A risk assessment system for hemodialysis patients based on objective assessment method

Abstract

The present invention provides a hemodialysis patient risk assessment system based on an objective assessment method, which relates to the field of medical data technology and includes: an effectiveness assessment unit for collecting basic clinical information of hemodialysis patients, calculating nutritional status indicators through objective assessment methods, and evaluating the effectiveness of objective assessment methods on nutritional status indicators using receiver operating characteristic curves; a model construction unit for constructing a risk assessment model for hemodialysis patients based on the effectiveness assessment results using matter-element extension variable weight theory and data fusion technology; and a risk assessment unit for performing individualized risk assessment on hemodialysis patients using the risk assessment model for hemodialysis patients to obtain risk assessment results. The present invention helps to efficiently allocate medical resources, identify high-risk patients early, reduce the occurrence of complications, and improve patient prognosis. It integrates multiple technologies and theories to enhance the innovation and flexibility of the assessment system.

Description

Technical Field

[0001] The present invention relates to the field of medical data technology, and in particular to a risk assessment system for hemodialysis patients based on an objective assessment method. Background Art

[0002] Hemodialysis, an important medical procedure for treating chronic renal failure, relies on a hemodialyzer to remove waste products and excess water from the blood. During this process, the patient's blood is pumped into the dialyzer, where the built-in semipermeable membrane plays a key role. Blood flows on one side of the membrane, while dialysate flows on the other. Waste products and excess water pass from the blood into the dialysate through the semipermeable membrane. At the same time, beneficial substances such as electrolytes in the dialysate can also pass through the membrane into the blood, thereby purifying the blood and maintaining electrolyte balance. The purified blood is then safely pumped back into the patient's body.

[0003] Historically, research on the risks of hemodialysis patients has been limited to observational analyses of single factors, attempting to explore the impact of a specific factor on patient prognosis. However, such studies often struggle to draw conclusions with clear intervention effects. With the rapid development of artificial intelligence (AI) technology, its application prospects in medicine are increasingly broad. Studies have begun to explore the use of AI methods in the management of hemodialysis patients, hoping to provide more precise and personalized treatment plans.

[0004] However, in the current medical assessment system, nutritional status is still primarily determined by the physician's subjective experience and the patient's subjective feelings, lacking scientific and objective assessment criteria. This subjectivity not only affects the accuracy of the assessment but also makes it difficult to ensure the efficacy and effectiveness of each assessment method.

[0005] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0006] In view of this, the present invention provides a hemodialysis patient risk assessment system based on an objective assessment method to solve the above-mentioned problems.

[0007] In order to solve the above problems, the specific technical solutions adopted by the present invention are as follows:

[0008] A risk assessment system for hemodialysis patients based on an objective assessment method, comprising:

[0009] The efficacy assessment unit is used to collect basic clinical information of hemodialysis patients, calculate nutritional status indicators through objective assessment methods, and evaluate the efficacy of objective assessment methods for nutritional status indicators using the receiver operating characteristic curve based on the Youden index;

[0010] A model building unit is used to build a risk assessment model for hemodialysis patients based on the effectiveness evaluation results using matter-element extension variable weight theory and data fusion technology;

[0011] a risk assessment unit, configured to perform individualized risk assessment on hemodialysis patients using a risk assessment model for hemodialysis patients, and obtain a risk assessment result;

[0012] Wherein, the performance evaluation unit includes:

[0013] A numerical calculation module is used to calculate the area under the curve for each receiver operating characteristic curve using a Monte Carlo algorithm of cross entropy sampling and dispersion sampling;

[0014] The Monte Carlo algorithm for calculating the area under the curve using cross-entropy sampling and dispersed sampling includes: setting a preliminary sampling distribution for random point generation based on the coordinate range of the receiver operating characteristic curve, and randomly generating a number of random points from the coordinate axis of each receiver operating characteristic curve; preliminarily determining the key area of ​​each random point based on the receiver operating characteristic curve, and calculating the cross entropy between the current sampling distribution and the receiver operating characteristic curve; updating the sampling distribution based on the cross entropy between the current sampling distribution and the receiver operating characteristic curve, and generating random points again; calculating the distance between all generated random points, and dynamically adjusting the sampling density using the dispersed sampling method; determining whether each random point is located below the receiver operating characteristic curve, counting the number of random points that meet preset conditions, and calculating the area under the curve.

[0015] Preferably, the performance evaluation unit further includes:

[0016] The objective scoring module is used to calculate the protein-energy consumption score, the elderly nutritional risk index score, the modified creatinine index score, and the malnutrition inflammatory response syndrome score based on the collected basic clinical information and multiple objective assessment methods to obtain an objective score set;

[0017] The curve construction module is used to calculate the sensitivity and specificity based on the Youden index and each objective score to construct the receiver operating characteristic curve;

[0018] The evaluation module is used to evaluate the effectiveness of the objective evaluation method for evaluating the nutritional status indicators of hemodialysis patients through correlation analysis based on the area under the curve.

[0019] Preferably, the calculation of sensitivity and specificity based on the Youden index and each objective score, the construction of a receiver operating characteristic curve and the determination of the optimal judgment threshold include:

[0020] For each customer score, the sensitivity and specificity at different thresholds are calculated at several pre-set thresholds;

[0021] Based on the calculated sensitivity and specificity, a preliminary receiver operating characteristic curve was constructed;

[0022] The Youden index at each threshold was calculated, and the threshold with the maximum Youden index was selected as the optimal judgment threshold. The preliminary receiver operating characteristic curve corresponding to the optimal judgment threshold was used as the final receiver operating characteristic curve.

[0023] Preferably, the evaluation of the effectiveness of the objective assessment method for nutritional status indicators of hemodialysis patients by correlation analysis based on the area under the curve includes:

[0024] The Pearson correlation coefficient method was used to analyze the Pearson correlation coefficient between the objective assessment method and the actual nutritional status;

[0025] Perform significance test on the obtained Pearson correlation coefficient;

[0026] The effectiveness of the objective evaluation method was analyzed based on the area under the curve of each objective evaluation method and combined with the Pearson correlation coefficient to obtain the effectiveness evaluation results.

[0027] Preferably, the model building unit includes:

[0028] The matter-element matrix construction module is used to construct a matter-element matrix for risk assessment of hemodialysis patients based on nutritional status indicators and using the matter-element extension theory;

[0029] The weight adjustment module is used to preliminarily determine the weight of each nutritional status indicator using the hierarchical analysis method, and dynamically adjust the weight of each nutritional status indicator through the state variable weight theory in combination with the performance evaluation results;

[0030] The risk assessment module is used to calculate the comprehensive risk assessment data of each patient based on the weight of each nutritional status indicator and using data fusion technology to fuse the nutritional status indicator data in the matter-element matrix with its corresponding weight;

[0031] The model building and optimization module is used to build and optimize the risk assessment model for hemodialysis patients based on comprehensive risk assessment data and through adaptive learning optimization method.

