Analysis method during dialysis and analysis device for dialysis

Abstract

The present invention provides an intradialysis analysis method and an analysis device for use in dialysis. The method comprises obtaining one or more input features. These input features include information related to dialysis machine operating parameters and the relationship between changes in the patient's current and previous data. Based on the input features, one or more prediction models are used to predict future data. This future data includes blood pressure information at future time points and predictions of intradialytic hypotension. This method enables highly accurate prediction of intradialytic hypotension.

Description

Technical Field

[0001] The present invention relates to a detection technology, and more particularly to an analysis method during dialysis and an analysis device for dialysis. Background Art

[0002] Intradialytic hypotension (IHD) is the most common complication in dialysis patients. It not only affects their quality of life but is also more likely to cause cardiac arrhythmias and chronic or acute cardiovascular and cerebrovascular ischemia. When IHD occurs, medical staff must immediately interrupt the patient's dialysis treatment and provide clinical management to avoid continued hypotension. Recurrent IHD can lead to inadequate uremic toxin clearance and dehydration, and can even worsen existing symptoms of uremia and heart failure over the long term, increasing mortality in dialysis patients.

[0003] Currently, there's no international consensus on the definition of intradialytic hypotension, which challenges the accuracy of clinical hypotension warning systems. Medical staff must maintain high focus throughout dialysis treatment. Besides real-time monitoring of patient dehydration rates and dialysis temperature, rapid changes in blood pressure still require timely response based on personal experience. However, without established standards for parameter adjustments, there's a risk of misjudgment. This suggests that existing intradialytic hypotension warning mechanisms still have flaws. Summary of the Invention

[0004] In view of this, an embodiment of the present invention provides an analysis method during dialysis and an analysis device for dialysis, which can provide an early warning of hypotension during dialysis.

[0005] The intradialysis analysis method of an embodiment of the present invention includes (but is not limited to) the following steps: obtaining one or more input features. These input features include information related to dialysis machine operating parameters and the relationship between the patient's current and previous data. Based on the input features, one or more prediction models are used to predict future data. This future data includes blood pressure information at a future time point and predictions of intradialytic hypotension.

[0006] An analysis device for dialysis according to an embodiment of the present invention includes (but is not limited to) a memory and a processor. The memory is used to store program code. The processor is coupled to the memory. The processor is configured to load and execute the program code to perform the following steps: obtaining one or more input features. These input features include information related to operating parameters of the dialysis machine and the relationship between current and previous data of the patient. Based on the input features, one or more prediction models are used to predict future data. This future data includes blood pressure information at a future time point and a prediction result of intradialytic hypotension.

[0007] Based on the above, the intradialytic analysis method and analysis device for dialysis according to embodiments of the present invention further consider new variables that influence intradialytic hypotension (e.g., the relationship between current and previous data) to improve prediction accuracy. This allows for early prediction of impending intradialytic hypotension in patients, notifying caregivers to take appropriate measures, thereby reducing dialysis interruptions, lowering patient mortality, and improving healthcare quality.

[0008] In order to make the above features and advantages of the present invention more clearly understood, embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 FIG. 1 is a block diagram of components of an analysis device according to an embodiment of the present invention.

[0010] Figure 2 FIG. 1 is a flow chart of an analysis method during dialysis according to an embodiment of the present invention.

[0011] Figure 3 FIG. 4 is a schematic diagram of analyzing training data for establishing a prediction model according to an embodiment of the present invention.

[0012] Figure 4 FIG. 4 is a schematic diagram of determining final future data according to an embodiment of the present invention.

[0013] Figure 5 FIG. 4 is a flow chart of resampling according to an embodiment of the present invention.

[0014] Description of the accompanying symbols:

[0015] 100. Analytical device;

[0016] 110. Memory;

[0017] 130, processor;

[0018] 150. Prompt device;

[0019] S210-S230, S510-S530, steps;

[0020] T, current time point;

[0021] t-1, t-2, previous time points;

[0022] t+1, a future time point;

[0023] △SBP1, △SBP2, systolic blood pressure changes;

[0024] SBP t , current systolic blood pressure;

[0025] SBPt-1 , SBP t-2 , SBP f , previous systolic blood pressure;

[0026] UR t-1 UR t-2 , previous ultrafiltration rate;

[0027] DT t-1 DT t-2 , previous dialysate temperature;

