Hemodialysis capacity load prediction method and terminal
By acquiring and analyzing clinical parameter data before and after hemodialysis, and using machine learning to establish a prediction model, the problem of inaccurate evaluation of capacity load after dialysis in the prior art is solved, and the accuracy of prediction and the optimization efficiency of dialysis scheme are improved.
Patent Information
- Application Number
- CN202411390101.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-08
- Publication Date
- 2025-01-03
AI Technical Summary
The prior art cannot accurately and comprehensively evaluate the volume load status of hemodialysis patients after dialysis, resulting in lag in the optimization of dialysis ultrafiltration scheme and insufficient accuracy of a single evaluation method.
By obtaining the clinical parameter data before and after hemodialysis, setting the label of whether the capacity load is overloaded, performing correlation analysis and screening, obtaining characteristic variables related to dialysis capacity load overload, and using machine learning to establish the optimal prediction of capacity load after dialysis Model.
It improves the accuracy of capacity load prediction after dialysis, can reflect the capacity load status more comprehensively and accurately, and guides the optimization of dialysis ultrafiltration scheme.
Smart Images
Figure CN119252489A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical data prediction, and in particular to a method and a terminal for predicting hemodialysis capacity load. Background Art
[0002] Ultrafiltration of excess water from the body through hemodialysis is an important means of managing volume overload in maintenance hemodialysis patients. Therefore, comprehensive and quantitative assessment of the volume load of hemodialysis patients to guide the formulation of a reasonable hemodialysis ultrafiltration program is the key to effectively managing volume overload. When patients have volume overload, their weight, vascular volume, or interstitial water will change accordingly, and the extent of these changes reflects the degree of volume load from different aspects. Currently, a variety of methods for assessing volume load are used in clinical practice, including interdialytic weight changes, laboratory tests such as brain natriuretic peptide (BNP), lung ultrasound, inferior vena cava ultrasound, and bioelectrical impedance analysis (BIS), which are to assess whether there is volume overload by assessing changes in one of these aspects. These methods have their own advantages and may complement each other. However, no single method can accurately and comprehensively assess the degree of volume load. Therefore, it is necessary to establish a method for assessing the multidimensional volume status of hemodialysis patients.
[0003] Some studies have explored the combination of different assessment methods, such as using BIS results combined with BNP measurement, to improve the accuracy of assessing volume overload. However, the clinical assessment of patient volume load is not limited to BIS and BNP. If more volume load indicators can be combined, it is expected to reflect the volume load more comprehensively and accurately. In addition, the current assessment method is limited to assessing the current volume load state of hemodialysis patients, and it is impossible to predict the post-dialysis volume load state. Therefore, the assessment results of volume load have a lag in guiding the optimization of dialysis ultrafiltration schemes. The existing invention patent "A multidimensional assessment system for hemodialysis volume load" (patent number CN202311382393.2) attempts to predict the post-dialysis volume load state based on the concentration of electrolytes in hemodialysis patients. However, the patent does not include the currently commonly used volume assessment parameters in clinical practice, and the concentration of electrolytes in hemodialysis patients is not a conventional indicator for assessing volume load. The patent lacks corresponding data analysis and theoretical basis for the reason that only electrolytes in hemodialysis patients are included in the analysis and other commonly used volume assessment parameters are excluded. This will be one-sided in the prediction of post-dialysis volume load and even affect the accuracy. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a method and a terminal for predicting hemodialysis volume load, which can improve the accuracy of post-dialysis volume load prediction.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:
[0006] A method for predicting hemodialysis volume load, comprising the steps of:
[0007] S1. Acquire clinical parameter data before and after historical hemodialysis, and set a label indicating whether the volume load is overloaded for the clinical parameter data;
[0008] S2. performing correlation analysis and screening on the clinical parameter data and their labels to obtain characteristic variables related to dialysis volume overload;
[0009] S3. Use the characteristic variables to perform machine learning to establish an optimal prediction model for predicting post-dialysis volume overload, and use the optimal prediction model to predict post-dialysis volume load.
[0010] In order to solve the above technical problems, another technical solution adopted by the present invention is:
[0011] A terminal for predicting hemodialysis capacity load comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step of the above-mentioned method for predicting hemodialysis capacity load is implemented.