[0032] Preferably, the method of preliminarily determining the weight of each nutritional status indicator by using the analytic hierarchy process, and dynamically adjusting the weight of each nutritional status indicator by using the state variable weight theory in combination with the performance evaluation results includes:

[0033] The nutritional status assessment problem of hemodialysis patients is decomposed into several levels, including the target level and the criterion level. The judgment matrix is ​​constructed by comparing the nutritional status indicators in the criterion level with each other.

[0034] The constant weight vector of each nutritional status index was calculated using the eigenvector method in the hierarchical analysis method;

[0035] Based on the performance evaluation results, the state vector is constructed and the state variable weight vector of each nutritional status indicator is calculated;

[0036] The weight of each nutritional status indicator is calculated using its constant weight vector and state variable weight vector.

[0037] Preferably, the calculation formula for calculating the weight of each nutritional status indicator using the constant weight vector and the state variable weight vector of each nutritional status indicator is:

[0038]

[0039] Where W j (X) represents the weight of each nutritional status indicator, w j represents the constant weight vector calculated by the AHP method, S j (X) represents the state variable weight vector of the j-th nutritional status indicator, and m represents the number of nutritional status indicators.

[0040] Preferably, the weight of each nutritional status indicator is based on the data fusion technology, and the nutritional status indicator data in the matter-element matrix are fused with the corresponding weight to calculate the comprehensive risk assessment data of each patient, including:

[0041] The matter-element matrix is ​​normalized using the minimum-maximum normalization method to obtain a normalized matter-element matrix;

[0042] The normalized matter-element matrix is ​​fused with the weights of each nutritional status indicator using the weighted average method to obtain a weighted normalized matrix.

[0043] Each row of the weighted normalized matrix is ​​summed to obtain the comprehensive risk assessment data for each patient.

[0044] Preferably, the risk assessment model for hemodialysis patients is constructed and optimized based on the comprehensive risk assessment data and through an adaptive learning optimization method, including:

[0045] The clinical status coefficient of each hemodialysis patient was calculated using the basic clinical information of the hemodialysis patients;

[0046] Based on the comprehensive risk assessment data and behavioral status coefficient of each hemodialysis patient, a risk assessment model for initial hemodialysis patients is constructed;

[0047] Based on the adaptive learning optimization method, the initial risk assessment model of hemodialysis patients was optimized in combination with gradient descent and cross entropy loss functions to obtain the final risk assessment model of hemodialysis patients.

[0048] Preferably, the calculating of the clinical status coefficient of each hemodialysis patient using the basic clinical information of the hemodialysis patient comprises:

[0049] Preprocessing the basic clinical information of hemodialysis patients, including data cleaning and data standardization;

[0050] Extract effective features from pre-processed clinical basic information to obtain effective feature vectors;

[0051] According to the extracted effective feature vectors, the clinical status coefficients of hemodialysis patients were calculated using the weighted summation method.

[0052] The beneficial effects of the present invention are:

[0053] 1. The present invention ensures the scientific nature and reliability of risk assessment. Through the material-element extension variable weight theory and data fusion technology, a dynamic and personalized risk assessment model is constructed, which can provide personalized risk assessment for each patient, help clinicians optimize treatment plans, improve patient management effects, contribute to the efficient allocation of medical resources, early identification of high-risk patients, reduce the occurrence of complications, and improve patient prognosis. The integration of multiple technologies and theories enhances the innovation and flexibility of the assessment system.

[0054] 2. The present invention calculates the area under the receiver operating characteristic curve (ROC) through the Monte Carlo algorithm. This method can provide a relatively accurate and flexible estimation method. By calculating the sensitivity and specificity at different thresholds, a preliminary ROC curve is constructed, and the final curve is further optimized, which helps to ensure the accuracy and effectiveness of the curve. Combined with the area under the curve and the Pearson correlation coefficient, a comprehensive evaluation of the effectiveness of the objective evaluation method is performed, which helps to more comprehensively understand the advantages and disadvantages of each evaluation method. Doctors can more accurately understand the effectiveness of different objective evaluation methods in assessing the nutritional status of hemodialysis patients.

[0055] 3. The present invention constructs a matter-element matrix for risk assessment of hemodialysis patients through the matter-element topological theory, which can systematically integrate and analyze multiple nutritional status indicators, provide a comprehensive data basis for risk assessment, and combine the performance evaluation results. The weights are dynamically adjusted through the state variable weight theory, so that the weights can be flexibly adjusted according to the actual status of the patient, thereby improving the accuracy and adaptability of risk assessment. Through data fusion technology, the nutritional status indicator data in the matter-element matrix are fused with their corresponding weights to calculate the comprehensive risk assessment data of each patient, realizing a comprehensive evaluation of multiple indicators. Based on the comprehensive risk assessment data, the risk assessment model is constructed and optimized through the adaptive learning optimization method, so that the model can continuously learn and adapt to new data, thereby improving the accuracy and stability of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. 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 work. In the drawings:

[0057] Figure 1 This is a principle block diagram of a hemodialysis patient risk assessment system based on an objective assessment method according to an embodiment of the present invention.

[0058] In the picture:

[0059] 1. Effectiveness evaluation unit; 2. Model building unit; 3. Risk assessment unit. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0061] According to an embodiment of the present invention, a risk assessment system for hemodialysis patients based on an objective assessment method is provided.

[0062] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, a hemodialysis patient risk assessment system based on an objective assessment method according to an embodiment of the present invention includes:

[0063] Efficacy evaluation unit 1 is used to collect basic clinical information of hemodialysis patients, calculate nutritional status indicators through objective evaluation methods, based on the Youden index, and use receiver operating characteristic curves to evaluate the effectiveness of objective evaluation methods for nutritional status indicators;

[0064] As a preferred embodiment, the performance evaluation unit 1 includes:

[0065] The objective scoring module is used to calculate the protein-energy consumption score, the elderly nutritional risk index score, the modified creatinine index score, and the malnutrition inflammatory response syndrome score based on the collected basic clinical information and multiple objective assessment methods to obtain an objective score set;

[0066] It should be added that the collected basic clinical information includes:

[0067] (1) General data collection: gender, age, height, dry weight, BMI, primary disease, dialysis age, and residual renal function.

