Device for predicting hemoglobin content of maintenance hemodialysis patient

Abstract

The invention discloses a device for predicting the hemoglobin content of a maintenance hemodialysis patient. The device for predicting the hemoglobin content of the maintenance hemodialysis patient comprises a linear regression module, and the linear regression module can predict the hemoglobin content according to the known hemoglobin content, diastolic pressure average value, average arterial pressure average value, systolic pressure average value and / or diastolic pressure variation standard deviation. Experiments prove that the device for predicting the hemoglobin content of the maintenance hemodialysis patient adopts an LR prediction model, MAE and RMSE of the LR prediction model are 9.844 and 13.342 respectively, and the device can be used for predicting the hemoglobin content of the maintenance hemodialysis patient.

Description

Technical Field

[0001] The invention relates to a device for predicting the hemoglobin content of maintenance hemodialysis patients in the field of biomedicine. Background Art

[0002] As a major global public health problem, chronic kidney disease (CKD) causes a great disease burden. When kidney tissue is continuously damaged and kidney function is gradually impaired, it will eventually develop into end-stage renal disease (ESRD), that is, the estimated glomerular filtration rate (eGFR) is less than 15ml / min / 1.73m 2 Currently, renal replacement therapy for ESRD includes hemodialysis, peritoneal dialysis and kidney transplantation. Maintenance hemodialysis (MHD) is the main renal replacement therapy for ESRD. MHD can relieve symptoms, improve quality of life and prognosis of patients by removing toxins and excess water from the body and correcting internal environmental disorders.

[0003] As early as 200 years ago, Richard Bright and other scholars discovered that there is a close connection between chronic kidney disease and anemia. Anemia is not only a common clinical manifestation of kidney disease, but also a complication and / or comorbidity of kidney disease. Anemia caused by various kidney diseases is defined as renal anemia (RA). It meets the anemia diagnosis of the World Health Organization, that is, the hemoglobin content of men, non-pregnant women and pregnant women at sea level is less than 130g / L, 120g / L and 110g / L respectively, and other non-renal anemia diseases (such as hemolytic anemia, etc.) are excluded. Renal anemia can be diagnosed. The main mechanism of renal anemia is the disorder of hypoxia-inducible factor-proline hydroxylase axis (HIF-PHD), which leads to reduced synthesis of erythropoietin (EPO) in the kidney, disorder of hepcidin-ferroportin system, insufficient hematopoietic material, shortened survival time of red blood cells, etc. In addition, regular blood tests and residual blood in the dialyzer can also cause MHD patients to lose some blood.

[0004] Hemoglobin is mainly responsible for transporting oxygen to various tissues and organs, while transporting away metabolites such as carbon dioxide. When the body is in an anemic state, tissue ischemia and hypoxia can cause progressive left ventricular hypertrophy and remodeling, left atrial enlargement and other cardiac changes through a series of neurohumoral mechanisms. In a study by Fukasawa et al., it was shown that increased red blood cell distribution width was closely related to all-cause mortality and cardiovascular mortality in hemodialysis patients. A retrospective study by Lin et al. also showed that hemodialysis patients with higher hemoglobin variability had a high risk of cardiovascular death. In addition, some scholars pointed out that the correlation curve between the pre-dialysis hemoglobin level of ESRD patients and the cardiovascular mortality rate after dialysis for 1 year was U-shaped, that is, too low or too high would increase the risk of death. A domestic observational follow-up study of 7210 CKD patients aged 45 years and above showed that anemia was associated with rapid deterioration of renal function. Cross-sectional studies have shown that symptoms such as pain, fatigue, and insomnia in MHD anemia patients will reduce the health-related quality of life of patients, and the quality of life has improved after the treatment of anemia.

[0005] Research on artificial intelligence (AI) is increasing rapidly, and AI can be embedded in medicine, biology, psychology, philosophy, education, economy, transportation, automation, etc. AI uses principles such as computing, mathematics, and logic to simulate human thinking processes and intelligent behaviors such as learning, thinking, reasoning, planning, and decision-making through computer programs. Machine learning (ML) is a branch of AI that completes designated tasks through data learning. In the medical field, medical data such as imaging images, pathological pictures, electronic medical record information, and test data can all be used as data sources for ML and can be used for disease diagnosis and prognosis assessment. In recent years, ML research on kidney diseases has continued to increase. For example, ML builds models to diagnose and manage chronic kidney disease and acute kidney injury. Helena et al. showed that machine learning can identify high-risk CKD patients who may develop into patients requiring renal replacement therapy. For dialysis patients, machine learning can be used to predict hemodynamic parameters (heart rate and blood pressure during dialysis), dialysis adequacy parameters (Kt / V), dry weight, anemia status, bleeding risk, etc., which can help medical staff provide individualized diagnosis and treatment and care for patients and promote the development of individualized dialysis plans.

