Cardiovascular complication risk prediction and dialysis method based on multi-dimensional data

Abstract

The invention discloses a cardiovascular complication risk prediction and dialysis method based on multi-dimensional data, which comprises a dialysate blending device, the dialysate blending device adds a potassium chloride solution into dialysate according to the blood potassium level of a patient to regulate and control the potassium ion concentration of the dialysate, and conveys the potassium ion concentration to a dialysis machine; the prediction and dialysis method comprises the following steps: firstly, collecting multi-dimensional data of a patient, and constructing a cardiovascular complication risk prediction model by utilizing big data analysis and machine learning technologies and combining the multi-dimensional data; predicting the possibility of cardiovascular events caused by blood potassium imbalance of the patient in a future period of time in advance; according to a prediction result, adjusting a dialysate potassium ion concentration strategy of the patient before dialysis; in the later treatment stage, the blood potassium level of the current patient is monitored, regulated and controlled, and individualized treatment is conducted according to the current blood potassium concentration. The potassium ion concentration strategy of the dialysate can be adjusted in advance, preventive treatment is implemented, and the treatment gate is moved forward.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dialysis, in particular to a cardiovascular complication risk prediction and dialysis method based on multi-dimensional data. BACKGROUND

[0002] The primary cause of death for maintenance hemodialysis (MHD) patients is cardiovascular complications, nearly 50% of which are sudden deaths, and sudden death accounts for 20-25% of the total number of deaths of hemodialysis patients. The risk of arrhythmia related to electrolyte imbalance, especially the use of low-potassium dialysate and large ultrafiltration volume to cause a large amount of potassium transfer in a short time, has a certain correlation with the sudden death of hemodialysis patients.

[0003] At present, most hemodialysis machines only have fixed concentrations of potassium ions dialysate of 2, 3, 4, and 5 mmol / L. This makes it impossible to provide patients with exclusive initial dialysate before use, and the initial dialysate does not match the patient, which can cause abnormal blood potassium concentration of the patient, and the concentration cannot be adjusted during dialysis, which is not conducive to the treatment of the patient. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a new supporting device for a hemodialysis machine, which can provide exclusive initial dialysate for different patients before dialysis to prevent the problem of abnormal blood potassium concentration of the patient caused by the mismatch between the initial dialysate and the patient, and monitor the blood potassium level during dialysis, and use different concentrations of dialysate for different patients according to the blood potassium level, especially for those patients with high potassium, to solve the problem that there is no corresponding concentration of dialysate for use.

[0005] To solve the above technical problems, the present application provides a cardiovascular complication risk prediction and dialysis method based on multi-dimensional data, comprising a dialysate preparation device, which adds potassium chloride solution to the dialysate according to the blood potassium level of the patient to regulate the potassium ion concentration of the dialysate, and delivers it to the dialysis machine; The prediction and dialysis method comprises the following steps: 1) Collecting multi-dimensional data of the patient, which includes historical dialysis data, basic disease information, and lifestyle data; 2) Using big data analysis and machine learning technology to construct a cardiovascular complication risk prediction model combined with the multi-dimensional data; 3) Based on the cardiovascular complication risk prediction model, predicting in advance the possibility of cardiovascular events caused by blood potassium imbalance in the patient in the future; 4) According to the prediction result, adjusting the potassium ion concentration strategy of the dialysate for the patient before dialysis; 5) In the treatment stage, monitor and regulate the current patient's blood potassium level, and individualize the treatment according to the current blood potassium concentration.

[0006] Further, the historical dialysis data includes the duration of each dialysis, the change of blood potassium level during dialysis, and the record of potassium ion concentration of dialysate; the basic disease information includes whether the patient has hypertension, diabetes, cardiovascular system related diseases and the severity of the disease; the life habit data includes the patient's dietary preference, exercise frequency and smoking and drinking situation.