[0068] (2) Collection of blood biochemical indicators: In this experiment, blood was regularly drawn from the blood purification center of a university affiliated hospital and sent to the laboratory department of this hospital for unified testing. The pre-dialysis indicators collected were: TIBC (total iron binding capacity), serum iron, ALB (albumin), prealbumin, uric acid, Scr (creatinine), blood urea, estimated glomerular filtration rate, CRP (C-reactive protein), neutrophil count, lymphocyte count, platelet count, hemoglobin, parathyroid hormone; post-dialysis indicators: post-dialysis BUN (reflects the dialysis effect; generally, the greater the decrease in BUN concentration, the better the dialysis effect in removing waste).

[0069] (3) Collection of manual measurement indicators

[0070] Thirty minutes before dialysis, patients performed a handgrip test three times using their stronger hand, and the maximum value was recorded in kilograms. Fifteen minutes after dialysis, the patient's height (m) and weight (kg) were measured barefoot while wearing light clothing. Waist circumference: Standing with feet 25-30 cm apart, a soft tape measure was placed around the body two cm above the navel. The two measurements were averaged and reported in centimeters. Hip circumference: Standing with legs together and arms hanging naturally, a tape measure was placed horizontally between the pubic symphysis and the most convex part of the gluteus maximus. The two measurements were averaged and reported in centimeters. Triceps skinfold thickness (TSF) was measured: The subject stood with their upper arm fully exposed. The measurer pinched the skin and subcutaneous tissue at the measured area with three fingers of their left hand (thumb, index finger, and middle finger). A skinfold thickness gauge was used to measure the thickness 1 cm below the pinch point. Three measurements were taken, and the average or identical value was reported in millimeters. Midarm circumference (MAC) measurement: Measure the fullest part of the upper arm when the upper limb is hanging naturally. Take the average of two measurements (in centimeters). All measurements are rounded to two decimal places.

[0071] Among them, the protein-energy expenditure score is used to assess the patient's protein and energy expenditure.

[0072] The MiniNutritional Assessment (MNA) score is used to assess the nutritional risk of elderly patients, especially for very old patients.

[0073] The modified creatinine index (MCI) is used to assess a patient's renal function and muscle mass, and is particularly suitable for dialysis patients because they may have muscle wasting.

[0074] The Nutritional Inflammation Score (NIS) is used to assess whether patients have a syndrome caused by malnutrition and inflammatory response, which is usually seen in critically ill patients and dialysis patients.

[0075] The curve construction module is used to calculate the sensitivity and specificity based on the Youden index and each objective score to construct the receiver operating characteristic curve;

[0076] As a preferred embodiment, the calculation of sensitivity and specificity based on the Youden index and each objective score, the construction of a receiver operating characteristic curve and the determination of the optimal judgment threshold include:

[0077] For each customer score, the sensitivity and specificity at different thresholds are calculated at several pre-set thresholds;

[0078] Based on the calculated sensitivity and specificity, a preliminary receiver operating characteristic curve was constructed;

[0079] Specifically, for each customer rating, it is necessary to calculate the sensitivity and specificity under several preset thresholds.

[0080] Sensitivity represents the true positive rate, or the proportion of samples correctly predicted as positive by the model to all samples that are actually positive. The formula is: Sensitivity = TP / (TP + FN), where TP = true positives (predicted positive and actually positive), and FN = false negatives (predicted negative but actually positive).

[0081] Specificity represents the true negative rate, or the proportion of samples correctly predicted as negative by the model to all samples that are actually negative. The formula is: Specificity = TN / (TN + FP), where TN = true negatives (predicted negative and actually negative), and FP = false positives (predicted positive and actually negative).

[0082] The receiver operating characteristic (ROC) curve was drawn based on the sensitivity and specificity calculated at different thresholds.

[0083] 1-specificity, ie, the false positive rate (1-specificity), is plotted on the abscissa.

[0084] Sensitivity (true positive rate) is plotted on the ordinate.

[0085] Each threshold corresponds to a point on the ROC curve. As the threshold changes, the sensitivity and specificity will also change accordingly, and eventually a curve connecting these points will be obtained.

[0086] The Youden index at each threshold was calculated, and the threshold with the maximum Youden index was selected as the optimal judgment threshold. The preliminary receiver operating characteristic curve corresponding to the optimal judgment threshold was used as the final receiver operating characteristic curve.

[0087] Among them, Youden index is an indicator used to select the optimal judgment threshold, and its calculation formula is: Youden index = sensitivity + specificity - 1;

[0088] Sensitivity and specificity often change with changes in threshold.

[0089] The higher the Youden index value, the better the judgment effect of the threshold. Its maximum value usually indicates the optimal threshold.

[0090] For example, based on the score, some thresholds are set, and the sensitivity and specificity at different thresholds are calculated. Then, the process of drawing the receiver operating characteristic curve is as follows:

[0091] 1. Calculate sensitivity and specificity:

[0092] At threshold 1, the sensitivity was 0.8 and the specificity was 0.9. At threshold 2, the sensitivity was 0.85 and the specificity was 0.85. At threshold 3, the sensitivity was 0.7 and the specificity was 0.95.

[0093] 2. Construct a preliminary ROC curve (receiver operating characteristic curve):

[0094] The sensitivity and specificity at thresholds 1, 2, and 3 are plotted on the ROC curve as points (assuming that threshold 1 corresponds to point (0.1, 0.8), threshold 2 corresponds to point (0.15, 0.85), and threshold 3 corresponds to point (0.05, 0.7)).

[0095] 3. Calculate the Youden index:

[0096] For a threshold of 1, Youden index = 0.8 + 0.9 - 1 = 0.7.

[0097] For a threshold of 2, Youden index = 0.85 + 0.85 - 1 = 0.7.

[0098] For threshold 3, Youden index = 0.7 + 0.95 - 1 = 0.65.

[0099] 4. Choose the best threshold:

[0100] The maximum Youden index is 0.7, corresponding to thresholds 1 and 2.

[0101] Assume that threshold 1 is selected as the optimal threshold.

[0102] 5. Determine the final ROC curve:

[0103] Using a threshold of 1, the sensitivity (0.8) and specificity (0.9) were recalculated.

[0104] Based on this, the final ROC curve is drawn.

[0105] A numerical calculation module is used to calculate the area under the curve for each receiver operating characteristic curve using a Monte Carlo algorithm of cross entropy sampling and dispersion sampling;

[0106] The method of calculating the area under the curve using the Monte Carlo algorithm of cross entropy sampling and distributed sampling includes:

[0107] Based on the coordinate range of the receiver operating characteristic curve, a preliminary sampling distribution is set for random point generation, and a number of random points are randomly generated from the coordinate axis of each receiver operating characteristic curve;

[0108] Specifically, the receiver operating characteristic curve usually uses sensitivity as the vertical axis and specificity as the horizontal axis, and its coordinate range is:

[0109] Sensitivity∈[0,1], specificity∈[0,1], the value range of these two indicators is [0,1].