[0028] BF t-1 , BF t-2 , previous blood flow;

[0029] SBP ml , previous average systolic blood pressure;

[0030] DBP ml , previous average diastolic blood pressure;

[0031] RP ml , previous pulse average;

[0032] SBP t+1 , future systolic blood pressure;

[0033] DT, training data;

[0034] ML 11 ~ML 1i ML 21 ~ML 2j ,Model;

[0035] PBP, predicted blood pressure;

[0036] P f , final prediction. DETAILED DESCRIPTION

[0037] Figure 1 is a block diagram of components of an analysis device 100 according to an embodiment of the present invention. Figure 1 The analysis device 100 includes (but is not limited to) a memory 110 and a processor 130. The analysis device 100 can be a hemodialysis machine, a control instrument, or any electronic device capable of computing a user's physiological data (e.g., a smartphone, tablet computer, server, cloud host, or computer host).

[0038] The memory 110 can be any type of random access memory (RAM), read-only memory (ROM), flash memory, hard disk drive (HDD), solid-state drive (SSD), or similar device. In one embodiment, the memory 110 is used to record program code, software modules, configurations, data (e.g., physiological parameters, biochemical test parameters, basic data, operational parameters, characteristics, data sets at various time points, predicted results, etc.), or files, and an embodiment thereof will be described in detail below.

[0039] The processor 130 is coupled to the memory 110 and may be a central processing unit (CPU), a graphics processing unit (GPU), or other programmable general-purpose or special-purpose microprocessor, a digital signal processor (DSP), a programmable controller (PLC), a field programmable gate array (FPGA), an application-specific integrated circuit (ASIC), a neural network accelerator, or other similar components or a combination of these components. In one embodiment, the processor 130 is used to perform all or part of the operations of the analysis device 100 and can load and execute various program codes, software modules, files, and data stored in the memory 110. In some embodiments, the functions of the processor 130 may be implemented by software or a chip.

[0040] In one embodiment, the analysis device 100 further includes a prompting device 150. Prompting device 150 can be a display, LED, speaker, buzzer, communication transceiver, other devices that provide visual or auditory effects, or a combination thereof. In some embodiments, prompting device 150 is used to generate an alarm. For example, it can display an alarm notification, flash a light, emit an alarm sound, or transmit an alarm message. However, the alarming method can be varied based on actual needs and is not limited by the present embodiment.

[0041] Hereinafter, the method according to the embodiment of the present invention will be described with reference to the various components and modules in the analysis device 100. The various processes of the method can be adjusted according to the implementation situation and are not limited thereto.

[0042] Figure 2This is a flow chart of an analysis method in dialysis according to one embodiment of the present invention. Figure 2 Processor 130 obtains one or more input features (step S210). Specifically, the input features are input data used for subsequent assessment of blood pressure information and / or intradialytic hypotension. Processor 130 may obtain the input features or their corresponding raw data via an input / output device, memory 110, an external storage device, or a network.

[0043] In one embodiment, the input features include operational parameters of the (hemo)dialysis machine and / or a change relationship between current data and previous data of a subject (eg, a dialysis patient or other user).

[0044] For example, Table (1) is an example of the operating parameters of a dialysis machine:

[0045] Table (1)

[0046]

[0047]

[0048] For example, Table (2) (related to physiological parameters) and Table (3) (related to biochemical test results) are examples of the relevant parameters of the subjects:

[0049] Table (2)

[0050]

[0051] Table (3)

[0052] Original domain name Variable Description domain name Data Definition Hb Hemoglobin hb number Hct Hematocrit hct number Albumin albumin albumin number P phosphorus p number K potassium k number

[0053] Current data refers to operational parameters, physiological parameters, biochemical test parameters, and / or other monitoring parameters measured at the current time point. It should be noted that the current time point of measurement may differ from the time point of processing parameters. In some embodiments, the most recent measurement time point may be used as the current time point, but this is not a limitation.

[0054] On the other hand, previous data refers to data such as operational parameters, physiological parameters, biochemical test parameters, and / or other monitoring parameters measured at one or more time points prior to the current time point. In other words, the previous time point is earlier than the current time point. For example, if the current time point is 12:00, the previous time point could be 11:00 or 11:30. It should be noted that the interval between two adjacent time points may be fixed or variable, depending on the user's needs.