[0007] Further, the process of constructing a cardiovascular complication risk prediction model using big data analysis and machine learning technology includes: 1) Preprocessing the collected multi-dimensional data, including data cleaning, missing value filling, and data standardization; In data cleaning, data that damages the learning of the prediction model is checked and deleted; In missing value filling, AI algorithm is introduced in the fitting interpolation method to reconstruct the missing values of the power generation data; 2) Extract features related to cardiovascular complication risk from the preprocessed multi-dimensional data; 3) Construct a time series prediction model of LSTM-Attention-LSTM, two LSTM models as the encoding end and decoding end of the time series prediction model; train the time series prediction model with the extracted features; 4) According to the trained time series prediction model, adjust the dialysate potassium ion concentration strategy for the current patient before dialysis according to the prediction result, and the dialysate preparation device adds potassium chloride solution in the dialysate in proportion and delivers it to the dialysis machine.

[0008] Further, the time series prediction model consists of an input layer, an encoding layer, an attention layer, a decoding layer, and an output layer, wherein the input layer first inputs the preprocessed multi-dimensional data, including three-dimensional time series data X = (X1, X2, X3.... Xt) composed of samples, time steps and features, which is encoded by the encoding layer, then enters the attention layer to calculate the attention weight, then is decoded by the decoding layer, and finally the output layer calculates the predicted sequence data Y = (Yt+1, Yt+2, Yt+3,..., Yt+L) of the next time step, where t is the length of the input sequence, and L is the time step.

[0009] Further, the dialysate preparation device comprises a box, a touch screen is arranged on the surface of the box, an ISE mixing pool, an electrode measuring mechanism, an electromagnetic valve, a peristaltic pump and a constant temperature box are arranged in the box.

[0010] Further, the ISE mixing tank is used for mixing the standard solution and the whole blood sample uniformly, and the electrode measuring mechanism measures the potassium ion concentration in the whole blood sample by using the selective electrode method.

[0011] Further, the thermostat is used for constant temperature treatment of the mixed dialysate, and the temperature is kept at 36-37 DEG C.

[0012] Further, the dialysate preparation device further comprises a PLC and a PID control module, the PLC is used for receiving the signal of the potassium ion concentration, and the PID control module adjusts the potassium ion dialysate with a concentration suitable for the patient according to the current potassium ion concentration.

[0013] Further, the formula for adjusting the blood potassium concentration by the PID control module is as follows: ; Wherein, is the potassium ion concentration of the dialysate; is the control increment, ; is the deviation amount at the nth sampling time, , , represents the potassium ion concentration in the blood of the patient, is the set value, which is used for representing the target blood potassium concentration; , , is the PID controller parameter.

[0014] Further, the dialysate concentration range is set as follows: ; The blood potassium safety range is set as follows: .

[0015] The beneficial effects of the present application are as follows: The present application utilizes big data analysis and machine learning technology, combines historical dialysis data, basic disease information, living habits and other multi-dimensional data of patients, constructs a cardiovascular complication risk prediction model, can predict the possibility of cardiovascular events caused by blood potassium imbalance in patients in the future, and adjusts the potassium ion concentration strategy of dialysate in advance, and implements preventive treatment, and moves the treatment to the front. The dialysate preparation device can monitor the blood potassium level during dialysis, use different concentrations of dialysate for different patients according to the blood potassium level, has wide applicability, and can be adjusted in real time to meet the treatment needs, especially for high potassium patients, which solves the problem of no corresponding concentration of dialysate ratio use. That is, the potassium ion in the current whole blood sample of the patient is monitored in real time, and the potassium ion concentration of the dialysate is automatically regulated to achieve the effect of individualization of the potassium ion concentration of the dialysate. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a structure diagram of the dialysate preparation device of the present application; Figure 2 is a flow chart of the human blood potassium concentration control of the present application. DETAILED DESCRIPTION

[0017] The present application will be further described below in combination with the drawings and specific embodiments, so that those skilled in the art can better understand the present application and implement it, but the embodiments are not as a limitation on the present application.