[0110] A preliminary sampling distribution can be a uniform distribution, which randomly generates points within the sensitivity and specificity coordinates, ensuring that each point has an equal probability of being sampled. In this case, the sampling points are randomly distributed within the sensitivity and specificity ranges, ensuring that each random point uniformly covers the entire two-dimensional space. The generation of random points is based on the previous sampling distribution. A random number generation method (such as a uniformly distributed random number generator) can be used to generate several points from the sensitivity and specificity coordinate axes.

[0111] According to the receiver operating characteristic curve, the key area of ​​each random point is preliminarily determined, and the cross entropy between the current sampling distribution and the receiver operating characteristic curve is calculated;

[0112] It should be noted that in the ROC curve, different regions contribute differently to the AUC. The area under the curve (i.e., the AUC value) is closely related to the sampling density of different regions. Generally, the more dramatic the change in the ROC curve, the greater the contribution to the AUC. Therefore, in this step, you need to:

[0113] Identifying key regions: The shape of the curve can be used to identify regions that contribute significantly to the AUC. For example, regions with large variations in sensitivity and specificity contribute more to the AUC. Generally speaking, the beginning of the curve (near x = 0 and y = 1) or the region near x = 1 and y = 0 has a significant impact on the area under the curve.

[0114] Calculate the importance of each random point: At each random point, determine whether the point is in the critical area of ​​the curve. This can be determined by the following criteria:

[0115] Whether the curve changes significantly near this point.

[0116] Whether the changes in sensitivity and specificity are large.

[0117] If a certain region contributes more to the AUC (for example, a region where the curve changes steeply), the sampling density of this region should be increased.

[0118] Cross entropy is used to measure the difference between two probability distributions. In Monte Carlo simulation, the cross entropy method is used to adjust the sampling distribution so that the samples are more concentrated in those areas that contribute more to the target (AUC). The formula for calculating cross entropy is:

[0119] H(p,q)=-∑ i p(x i )logq(x i );

[0120] Where H(p,q) represents the cross entropy, p(x i ) represents the target distribution (i.e., the true probability distribution of the curve, based on the true shape of the receiver operating characteristic curve), q(x i ) represents the pre-sampling distribution.

[0121] In AUC estimation, the target distribution p(xi) can be determined by the shape of the ROC curve, while the current sampling distribution q(xi) is a distribution preliminarily set within the range of sensitivity and specificity.

[0122] Based on the cross entropy between the current sampling distribution and the receiver operating characteristic curve, the sampling distribution is updated and random points are generated again;

[0123] It should be noted that in order to reduce the cross entropy, the sampling distribution q(x) needs to be adjusted according to the currently calculated cross entropy value. Generally, the goal of this adjustment is to make the sampling more concentrated in the key area below the curve. General methods for updating the sampling distribution include:

[0124] Weighted sampling: Based on the cross entropy results, the sampling density of those areas that contribute more to the AUC can be increased. For example, an area with a large cross entropy value may mean that there are too few samples in that area, so the number of sampling points in that area needs to be increased.

[0125] Importance Sampling: In importance sampling, the updated sampling distribution is weighted according to the target distribution, ensuring that more sampling points are allocated to areas that contribute more to the AUC. For example, if the curve has a sharp change in some areas, the sampling distribution will increase the sampling density in these areas.

[0126] The specific update method is usually based on the following steps:

[0127] Compute the difference between the current sampling distribution q(x) and the target distribution p(x).

[0128] The sampling distribution is updated through gradient descent or other optimization methods to gradually approach the target distribution. The optimization process gradually concentrates the sampling distribution in areas with large changes in sensitivity and specificity, thereby optimizing the estimation of AUC.

[0129] After updating the sampling distribution, random points are generated again based on the new distribution. The new sampling points will be more concentrated in the area below the curve that contributes more to the AUC, thereby improving the accuracy of the estimation.

[0130] New random points are generated in a similar way as before, except this time they are more likely to be in high-density areas of the sampled distribution.

[0131] Since cross entropy minimization is an iterative process, it is necessary to continuously update the sampling distribution and generate new random points until the following conditions are met:

[0132] Convergence: The algorithm is considered to have converged when the cross entropy difference between the distribution of newly generated random points and the target distribution becomes very small, or when the calculated AUC value no longer changes significantly.

[0133] Accuracy meets requirements: By repeatedly adjusting the sampling distribution, the estimation accuracy of AUC can be gradually improved until it reaches the preset error range.

[0134] Calculate the distance between all generated random points and dynamically adjust the sampling density using the dispersed sampling method;

[0135] Specifically, the goal of the scattered sampling method is to ensure uniformity of sampling points throughout the coordinate space, avoiding excessive concentration of sampling points in some areas and insufficient sampling points in other areas. Based on the calculation results of the distance matrix, the following steps can be taken to dynamically adjust the sampling density:

[0136] Dense Area Identification: For point pairs with small distances, these points are too close together, and the sampling density in these areas can be appropriately reduced. These dense areas can be identified by calculating the distances between all points. A threshold can be set; when the distance between two points is less than the threshold, they are considered to be in a dense area, and the sampling density in this area needs to be reduced.

[0137] Sparse region identification: For point pairs with large distances, the area between these points may be relatively empty, and more sampling points are needed in these areas. If the distance between points in some areas is greater than a certain threshold, it can be considered that these areas are undersampled and the density of sampling points should be increased.

[0138] Dynamically adjust the sampling distribution: According to the dense and sparse areas identified above, the sampling distribution is dynamically adjusted. The specific method is:

[0139] Reduce sampling density in dense areas: If there are too many points in a certain area, you can reduce its density by reducing the sampling probability of these areas. For example, you can reduce the sampling weight of the area or reduce the number of random points in these areas.

[0140] Increase the sampling density of sparse areas: If there are insufficient sampling points in a certain area, you can increase the sampling density of that area and increase the number of random points in that area. This can be achieved by increasing the sampling probability of that area or increasing the number of points in that area through importance sampling.

[0141] Additionally, practical steps to implement a dispersed sampling method include:

[0142] (1) Calculate the distance between points: First, calculate the distance between all points using Euclidean distance and construct a distance matrix.

[0143] (2) Identify dense and sparse areas: Based on the distance matrix, determine which areas are dense areas and which are sparse areas.

[0144] (3) Adjusting sampling density: After identifying dense and sparse areas, the sampling distribution is dynamically adjusted so that the number of sampling points is greater in sparse areas and less in dense areas.

[0145] (4) Generate new random points: Based on the updated sampling distribution, regenerate random points so that the newly generated points are distributed more evenly, reducing clustered areas and filling sparse areas.