[0055] In one embodiment, based on clinical experience, the relationship between previous and current data influences blood pressure changes. The relationship can be the difference in values ​​between two adjacent time points. For example, the relationship can include changes in ultrafiltration rate, conductivity, dialysate temperature, and / or blood flow between two adjacent time points.

[0056] The change relationship can be the change of data between t and t-1 or the change between t-1 and t-2, where t is the current time point and t-1 and t-2 are previous time points. Figure 3 This is a schematic diagram of training data analysis for establishing a prediction model according to an embodiment of the present invention. Figure 3 , assuming that the current time point is t, the previous time points are t-1, t-2, and the future time point is t+1. The processor 130 can obtain or calculate the systolic blood pressure change ΔSBP1 between the current time point t and the previous time point t-1 (for example, the current systolic blood pressure SBP t - Previous systolic blood pressure (SBP) t-1 ) and / or the systolic blood pressure change ΔSBP2 between the previous time points t-1 and t-2 (e.g., the previous systolic blood pressure SBP t-1 - Previous systolic blood pressure (SBP) t-2 ). Similarly, the processor 130 may convert the previous ultrafiltration rate UR t-1 UR t-2 The difference between the two changes in ultrafiltration rate is the previous dialysate temperature DT t-1 DT t-2 The difference between the dialysate temperature and / or the previous blood flow BF t-1 , BF t-2 The difference between them is taken as the blood flow change.

[0057] In another embodiment, the change relationship can be the dehydration rate difference, venous pressure difference, dialysis time difference, dialysate flow difference, systolic pressure difference and / or pulse difference between the current time point t and the previous time point t-1, between the previous time point t-1 and the previous time point t-2, or between two other adjacent time points.

[0058] In addition to the feature of the changing relationship, in one embodiment, the monitoring parameters further include previous data. The previous data includes blood pressure information before the current dialysis operation and / or statistical information of the previous dialysis operation. Figure 3 For example, the blood pressure information before the current dialysis operation is the first previous systolic blood pressure SBP measured before the start of the dialysis operation. fThe statistical information of the previous dialysis session is the average systolic blood pressure (SBP) of the previous or more previous dialysis sessions. ml , previous diastolic blood pressure average DBP ml and the previous pulse average RP ml It should be noted that the statistical information may also be the median, mode, or other statistical indicators, and the embodiments of the present invention are not limited thereto. In addition, the previous data may further include parameters such as the previous blood pressure of the current dialysis operation, the previous dialysate temperature, etc.

[0059] In one embodiment, the input features include basic data of the subject. For example, Table (4) is an example of basic data:

[0060] Table (4)

[0061]

[0062]

[0063] It should be noted that the basic data may further include medication records and / or medical history.

[0064] In one embodiment, the input features include external data, such as climate, temperature, humidity and other environmental parameters.

[0065] In one embodiment, the processor 130 converts the operating parameters of the dialysis machine, the measured data related to the subject (e.g., physiological parameters, biochemical test parameters, or basic data) and / or external data into input features. That is, variable conversion. The input features conform to the input format of the prediction model used in subsequent evaluation. In one embodiment, the processor 130 can perform field definition / description, quantity ratio processing, judgment of missing values, conversion of categorical data into a unified digital format (e.g., binary, decimal or hexadecimal format), calculation of change relationships, etc. on the aforementioned parameters or data to generate input features. For example, Table (5), Table (6) and Table (7) are examples of converted input features:

[0066] Table (5)

[0067]

[0068] Table (6)

[0069]

[0070]

[0071] Table (7)

[0072]

[0073]

[0074] In one embodiment, during the process of training the prediction model, the processor 130 may perform outlier processing on the aforementioned parameters or data to exclude outliers in an unreasonable range. For example, Table (8) is an example illustrating the exclusion range corresponding to the parameters:

[0075] Table (8)

[0076] domain name Exclusion range time >300 sbp <30,>300 dbp <30,>300 temperature <32,>40 temp(℃) <32,>40 conductivity <10,>20 uf <0,>3 target_uf <0,>10 dryweight <30,>200 blood_flow <0,>700 pulse <30,>200

[0077] For example, the processor 130 deletes the data of the pulse being 20.