[0018] Referring to Figure 1 , an embodiment of the cardiovascular complication risk prediction and dialysis method based on multi-dimensional data of the present application needs to be implemented with a dialysate preparation device. The dialysate preparation device is divided into two parts, which are a blood potassium prediction module and an automatic regulation module: The blood potassium prediction module combines historical dialysis data, basic disease information, living habits and other multi-dimensional data of patients, obtains prediction values of nine variables of dialysis duration, dialysis interval time, dialysate potassium ion concentration use record, hypertension, diabetes, cardiovascular system related diseases and disease severity, dietary preference, exercise frequency, smoking and drinking through a time series prediction model, constructs a cardiovascular complication risk prediction model, predicts the possibility of cardiovascular events caused by blood potassium imbalance in patients in the future, and selects an initial potassium ion concentration dialysate to implement preventive treatment; The automatic regulation module comprises two parts, namely a whole blood potassium ion monitoring module and a dialysate potassium ion regulation module; a peristaltic pump 4 is arranged in the whole blood potassium ion monitoring module to draw out the A standard solution and the B standard solution and mix them with the whole blood sample; an ISE mixing pool 1 is arranged in the whole blood potassium ion monitoring module to uniformly mix the standard solution and the whole blood sample; an electrode measurement mechanism 2 is further arranged in the whole blood potassium ion monitoring module to react with the whole blood sample to be detected and separately measure the potassium ion concentration in the whole blood sample; the dialysate potassium ion regulation module is used to obtain the whole blood potassium ion monitoring data, control the movement of the potassium chloride solution valve, realize accurate control of important parameters and reach the standard; and a host computer is used to realize the man-machine interaction function and display real-time parameters, specifically a touch screen 3.

[0019] Specifically, the measurement principle of the whole blood potassium ion monitoring module is the ion selective electrode method, and the measurement method is an indirect method. After the standard solution or sample solution is added, the measurement is automatically completed by the arranged electrolyte module. The ion selective electrode method is based on the ion selective electrode (ISE) and is used to measure the activity or concentration of specific ions in a solution. When an ISE with a specific ion selective membrane is placed in a solution containing ions to be measured, these ions to be measured will pass through the semi-permeable membrane into the electrode body, react with the internal reference electrode, and generate a small but stable potential difference. This small but stable potential difference can be used to calculate the concentration of the target ion in the solution. The dialysate potassium ion regulation module can measure the data of the whole blood potassium ion in real time, and then transmit it to the PLC controller. According to the algorithm set in the system, the electromagnetic valve 5 controlled by the PLC is opened in proportion, so that the potassium chloride solution is added and delivered to the dialysis machine.

[0020] Since the temperature of the dialysate is normal, the normal temperature will cause physical discomfort to the patient during use, and therefore a thermostat 6 is further arranged to perform constant temperature treatment on the mixed dialysate to ensure that the temperature is between 36℃ and 37℃.

[0021] Based on the above device structure, the method of the present application is as follows: First, multi-dimensional data of the patient is collected, including historical dialysis data, basic disease information and lifestyle data; the historical dialysis data includes the duration of each dialysis, the change of the blood potassium level during the dialysis process and the record of the use of the dialysate potassium ion concentration; the basic disease information includes whether the patient has hypertension, diabetes, cardiovascular system related diseases and the severity of the diseases; the lifestyle data includes the patient's dietary preference, exercise frequency, smoking and drinking.

[0022] Subsequently, a cardiovascular complication risk prediction model is constructed by using big data analysis and machine learning technology in combination with the multi-dimensional data; and based on the cardiovascular complication risk prediction model, the possibility of the patient triggering a cardiovascular event due to blood potassium imbalance in a future period of time is predicted in advance; The modeling process includes: First, the collected multi-dimensional data is pre-processed, including data cleaning, missing value filling, and data standardization. The acquisition of the blood potassium concentration curve depends on the quality of the original sample. Abnormal data in the data can distort the potential statistical relationship in the power curve, thereby impairing the ability of the prediction model to learn effective patterns. Therefore, early detection and removal of such data is crucial to maintaining the reliability of the model. If the dialysis duration exceeds 10 hours or is less than 2 hours, it needs to be checked.