[0146] Determine whether each random point is below the receiver operating characteristic curve, count the number of random points that meet the preset conditions, and calculate the area under the curve.

[0147] It should be noted that for each randomly generated point, assuming its coordinates are (x, y), determining whether the point is below the curve includes:

[0148] If the y-coordinate of a random point (1-specificity value) is less than or equal to the y-value on the curve (corresponding to a specific sensitivity value x), the point is considered to be below the curve.

[0149] For example, for a particular random point, if the y-coordinate of the point is less than or equal to the y-value of the curve at the x-coordinate, the point is considered to be below the curve.

[0150] The area under the receiver operating characteristic curve was calculated by dividing the number of random points below the curve by the total number of random points.

[0151] The evaluation module is used to evaluate the effectiveness of the objective evaluation method for evaluating the nutritional status indicators of hemodialysis patients through correlation analysis based on the area under the curve.

[0152] As a preferred embodiment, the evaluation of the effectiveness of the objective evaluation method for evaluating the nutritional status index of hemodialysis patients by correlation analysis based on the area under the curve includes:

[0153] The Pearson correlation coefficient method was used to analyze the Pearson correlation coefficient between the objective assessment method and the actual nutritional status;

[0154] It should be noted that the Pearson correlation coefficient is an indicator that measures the strength and direction of the linear relationship between two variables. Its value ranges from -1 to +1, and its specific meaning is as follows:

[0155] r=1: indicates perfect positive correlation;

[0156] r=0: indicates no linear correlation;

[0157] r=-1: indicates perfect negative correlation.

[0158] Perform significance test on the obtained Pearson correlation coefficient;

[0159] The effectiveness of the objective evaluation method was analyzed based on the area under the curve of each objective evaluation method and combined with the Pearson correlation coefficient to obtain the effectiveness evaluation results.

[0160] Specifically, the relationship between each objective scoring method (assessment tool) and actual nutritional status was explained based on the calculated correlation coefficient.

[0161] High correlation (positive correlation): If the correlation coefficient is close to +1, it means that the assessment tool is highly consistent with the actual nutritional status, indicating that the assessment method is clinically reliable and can effectively assess the patient's nutritional status.

[0162] Low correlation: If the correlation coefficient is close to 0, it means that there is no obvious relationship between the assessment tool and the actual nutritional status, suggesting that the assessment method may not be suitable for this patient group and may need to be replaced or optimized.

[0163] Negative correlation: If the correlation coefficient is close to -1, it means that there is an inverse relationship between the score of the assessment tool and the actual nutritional status, indicating that the tool may not be suitable for assessing nutritional status.

[0164] In addition, the AUC values ​​(area under the curve) and correlation analysis results of each scoring tool were combined to comprehensively evaluate the effectiveness of each assessment method, including:

[0165] (1) High AUC value and high correlation coefficient: If the AUC value of the assessment tool is close to 1 and the Pearson correlation coefficient is close to 1 (or has a significant positive correlation), it means that the tool can not only effectively distinguish positive and negative cases, but also has a strong linear relationship with the actual nutritional status. In other words, the assessment tool has good classification performance in practical application and can reliably reflect the nutritional status of patients.

[0166] (2) High AUC value and low correlation coefficient: If the AUC value of the assessment tool is high but the Pearson correlation coefficient is low or close to 0, it means that the tool can effectively distinguish between positive and negative cases, but the linear relationship between its score and the actual nutritional status is weak. This may mean that the tool performs well in classification but cannot well reflect the actual nutritional status. In this case, the assessment tool has good classification ability but needs further optimization to improve the correlation with the actual nutritional status.

[0167] (3) Low AUC value and high correlation coefficient: If the AUC value of the assessment tool is low but the Pearson correlation coefficient is high, it means that the classification ability of the assessment tool is weak and cannot distinguish positive and negative cases well, but it has a strong linear relationship with the actual nutritional status. In this case, the assessment tool may reflect the nutritional status well through continuous variables (such as scores), but lacks sufficient discrimination ability, which may lead to misdiagnosis in clinical application.

[0168] (4) Low AUC value and low correlation coefficient: If the AUC value of the assessment tool is low and the Pearson correlation coefficient is close to 0, it means that the assessment tool not only performs poorly in classification but also has almost no relationship with the actual nutritional status. In this case, the assessment tool is not suitable for clinical assessment and may need to be replaced or improved.

[0169] Model building unit 2 is used to build a risk assessment model for hemodialysis patients based on the effectiveness evaluation results using matter-element extension variable weight theory and data fusion technology;

[0170] As a preferred embodiment, the model building unit 2 includes:

[0171] The matter-element matrix construction module is used to construct a matter-element matrix for risk assessment of hemodialysis patients based on nutritional status indicators and using the matter-element extension theory;

[0172] It should be noted that the Extension Theory of Matter Element is a comprehensive evaluation theory primarily used for analyzing multi-dimensional, multi-factor problems. It is particularly well-suited for evaluating complex systems, especially those containing multiple uncertainties.

[0173] The matter-element structure in matter-element extension theory usually consists of three parts: object (Element), attribute (Attribute) and weight (Weight).

[0174] Object: Here, the object of the object of the object element is each hemodialysis patient, or a group of indicators of the patient (such as protein energy consumption score, elderly nutritional risk index score, etc.).

[0175] Attributes: The attributes of each object are nutritional status indicators of hemodialysis patients, such as protein-energy expenditure score, elderly nutritional risk index score, modified creatinine index score, and malnutrition inflammatory response syndrome score;

[0176] Weight: Each attribute's weight represents its contribution to the overall assessment. Weights are typically derived from expert experience or statistical analysis. Weights can be pre-set or determined using data analysis methods such as the Analytic Hierarchy Process (AHP).

[0177] The key to constructing a matter-element matrix is ​​to express the evaluation indicators of each patient in the form of a matter-element matrix, where the elements in the matrix reflect the values ​​of the patient's various nutritional status indicators.

[0178] Set the value range of each attribute to [a, b] and determine the evaluation level corresponding to each attribute. The values ​​of each attribute can be filled in the matrix based on the patient's actual data (such as scores, blood biochemical indicators, etc.). Convert these indicators into normalized scores, for example, each indicator can be normalized to the range [0, 1]. Then, construct a matter-element matrix for each patient, with each row representing a patient and each column representing an evaluation indicator.