[0078] Please refer to Figure 2 , the processor 130 predicts future data using one or more prediction models based on the input features (step S230). Specifically, the future data includes blood pressure information at a future time point and the prediction results of hypotension during dialysis. Figure 3 For example, the future time point t+1 is 12:30, and the current time point t is 12:00. In one embodiment, the blood pressure information at the future time point is the future systolic blood pressure (e.g. Figure 3 In some embodiments, the processor 13 may determine and / or compare a hypotension threshold based on the blood pressure information for use in assessing intradialytic hypotension. In one embodiment, the predicted intradialytic hypotension result is the probability of intradialytic hypotension occurring. In another embodiment, the predicted intradialytic hypotension result is the probability of intradialytic hypotension occurring or not occurring.

[0079] Notably, the Chang Gung University School of Medicine's 2011 publication, "Intradialytic Hypotension" (authors: Wu Mengtai, Yang Zhichao, Lin Jingkun, and Li Jiande), demonstrated that intradialytic hypotension is associated with parameters such as systolic and diastolic function, blood volume fluctuations, pulse output, and dialysate temperature. Furthermore, the 2015 publication in the journal Clinical Epidemiology, "Association of Mortality Risk with Various Definitions of Intradialytic Hypotension and Mortality Risk," further defined the Fall20Nadir90 threshold for hypotension as (predialysis systolic pressure minus intradialytic nadir blood pressure) ≥ 20 mmHg and intradialytic nadir blood pressure < 90 mmHg.

[0080] In one embodiment, the hypotension threshold value for evaluating intradialytic hypotension can be based on Document 2. For example, the patient's systolic blood pressure before dialysis is 120 mm-Hg. Therefore, during dialysis, if the systolic blood pressure meets the above-mentioned Fall20Nadir90 condition (i.e., the systolic blood pressure is lower than 90 mm-Hg and is 20 mm-Hg different from the pre-dialysis systolic blood pressure), it can be recorded as an intradialytic blood pressure drop (hypotension) event. During dialysis, if the measured systolic blood pressure is greater than 90 mm-Hg, it is a normal event with no pressure drop (i.e., no intradialytic hypotension). In other embodiments, the definition of intradialytic hypotension can still be changed according to actual needs, and the present invention is not limited thereto.

[0081] In one embodiment, the prediction model is established by one or more machine learning algorithms. The machine learning algorithm can be a regression analysis algorithm, an eXtreme Gradient Boosting (XGboost) algorithm, a Light Gradient Boosting Machine (LightGBM), a Bootstrap Aggregating (Bagged) algorithm, a neural network algorithm, a LASSO algorithm, a Random Forest algorithm, a Support Vector Regression algorithm, or other algorithms. The machine learning algorithm can analyze training data / samples to obtain patterns from them, thereby predicting unknown data based on the patterns. The prediction model is the machine learning model constructed after learning, and is used to infer the data to be evaluated.

[0082] It should be noted that the training data of the prediction model is the same as or related to the parameters or data types corresponding to the aforementioned input features. For example, the operating parameters of the dialysis machine, the physiological state of the subject, basic data, and / or external data. In some embodiments, the training data further includes actual data (i.e., future blood pressure information and / or whether intradialytic hypotension actually occurs). The document 3 "Standard operation procedures (SOPs) for the management of a patient's dialysis care" proposed by the University Hospitals Birmingham in 2017 illustrates the correlation between the input features of an embodiment of the present invention and the predicted future data.

[0083] In one embodiment, the processor 130 determines the final future data based on the future data predicted by multiple prediction models. The machine learning algorithms used by these prediction models may be the same or different, and the final future data also includes blood pressure information at a future time point and the prediction result of hypotension during dialysis. For example, Figure 4 This is a schematic diagram of determining the final future data according to an embodiment of the present invention. Figure 4 , assuming that the prediction model includes the first-class model and there are i first-class models ML 11 ~ML 1i (i is a positive integer greater than one). These first-class models ML 11 ~ML 1i After sampling the same, related or similar training data TD but training based on different machine learning algorithms (e.g., regression analysis, XGBoost, neural network system, random forest, LASSO, support vector regression, neural network, etc.), the processor 130 uses these first-class models ML 11 ~ML 1i Predict future data P respectively 11 ~P 1i , and based on these future data P 11 ~P 1i Determine the final future data P f For example, to determine the blood pressure information in the future data P 11 ~P 1i The prediction of blood pressure PBP (i.e., the first type of model ML 11 ~ML 1i The predicted blood pressure PBP can be used as one of the final future data.