[0023] The present application adopts a data reconstruction method based on AI algorithm, which reconstructs the missing values of power generation data by introducing AI algorithm in the fitting interpolation method. The results obtained by the data reconstruction method based on statistics often depend on only a single factor causing the change of blood potassium concentration to effectively interpolate small missing data. A large amount of high-quality labeled data is used for model training, and the hyperparameter setting is more sensitive. These generative models provide enhanced flexibility.

[0024] Subsequently, features related to cardiovascular complication risk are extracted from the pre-processed data. The factors that induce cardiovascular disease problems caused by blood potassium imbalance are various, mainly including potassium ion concentration in blood, historical dialysis data, basic disease information, and lifestyle data. These data have obvious spatiotemporal and multi-dimensional characteristics, and the feature extraction of the system can reveal the hidden patterns in the data. By revealing the potential structure across time, space, and physical conditions, feature extraction converts the original data into an operable input for the prediction model. Effective feature extraction is crucial for enhancing the robustness and generalization ability of the prediction model.

[0025] Model input variables: Variable type Specific variable Data processing method Historical dialysis data Dialysis duration (hours), dialysis interval (days), historical dialysis K+ concentration (mmol / L) Standardization, among which the historical dialysis K+ concentration adopts weighted average (the next weight accounts for 60%) Disease information Convert binary variables such as hypertension and diabetes into 0 / 1; convert disease severity into ordinal classification (e.g. 0 = none, 1 = mild, 2 = moderate, 3 = severe) Ordinal classification coding Lifestyle data Dietary preferences (e.g. low potassium / high potassium diet), exercise frequency (often / never), smoking (smoker / non-smoker), alcohol consumption (alcohol consumption / non-alcohol consumption) are converted into ordinal classification according to the degree (e.g. 0 = none, 1 = mild, 2 = moderate, 3 = severe). Segmented processing of continuous variables Finally, a suitable machine learning algorithm is selected to train a cardiovascular complication risk prediction model using the extracted features; the present application adopts a time series prediction model LSTM-attention-LSTM. The model uses two LSTM models as the encoding end and the decoding end, and introduces an attention mechanism between the encoding and decoding ends. The model has two obvious characteristics: first, it uses the attention mechanism to calculate the mutual relationship between sequence data, overcoming the shortcoming that the encoding and decoding sides cannot obtain long enough input sequences; second, it is suitable for long time step sequence prediction.

[0026] According to the prediction result, the potassium ion concentration strategy of the dialysate for the patient is adjusted in advance, and preventive treatment is implemented; through the algorithm set by the internal budget of the system, potassium chloride solution is added in proportion, and is delivered to the dialysis machine. Thus, the potassium ion concentration of the dialysate is automatically regulated and controlled.

[0027] Model output variables: Predicted potassium range (mmol / L) Risk level Recommended dialysis K+ concentration (mmol / L) Adjustment basis <3.0 Severe hypokalemia 3.0-3.5 (gradually increase) Avoid sudden increase of blood potassium leading to arrhythmia, use stepwise increase of concentration 3.0-3.4 Low potassium 2.5-3.0 Mild hypokalemia, maintain safe concentration while preventing overcorrection 3.5-5.0 Normal Maintain current concentration (2.0-3.0) Adjust according to historical trends (e.g. recent increase in blood potassium by 0.5 mEq / L) 5.1-5.5 Critical high potassium 1.5-2.0 Prioritize reducing blood potassium while monitoring ECG to prevent Q-T interval prolongation >5.5 Severe high potassium 1.0-1.5 or no potassium dialysate Emergency potassium reduction, combined with potassium-lowering resin (e.g. polystyrene sulfonate sodium) The LSTM-attention-LSTM model structure consists of an input layer, an encoding layer, an attention layer, a decoding layer, and an output layer. The input layer first inputs the pre-processed three-dimensional time series data X = (X1, X2, X3,..., Xt) of the [sample, time step, feature] formula, which is encoded by the encoding layer, then the attention weight is calculated by the attention layer, and then the data is decoded by the decoding layer, and finally the predicted sequence data Y = (Yt+1, Yt+2, Yt+3,..., Yt+L) of the next time step is calculated by the output layer, where t is the length of the input sequence and L is the time step.