[0006] The anemia control model (ACM) is a supplement to the current machine learning for anemia management in hemodialysis patients. The artificial neural network model based on the patient's clinical data to predict the hemoglobin level and the machine learning algorithm to determine the optimal erythropoiesis stimulant (ESA) dose constitute the components of ACM. After the use of ACM, the patient's hemoglobin level can be observed to increase and the dose of ESAs can be reduced. Compared with ACM, the hemoglobin prediction model for hemodialysis patients based on deep recurrent neural network proposed by Oscar et al. has a data source that is not limited to the first 3 months of prediction, but the entire course of the disease, while improving the accuracy of prediction. In addition, by scanning the dialysis waste and using the near-infrared spectral information obtained by the spectrometer, the artificial neural network algorithm can be used to non-invasively identify hemoglobin, red blood cells, iron and other substances in dialysis patients. The data learned by the existing anemia prediction model comes from blood test data and near-infrared spectral data of dialysis waste, lacking dialysis-related data and not considering the potential impact of dialysis conditions on hemoglobin. Summary of the invention

[0007] The present invention collects data such as previous hemoglobin values, dialysis records, and demographic characteristics of maintenance hemodialysis patients to build a machine learning model to predict the hemoglobin value of the next routine test, and analyzes and compares the performance of different models to facilitate the management of anemia in patients.

[0008] The technical problem to be solved by the present invention is how to predict the hemoglobin content of maintenance hemodialysis patients.

[0009] To solve the above technical problems, the present invention first provides a device for predicting the hemoglobin content of maintenance hemodialysis patients, wherein the device includes a linear regression module, which can predict the hemoglobin content based on known hemoglobin content, average diastolic blood pressure, average mean arterial pressure, average systolic blood pressure and / or standard deviation of diastolic blood pressure variation.

[0010] In the above device, the known hemoglobin content may be the hemoglobin content in the blood during the previous maintenance hemodialysis.

[0011] The average diastolic pressure value may be an average diastolic pressure value of the previous four consecutive times.

[0012] The average mean arterial pressure value may be an average mean arterial pressure value of the preceding four consecutive times.

[0013] The average systolic blood pressure value may be an average systolic blood pressure value of the previous four consecutive times.

[0014] The diastolic pressure variation standard deviation may be the diastolic pressure variation standard deviation of the previous four consecutive times.

[0015] The predicted hemoglobin level may be the hemoglobin level in the blood during the maintenance hemodialysis after the first four consecutive times.

[0016] In the above device, the linear regression module can use a linear regression algorithm to predict the hemoglobin content.

[0017] Furthermore, the linear regression module can obtain the predicted value Hb(t) of the hemoglobin content according to Formula 1, ; Among them, X 1 is the known hemoglobin content Hb(t-1); X 2 :Mean arterial pressure average value MEAN_AP_mean; X 3 :Systolic blood pressure mean SBP_mean; X 4 : Diastolic blood pressure mean DBP_mean; X 5 : Standard deviation of diastolic blood pressure variation DBP_diff_std.

[0018] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer-readable storage medium is a computer-readable storage medium for predicting the hemoglobin content of maintenance hemodialysis patients, and the computer program of the computer-readable storage medium for predicting the hemoglobin content of maintenance hemodialysis patients enables a computer to run the module of any one of the devices described in claims 1-8.

[0019] The use of the device in the preparation of a product for predicting the hemoglobin content of maintenance hemodialysis patients also falls within the protection scope of the present invention.

[0020] Experiments have shown that the device for predicting the hemoglobin content of maintenance hemodialysis patients of the present invention adopts the LR prediction model, and its MAE and RMSE are 9.844 and 13.342 respectively. The device of the present invention can be used to predict the hemoglobin content of maintenance hemodialysis patients.

[0021] The present invention is further described in detail below in conjunction with specific embodiments, and the examples provided are only for illustrating the present invention, rather than for limiting the scope of the present invention. The examples provided below can be used as a guide for further improvements by those of ordinary skill in the art, and do not constitute a limitation of the present invention in any way. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Modeling process for machine learning.

[0023] Figure 2 This is the 5-fold cross validation process for Lasso regression.