[0084] In one embodiment, processor 130 determines statistical indicators of future data predicted by multiple prediction models based on ensemble learning, and these statistical indicators serve as the final future data. These prediction models utilize different machine learning algorithms. For example, processor 130 aggregates and averages future data (e.g., blood pressure values) derived from prediction models based on multivariate linear regression, LASSO, random forest, support vector regression, and neural networks to produce the final future data. More specifically, processor 130 uses a bagging concept to average (or vote) data to obtain a more stable (e.g., lower variance) average performance. It also uses a stacking concept to combine predictions from different prediction models. Bagging literally means dividing data into multiple bags and then combining the results from each bag. Algorithmically, training data is repeatedly sampled (with replacement) to generate multiple subsets. Multiple models are then sequentially built, and the results from all models are then aggregated. If you want to predict a regression problem, you can average all the results; if it is a classification problem, you can use voting to determine the category that appears the most times. Compared with individual prediction models, the combination of several models can improve the prediction accuracy. Figure 4 For example, the processor 130 uses ensemble learning to determine the first type of model ML 11 ~ML 1i The predicted future data P 11 ~P 1i Statistical indicators for predicting blood pressure PBP.

[0085] In one embodiment, in addition to the one or more first-category models described above for predicting blood pressure information in future data, embodiments of the present invention further provide one or more second-category models for predicting intradialytic hypotension. It is worth noting that directly using the Fall20Nadir90 condition to assess intradialytic hypotension in the blood pressure information predicted by the first-category models may result in excessive false alarms within the confidence interval. Therefore, in addition to the aforementioned dialysis machine operating parameters, the patient's physiological state, basic data, and / or external data, the predicted blood pressure information needs to be additionally considered. The processor 130 may determine the blood pressure information in the future data predicted by the first-category models, add the blood pressure information at the future time point predicted by the first-category models to the input features of the second-category models (i.e., using the blood pressure information predicted by the first-category models as one of the input features of the second-category models), and determine the prediction result for intradialytic hypotension in the future data predicted by the second-category models based on the newly added input features of the blood pressure information at the future time point (i.e., predicting whether intradialytic hypotension will occur at the future time point). That is, the input features for the second type of model include, in addition to the aforementioned operating parameters of the dialysis machine, the physiological state of the patient, basic data, and / or external data, the blood pressure information at a future time point predicted by the first type of model. In addition, after considering the predicted blood pressure information, the processor 130 can use the Fall20Nadir90 condition or other definitions to judge intradialytic hypotension. In one embodiment, when training the second type of model, labeling can be performed based on the operating parameters of the dialysis machine, the physiological state of the patient, basic data, external data, and blood pressure information of each sample. After labeling whether an intradialytic hypotension event has occurred, each sample can be used as a training sample for training the second type of model. During the labeling process, the Fall20Nadir90 condition or other definitions of intradialytic hypotension can be used to determine whether intradialytic hypotension has occurred for each sample. The blood pressure information used for training includes actual blood pressure information and may also include predicted blood pressure information.

[0086] by Figure 4 For example, the processor 130 determines which first-class models ML 11 ~ML 1i The predicted future data P 11 ~P 1i The processor 130 will be based on those first type models ML 11 ~ML 1i The statistical indicator of blood pressure PBP obtained is used as the second type of model ML 21 ~ML 2j (j is a positive integer greater than one) is one of the input features. These j second-class models ML 21~ML 2j The second model ML can be established based on different machine learning algorithms (e.g., regression analysis, XGBoost, neural network system, random forest, LASSO, support vector regression, neural network, etc.). Then, the processor 130 again determines those second models ML based on the input features including the predicted blood pressure PBP and based on ensemble learning. 21 ~ML 2j The prediction result of intradialytic hypotension in the predicted future data. For example, the majority vote of ensemble learning is used to determine whether intradialytic hypotension occurs. In this way, the final prediction P f In one embodiment, each second-class model outputs a prediction result of intradialytic hypotension as 0 or 1, wherein 0 represents no intradialytic hypotension and 1 represents intradialytic hypotension, and the prediction result of each second-class model is judged by majority voting of ensemble learning. For example, if the number of outputs of 1 from multiple second-class models is greater than the number of outputs of 0, the processor 130 will judge that intradialytic hypotension has occurred. In another embodiment, the processor 130 may integrate the predicted blood pressure PBP into the final predicted PBP. f , as future data.

