the damage rhythm characteristics; and high-intensity event characteristics are extracted by combining a preset damage rate threshold to characterize the acute peak. This overcomes the limitation of using only a single total indicator (cumulative dose).

[0061] In one embodiment, after constructing the damage integral kernel function based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations, the method further includes: Step S501: Obtain a blood sample from the target patient and detect the concentration of biomarkers related to endothelial injury and oxidative stress in the blood sample.

[0062] For example, at or near the start of a preset monitoring period, fasting peripheral venous blood samples from the target patient are obtained. Enzyme-linked immunosorbent assay (ELISA) or chemiluminescence immunoassay (CLIA) can be used to accurately quantify a preset set of biomarkers in plasma or serum that are directly related to the pathological processes of vascular endothelial injury and oxidative stress. These biomarkers may include, but are not limited to: soluble intercellular adhesion molecule-1 (sICAM-1) as a marker of endothelial activation; 8-iso-prostaglandin F2α (8-iso-PGF2α) reflecting the degree of lipid peroxidation; and soluble receptor for advanced glycation end products (sRAGE) associated with the advanced glycation end products (AGEs) pathway. Acquire and store the concentration values ​​of these biomarkers to form a biomarker concentration vector. ,in, The number of selected biomarkers; ELISA is an immunoassay technique based on the principle of specific antigen-antibody binding, combined with the signal amplification characteristics of enzyme-catalyzed reactions. It achieves quantitative or qualitative analysis through "antibody capture of target substances + enzyme-catalyzed colorimetric reaction of content"; CLIA is an in vitro detection technique that combines the specificity of immune reactions and the high sensitivity of chemiluminescence signals. It locates target substances through "specific antigen-antibody binding" and then amplifies the signal through "chemiluminescence reaction" to achieve quantitative detection of trace biomolecules; precise quantification refers to the accurate and repeatable numerical determination of specific attributes of a target object, such as substance content and concentration, through standardized methods, precise instruments, or rigorous calculations.

[0063] Step S502: Based on the preset calibration mapping relationship, the individualized damage amplitude coefficient in the damage integral kernel function is calibrated according to the concentration of biomarkers to obtain the calibrated individualized damage amplitude coefficient; wherein, the calibrated individualized damage amplitude coefficient is used for the calculation of the damage integral kernel function.

[0064] Among them, the preset calibration mapping relationship refers to a set of fixed mapping rules established in advance through experiments, data training or clinical validation. In essence, it establishes a one-to-one correspondence between two different dimensions of parameters, namely biomarker concentration and individualized damage amplitude coefficient A, which can be directly called. For example, it can be pre-stored in the form of formulas, data tables, or trained simple algorithms. No real-time calculation is required; the result can be matched simply by inputting data.

[0065] For example, by invoking a pre-defined calibration mapping model, the baseline lesion sensitivity coefficient of the target patient can be calculated using a linear regression model or empirical formula based on the patient's baseline demographic and clinical characteristics (e.g., age, duration of diabetes). This coefficient represents the average expected sensitivity in the absence of specific biomarker information. The concentration of each biomarker... Standardization was performed: the mean of the biomarker was obtained from a large-scale reference population database. and standard deviation Calculate its standardized score Its calculation formula is Each standardized score and its corresponding preset physiological weight coefficient are then assigned. Multiply and sum to obtain the calibrated individualized damage amplitude coefficient, the formula of which is: ,Should The value is the final kernel function used to construct the patient-specific damage integral. The amplitude coefficients. Among them, the linear regression model is a quantitative analysis model based on statistical theory. Its core is to explore the linear relationship between independent and dependent variables. Essentially, it fits the optimal linear equation to explain or predict changes in the dependent variable. Empirical formulas are approximate expressions based on experimental data or practical experience. Their core is to describe the relationship between variables using concise formulas, without strictly relying on statistical theory or physical laws. Essentially, they are empirical fittings to phenomena. A large-scale reference population database refers to a database that covers a sufficiently large and representative healthy population or a specific target population, such as a group based on regional population, specific age group / gender, excluding groups with definite diseases or confounding factors. Preset physiological weight coefficients. It can be pre-calibrated by performing multivariate regression analysis on historical cohort data to reflect the relative contribution of each biomarker to the overall sensitivity to endothelial injury. Multivariate regression analysis is a statistical method to explore the linear relationship between multiple independent variables (explanatory variables) and one dependent variable (outcome variable). By controlling for the influence of other variables, it can accurately quantify the independent effect of each independent variable on the dependent variable.