[0028] The input of the model includes time series X, time step L, and training times n. The intermediate layer is processed by the encoding layer, the attention layer, and the decoding layer, and the final output is the predicted time series Y, the root mean square error (RMSE), and the mean absolute percentage error (MAPE). Taking historical dialysis data as an example: suppose there is a set of data dialysis duration X (X1, X2, X3) Encoding layer: The LSTM-attention-LSTM model first encodes the input 3D vector X (X1, X2, X3) by LSTM, and the encoding layer calculation is as follows: hi=f(Xi, hi−1)1≤ i ≤ t where hi represents the hidden state calculated by the encoding layer at time i, hi-1 represents the hidden state at time i-1, t represents the time step, and f represents the calculation function of the input gate, the forget gate, and the output gate in the LSTM model. i Attention layer:

[0029] After the encoding layer, the output vector is discarded and batch normalization is performed. Batch normalization uses the mean and standard deviation of the small batch to constantly adjust the intermediate output of the model, making the intermediate output of each layer more stable. Both are used to prevent model overfitting. The next layer is the attention layer, which calculates the attention weight of the matrix i of the sequence j ij . Then, the output vector α is obtained according to the attention weight.

[0030] ​With the introduction of the attention mechanism, the varying degrees of influence of a sequence at a certain time step on the prediction of the sequence at the current time step will be reflected in the time series prediction. Taking the previous example, we first input a time series dialysis duration X (X1, X2, X3) with a time step of 3. First, in the coding layer, the sequence X is calculated through the input gate, forget gate, and output gate to obtain the output vector hi at time step i.

[0031] Since this invention uses LSTM as the decoder, when predicting the value at time yi, it can know the output value Hi of the hidden layer node obtained by the decoder before yi. Therefore, the state Hi of the hidden layer node at time i can be compared with the state hj of the hidden layer node obtained by the encoder at time j in the input sequence.

[0032] For example, calculating the attention weights of sequence X (X1, X2, X3). α 3j {(X1, 0.2), (X2, 0.2), (X3, 0.2)} represents the attention allocation model assigns to each time point in the predicted time-point sequence X3. Similarly, each time-point sequence in the target sequence learns the attention allocation probability information of its corresponding original sequence. Thus, when generating each time-point sequence yi, the previously fixed semantic vector... C It will be Replaced, and It will continuously change based on the currently generated sequence. Therefore, for the probability distribution {(X1,0.2),(X2,0.2),(X3,0.2)}, the information corresponding to the time point sequence X3 is as follows: = g(0.2* h1, 0.2 *h2, 0.2* h3, ) Where h1, h2, and h3 are the output values ​​of the coding layer at each time point, and g function represents the encoder's transform function; The g function is essentially a weighted summation function. Let hi be the attention layer output vector at time i, and hj be the hidden state at time j. α ij Let t represent the attention weight from time j to time i, and t be the time step.

[0033] Decoding layer: Obtained through attention mechanisms The input sequence is fed into the next decoding layer, where yi is calculated by the decoding layer, and the expression is: in fi , ii and Oi respectively represent the forget gate, the input gate and the output gate, Li and C i represent the more new cell state, C i −1 is the cell state at the i - 1 moment, δ is the sigmoid activation function. W f , W i , W c and W o are the weights of the input, b f , b i , b c and b o respectively represent the bias value of the input. y i −1 is the predicted output value of the decoder at the i - 1 hi is the hidden state of the decoder at the i moment, is the output vector of the attention layer at the i moment. The predicted output value yi at the i moment is obtained from the decoder output, and yi is calculated through the relu activation function of the fully connected layer to obtain the final prediction sequence Y (Y 1, Y 2, Y 3).