[0087] It should be noted that, in other embodiments, the first type of model may also predict both blood pressure information at future time points and the prediction results of intradialytic hypotension.

[0088] Each prediction model can also be retrained periodically or based on specific events, and parameters can be adjusted accordingly. In addition, medical data may have imbalanced classification problems. For example, Table (9) is an example of the relationship between the predicted results and the actual data of hypotension during dialysis:

[0089] Table (9)

[0090]

[0091] The actual number of events without intradialytic hypotension in Table (9) (e.g., 14821+605=15426) is significantly greater than the actual number of events with intradialytic hypotension (e.g., 870+1626=2496), so it is considered "unbalanced data."

[0092] In order to correct “imbalanced data” and improve prediction accuracy and sensitivity, embodiments of the present invention may further resample the training data. Figure 5 is a flow chart of resampling according to an embodiment of the present invention. Figure 5Processor 130 resamples the training data used in the prediction model for predicting intradialytic hypotension (e.g., the second type of model) based on the future data and the corresponding actual data, and establishes a new prediction model based on the resampled training data (step S510). For example, processor 130 determines whether the number of different prediction results in the actual data is unbalanced (e.g., the number difference or number ratio exceeds a corresponding threshold). Furthermore, resampling may involve duplicating and / or deleting training samples / data corresponding to a specific prediction result.

[0093] In one embodiment, the processor 130 can adjust the number of one or more positive samples and one or more negative samples in those training data according to the quantity ratio. This quantity ratio is the ratio of the number of those positive samples and those negative samples. For example, 1:1, 7:8 or 10:12. Positive samples are samples related to the occurrence of intradialysis hypotension (i.e., actually having intradialysis hypotension) in the actual data, and negative samples are samples related to the absence of intradialysis hypotension (i.e., actually not having intradialysis hypotension) in the actual data. In other words, positive samples are samples known to actually have intradialysis hypotension in those training data (e.g., the operating parameters of the dialysis machine, the physiological state-related parameters of the subject to be tested, basic data, etc.). And negative samples are samples known to actually have no intradialysis hypotension in those training data (e.g., the operating parameters of the dialysis machine, the physiological state-related parameters of the subject to be tested, basic data, etc.). The processor 130 can resample the training data so that the ratio of the resampled positive samples and negative samples is the same or close to the set quantity ratio.

[0094] In one embodiment, the processor 130 may duplicate positive samples and / or negative samples, and / or delete positive samples and / or negative samples, so that the ratio of positive samples to negative samples after resampling is the same or close to the set ratio. Duplicating data may result in two or more identical samples in the training data. Deleting data may reduce the amount of training data.

[0095] Taking Table (9) as an example, assuming that the expected ratio is 1:1. For the oversampling method, the processor 130 may randomly copy at least one of the positive samples. For the undersampling method, the processor 130 may randomly delete at least one of the negative samples. For the combined oversampling and undersampling method, the processor 130 may randomly copy positive samples and randomly delete negative samples according to a ratio of 0.5.

[0096] For the resampled training data, the processor 130 can retrain and establish a new prediction model by itself or provide it to an external cloud host. Then, the processor 130 can input the input features corresponding to the current time point (the time point after the prediction by the prediction model) into the new prediction model, and predict the future data again based on the new prediction model (step S530). For example, the processor 130 inputs the input features corresponding to 11:30 into the prediction model to predict the future data at 12:00. After the actual data at 12:00 is known (i.e., whether there is hypotension during dialysis), the processor 130 uses the actual data obtained at 12:00 and other data such as operational parameters, physiological parameters and external data as training data and resamples them, for example, by resampling using the above-mentioned oversampling method, undersampling method or a combination of oversampling and undersampling methods to establish a new prediction model. Then, the processor 130 uses the new prediction model to determine the future data corresponding to the input features at 12:00 (e.g., predict the future data at 12:30).