[0066] In this embodiment, by acquiring objective biomarker concentration data reflecting the current endothelial microenvironment state of the target patient, and based on a preset calibration mapping relationship and calibration formula, a calibrated individualized damage amplitude coefficient is obtained. This enables the damage accumulation model based on blood glucose fluctuations to no longer be a fixed formula applicable to all patients, but rather a sensitive system that can be dynamically calibrated according to the real-time endogenous biological signals of individual patients.

[0067] In one embodiment, the trained risk prediction model is obtained through the following method: Step S601: Collect longitudinal data of the historical patient cohort and construct multiple observation records for each patient based on the longitudinal data; wherein, the observation records include dynamic damage dose feature vector, static clinical feature variables, monitoring period time interval, fistula failure event status and fistula failure event time information for the corresponding monitoring period.

[0068] For example, complete longitudinal data of a historical patient cohort is collected. This data comes from multiple heterogeneous systems, including data warehouses of Electronic Medical Record (EMR), Laboratory Information System (LIS), and Continuous Glucose Monitor (CGM). For each patient in the cohort, multidimensional information is extracted over a long observation period: raw data from multiple CGM cycles arranged chronologically throughout the observation period are extracted; baseline static characteristics at the start of the observation period, such as age, duration of diabetes, and fistula location, are extracted; fistula failure event information is precisely extracted from follow-up records, including whether failure occurred (event status) and the exact time of occurrence (event time). For patients who did not experience failure, the follow-up time at which the fistula function was last confirmed to be normal is recorded (censored time). Based on this raw data, for each complete CGM cycle for each patient, such as each consecutive 14-day segment, the corresponding dynamic damage dose feature vector is calculated. ; this vector and the patient's static clinical characteristic variable vector The start time of this monitoring cycle and end time The event status and event time during the follow-up period after this cycle are associated and encapsulated. A structured training dataset is generated for the entire historical cohort, where each observation record represents a patient's exposure and subsequent outcome within a specific time interval (one monitoring cycle). EMR is the core system used by medical institutions for digital storage and management of patient medical information, replacing traditional paper medical records and serving as a fundamental tool for medical informatization; LIS is an information management system specifically designed for hospital laboratories such as laboratory and pathology departments, with the core objective of achieving "digitalization of the testing process + efficient flow of results"; the data warehouse is the core of integrated storage and management of multi-source medical data in medical scenarios, its core value being to break down data silos between different medical systems / equipment, providing unified data support for subsequent clinical decision-making, patient management, and medical research; association and encapsulation refers to packaging multi-dimensional medical data into a unified data unit after associating them according to clinical logic; clinical logic is a systematic thinking framework for medical staff to rationally think, judge, and make decisions based on medical knowledge, clinical experience, and the actual situation of the patient during the diagnosis and treatment process.

[0069] Step S602: Set the time-varying Cox proportional hazards model as the basic framework of the arteriovenous fistula failure risk prediction model; wherein, the hazard function of the arteriovenous fistula failure risk prediction model is:

[0070] in, In time The risk of arteriovenous fistula failure. For time, For the baseline risk function, In time The corresponding dynamic damage dose feature vector, This is a vector of static clinical characteristic variables. Let be the first coefficient vector to be estimated. Let be the second coefficient vector to be estimated.

[0071] The Time-Varying Cox Proportional Hazards Model (TVOH) is an extension of the classic Cox proportional hazards model, used for survival analysis, such as studying patient survival time and equipment failure time. It overcomes the key limitations of the classic model by allowing covariates (factors affecting survival) to change over time, making it more relevant to real-world scenarios where variables change dynamically. The classic Cox model is used to analyze the relationship between the time of an event (such as death, relapse, or failure) and one or more covariates. Its key feature is that it does not pre-assume the specific form of the baseline hazard function (hence it is semi-parametric), but only assumes that the multiplicative effect of each covariate on the risk remains constant, thus satisfying the proportional hazards assumption. For any point in time from the starting point of observation; To provide current time-varying features and static features Under the conditions, in time The instantaneous risk rate of fistula failure; The baseline risk function is a nonparametric part that represents the pattern of the underlying risk over time when all features take zero values; its shape is estimated directly from the data. In time The dynamic damage dose feature vector calculated for the most recent monitoring period is a covariate that changes over time. This is a vector of static clinical characteristic variables for patients, which does not change over time; and These are the parameters to be estimated in the model, namely the first coefficient vector and the second coefficient vector, which are used to quantify the contribution of each dimension in the dynamic injury dose characteristics and static clinical characteristics to the log-risk ratio (logarithm of the risk ratio).