[0034] Finally, in the treatment stage, the current patient's blood potassium level is monitored and regulated, and individualized treatment is performed according to the current blood potassium concentration.

[0035] The treatment strategy adjustment module described above is provided with a PLC, which can receive an analog signal from the whole blood potassium ion real-time monitoring module, and is provided with a PID control module, which can adjust the high-concentration potassium ion dialysate suitable for the patient according to the potassium ion concentration in the current whole blood sample.

[0036] As a classic strategy in control engineering, the PID controller has high maturity and performs well in many applications. After long-term engineering refinement, this control strategy has established itself as a model of high efficiency, stability and reliability, with a complete control technology and standard system architecture. The PID control law is: wherein: — the output of the controller; — the proportional coefficient; — the deviation.

[0037] In the present application, the incremental PID control algorithm is selected for the consideration of reducing error accumulation, improving system stability and reducing the influence of false action. The incremental PID controller is only related to the incremental change of control action, i.e. the difference between the current and the previous time control output. This method eliminates the cumulative influence of historical deviation, thereby reducing the accumulation of errors.

[0038] wherein: — the control amount calculated at the first sampling time; — the control amount calculated at the second sampling time; — the deviation amount at the first sampling time; — the deviation amount at the second sampling time; — the deviation amount at the third sampling time; = — the integral coefficient; — the differential coefficient.

[0039] The human body blood potassium concentration control flow chart is shown in Figure 2 Therefore, the formula for adjusting the blood potassium concentration by the PID control module in the present application is: The controlled quantity: the potassium ion concentration in the blood of the patient: ; The set value: the target blood potassium concentration: ; The deviation: ; The control amount: ; The control increment: the potassium ion concentration of the dialysate (because changing the potassium concentration of the dialysate can affect the clearance rate of blood potassium) ; The control increment: ; The final control formula: wherein: , , are the controller parameters (which need to be adjusted). ​Each sampling update: save e(k-2)=e(k-1), e(k-1)=e(k).

[0040] The treatment strategy adjustment module can obtain real-time monitoring data of whole blood potassium ions, open or close the peristaltic pump and the one-way valve according to an algorithm set by the system internally, thereby automatically delivering the dialysate containing potassium chloride to the dialyzer; and is internally provided with a touch screen, can realize the man-machine interaction function on the upper computer, and display real-time parameters, and is convenient to use. The above-mentioned examples are only preferred examples for fully illustrating the present application, and the protection scope of the present application is not limited thereto. The equivalent substitutions or transformations made by the person skilled in the art on the basis of the present application are within the protection scope of the present application.

Claims

1. A method for predicting cardiovascular complication risk and dialysis based on multi-dimensional data, characterized in that, The device includes a dialysate preparation unit, which adds potassium chloride solution to the dialysate according to the patient's blood potassium level to adjust the potassium ion concentration of the dialysate and delivers it to the dialysis machine. Predictive and dialysis methods include the following steps: 1) Collect multi-dimensional data of patients, including historical dialysis data, basic disease information, and lifestyle data; 2) Utilize big data analytics and machine learning techniques, combined with the aforementioned multi-dimensional data, to construct a cardiovascular complication risk prediction model; 3) Based on the cardiovascular complication risk prediction model, predict in advance the likelihood of patients experiencing cardiovascular events due to potassium imbalance in the future; 4) Based on the prediction results, adjust the potassium ion concentration strategy of the dialysate for this patient before dialysis; 5) During the treatment phase, monitor and regulate the patient's current blood potassium level in real time, and provide individualized treatment based on the current blood potassium concentration.