[0097] In some embodiments, the processor 130 may further use sensitivity (Sensitivity), false omission rate (FOR), specificity (Specificity), and / or false positive rate (FPR) to evaluate the performance of the prediction model. Sensitivity is the proportion of cases where the prediction model predicts intradialytic hypotension among all cases where intradialytic hypotension actually occurs. Therefore, the higher the sensitivity, the better the performance. The false omission rate is the proportion of cases where the prediction model predicts no intradialytic hypotension among all cases where intradialytic hypotension actually occurs. Therefore, the lower the false omission rate, the better the performance. Specificity is the proportion of cases where the prediction model predicts no intradialytic hypotension among all cases where intradialytic hypotension actually does not occur. Therefore, the higher the specificity, the better the performance. In addition, the false positive rate is the proportion of cases where the prediction model predicts intradialytic hypotension among all cases where intradialytic hypotension actually does not occur. Therefore, the lower the false positive rate, the better the performance. Experiments have shown that compared with the prediction model, the resampled new prediction model can effectively improve the values ​​of the aforementioned indicators.

[0098] In addition to predicting future data, in one embodiment, the processor 130 can issue an alert notification via the prompt device 150 based on the predicted result of intradialytic hypotension. For example, if the predicted result indicates an intradialytic hypotension event or the probability of occurrence exceeds a corresponding threshold, the processor 130 can display a warning message, sound an alarm, or transmit a message to the nurse counter. For another example, if the predicted result indicates no intradialytic hypotension event or the probability of occurrence does not exceed a corresponding threshold, no alert notification may be issued.

[0099] In summary, in the analysis method during dialysis and the analysis device for dialysis of the embodiment of the present invention, more new variables that affect the change of blood pressure during dialysis (for example, change relationship, previous data, etc.) are taken into account, and the blood pressure information at future time points and the presence or absence of hypotension during dialysis are directly estimated by fusing multiple prediction models. In order to conform to the format of the prediction model, the original data or parameters can be further calculated and / or converted into input features that conform to the model. The model is corrected by resampling to improve the accuracy, sensitivity and specificity of the prediction. In addition, in order to meet the needs of practical applications, the definition of hypotension during dialysis can be adjusted.

[0100] Although the present invention has been disclosed above with reference to the embodiments, they are not intended to limit the present invention. Any person skilled in the art may make slight changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the scope of the appended patent applications.

Claims

1. A method for analyzing during dialysis, characterized in that include: Obtaining at least one input feature, wherein the at least one input feature includes a change relationship between an operating parameter of a dialysis machine and current data and previous data of a subject; and Predicting future data using at least one prediction model based on the at least one input feature, wherein the future data includes blood pressure information at a future time point and a prediction result of intradialytic hypotension; The at least one prediction model includes a plurality of prediction models, and the plurality of prediction models include a plurality of first-type models and a plurality of second-type models; The step of predicting future data using at least one prediction model based on the at least one input feature includes: determining the blood pressure information in the future data predicted by the plurality of first-type models; as well as Prediction results of the intradialytic hypotension in the future data predicted by the plurality of second-type models are determined according to the at least one input feature and the blood pressure information.

2. The method for analyzing during dialysis according to claim 1, characterized in that The step of predicting the future data using the at least one prediction model according to the at least one input feature includes: Final future data is determined based on the future data predicted by the multiple prediction models, wherein the final future data includes the blood pressure information at the future time point and the prediction result of the intradialytic hypotension.

3. The method for analyzing during dialysis according to claim 2, characterized in that The step of determining the final future data according to the future data predicted by the multiple prediction models includes: A statistical indicator of the future data predicted by the multiple prediction models is determined based on an ensemble learning, wherein the multiple prediction models use different machine learning algorithms.

4. The method for analyzing during dialysis according to claim 1, wherein The change relationship includes at least one of an ultrafiltration rate change, a conductivity change, a dialysate temperature change, a blood flow change, a dehydration rate difference, a venous pressure difference, a dialysis time difference, a dialysate flow difference, a systolic pressure difference, and a pulse difference. The change relationship is a change in data between t and t-1 or a change between t-1 and t-2, where t is the current time point, and t-1 and t-2 are previous time points.

5. The method for analyzing during dialysis according to claim 1, characterized in that The previous data includes at least one of blood pressure information before the current dialysis procedure and statistical information of previous dialysis procedures.

6. The method for analyzing during dialysis according to claim 1, characterized in that Also includes: resampling a plurality of training data used by the at least one prediction model for predicting the intradialytic hypotension prediction result according to the future data and corresponding actual data; Building a new prediction model based on the resampled training data; and The future data is predicted again using the new prediction model.