[0024] Figure 3 This is a performance comparison between the LR prediction model and the Lasso regression prediction model.

[0025] Figure 4 is the variable coefficient of the LR prediction model.

[0026] Figure 5 is the variable coefficient of the Lasso regression prediction model. DETAILED DESCRIPTION

[0027] The experimental methods in the following examples, unless otherwise specified, are all conventional methods, and are performed according to the techniques or conditions described in the literature in the field or according to the product instructions. The materials, reagents, instruments, etc. used in the following examples, unless otherwise specified, can all be obtained from commercial channels.

[0028] Embodiment 1, 1 Research Methods 1.1 Data Source This is a retrospective study, which included patients who underwent maintenance hemodialysis in multiple blood purification centers in Shenzhen from August 28, 2014 to April 20, 2021. This invention was approved by the patients. During dialysis, patients underwent routine testing every 3 months, dialysis 2-3 times a week, and dialysis for 4 hours each time. Inclusion criteria: (1) Age ≥ 18 years old; (2) Hemoglobin testing before and during maintenance hemodialysis, and the number of hemoglobin tests during dialysis was greater than 5 times. A total of 2104 patients were finally included in the study. The data involved the patient's age, gender, systolic blood pressure (SBP), diastolic blood pressure (DBP) and hemoglobin (Hb) values ​​in the dialysis records. The patient's age, gender, systolic and diastolic blood pressure and their related variables, and 5 consecutive hemoglobin values ​​were recorded as one data record.

[0029] The hold-out method was used to randomly divide the 8159 data records of the patients included in the study into a training dataset and a testing dataset in a ratio of 80%:20%. The training set had 6527 records (80%) and the testing set had 1632 records (20%).

[0030] 1.2 Data Preprocessing The Hb values ​​of 5 consecutive routine tests of the patients during any period of time during dialysis were selected and recorded as Hb(t-4), Hb(t-3), Hb(t-2), Hb(t-1), and Hb(t) in sequence. Hb(t) was defined as the prediction target, that is, the hemoglobin value of the next routine test, and Hb at other times was used as the model variable. Then, the blood pressure in the dialysis records (blood pressure measured before dialysis, during dialysis, and during dialysis) was processed to obtain new model variables (see Table 1), where the variation was defined as the difference between the actual value and the mean value. In addition, the definition of intradialytic hypotension (IDH) followed the 2002 K / DOQI guidelines and the European best practice guidelines, that is, a decrease in SBP ≥20 mmHg or a decrease in mean arterial pressure (MAP) ≥10 mmHg during dialysis. Finally, a total of 10 patients were included in the study.

[0031] Table 1. Definition and assignment of model variables

[0032] 1.3 Model selection The predicted target hemoglobin value of the present invention is a continuous variable, which is performed using a regression prediction model in machine learning. As traditional machine learning algorithms, linear regression and least absolute shrinkage and selection operator regression are usually used to predict clinical data and perform well in regression analysis. In the present invention, the inventors use two machine learning algorithms, linear regression and least absolute shrinkage and selection operator regression, to establish a prediction model for the next routine hemoglobin level test. Data analysis was performed by using the Linear Regression package and LassoLarsCV package (Scikit-learn: Machine Learning in Python, Pedregosa et al., JMLR 12, pp. 2825-2830, 2011.) in the machine learning algorithm library Scikit-learn 1.0.2 implemented in Python language, as shown in FIG. Figure 1 shown.

[0033] 1.4 Linear Regression Linear regression (LR) is the most widely used model in prediction models. It constructs a high-dimensional linear function to fit all data points as much as possible (Formula 1), and at the same time constructs a loss function, using the least squares method to minimize the sum of squared errors between the predicted target value and the actual value (Formula 2). Linear regression is simple, easy to implement and understand, with low computational complexity and fast calculation speed. However, when a large number of variables are included in the model, if there is a correlation between the independent variables or the independent variables are not independent and influence each other, it is a phenomenon of multicollinearity, which will lead to unstable analysis results and the inability to find the true relationship between the independent variables and the dependent variable.