[0179] The weight adjustment module is used to preliminarily determine the weight of each nutritional status indicator using the hierarchical analysis method, and dynamically adjust the weight of each nutritional status indicator through the state variable weight theory in combination with the performance evaluation results;

[0180] As a preferred embodiment, the weight of each nutritional status indicator is preliminarily determined by using the analytic hierarchy process, and the weight of each nutritional status indicator is dynamically adjusted by the state variable weight theory in combination with the performance evaluation results, including:

[0181] The nutritional status assessment problem of hemodialysis patients is decomposed into several levels, including the target level and the criterion level. The judgment matrix is ​​constructed by comparing the nutritional status indicators in the criterion level with each other.

[0182] It should be noted that the nutritional status assessment problem of hemodialysis patients is decomposed hierarchically, usually including the target layer and the criterion layer (and possible sub-criteria layer). The specific decomposition is as follows:

[0183] Target layer: Nutritional status assessment of hemodialysis patients (i.e., assessment of patients' nutritional status and risks).

[0184] Criteria layer: Various nutritional status indicators are used to assess the patient's nutritional status. Common indicators include: protein-energy expenditure score, elderly nutritional risk index score, modified creatinine index score, and malnutrition inflammatory response syndrome score.

[0185] The constant weight vector of each nutritional status index was calculated using the eigenvector method in the hierarchical analysis method;

[0186] Specifically, according to the judgment matrix, the constant weight vector of each nutritional status indicator is calculated by the eigenvector method. The steps of the eigenvector method are as follows:

[0187] Construct a judgment matrix: Normalize the above judgment matrix, that is, divide each element by the sum of the column.

[0188] Compute Eigenvectors: Compute the eigenvectors of a matrix, usually by solving an eigenvalue problem.

[0189] Normalized eigenvector: Normalize the calculated eigenvector so that the sum of the eigenvectors is 1, and obtain the constant weight vector of each indicator.

[0190] Based on the performance evaluation results, the state vector is constructed and the state variable weight vector of each nutritional status indicator is calculated;

[0191] In state-variable weight theory, the state vector represents the current evaluation status of each nutritional status indicator. The calculation of the state vector requires the combination of each patient's actual data or performance evaluation results.

[0192] The performance score of each nutritional status indicator is used as an element of the state vector. Generally, the state vector can be expressed as S = [s1, s2, s3, s4, s5, s6], where s1, s2, ..., s6 are the performance scores of each nutritional status indicator respectively.

[0193] The weight of each nutritional status indicator is calculated using its constant weight vector and state variable weight vector.

[0194] As a preferred embodiment, the calculation formula for calculating the weight of each nutritional status indicator using the constant weight vector and the state variable weight vector of each nutritional status indicator is:

[0195]

[0196] Where W j (X) represents the weight of each nutritional status indicator, w j represents the constant weight vector calculated by the AHP method, S j (X) represents the state variable weight vector of the j-th nutritional status indicator, and m represents the number of nutritional status indicators.

[0197] The risk assessment module is used to calculate the comprehensive risk assessment data of each patient based on the weight of each nutritional status indicator and using data fusion technology to fuse the nutritional status indicator data in the matter-element matrix with its corresponding weight;

[0198] As a preferred embodiment, the weight of each nutritional status indicator is based on the use of data fusion technology to fuse the nutritional status indicator data in the matter-element matrix with its corresponding weight to calculate the comprehensive risk assessment data of each patient, including:

[0199] The matter-element matrix is ​​normalized using the minimum-maximum normalization method to obtain a normalized matter-element matrix;

[0200] The normalized matter-element matrix is ​​fused with the weights of each nutritional status indicator using the weighted average method to obtain a weighted normalized matrix.

[0201] Each row of the weighted normalized matrix is ​​summed to obtain the comprehensive risk assessment data for each patient.

[0202] The model building and optimization module is used to build and optimize the risk assessment model for hemodialysis patients based on comprehensive risk assessment data and through adaptive learning optimization method.

[0203] As a preferred embodiment, the risk assessment model for hemodialysis patients is constructed and optimized based on the comprehensive risk assessment data and through an adaptive learning optimization method, including:

[0204] The clinical status coefficient of each hemodialysis patient was calculated using the basic clinical information of the hemodialysis patients;

[0205] As a preferred embodiment, the calculation of the clinical status coefficient of each hemodialysis patient using the basic clinical information of the hemodialysis patient includes:

[0206] Preprocessing the basic clinical information of hemodialysis patients, including data cleaning and data standardization;

[0207] It should be noted that the purpose of data cleaning is to remove noise and outliers in the data to ensure data quality. Data cleaning usually includes the following operations:

[0208] Handling missing values: Missing values ​​can be handled by filling in the values ​​(e.g. using the mean, mode or median) or by removing rows with missing data.

[0209] Handling outliers: Detect and remove obvious outliers through statistical methods (such as z-score normalization method).

[0210] Deduplication: Checks the data set for duplicate records and deletes the duplicate data.

[0211] Format standardization: Ensure that data formats are consistent, such as standardizing dates to the same format and strings to lowercase or uppercase.

[0212] Data standardization is the process of converting basic clinical information of different dimensions or scales (such as age, blood pressure, weight, etc.) into the same scale to facilitate subsequent analysis.

[0213] Extract effective features from pre-processed clinical basic information to obtain effective feature vectors;

[0214] Specifically, once data preprocessing is complete, effective features that have a significant impact on the clinical status of hemodialysis patients can be extracted. These features will serve as input for the subsequent calculation of clinical status coefficients. These include:

[0215] (1) Feature selection: The purpose of feature selection is to select the most influential features for clinical status assessment from a large amount of clinical information. The following methods can be used:

[0216] Expert experience method: important features are selected based on the knowledge and experience of clinicians, such as age, blood pressure, weight, dialysis type, dialysis frequency, etc.

[0217] Correlation analysis: By calculating the correlation coefficient between features (such as the Pearson correlation coefficient), features with a high correlation with the target variable (such as clinical status) are screened out.

[0218] Statistical analysis methods: Select features with strong significance through statistical analysis methods such as chi-square test and t-test.

[0219] Machine learning algorithms: such as random forest and LASSO regression, are used to rank feature importance and select features with greater influence.

[0220] (2) Feature vector construction: After extracting valid features from the preprocessed data, they are combined into a valid feature vector. The valid feature vector of each patient can be expressed as: Fi = [f1, f2, f3, ..., fn];

[0221] Among them: f1, f2,…, fn are the effective features of the i-th patient, which are usually standardized clinical data.

[0222] According to the extracted effective feature vectors, the clinical status coefficients of hemodialysis patients were calculated using the weighted summation method.

[0223] It should be noted that each feature has a different impact on the patient's clinical status, so a weight needs to be assigned to each feature. The weight can be obtained through the following methods:

[0224] Expert scoring method: Different weights are assigned to each feature based on the expert's clinical experience and importance ranking.

[0225] Statistical analysis method: Use correlation analysis, regression analysis and other methods to calculate weights based on the relationship between characteristics and clinical status.