[0012] The beneficial effects of the present invention are: obtaining clinical parameter data before and after historical hemodialysis, and setting a label for the clinical parameter data to indicate whether the volume load is overloaded, so as to perform overload analysis through the clinical parameter data of the existing primary dialysis unit; performing correlation analysis and screening on the clinical parameter data and its labels, and obtaining characteristic variables related to dialysis volume load overload, that is, obtaining relevant characteristic variables through layer-by-layer screening; using the characteristic variables for machine learning, establishing an optimal prediction model for predicting post-dialysis volume load overload, and using the optimal prediction model to predict post-dialysis volume load. In this way, the accuracy of post-dialysis volume load prediction is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 A flowchart of a method for predicting hemodialysis volume load according to an embodiment of the present invention;
[0014] Figure 2 This is a schematic diagram of a terminal for predicting hemodialysis capacity load according to an embodiment of the present invention;
[0015] Figure 3 A flowchart of a prediction model construction of a method for predicting hemodialysis volume load according to an embodiment of the present invention;
[0016] Figure 4 The first receiver operating characteristic curve graph corresponding to the random forest prediction model validation data set (AUC = 0.96);
[0017] Figure 5 The first receiver operating characteristic curve graph corresponding to the neural network prediction model validation data set (AUC = 0.86);
[0018] Figure 6 The first receiver operating characteristic curve graph corresponding to the decision tree prediction model validation data set (AUC = 0.84);
[0019] Figure 7 The first receiver operating characteristic curve graph corresponding to the Bayesian prediction model validation data set (AUC = 0.88);
[0020] Figure 8 The first receiver operating characteristic curve graph corresponding to the validation data set of the gradient boosting algorithm prediction model (AUC = 0.81);
[0021] Fig. 9 The first receiver operating characteristic curve graph corresponding to the support vector machine prediction model validation data set (AUC = 0.87);
[0022] Fig.10 This is the first receiver operating characteristic curve graph corresponding to the K-nearest neighbor algorithm prediction model validation data set (AUC=0.86).
[0023] Description of labels:
[0024] 1. A terminal for predicting hemodialysis capacity load; 2. A memory; 3. A processor. DETAILED DESCRIPTION
[0025] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.
[0026] Please refer to Figure 1 The embodiment of the present invention provides a method for predicting hemodialysis volume load, comprising the steps of:
[0027] S1. Acquire clinical parameter data before and after historical hemodialysis, and set a label indicating whether the volume load is overloaded for the clinical parameter data;
[0028] S2. performing correlation analysis and screening on the clinical parameter data and their labels to obtain characteristic variables related to dialysis volume overload;
[0029] S3. Use the characteristic variables to perform machine learning to establish an optimal prediction model for predicting post-dialysis volume overload, and use the optimal prediction model to predict post-dialysis volume load.
[0030] From the above description, it can be seen that the beneficial effects of the present invention are: obtaining clinical parameter data before and after historical hemodialysis, and setting a label for the clinical parameter data to indicate whether the volume load is overloaded, so as to perform overload analysis through the clinical parameter data of the existing primary dialysis unit; performing correlation analysis and screening on the clinical parameter data and its labels, and obtaining characteristic variables related to dialysis volume load overload, that is, obtaining relevant characteristic variables through layer-by-layer screening; using the characteristic variables for machine learning, and establishing an optimal prediction model for predicting post-dialysis volume load overload, so as to use the optimal prediction model to predict post-dialysis volume load. In this way, the accuracy of post-dialysis volume load prediction is improved.
[0031] Further, step S1 includes:
[0032] Acquiring clinical parameter data before and after historical hemodialysis, wherein the clinical parameter data includes dialysis parameters and laboratory test data before dialysis, and bioelectrical impedance analysis data and ultrasound evaluation data before and after dialysis;
[0033] Provide a method for judging whether the dialysis parameters and laboratory test data before dialysis in the clinical data meet preset conditions, whether the bioelectrical impedance analysis data before and after dialysis meet preset overhydration conditions, and whether the ultrasound assessment data after dialysis meets preset capacity overload conditions. If all conditions are met, a capacity overload label is set; otherwise, a capacity not overload label is set.
[0034] From the above description, it can be seen that by obtaining dialysis parameters and laboratory test data before dialysis, as well as bioelectrical impedance analysis data and ultrasound assessment data before and after dialysis, a variety of commonly used clinical indicators for assessing volume load can be used to assess whether the volume load is overloaded; and a variety of volume assessment methods can be combined to predict whether there is volume load overload after dialysis, and it is also convenient to explore the optimal volume load state. The previous scheme was based on a comprehensive analysis of the patient's dialysis parameters over a period of time to determine the appropriate state or simulated the volume state based on multiple dialysis monitoring parameters. In fact, it is impossible to truly quantify the volume state after dialysis and cannot solve the problem of improving the patient's volume load overload.