[0034] 1.5 Least Absolute Shrinkage and Selection Operator Regression Least absolute shrinkage and selection operator (Lasso) regression, referred to as Lasso regression, was first proposed by scholar Robert Tibshirani. It adds a penalty function to linear regression to form a new loss function (see formula 3). The penalty function is an L1 regularization term containing an adjustment parameter λ, which constrains the mean square error (MSE, the sum of squares of the error between the predicted target value and the actual value) (see formula 4). The optimal adjustment parameter λ is obtained by the least angle regression method (Least Angle Regression, Lars), and the loss function is minimized. At the same time, the coefficients of the variables are compressed so that the coefficients of some variables are 0, and the variables are selected and reduced in dimension. The variables with the greatest contribution and the variables that do not overly affect each other are selected from a large number of variables. Lasso regression adds regularization to enhance the generalization ability of the model, and can solve the problems of multicollinearity and overfitting in linear regression, thereby improving the accuracy of model prediction. In the model training process of the present invention, the default penalty function λ value in the LassoLarsCV package is selected, that is, λ=0.003.

[0035] 1.6 Model evaluation indicator selection To evaluate the model, the present invention selects the coefficient of determination (R 2 ), mean absolute error (MAE) and root mean square error (RMSE) as evaluation indicators. The evaluation model is divided into two aspects. On the one hand, the degree of fit of the evaluation model is the determination coefficient R 2 (See Formula 5) is a fitting index that reflects the fitting effect of the linear model on the data. 2The closer it is to 1, the closer the predicted value is to the actual value, and the better the regression fit is. On the other hand, it is to evaluate the generalization ability of the model, that is, the ability of the model trained by the machine learning algorithm to recognize unprocessed data. MAE and RMSE are usually used to evaluate the generalization ability of the model. The mean absolute error (MAE) (see formula 6) is used to measure the average value of the absolute error between the predicted value and the true value. The smaller the MAE, the closer the predicted value is to the actual value, indicating that the model is better. The root mean square error (RMSE) (see formula 7) is also the most common indicator for evaluating the performance of regression prediction models. It is used to measure the deviation between the predicted value and the true value. It is sensitive to outliers. The smaller the RMSE, the higher the model accuracy.

[0036] In addition to using the hold-out method to randomly divide the included patients into a training set and a test set, the present invention also uses K-fold cross validation (CV) to randomly and evenly split the data set into K subsets, 1 validation set (cross validation set) and K-1 training sets, and repeats K times. This can ensure that each sample data can be used for model training and also participate in model validation. The data set is split multiple times to reduce the randomness of the results, improve the accuracy of the model, and avoid overfitting and underfitting.

[0037] When evaluating the linear regression model, calculate the R of the training set separately 2 , test set (R 2 , MAE, RMSE) as evaluation indicators. When evaluating the Lasso regression model, the present invention first adopts the 5-fold cross validation method (see Figure 2 ), that is, the training set is randomly divided into 5 subsets of equal size, one of which is used as the validation set to validate the model, and the remaining 4 subsets are combined as the training set to train the model. The cross-validation process is repeated 5 times to obtain 5 models. The variable coefficients of the 5 models are averaged as the final coefficient of the variable and the final model is trained; then the evaluation index is calculated. For the test set, the patient's data records are put into the data model, and the predicted value is compared with the actual value to calculate R 2 , MAR, MSE.

[0038] .

[0039] 2 Results 2.1 Model Performance Comparison The present invention selects the determination coefficient R 2 , MAE and RMSE are used as performance evaluation indicators of the prediction model. (1) Determination coefficient R 2 : In the training set, the R of the LR prediction model 2The R of the Lasso regression prediction model is 0.625. 2 is 0.621, indicating that the LR prediction model has a higher degree of fit in the training set than the Lasso regression prediction model. 2 The R of the Lasso regression prediction model is 0.621. 2 is 0.619, indicating that the LR prediction model has a higher degree of fit in the test set than the Lasso regression prediction model. 2 In the training set, the MAE and RMSE are higher than those in the test set. (2) MAE and RMSE: The MAE and RMSE of the LR prediction model are 9.844 and 13.342, respectively, and the MAE and RMSE of the Lasso regression prediction model are 9.904 and 13.384, respectively. The generalization ability of the LR prediction model is higher than that of the Lasso regression prediction model (see Table 2, Figure 3 ).

[0040] Table 2. Performance comparison of LR prediction model and Lasso regression prediction model

[0041] 2.2 Analysis of model prediction results The present invention includes 21 variables in total. The coefficients of each variable in the LR prediction model and the Lasso regression prediction model are shown in Table 3. Figure 4 , Figure 5 . In the LR prediction model, the top five variables most correlated with Hb(t) were Hb(t-1), MEAN_AP_mean, SBP_mean, DBP_mean, and DBP_diff_std. In the Lasso regression prediction model, the top five variables most correlated with Hb(t) were Hb(t-1), Hb(t-4), Hb(t-2), Hb(t-3), and DBP_mean. The top five correlated variables of both the LR prediction model and the Lasso regression prediction model were Hb(t-1) and DBP_mean, of which the Hb(t-1) coefficient was positive in both models, while the DBP_mean coefficient was positive in the LR prediction model and negative in the Lasso regression prediction model. In addition, compared with dialysis records and previous hemoglobin values, demographic characteristics (age, gender) were relatively less valuable in predicting Hb(t).