7. The method for analyzing during dialysis according to claim 6, characterized in that The step of resampling the plurality of training data used by the at least one prediction model comprises: Adjusting the number of at least one positive sample and at least one negative sample in the plurality of training data according to a quantity ratio, wherein the quantity ratio is an expected ratio of the at least one positive sample to the at least one negative sample in quantity, the at least one positive sample is a sample in the actual data associated with the occurrence of intradialytic hypotension, and the at least one negative sample is a sample in the actual data associated with the non-occurrence of intradialytic hypotension; replicating at least one of the at least one positive sample and the at least one negative sample; and / or At least one of the at least one positive sample and the at least one negative sample is deleted.

8. The method for analyzing during dialysis according to claim 1, characterized in that Also includes: A warning notification is issued based on the blood pressure information at the future time point or the prediction result of intradialytic hypotension.

9. The method for analyzing during dialysis according to claim 1, characterized in that Also includes: The operating parameters of the dialysis machine and / or the data related to the patient are converted into the at least one input feature, wherein the at least one input feature conforms to the input format of the at least one prediction model.

10. An analytical device for dialysis, characterized in that include: a memory for storing a program code; as well as a processor, coupled to the memory and configured to load and execute the program code to perform: Obtaining at least one input feature, wherein the at least one input feature includes a change relationship between an operating parameter of a dialysis machine and current data and previous data of a subject; and Predicting future data using at least one prediction model based on the at least one input feature, wherein the future data includes blood pressure information at a future time point and a prediction result of intradialytic hypotension; The at least one prediction model includes a plurality of prediction models, the plurality of prediction models including a plurality of first-type models and a plurality of second-type models, and the processor is further configured to: determining the blood pressure information in the future data predicted by the plurality of first-type models; as well as Prediction results of the intradialytic hypotension in the future data predicted by the plurality of second-type models are determined according to the at least one input feature and the blood pressure information.

11. The analytical device for dialysis according to claim 10, characterized in that The processor is further configured to: Final future data is determined based on the future data predicted by the multiple prediction models, wherein the final future data includes the blood pressure information at the future time point and the prediction result of the intradialytic hypotension.

12. The analytical device for dialysis according to claim 11, characterized in that The processor is further configured to: A statistical indicator of the future data predicted by the multiple prediction models is determined based on an ensemble learning, wherein the multiple prediction models use different machine learning algorithms.

13. The analytical device for dialysis according to claim 10, characterized in that The change relationship includes at least one of an ultrafiltration rate change, a conductivity change, a dialysate temperature change, a blood flow change, a dehydration rate difference, a venous pressure difference, a dialysis time difference, a dialysate flow difference, a systolic pressure difference, and a pulse difference. The change relationship is a change in data between t and t-1 or a change between t-1 and t-2, where t is the current time point, and t-1 and t-2 are previous time points.

14. The analytical device for dialysis according to claim 10, characterized in that The previous data includes at least one of blood pressure information before the current dialysis procedure and statistical information of previous dialysis procedures.

15. The analytical device for dialysis according to claim 10, characterized in that The processor is further configured to: resampling a plurality of training data used by the at least one prediction model for predicting the intradialytic hypotension prediction result according to the future data and corresponding actual data; Building a new prediction model based on the resampled training data; and The future data is predicted again using the new prediction model.

16. The analytical device for dialysis according to claim 15, characterized in that The processor is further configured to: Adjusting the number of at least one positive sample and at least one negative sample in the plurality of training data according to a quantity ratio, wherein the quantity ratio is an expected ratio of the at least one positive sample to the at least one negative sample in quantity, the at least one positive sample is a sample in the actual data associated with the occurrence of intradialytic hypotension, and the at least one negative sample is a sample in the actual data associated with the non-occurrence of intradialytic hypotension; replicating at least one of the at least one positive sample and the at least one negative sample; and / or At least one of the at least one positive sample and the at least one negative sample is deleted.

17. The analytical device for dialysis according to claim 10, characterized in that Also includes: a prompt device coupled to the processor, wherein the processor is further configured to: A warning notification is issued through the prompting device according to the prediction result of the intradialytic hypotension.

18. The analytical device for dialysis according to claim 10, characterized in that The processor is further configured to: The operating parameters of the dialysis machine and / or the data related to the patient are converted into the at least one input feature, wherein the at least one input feature conforms to the input format of the at least one prediction model.

Images