[0035] Further, step S2 includes:
[0036] Performing correlation analysis and screening on the clinical parameter data and their labels to obtain a feature vector related to dialysis volume overload;
[0037] The Lasso regression algorithm is used to screen out characteristic variables with non-zero coefficients for the characteristic vector.
[0038] Furthermore, correlation analysis and screening are performed on the clinical parameter data and their labels to obtain feature vectors related to dialysis volume overload, including:
[0039] Using univariate analysis to perform correlation analysis on the clinical parameter data and their labels, and obtaining a preliminary screening feature vector related to post-dialysis volume overload;
[0040] The parameter data without multicollinearity in the initial screening feature vector were included in the multivariate logistic regression analysis to screen out the potential feature vector for predicting post-dialysis volume overload.
[0041] From the above description, we can know that we first compare the groups, that is, divide the patients into two groups according to whether there is post-dialysis volume overload, and compare the differences in pre-dialysis indicators between the two groups by univariate analysis, and preliminarily screen out indicators with statistical differences P<0.05. Then, multicollinearity analysis is performed on the initial screening indicators to eliminate parameters with multicollinearity, and then multivariate logistic regression is performed on the remaining parameters to further screen out parameters with potential predictive value. Finally, these parameters with potential predictive value are further subjected to Lasso regression analysis, and the weights of several parameters are compared to further screen out parameters formally used for machine learning modeling. Unlike previous studies that only included all existing parameters into the modeling, pre-screening of modeling parameters can not only remove confounding factors and make the model more stable, but also further control the number of modeling variables, reduce the number of cases required for modeling, and improve the efficiency of subsequent modeling success.
[0042] Furthermore, in step S3, using the characteristic variables to perform machine learning to establish an optimal prediction model for predicting post-dialysis volume overload includes:
[0043] Randomly dividing the clinical parameter data into a training cohort and a test cohort, using at least two machine learning algorithms to analyze potential feature vectors in the training cohort, and establishing a corresponding training model;
[0044] The test cohort is substituted into the training model for model verification and model evaluation, and the training model with the best evaluation result is taken as the optimal prediction model.
[0045] From the above description, it can be seen that a variety of machine learning methods are used to establish the optimal model for predicting post-dialysis volume overload, and a test cohort is used to verify and evaluate the model. The training model with the best evaluation result is used as the optimal prediction model to further ensure the accuracy of the prediction model in predicting post-dialysis volume load.
[0046] Please refer to Figure 2 Another embodiment of the present invention provides a hemodialysis capacity load prediction terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements each step of the above-mentioned hemodialysis capacity load prediction method when executing the computer program.
[0047] The above-mentioned method and terminal for predicting hemodialysis volume load of the present invention are suitable for improving the accuracy of post-dialysis volume load prediction, and are described below through specific implementation methods:
[0048] Embodiment 1
[0049] Please refer to Figure 1 and Figure 3 , a method for predicting hemodialysis volume load, comprising the steps of:
[0050] S1. Obtain clinical parameter data before and after historical hemodialysis, and set a label indicating whether the volume load is overloaded for the clinical parameter data.
[0051] In this example, eligible hemodialysis patients were enrolled, and each patient's primary dialysis unit was evaluated.
[0052] S11. Acquire clinical parameter data before and after historical hemodialysis, wherein the clinical parameter data include dialysis parameters and laboratory test data before dialysis, and bioelectrical impedance analysis data and ultrasound evaluation data before and after dialysis.
[0053] Specifically, the dialysis parameters before dialysis include but are not limited to: dry weight, vascular access, ultrafiltration volume (UFV), pre-dialysis weight, and blood pressure; laboratory test data include but are not limited to: hemoglobin (Hb), hematocrit (HCT), urea nitrogen (BUN), creatinine (Cr), potassium (K), phosphorus, calcium (Ca) and sodium (Na), albumin (ALB), prealbumin (PA), cholesterol (Ch), triglycerides (TG), serum iron (Fe), parathyroid hormone (PTH), and brain natriuretic peptide (BNP). In addition, the patient's general condition data, such as gender, age, and diabetes, should be obtained.
[0054] The bioelectrical impedance analysis data before and after dialysis included: pre-dialysis over-water (OH), total body water (TBW), extracellular water (ECW), intracellular water (ICW), ECW / ICW ratio (E / I), and post-dialysis over-water (OH).