[0226] Machine learning model: Calculate the relative importance of features through machine learning algorithms (such as decision trees, random forests, etc.) and assign weights to each feature.

[0227] Through this weighted summation method, the basic clinical information of each patient can be converted into a clinical status coefficient, which reflects the overall clinical status of the patient.

[0228] Based on the comprehensive risk assessment data and behavioral status coefficient of each hemodialysis patient, a risk assessment model for initial hemodialysis patients is constructed;

[0229] Specifically, first, the comprehensive risk assessment data and behavioral state coefficient of each patient are used as input features. Assume that the data of each patient contains: X i =[R i ,B i ];

[0230] Among them, R i is the comprehensive risk assessment data of the i-th patient, B i is the behavioral state coefficient of the i-th patient. These two data items can be combined to form a feature vector X i , used as input into the risk assessment model.

[0231] In addition, the initial risk assessment model can be constructed using common machine learning algorithms such as linear regression, support vector machine (SVM), and logistic regression. To simplify the problem, the logistic regression model is used as an example to illustrate how to construct an initial risk assessment model based on input data.

[0232] Assume that the goal is to predict the risk level of each patient (such as low, medium, or high risk). For the i-th patient, the risk probability output by the model is Pi, which is calculated as follows:

[0233] P i =σ(W T X i +b);

[0234] Among them, σ(z) represents the sigmoid function in logistic regression, W represents the weight vector, which represents the weight of each feature (such as comprehensive risk assessment data and behavioral state coefficient), X i represents the input feature vector (i.e., the patient's comprehensive risk assessment data and behavioral state coefficient), and b represents the bias term.

[0235] Based on the adaptive learning optimization method, the initial risk assessment model of hemodialysis patients was optimized in combination with gradient descent and cross entropy loss functions to obtain the final risk assessment model of hemodialysis patients.

[0236] Specifically, after the initial model is built, an optimization algorithm is needed to improve the model. The most commonly used method is gradient descent. The goal of gradient descent is to calculate the gradient of the loss function with respect to the model parameters through backpropagation and to reduce the value of the loss function by iteratively updating the parameters.

[0237] The cross-entropy loss function is used to calculate the difference between the model's predicted value and the true value. It is applicable to classification problems and optimizes the model's weights and biases by minimizing the cross-entropy loss function through the gradient descent method.

[0238] To further optimize model performance, you can introduce adaptive learning rate optimization methods, such as the Adam optimization algorithm. The Adam algorithm combines the advantages of gradient descent and adaptive learning rate adjustment to effectively accelerate convergence. Using the Adam optimization algorithm, the model dynamically adjusts the learning rate of each parameter, thereby finding the optimal solution more quickly.

[0239] By collecting a training set and then using the cross-entropy loss function and gradient descent (or the Adam optimization algorithm) to train the initial model, an optimized risk assessment model for hemodialysis patients is ultimately obtained. The optimized model can more accurately predict the patient's risk level and adaptively adjust to different input features.

[0240] The risk assessment unit 3 is used to perform individualized risk assessment on the hemodialysis patient using the risk assessment model for the hemodialysis patient to obtain a risk assessment result;

[0241] Specifically, using the risk assessment model for hemodialysis patients to conduct individualized risk assessment for the patient and obtain the assessment results includes the following steps:

[0242] Step 1: Prepare input data;

[0243] First, it is necessary to obtain real-time basic clinical information, nutritional status indicators, and other factors that may affect risk for each hemodialysis patient. This information may include but is not limited to:

[0244] The patient's basic information such as age, gender, weight, height, and dialysis history.

[0245] Various health-related biochemical indicators (such as blood sugar, blood pressure, serum albumin, creatinine, etc.).

[0246] The comprehensive risk assessment data and behavioral state coefficients obtained by the calculation method mentioned in the present invention.

[0247] Step 2: Standardize input data;

[0248] Before inputting data for risk assessment, different characteristics (such as age, weight, nutritional status indicators, etc.) need to be standardized to eliminate scale differences so that different characteristics have the same influence in the model.

[0249] Step 3: Calculate individualized risk assessment results;

[0250] The trained risk assessment model is used to predict the input data of each patient. For each patient, the input features (such as comprehensive risk assessment data, behavioral state coefficient, etc.) will be calculated through the model to obtain a predicted value (such as the probability of disease risk or risk level).

[0251] If the predicted value is close to 1, it means the patient is at higher risk (eg, likely in the high-risk dialysis group).

[0252] If the predicted value is close to 0, it means the patient is at low risk (eg, belongs to the low-risk dialysis group).

[0253] Based on the risk probability output by the model, the patient's risk can be classified into low risk, medium risk and high risk.

[0254] In summary, with the help of the above technical solutions of the present invention, the present invention ensures the scientificity and reliability of risk assessment, and constructs a dynamic and personalized risk assessment model through the material element extension variable weight theory and data fusion technology, which can provide personalized risk assessment for each patient, and help clinicians optimize treatment plans, improve patient management effects, and contribute to the efficient allocation of medical resources, early identification of high-risk patients, reduce the occurrence of complications, and improve patient prognosis. The integration of multiple technologies and theories enhances the innovation and flexibility of the assessment system. The present invention calculates the area under the receiver operating characteristic curve by the Monte Carlo algorithm. This method can provide a relatively accurate and flexible estimation method. By calculating the sensitivity and specificity under different thresholds, a preliminary ROC curve is constructed, and the final curve is further optimized to help ensure the accuracy and effectiveness of the curve. Combined with the area under the curve and the Pearson correlation coefficient, the effectiveness of the objective assessment method is comprehensively evaluated, which helps to more comprehensively understand the advantages and disadvantages of each assessment method. Doctors can more accurately understand the effectiveness of different objective assessment methods in the nutritional status assessment of hemodialysis patients. The present invention constructs a matter-element matrix for risk assessment of hemodialysis patients through matter-element extension theory, which can systematically integrate and analyze multiple nutritional status indicators, provide a comprehensive data basis for risk assessment, and dynamically adjust the weights through state variable weight theory in combination with the performance assessment results, so that the weights can be flexibly adjusted according to the actual status of the patient, thereby improving the accuracy and adaptability of risk assessment. Through data fusion technology, the nutritional status indicator data in the matter-element matrix are fused with their corresponding weights to calculate the comprehensive risk assessment data of each patient, realizing a comprehensive evaluation of multiple indicators. Based on the comprehensive risk assessment data, a risk assessment model is constructed and optimized through an adaptive learning optimization method, so that the model can continuously learn and adapt to new data, thereby improving the accuracy and stability of prediction.

[0255] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, optical storage, etc.) containing computer-usable program code.