2. The method for predicting cardiovascular complication risk and dialysis based on multi-dimensional data as described in claim 1, characterized in that, The historical dialysis data includes the duration of each dialysis session, changes in blood potassium levels during dialysis, and records of dialysate potassium concentration usage; the underlying disease information includes whether the patient has hypertension, diabetes, cardiovascular-related diseases, and the severity of those diseases; the lifestyle data includes the patient's dietary preferences, exercise frequency, and smoking and drinking habits.

3. The method for predicting cardiovascular complication risk and dialysis based on multi-dimensional data as described in claim 1, characterized in that, The process of constructing a cardiovascular complication risk prediction model using big data analytics and machine learning techniques includes: 1) Preprocess the collected multi-dimensional data, including data cleaning, missing value imputation, and data standardization; During data cleaning, the data learned by the damage prediction model is checked and deleted; In missing value imputation, AI algorithms are introduced into the fitting interpolation method to reconstruct missing values ​​in power generation data; 2) Extract features related to the risk of cardiovascular complications from the preprocessed multidimensional data; 3) Construct an LSTM-Attention-LSTM time series prediction model, with two LSTM models serving as the encoder and decoder of the time series prediction model; train the time series prediction model using the extracted features; 4) Make predictions based on the trained time series prediction model. Based on the prediction results, adjust the potassium ion concentration strategy of the dialysate for the current patient before dialysis. The dialysate preparation device adds potassium chloride solution to the dialysate in proportion and delivers it to the dialysis machine.

4. The method for predicting cardiovascular complication risk and dialysis based on multi-dimensional data as described in claim 3, characterized in that, The time series prediction model consists of an input layer, an encoding layer, an attention layer, a decoding layer, and an output layer. The input layer first takes in preprocessed multi-dimensional data, including three-dimensional time series data X = (X1, X2, X3, ..., Xt) composed of samples, time steps, and features. This data is encoded by the encoding layer, then enters the attention layer to calculate attention weights, and then passes through the decoding layer for decoding. Finally, the output layer calculates the predicted sequence data Y = (Yt+1, Yt+2, Yt+3, ..., Yt+L) for the next time step, where t is the length of the input sequence and L is the time step.

5. The method for predicting cardiovascular complication risk and dialysis based on multi-dimensional data as described in claim 1, characterized in that, The dialysate preparation device includes a box, the surface of which is equipped with a touch screen, and the box contains an ISE mixing tank, an electrode measuring mechanism, a solenoid valve, a peristaltic pump, and a constant temperature chamber.

6. The method for predicting cardiovascular complication risk and dialysis based on multi-dimensional data as described in claim 5, characterized in that, The ISE mixing chamber is used to mix the standard solution and whole blood sample evenly, and the electrode measurement mechanism uses the selective electrode method to measure the potassium ion concentration in the whole blood sample.

7. The method for predicting cardiovascular complication risk and dialysis based on multi-dimensional data as described in claim 5, characterized in that, The constant temperature chamber is used to maintain the temperature of the mixed dialysate at 36℃-37℃.

8. The method for predicting cardiovascular complication risk and dialysis based on multi-dimensional data as described in claim 5, characterized in that, The dialysate preparation device also includes a PLC and a PID control module. The PLC is used to receive the potassium ion concentration signal, and the PID control module prepares a potassium ion dialysate with a concentration suitable for the patient based on the current potassium ion concentration.

9. The method for predicting cardiovascular complication risk and dialysis based on multi-dimensional data as described in claim 8, characterized in that, The formula for adjusting blood potassium concentration using the PID control module is: ; in, This refers to the potassium ion concentration in the dialysate. To control the increment, ; For the first The deviation at the next sampling time. , This indicates the concentration of potassium ions in the patient's blood. This is a set value used to represent the target blood potassium concentration; , , These are the parameters for the PID controller.

10. The method for predicting cardiovascular complication risk and dialysis based on multi-dimensional data as described in claim 9, characterized in that, Set the dialysate concentration range: ; Setting a safe range for blood potassium: .

Images