[0055] Ultrasound evaluation data before and after dialysis included: obtaining lung ultrasound (US) B-line measurement parameters before and after dialysis and inferior vena cava measurement parameters, such as the inspiratory inferior vena cava diameter index (IVCDImin) and expiratory inferior vena cava diameter index (IVCDImax).
[0056] S12. Provide a method for judging whether the dialysis parameters and laboratory test data before dialysis in the clinical data meet preset conditions, whether the bioelectrical impedance analysis data before and after dialysis meet preset overhydration conditions, and whether the ultrasound assessment data after dialysis meets preset capacity overload conditions. If all conditions are met, set a capacity overload label; otherwise, set a capacity not overload label.
[0057] Specifically, the assessment of post-dialysis volume overload can be based on the following criteria: failure to reach the dry weight target, post-dialysis BIS indicating overhydration, or post-dialysis lung ultrasound indicating pulmonary interstitial edema. Therefore, based on this criterion, the post-dialysis weight, post-dialysis BIS and post-dialysis lung ultrasound results of the enrolled patients were comprehensively analyzed to evaluate whether the patients had post-dialysis volume overload.
[0058] S2. Perform correlation analysis and screening on the clinical parameter data and their labels to obtain characteristic variables related to dialysis volume overload.
[0059] S21. Use univariate analysis to perform correlation analysis on the clinical parameter data and their labels to obtain a preliminary screening feature vector related to post-dialysis volume overload.
[0060] Specifically, univariate analysis was used to perform significant correlation analysis on the clinical information in the clinical parameter data and the post-dialysis volume overload, and multiple initial screening characteristic variables related to post-dialysis volume overload were obtained, wherein the multiple initial screening characteristic variables obtained included: age, diabetes, systolic blood pressure, Hb, HCT, Cr, Ch, IVCDImax, IVCDImin, US-B line, OH, UFV / pre-dialysis weight ratio, TBW, ECW, and E / I.
[0061] S22. Incorporating parameter data without multicollinearity in the initial screening feature vector into multivariate logistic regression analysis to screen out potential feature vectors for predicting post-dialysis volume overload.
[0062] Specifically, the indicators of age, systolic blood pressure, Hb, HCT, Cr, Ch, IVCDImax, IVCDImin, US-B lines, OH, and UFV / predialysis weight ratio without multicollinearity (tolerance>0.1 and variance inflation factor<10) in the initial screening characteristic variables were included in the multivariate logistic regression analysis, and the potential characteristic variables for predicting postdialysis volume overload were further screened out: age, UFV / predialysis weight ratio, OH (P<0.001), and predialysis US-B lines.
[0063] S23, using the Lasso regression algorithm to select characteristic variables with non-zero coefficients for the characteristic vector.
[0064] Specifically, in order to establish a prediction model for postdialysis volume overload, the LASSO regression algorithm was further used to screen out characteristic variables with non-zero coefficients for the training cohort: UFV / predialysis body weight ratio, OH (P < 0.001), and predialysis US-B line.
[0065] Therefore, in step S2, firstly compare the groups, that is, divide the patients into two groups according to whether there is post-dialysis volume overload, compare the differences in pre-dialysis indicators between the two groups by univariate analysis, and preliminarily screen out indicators with statistical differences P<0.05, and then perform multicollinearity analysis on the initial screening indicators to eliminate the parameters with multicollinearity, and then further perform multivariate logistic regression on the remaining parameters to further screen out parameters with potential predictive value, and finally perform Lasso regression analysis on these parameters with potential predictive value, compare the weights of several parameters, and further screen out the parameters formally used for machine learning modeling, and then use machine learning methods to establish a prediction model. Different from previous studies that only included all existing parameters into the modeling, the preliminary screening of modeling parameters can not only remove confounding factors and make the model more stable, but also further control the number of modeling variables, reduce the number of cases required for modeling, and improve the efficiency of subsequent modeling success.
[0066] S3. Use the characteristic variables to perform machine learning to establish an optimal prediction model for predicting post-dialysis volume overload, and use the optimal prediction model to predict post-dialysis volume load.
[0067] S31. Randomly divide the clinical parameter data into a training cohort and a test cohort, use at least two machine learning algorithms to analyze the potential feature vectors in the training cohort, and establish a corresponding training model.