[0256] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A risk assessment system for hemodialysis patients based on an objective assessment method, characterized in that: include: The efficacy assessment unit is used to collect basic clinical information of hemodialysis patients, calculate nutritional status indicators through objective assessment methods, and evaluate the efficacy of objective assessment methods for nutritional status indicators using the receiver operating characteristic curve based on the Youden index; A model building unit is used to build a risk assessment model for hemodialysis patients based on the effectiveness evaluation results using matter-element extension variable weight theory and data fusion technology; a risk assessment unit, configured to perform individualized risk assessment on hemodialysis patients using a risk assessment model for hemodialysis patients, and obtain a risk assessment result; Wherein, the performance evaluation unit includes: A numerical calculation module is used to calculate the area under the curve for each receiver operating characteristic curve using a Monte Carlo algorithm of cross entropy sampling and dispersion sampling; The Monte Carlo algorithm for calculating the area under the curve using cross-entropy sampling and dispersed sampling includes: setting a preliminary sampling distribution for random point generation and generating a number of random points; preliminarily determining a key area for each random point, and calculating the cross entropy between the current sampling distribution and the receiver operating characteristic curve; updating the sampling distribution based on the cross entropy between the current sampling distribution and the receiver operating characteristic curve, and generating random points again; calculating the distance between all generated random points, and dynamically adjusting the sampling density using dispersed sampling; counting the number of random points that meet preset conditions, and calculating the area under the curve; The objective scoring module is used to calculate the protein-energy consumption score, the elderly nutritional risk index score, the modified creatinine index score, and the malnutrition inflammatory response syndrome score based on the collected basic clinical information and multiple objective assessment methods to obtain an objective score set; The curve construction module is used to calculate the sensitivity and specificity based on the Youden index and each objective score to construct the receiver operating characteristic curve; An evaluation module is used to evaluate the effectiveness of objective evaluation methods for evaluating nutritional status indicators of hemodialysis patients through correlation analysis based on the area under the curve; The Youden index is used to calculate the sensitivity and specificity of each objective score, construct the receiver operating characteristic curve, and determine the optimal judgment threshold, including: For each customer score, the sensitivity and specificity at different thresholds are calculated at several pre-set thresholds; Based on the calculated sensitivity and specificity, a preliminary receiver operating characteristic curve was constructed; The Youden index at each threshold was calculated, and the threshold with the maximum Youden index was selected as the optimal judgment threshold. The preliminary receiver operating characteristic curve corresponding to the optimal judgment threshold was used as the final receiver operating characteristic curve. The effectiveness of the objective assessment method for evaluating the nutritional status index of hemodialysis patients by correlation analysis based on the area under the curve includes: The Pearson correlation coefficient method was used to analyze the Pearson correlation coefficient between the objective assessment method and the actual nutritional status; Perform significance test on the obtained Pearson correlation coefficient; The effectiveness of the objective evaluation method was analyzed based on the area under the curve of each objective evaluation method and combined with the Pearson correlation coefficient to obtain the effectiveness evaluation results; The model building unit includes: The matter-element matrix construction module is used to construct a matter-element matrix for risk assessment of hemodialysis patients based on nutritional status indicators and using the matter-element extension theory; The weight adjustment module is used to preliminarily determine the weight of each nutritional status indicator using the hierarchical analysis method, and dynamically adjust the weight of each nutritional status indicator through the state variable weight theory in combination with the performance evaluation results; The risk assessment module is used to calculate the comprehensive risk assessment data of each patient based on the weight of each nutritional status indicator and using data fusion technology to fuse the nutritional status indicator data in the matter-element matrix with its corresponding weight; A model building and optimization module is used to build and optimize the risk assessment model for hemodialysis patients based on comprehensive risk assessment data and through adaptive learning optimization methods; The method of using the analytic hierarchy process to preliminarily determine the weight of each nutritional status indicator and, in combination with the performance evaluation results, dynamically adjusting the weight of each nutritional status indicator through the state variable weight theory includes: The nutritional status assessment problem of hemodialysis patients is decomposed into several levels, including the target level and the criterion level. The judgment matrix is ​​constructed by comparing the nutritional status indicators in the criterion level with each other. The constant weight vector of each nutritional status index was calculated using the eigenvector method in the hierarchical analysis method; Based on the performance evaluation results, the state vector is constructed and the state variable weight vector of each nutritional status indicator is calculated; The weight of each nutritional status indicator is calculated using its constant weight vector and state variable weight vector.

2. A hemodialysis patient risk assessment system based on an objective assessment method according to claim 1, characterized in that: The calculation formula for calculating the weight of each nutritional status indicator using the constant weight vector and the state variable weight vector of each nutritional status indicator is: Where W j (X) represents the weight of each nutritional status indicator, w j represents the constant weight vector calculated by the AHP method, S j (X) represents the state variable weight vector of the j-th nutritional status indicator, and m represents the number of nutritional status indicators.

3. A hemodialysis patient risk assessment system based on an objective assessment method according to claim 2, characterized in that: Based on the weight of each nutritional status indicator, the data fusion technology is used to fuse the nutritional status indicator data in the matter-element matrix with its corresponding weight to calculate the comprehensive risk assessment data of each patient, including: The matter-element matrix is ​​normalized using the minimum-maximum normalization method to obtain a normalized matter-element matrix; The normalized matter-element matrix is ​​fused with the weights of each nutritional status indicator using the weighted average method to obtain a weighted normalized matrix. Each row of the weighted normalized matrix is ​​summed to obtain the comprehensive risk assessment data for each patient.

4. A hemodialysis patient risk assessment system based on an objective assessment method according to claim 3, characterized in that: The risk assessment model for hemodialysis patients is constructed and optimized based on comprehensive risk assessment data and through an adaptive learning optimization method, including: The clinical status coefficient of each hemodialysis patient was calculated using the basic clinical information of the hemodialysis patients; Based on the comprehensive risk assessment data and behavioral status coefficient of each hemodialysis patient, a risk assessment model for initial hemodialysis patients is constructed; Based on the adaptive learning optimization method, the initial risk assessment model of hemodialysis patients was optimized in combination with gradient descent and cross entropy loss functions to obtain the final risk assessment model of hemodialysis patients.

5. A hemodialysis patient risk assessment system based on an objective assessment method according to claim 4, characterized in that: The method of calculating the clinical status coefficient of each hemodialysis patient using the basic clinical information of the hemodialysis patient includes: Preprocessing the basic clinical information of hemodialysis patients, including data cleaning and data standardization; Extract effective features from pre-processed clinical basic information to obtain effective feature vectors; According to the extracted effective feature vectors, the clinical status coefficients of hemodialysis patients were calculated using the weighted summation method.

Images