[0042] Table 3. LR prediction model and Lasso regression prediction model coefficients

[0043] 2.3 Prediction model construction The present invention includes 2104 patients, with a total of 8159 data records for model construction. The top five variables with coefficient weights are selected for predictive model construction.

[0044] In the linear regression model, the intercept is 98.707818, and the constructed model is Hb(t)=98.707818+13.405X 1 +10.756X 2 -6.152X 3 -5.062X 4 +2.919X 5 , where X 1 :Hb(t-1),X 2 :MEAN_AP_mean, X 3 :SBP_mean,X 4 :DBP_mean,X 5 :DBP_diff_std,λ=0.003,R 2 It is 0.621, MAE and RMSE are 9.844 and 13.342 respectively.

[0045] In the Lasso regression model, the intercept is 98.707813, and the constructed model is Hb(t)=98.707813+13.47X 1 +2.205X 2 +1.209X 3 +0.796X 4 +0.414X 5 , where X 1 :Hb(t-1),X 2 :Hb(t-4),X 3 :Hb(t-2),X 4 :Hb(t-3),X 5 :DBP_mean,λ=0.003,R 2 It is 0.619, MAE and RMSE are 9.904 and 13.384 respectively.

[0046] 3 Conclusion The present invention constructs a hemoglobin prediction model based on machine learning to predict the hemoglobin value of the next routine test, and explores the factors affecting the hemoglobin of maintenance hemodialysis patients by related parameters.

[0047] The present invention constructs a hemoglobin prediction model based on two machine learning algorithms, linear regression (LR), least absolute shrinkage and selection operator (Lasso) regression, and compares the performance of the model and the predictive value of the model variables for Hb. All variables in the LR prediction model are valuable for predicting hemoglobin in maintenance hemodialysis patients, while the Lasso regression prediction model selects 10 variables from 21 variables that have certain value in predicting hemoglobin in maintenance hemodialysis patients. By comparing the determination coefficient R 2 , MAE and RMSE found that the fitting degree and generalization ability of the LR prediction model are higher than those of the Lasso regression model.

[0048] Hb(t-1) has predictive value for Hb in both the LR model and the Lasso regression model, and its weight ranking ranks first, that is, the most important predictor of the Hb value of the next routine test of MHD patients is Hb(t-1). Compared with other previous Hb, Hb(t-1) is short from Hb(t), which is 3 months in the present invention. Hb(t-4), Hb(t-2) and Hb(t-3) are ranked after Hb(t-1) in weight ranking in previous Hb values, and their distances from Hb(t) are 12 months, 6 months and 9 months respectively. In addition, studies have shown that after 3 months, 6 months, 9 months and 12 months of treatment with Shengxuening or Shengxuening combined with recombinant human erythropoietin for patients with renal anemia undergoing dialysis, the patients' Hb showed a gradual upward trend, and Hb was higher than before treatment. As the treatment time increases, the patient's anemia condition gradually improves. The treatment time is the longest when Hb(t-1) is detected. Combined with the fact that the detection time of Hb(t-1) is on the red blood cell life time axis from the detection time of Hb(t), Hb(t-1) has a greater predictive value than other previous Hb values. Therefore, when predicting Hb(t), the Hb(t-1) value can be used as a reference, and other influencing factors can be combined at the same time.

[0049] Mean arterial pressure (MEAN_AP) is the average value of arterial blood pressure in a complete cardiac cycle, reflecting cardiac function and the resistance of peripheral large arteries, and also reflecting the blood supply of tissues and organs. In the LR model of the present invention, the weight of the MEAN_AP_mean coefficient ranks second, second only to Hb(t-1), and MEAN_AP has a certain value in predicting Hb. However, in the Lasso regression model, the weight of the MEAN_AP_mean coefficient is 0. Considering that MEAN_AP is a value calculated from SBP and DBP, it will be affected by SBP and DBP, so the LR model of the present invention has a multicollinearity problem.