[0072] For example, based on the two main characteristics of the arteriovenous fistula (AVF) failure risk prediction problem, a time-varying Cox proportional hazards model is selected and configured as the core algorithm framework for the AVF failure risk prediction model. The mathematical expression of the risk function of this model is as described above. By setting this model framework, a quantitative correlation model between multi-dimensional dynamic injury dose characteristics and AVF failure risk is clearly established. The two main characteristics of the AVF failure risk prediction problem are: first, risk prediction needs to consider factors that change over time; second, it needs to process time-event data until the event occurs or is censored.

[0073] Step S603: Based on the observation records, train the arteriovenous fistula failure risk prediction model to obtain the trained arteriovenous fistula failure risk prediction model.

[0074] For example, based on the constructed structured observation record dataset, the parameters of the time-varying Cox proportional hazards model are estimated and trained by maximizing the partial likelihood function. All observation records are sorted and organized according to the end time of their respective monitoring periods (i.e., the risk exposure update time point) and the final event occurrence or censoring time. In the iterative calculation, for each failure event time point, patients who are "still under follow-up and have not experienced an event" (i.e., the risk set) are selected. The relative risk of all patients in the risk set is calculated according to the model formula to reflect the relative probability of the patient experiencing the event; and the coefficient vector is adjusted through an optimization algorithm. and The value of maximizes the conditional probability of patients who observe actual events within the risk set. To prevent overfitting and improve model generalization, a regularization term is introduced into the objective function, and the regularization strength parameter is determined through cross-validation. Efficient optimization algorithms (such as coordinate descent) can be used for iterative solving until the model coefficients converge to stable values. After training, a definite coefficient vector estimate is obtained. and and an estimated baseline risk function. These elements together constitute a trained arteriovenous fistula failure risk prediction model. This model is able to receive dynamic and static feature vectors from new patients and output an estimate of their risk rate over time.

[0075] In this context, "follow-up" refers to patients who, due to specific health needs (such as post-operative / post-treatment monitoring), are included in a regular tracking and management system by a medical institution or professional team and are currently still within the planned monitoring cycle, with the follow-up process not yet completed; maximizing the partial likelihood function aims to ensure that patients who actually experience arteriovenous fistula failure exhibit the highest conditional probability in the model calculation, thus ensuring that the model closely reflects the real data patterns; censoring time refers to the time during which follow-up ceases but no event is observed (e.g., the study ends); parameter estimation is the process of calculating the optimal values ​​of model parameters using known data, usually training data, based on certain mathematical principles, such as maximum likelihood estimation; training is the complete process of iteratively optimizing model parameters to continuously improve performance. The process, in essence, involves repeatedly performing parameter estimation, error verification, and parameter adjustment loops until the model's error on the training data reaches the expected level or meets the stopping condition, such as exhausting the number of iterations or the error no longer decreasing. Regularization is a core technical component in machine learning and statistical modeling used to prevent overfitting. It is usually appended to the model's loss function to form a new optimization objective, i.e., "loss function + regularization term". Cross-validation is a commonly used model evaluation and parameter selection method in machine learning. Its core logic is that the traditional "training set - test set" split has limitations: if the test set is lucky (matching data the model is good at), the evaluation result will be too high; if it's unlucky (all data the model is not good at), the result will be... Low performance and high randomness: Cross-validation reduces randomness by splitting the data multiple times, training and testing multiple times, and using average performance to more objectively reflect the model's generalization ability; Optimization algorithms are a class of mathematical methods and computational processes used to find solutions in the solution space of a specific problem that minimize or maximize the objective function (such as error); Coordinate descent is an iterative optimization algorithm whose core idea is to decompose the optimization problem of high-dimensional variables into multiple low-dimensional (usually one-dimensional) subproblems, and gradually approach the minimum value of the objective function by optimizing the variables of each dimension one by one; Iterative solution is a numerical computation method that gradually approaches the true answer to a problem by repeatedly executing fixed steps. Its core idea is to start from an initial guess... Starting with the initial calculation, the model uses the results of each step to correct the previous step until the result meets the accuracy requirements. Convergence to a stable value means that during iterative training, the model continuously adjusts its coefficients (such as the weights in linear regression) to reduce prediction errors. When the iteration reaches a certain stage, the update magnitude of the coefficients will become smaller and smaller, eventually no longer changing significantly and remaining in a relatively fixed numerical range. The partial likelihood function is a parameter estimation tool in statistics used to process data containing truncated data or time-dependent variables. The core idea is to infer parameters without relying on the probability distribution of complete data, but only using the directly observable sequence of events or relative risk information in the data. The most typical application scenario is survival analysis, such as medical follow-up studies and equipment lifespan studies.