[0068] Specifically, the clinical parameter data were randomly divided into a training cohort and a test cohort in a ratio of 7:3. In this embodiment, 7 machine learning algorithms were used to analyze the potential predictive factors in the training cohort, including random forest, naive Bayes, support vector machine, K-nearest neighbor algorithm, decision tree, gradient boosting algorithm and neural network analysis, to establish a training model. The test cohort was substituted into the training model for model verification. The area under the receiver operating characteristic (ROC) curve (AUC), accuracy, precision, recall rate and F1 score (F1 value is the harmonic mean of precision and recall rate) were calculated, please refer to Figures 4 to 10 , to evaluate the performance of the model. Finally, the best performing random forest model was selected as the prediction model.
[0069] In addition, the prediction model code will be further improved to achieve continuous training and optimization of the prediction model by continuously increasing the sample size on the existing basis, so as to establish a more stable prediction model.
[0070] S32, substituting the test queue into the training model for model verification and model evaluation, and taking the training model with the best evaluation result as the optimal prediction model.
[0071] Specifically, the patient's planned dialysis ultrafiltration regimen, i.e., UFV, pre-dialysis weight, pre-dialysis OH, and pre-dialysis US-B line, are obtained before dialysis and input into the prediction model for analysis. This can predict the post-dialysis volume overload, i.e., under the current conditions, whether there is volume overload after dialysis if dialysis is performed with the current UFV ultrafiltration regimen.
[0072] Therefore, the modeling of this embodiment is based on the targeted initial screening of multiple dimensional parameters. Compared with the multiple parameters that were included without initial screening in the previous scheme, the prediction weight for the result is greater and the confounding factors are smaller. Compared with the previous scheme, more than one machine learning algorithm is used, and the best of the models established by multiple algorithms is selected, so the results are more reliable and accurate. In addition, this scheme is more efficient than the previous scheme. The previous scheme needs to summarize the measurement parameters of multiple dialysis to obtain the evaluation model of the patient's capacity load. According to dialysis 3 times a week, it takes several weeks to achieve modeling; while this scheme analyzes the parameters before dialysis once to build a model, and only needs to measure the relevant parameters before this dialysis to obtain the prediction result of the capacity load after dialysis, which is more efficient in terms of the time to obtain the result.
[0073] Embodiment 2
[0074] Please refer to Figure 2 A hemodialysis capacity load prediction terminal 1 includes a memory 2, a processor 3, and a computer program stored in the memory 2 and executable on the processor 3. When the processor 3 executes the computer program, each step of a hemodialysis capacity load prediction method of embodiment 1 is implemented.
[0075] In summary, the present invention provides a method and terminal for predicting hemodialysis capacity load. By acquiring dialysis parameters and laboratory examination data before dialysis, as well as bioelectrical impedance analysis data and ultrasound evaluation data before and after dialysis, it can use a variety of commonly used clinical indicators for evaluating capacity load to evaluate whether the capacity load is overloaded; and it combines a variety of capacity evaluation methods to predict whether there is a capacity load overload after dialysis, and it is also convenient to explore the optimal capacity load state. Through inter-group comparison, that is, the patients were divided into two groups according to whether there was post-dialysis volume overload, and the differences in pre-dialysis indicators between the two groups were compared by univariate analysis, and the indicators with statistical differences P<0.05 were preliminarily screened out. Then, the initial screening indicators were subjected to multicollinearity analysis, and the parameters with multicollinearity were eliminated. The remaining parameters were further subjected to multivariate logistic regression to further screen out parameters with potential predictive value. Finally, these parameters with potential predictive value were further subjected to Lasso regression analysis, and the weights of several parameters were compared to further screen out the parameters formally used for machine learning modeling. Then, a variety of machine learning methods were used to establish and select the optimal prediction model. Unlike previous studies that only included all existing parameters into the modeling, the pre-screening of modeling parameters can not only remove confounding factors and make the model more stable, but also further control the number of modeling variables, reduce the number of cases required for modeling, and improve the efficiency of subsequent modeling success.
[0076] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A method for predicting hemodialysis volume load, characterized in that: Includes steps: S1. Obtaining clinical parameter data before and after historical hemodialysis, and setting a label indicating whether the volume load is overloaded for the clinical parameter data; S2. performing correlation analysis and screening on the clinical parameter data and their labels to obtain characteristic variables related to dialysis volume overload; S3. Use the characteristic variables to perform machine learning to establish an optimal prediction model for predicting post-dialysis volume overload, and use the optimal prediction model to predict post-dialysis volume load.