[0050] In the LR model of the present invention, the coefficient weight ranking of SBP_mean and DBP_mean follows closely behind MEAN_AP_mean, except that SBP and DBP are negative predictors of Hb. However, in the Lasso regression model, the predictive value of SBP for Hb is the lowest, and the predictive value of DBP for Hb increases compensatorily. Current studies have shown that there is a complex relationship between Hb and blood pressure. The predictive value of SBP and DBP for Hb in the LR model and Lasso regression model of the present invention exists, but considering that the present invention does not include the patient's blood pressure regulation and improvement of anemia treatment, and as an observational study, it is impossible to clarify the interaction mechanism between Hb and blood pressure.

[0051] In terms of demographic characteristics, in the LR model of the present invention, although the gender variable ranks low in the coefficient weight ranking, it still makes a certain contribution to the prediction of Hb. Gender can have predictive value for Hb, and more attention should be paid to women in the anemia management process of dialysis patients.

[0052] In the LR model of the present invention, the coefficient weight of age is 0.289, which has a lower predictive value for Hb than gender, indicating that with the increase of age, the patient's Hb level gradually increases, but this conclusion is different from other studies. In the Lasso regression model of the present invention, the coefficient of the age variable is compressed to 0, indicating that there is multicollinearity between age and other variables, so other coefficients need to be combined when using age to predict Hb.

[0053] The coefficient of determination (R 2 , MAER, MSE) and model coefficients, the present invention recommends a linear regression module algorithm with better performance for predicting hemoglobin content, namely: ; Among them, X 1 is the known hemoglobin content Hb(t-1); X 2 :Mean arterial pressure average value MEAN_AP_mean; X 3 :SBP_mean; X 4 : Diastolic blood pressure average DBP_mean; X 5 : Standard deviation of diastolic blood pressure variation DBP_diff_std.

[0054] The present invention has been described in detail above. For those skilled in the art, without departing from the purpose and scope of the present invention, and without the need to carry out unnecessary experimental conditions, the present invention can be implemented in a wide range under equivalent parameters, concentrations and conditions. Although the present invention provides specific embodiments, it should be understood that the present invention can be further improved. In a word, according to the principles of the present invention, the application is intended to include any changes, uses or improvements to the present invention, including departure from the disclosed scope in the application, and changes made with conventional techniques known in the art.

Claims

1. A device for predicting the hemoglobin content of a maintenance hemodialysis patient, comprising a linear regression module, wherein the linear regression module can predict the hemoglobin content based on known hemoglobin content, diastolic blood pressure mean value, mean arterial pressure mean value, systolic blood pressure mean value and / or diastolic blood pressure variation standard deviation.

2. The device according to claim 1, characterized in that: The known hemoglobin content is the hemoglobin content in the blood during the previous maintenance hemodialysis.

3. The device according to claim 1 or 2, characterized in that: The average diastolic blood pressure value is the average diastolic blood pressure value of the previous four consecutive times.

4. The device according to any one of claims 1 to 3, characterized in that: The average mean arterial pressure is the average mean arterial pressure of the previous four consecutive times.

5. The device according to any one of claims 1 to 4, characterized in that: The average systolic blood pressure value is the average systolic blood pressure value of the previous 4 consecutive times.

6. The device according to any one of claims 1 to 5, characterized in that: The standard deviation of diastolic blood pressure variation is the standard deviation of diastolic blood pressure variation for the previous four consecutive times.

7. The device according to any one of claims 1 to 6, characterized in that: The predicted hemoglobin content is the hemoglobin content in the blood during the maintenance hemodialysis after the first four consecutive times.

8. The device according to any one of claims 1 to 7, characterized in that: The linear regression module uses a linear regression algorithm to predict the hemoglobin content; Furthermore, the linear regression module obtains the predicted value Hb(t) of the hemoglobin content according to Formula 1, ； Among them, X1 is the known hemoglobin content Hb(t-1); X2: mean arterial pressure MEAN_AP_mean; X3: systolic blood pressure mean SBP_mean; X4: diastolic blood pressure mean DBP_mean; X5: Diastolic blood pressure variation standard deviation DBP_diff_std.

9. A computer-readable storage medium storing a computer program, characterized in that: The computer-readable storage medium is a computer-readable storage medium for predicting the hemoglobin content of maintenance hemodialysis patients, and the computer program of the computer-readable storage medium for predicting the hemoglobin content of maintenance hemodialysis patients enables a computer to run the module of any one of claims 1-8.

10. Use of the device according to any one of claims 1 to 8 in the preparation of a product for predicting the hemoglobin content of maintenance hemodialysis patients.

Images