[0076] In this embodiment, a structured training set containing time-varying covariates, static covariates, and precise survival time is extracted and constructed from historical data; a survival analysis mathematical model framework capable of integrating time-varying and static features is initialized; and the quantitative weighting relationship between all features and the risk of arteriovenous fistula failure in the model is determined through a partial likelihood-based, regularized optimization process based on the training data. This allows the dynamic biological process of cumulative dose of endothelial injury induced by blood glucose fluctuations to be quantified into predictive factors that can be incorporated into survival analysis. The final model can provide individualized dynamic risk predictions that are both rooted in pathophysiological mechanisms and validated by real-world data, providing an unprecedented, time-varying quantitative decision-making tool for prospective clinical interventions.

[0077] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0078] Based on the same inventive concept, this application also provides a system for predicting the state of an arteriovenous fistula (AVF) to implement the aforementioned method. The solution provided by this system is similar to the implementation described in the above method. Therefore, the specific limitations of one or more embodiments of the AVF state prediction system provided below can be found in the limitations of the AVF state prediction method described above, and will not be repeated here.

[0079] In one exemplary embodiment, such as Figure 3 As shown, an arteriovenous fistula status prediction system 300 is provided, comprising: The data acquisition module 301 is used to acquire continuous dynamic blood glucose monitoring data of the target patient within a preset monitoring period; and to perform standardized preprocessing on the continuous dynamic blood glucose monitoring data to generate blood glucose concentration time series data. The feature extraction module 302 is used to obtain multi-dimensional dynamic injury dose characteristics of the target patient based on blood glucose concentration time series data; wherein, the multi-dimensional dynamic injury dose characteristics include baseline total characteristics, injury fluctuation characteristics, injury rhythm characteristics, and injury peak characteristics; The vector generation module 303 is used to combine multi-dimensional dynamic damage dose features to generate a dynamic damage dose feature vector. The state prediction module 304 is used to input the dynamic damage dose feature vector and the static clinical feature variables of the target patient into the trained arteriovenous fistula failure risk prediction model, and output the risk probability of the target patient experiencing arteriovenous fistula failure within a preset future time window; wherein, the static clinical feature variables are stable patient baseline features.

[0080] In one embodiment, the feature extraction module 302 is further configured to: Based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations and blood glucose concentration time series data, the instantaneous endothelial injury rate data were calculated. Within the time range of the preset monitoring period, the instantaneous endothelial injury rate data are numerically integrated to generate the cumulative injury dose; wherein, the cumulative injury dose is used to characterize the total load borne by intravascular cells within the preset monitoring period. Based on transient endothelial injury rate data and cumulative injury dose, multidimensional dynamic injury dose characteristics of the target patient are generated.

[0081] In one embodiment, the feature extraction module 302 is further configured to: Based on blood glucose concentration time series data, the rate of change of blood glucose and the degree of deviation of blood glucose at each sampling time point are calculated, and based on each rate of change of blood glucose and each degree of deviation of blood glucose, a blood glucose change rate sequence and a blood glucose deviation sequence are generated. Based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations, a damage integral kernel function is constructed. The specific expression of the damage integral kernel function is as follows:

[0082] in, For the damage integral kernel function, For time delay variables; This is an individualized damage amplitude coefficient; Weights for fast response components; The fast response time constant; The slow response time constant; It is a unit step function; Based on the blood glucose change rate sequence, blood glucose deviation sequence, and damage activation function, the damage activation intensity at each sampling time point is calculated, and a damage activation intensity sequence is generated based on each damage activation intensity; wherein, the damage activation intensity is proportional to the power of the absolute value of the blood glucose change rate. The instantaneous endothelial damage rate data are obtained by convolving the damage excitation intensity sequence and the damage integral kernel function.