2. A method for predicting hemodialysis volume load according to claim 1, characterized in that: Step S1 includes: Acquiring clinical parameter data before and after historical hemodialysis, wherein the clinical parameter data includes dialysis parameters and laboratory test data before dialysis, and bioelectrical impedance analysis data and ultrasound evaluation data before and after dialysis; Provide a method for judging whether the dialysis parameters and laboratory test data before dialysis in the clinical data meet preset conditions, whether the bioelectrical impedance analysis data before and after dialysis meet preset overhydration conditions, and whether the ultrasound assessment data after dialysis meets preset capacity overload conditions. If all conditions are met, a capacity overload label is set; otherwise, a capacity not overload label is set.
3. The method for predicting hemodialysis volume load according to claim 1, characterized in that: Step S2 includes: Performing correlation analysis and screening on the clinical parameter data and their labels to obtain a feature vector related to dialysis volume overload; The Lasso regression algorithm is used to screen out characteristic variables with non-zero coefficients for the characteristic vector.
4. A method for predicting hemodialysis volume load according to claim 3, characterized in that: Correlation analysis and screening of the clinical parameter data and their labels are performed to obtain feature vectors related to dialysis volume overload, including: Using univariate analysis to perform correlation analysis on the clinical parameter data and their labels, and obtaining a preliminary screening feature vector related to post-dialysis volume overload; The parameter data without multicollinearity in the initial screening feature vector were included in the multivariate logistic regression analysis to screen out the potential feature vector for predicting post-dialysis volume overload.
5. A method for predicting hemodialysis volume load according to claim 4, characterized in that: In step S3, the feature variables are used to perform machine learning to establish an optimal prediction model for predicting post-dialysis volume overload, including: Randomly dividing the clinical parameter data into a training cohort and a test cohort, using at least two machine learning algorithms to analyze potential feature vectors in the training cohort, and establishing a corresponding training model; The test cohort is substituted into the training model for model verification and model evaluation, and the training model with the best evaluation result is taken as the optimal prediction model.
6. A terminal for predicting hemodialysis capacity load, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the following steps are implemented: S1. Obtaining clinical parameter data before and after historical hemodialysis, and setting a label indicating whether the volume load is overloaded for the clinical parameter data; S2. performing correlation analysis and screening on the clinical parameter data and their labels to obtain characteristic variables related to dialysis volume overload; S3. Use the characteristic variables to perform machine learning to establish an optimal prediction model for predicting post-dialysis volume overload, and use the optimal prediction model to predict post-dialysis volume load.
7. A hemodialysis capacity load prediction terminal according to claim 6, characterized in that: Step S1 includes: Acquiring clinical parameter data before and after historical hemodialysis, wherein the clinical parameter data includes dialysis parameters and laboratory test data before dialysis, and bioelectrical impedance analysis data and ultrasound evaluation data before and after dialysis; Provide a method for judging whether the dialysis parameters and laboratory test data before dialysis in the clinical data meet preset conditions, whether the bioelectrical impedance analysis data before and after dialysis meet preset overhydration conditions, and whether the ultrasound assessment data after dialysis meets preset capacity overload conditions. If all conditions are met, a capacity overload label is set; otherwise, a capacity not overload label is set.
8. A hemodialysis capacity load prediction terminal according to claim 6, characterized in that: Step S2 includes: Performing correlation analysis and screening on the clinical parameter data and their labels to obtain a feature vector related to dialysis volume overload; The Lasso regression algorithm is used to screen out characteristic variables with non-zero coefficients for the characteristic vector.
9. A hemodialysis capacity load prediction terminal according to claim 8, characterized in that: Correlation analysis and screening of the clinical parameter data and their labels are performed to obtain feature vectors related to dialysis volume overload, including: Using univariate analysis to perform correlation analysis on the clinical parameter data and their labels, and obtaining a preliminary screening feature vector related to post-dialysis volume overload; The parameter data without multicollinearity in the initial screening feature vector were included in the multivariate logistic regression analysis to screen out the potential feature vector for predicting post-dialysis volume overload.
10. A hemodialysis capacity load prediction terminal according to claim 9, characterized in that: In step S3, the feature variables are used to perform machine learning to establish an optimal prediction model for predicting post-dialysis volume overload, including: Randomly dividing the clinical parameter data into a training cohort and a test cohort, using at least two machine learning algorithms to analyze potential feature vectors in the training cohort, and establishing a corresponding training model; The test cohort is substituted into the training model for model verification and model evaluation, and the training model with the best evaluation result is taken as the optimal prediction model.