[0083] In one embodiment, the feature extraction module 302 is further configured to: The cumulative damage dose is determined as the baseline total dose characteristic; Calculate the standard deviation and coefficient of variation of instantaneous endothelial injury rate data within a preset monitoring period, and determine the standard deviation and coefficient of variation as the characteristics of injury variability; The day is divided into multiple preset time periods. The proportion of damage dose in each preset time period to the total daily damage dose is calculated, and the proportion is determined as the damage rhythm characteristic. The number of peaks, average peak height, and total duration of instantaneous endothelial injury rate exceeding a preset injury rate threshold are statistically analyzed, and the number of peaks, average peak height, and total duration are determined as injury peak characteristics.

[0084] In one exemplary embodiment, the system further includes: The sample acquisition module is used to acquire blood samples from target patients and detect the concentrations of biomarkers related to endothelial injury and oxidative stress in the blood samples. The coefficient calibration module is used to calibrate the individualized damage amplitude coefficient in the damage integral kernel function based on a preset calibration mapping relationship and the concentration of biomarkers, thereby obtaining the calibrated individualized damage amplitude coefficient; wherein, the calibrated individualized damage amplitude coefficient is used for the calculation of the damage integral kernel function.

[0085] In one embodiment, the state prediction module 304 is further configured to: Longitudinal data from historical patient cohorts were collected, and multiple observation records were constructed for each patient based on the longitudinal data. The observation records included dynamic injury dose feature vectors, static clinical feature variables, monitoring period time intervals, fistula failure event status, and fistula failure event time information for the corresponding monitoring period. The time-varying Cox proportional hazards model is used as the basic framework for the arteriovenous fistula (AVF) failure risk prediction model; the hazard function of the AVF failure risk prediction model is:

[0086] in, In time The risk of arteriovenous fistula failure. For time, For the baseline risk function, In time The corresponding dynamic damage dose feature vector, This is a vector of static clinical characteristic variables. Let be the first coefficient vector to be estimated. Let the second coefficient vector be the one to be estimated. Based on observation records, the arteriovenous fistula failure risk prediction model was trained to obtain a well-trained arteriovenous fistula failure risk prediction model.

[0087] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the arteriovenous fistula state prediction method as described above.

[0088] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0089] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0090] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for predicting the status of arteriovenous fistulas, characterized in that, The method includes: Acquire continuous dynamic blood glucose monitoring data of the target patient within a preset monitoring period; perform standardized preprocessing on the continuous dynamic blood glucose monitoring data to generate blood glucose concentration time series data; Based on the blood glucose concentration time series data, the multidimensional dynamic injury dose characteristics of the target patient are obtained; wherein, the multidimensional dynamic injury dose characteristics include baseline total characteristics, injury variability characteristics, injury rhythm characteristics, and injury peak characteristics. The multi-dimensional dynamic damage dose features are combined to generate a dynamic damage dose feature vector. The dynamic injury dose feature vector and the static clinical feature variables of the target patient are input into the trained arteriovenous fistula failure risk prediction model, and the risk probability of the target patient experiencing arteriovenous fistula failure within a preset future time window is output; wherein, the static clinical feature variables are stable patient baseline features.

2. The method according to claim 1, characterized in that, The step of obtaining the multidimensional dynamic damage dose characteristics of the target patient based on the blood glucose concentration time series data includes: Based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations and the blood glucose concentration time series data, the instantaneous endothelial injury rate data were calculated. Within the time range of the preset monitoring period, the instantaneous endothelial injury rate data is numerically integrated to generate a cumulative injury dose; wherein, the cumulative injury dose is used to characterize the total load borne by the intravascular cells within the preset monitoring period; Based on the instantaneous endothelial injury rate data and the cumulative injury dose, the multidimensional dynamic injury dose characteristics of the target patient are generated.

3. The method according to claim 2, characterized in that, The transient endothelial injury rate data, calculated based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations and the blood glucose concentration time series data, includes: Based on the blood glucose concentration time series data, calculate the blood glucose change rate and blood glucose deviation at each sampling time point, and generate a blood glucose change rate sequence and a blood glucose deviation sequence based on each of the blood glucose change rates and blood glucose deviations. Based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations, a damage integral kernel function is constructed. The specific expression of the damage integral kernel function is as follows: ； in, For the damage integral kernel function, For time delay variables; This is an individualized damage amplitude coefficient; Weights for fast response components; The fast response time constant; The slow response time constant; It is a unit step function; Based on the blood glucose change rate sequence, the blood glucose deviation sequence, and the damage activation function, the damage activation intensity at each sampling time point is calculated, and a damage activation intensity sequence is generated based on each damage activation intensity; wherein, the damage activation intensity is proportional to the power of the absolute value of the blood glucose change rate; The instantaneous endothelial damage rate data are obtained by convolving the damage excitation intensity sequence and the damage integral kernel function.

4. The method according to claim 2, characterized in that, The generation of the multidimensional dynamic injury dose characteristics of the target patient based on the instantaneous endothelial injury rate data and the cumulative injury dose includes: The cumulative damage dose is determined as the baseline total dose characteristic; Calculate the standard deviation and coefficient of variation of the instantaneous endothelial injury rate data within the preset monitoring period, and determine the standard deviation and coefficient of variation as the injury variability characteristics; A single day is divided into multiple preset time periods, and the proportion of the damage dose in each preset time period to the total daily damage dose is calculated, and the proportion is determined as the damage rhythm characteristic. The number of peaks, average peak height, and total duration of the instantaneous endothelial injury rate data exceeding a preset injury rate threshold are statistically analyzed, and the number of peaks, average peak height, and total duration are determined as injury peak characteristics.

5. The method according to claim 3, characterized in that, After constructing the damage integral kernel function based on the biological response characteristics of vascular endothelial cells to blood glucose fluctuations, the method further includes: Blood samples were obtained from the target patient, and the concentrations of biomarkers related to endothelial injury and oxidative stress in the blood samples were detected. Based on a preset calibration mapping relationship, the individualized damage amplitude coefficient in the damage integral kernel function is calibrated according to the concentration of the biomarker, to obtain the calibrated individualized damage amplitude coefficient; wherein, the calibrated individualized damage amplitude coefficient is used for the calculation of the damage integral kernel function.

6. The method according to claim 1, characterized in that, The trained risk prediction model was obtained through the following method: Collect longitudinal data from historical patient cohorts and construct multiple observation records for each patient based on the longitudinal data; wherein, the observation records include the dynamic injury dose feature vector, the static clinical feature variables, the monitoring period time interval, the fistula failure event status, and the fistula failure event time information for the corresponding monitoring period; The time-varying Cox proportional hazards model is used as the basic framework for the arteriovenous fistula (AVF) failure risk prediction model; wherein, the hazard function of the AVF failure risk prediction model is: ； in, In time The risk of arteriovenous fistula failure. For time, For the baseline risk function, In time The corresponding dynamic damage dose feature vector, Let the static clinical feature variable vector be... Let be the first coefficient vector to be estimated. Let the second coefficient vector be the one to be estimated. Based on the observation records, the arteriovenous fistula failure risk prediction model is trained to obtain the trained arteriovenous fistula failure risk prediction model.

7. A system for predicting the status of arteriovenous fistulas, characterized in that, The system includes: The data acquisition module is used to acquire continuous dynamic blood glucose monitoring data of the target patient within a preset monitoring period; and to perform standardized preprocessing on the continuous dynamic blood glucose monitoring data to generate blood glucose concentration time series data. The feature extraction module is used to obtain multi-dimensional dynamic injury dose characteristics of the target patient based on the blood glucose concentration time series data; wherein, the multi-dimensional dynamic injury dose characteristics include baseline total characteristics, injury variability characteristics, injury rhythm characteristics, and injury peak characteristics; The vector generation module is used to combine the multi-dimensional dynamic damage dose features to generate a dynamic damage dose feature vector. The state prediction module is used to input the dynamic damage dose feature vector and the static clinical feature variables of the target patient into the trained arteriovenous fistula failure risk prediction model, and output the risk probability of the target patient experiencing arteriovenous fistula failure within a preset future time window; wherein, the static clinical feature variables are stable patient baseline features.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images