Chronically ill patient re-visit auxiliary early warning data processing method based on time series analysis

Abstract

The present application relates to the technical field of data processing, in particular to a chronic disease patient re-consultation auxiliary early warning data processing method based on time series analysis, comprising: randomly dividing a historical monitoring data sequence into a state transition sample sequence and a order verification sample sequence; and combining the two to traverse Markov order for evaluation and determination, and marking the Markov sample order of different chronic disease patients; then based on the Markov sample order, comparing the similarity degree of the disease time series auxiliary vector between the target patient and other chronic disease patients, and screening out a number of time series dynamic similar patients; based on the time series dynamic similar patients, constructing a time series state reference matrix and a time series data monitoring trend matrix, and performing synergistic fusion to predict the final monitoring data of the target patient; and according to the final monitoring data, intelligently generating a re-consultation auxiliary early warning scheme. The present application improves the accuracy of data prediction and improves the re-consultation auxiliary early warning effect for chronic disease patients.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method for processing data for auxiliary early warning of follow-up visits for patients with chronic diseases based on time-series analysis. Background Technology

[0002] Monitoring patient pathology data for follow-up visits and early warning is fundamentally because relapses or early complications after treatment of serious illnesses such as malignant tumors are often asymptomatic, and relying solely on patient complaints for follow-up visits can easily delay diagnosis. However, monitoring data from patients with chronic diseases can reveal abnormalities at the cellular level to some extent, such as histopathological features, circulating tumor DNA, and specific biomarkers. By dynamically tracking these indicators and using intelligent algorithms to build individualized early warning models, subtle trends of malignant transformation can be detected in advance, issuing warnings weeks or even months before clinical manifestations appear. This improves the accuracy of follow-up visits, avoids unnecessary and frequent visits, and assists doctors in quickly triaging high-risk groups, improving medical efficiency. It also compensates for patients' insufficient awareness of occult lesions and the lag in imaging examinations for detecting small lesions, shifting from passive waiting to active monitoring. The resulting follow-up visit early warning system can significantly advance the intervention window, thereby significantly improving patient prognosis.

[0003] Current technologies typically utilize neural networks to collect a large number of patient data samples as training samples. Then, leveraging the inherent randomness and regularity of low-order Markov algorithms, the training samples are optimized to establish an individualized early warning model. However, in real-world scenarios, acquiring a large number of patient data samples is costly and difficult, often resulting in insufficient training data to support the optimization processing of low-order Markov algorithms. This leads to increased data discrepancies between the data sample size and the algorithm's capabilities, reducing the accuracy of predictions and diminishing the effectiveness of early warning systems for chronic disease patients during follow-up visits. Summary of the Invention

[0004] This invention provides a data processing method for assisting early warning of follow-up visits for patients with chronic diseases based on time series analysis, in order to solve the existing problems: the conflict between the insufficient amount of existing data samples and the increasing contradiction between the prediction effect of low-order Markov algorithms based on data samples, which reduces the accuracy of data prediction and reduces the effect of assisting early warning of follow-up visits for patients with chronic diseases.

[0005] The present invention provides a time-series analysis-based method for processing auxiliary early warning data for follow-up visits of patients with chronic diseases, which adopts the following technical solution: Includes the following steps: Obtain historical monitoring data sequences for patients with different chronic diseases; The historical monitoring data sequence is randomly divided into a state transition sample sequence and an order verification sample sequence. The state transition matrix is ​​constructed by traversing the Markov order through the state transition sample sequence, and the state transition probability of the Markov order is calculated in different traversal processes. The state transition probability is evaluated and determined by the order verification sample sequence, and the Markov sample order of different chronic disease patients is marked in all Markov orders. Taking any chronic disease patient as the target patient, the historical monitoring data sequence of the target patient is traversed through the Markov sample order to construct the historical state window of the target patient and calculate several symptom time-series auxiliary vectors of the target patient; by comparing the similarity of the symptom time-series auxiliary vectors of the target patient with those of other chronic disease patients, several time-series dynamic similar patients of the target patient are selected from other chronic disease patients. Based on the historical status window of the target patient, the status transition matrices of all time-series dynamic similar patients within the same time range are integrated to construct several time-series status reference matrices for the target patient; based on the historical monitoring data sequence of the target patient, the trend of the target patient's own monitoring data affecting the disease status is analyzed to construct the time-series data monitoring trend matrix for the target patient; the time-series status reference matrices and the time-series data monitoring trend matrices are synergistically fused to predict the final monitoring data of the target patient; Based on the final monitoring data, the current patient's follow-up visit level is assessed, and an intelligent follow-up visit auxiliary early warning plan is generated.

[0006] Preferably, the method for obtaining the state transition probability is as follows: For any chronic disease patient, a predefined range of Markov order traversal is defined. The state transition sample sequence of the chronic disease patient is traversed by Markov order to obtain the state transition sequence of the chronic disease patient in different traversal processes. The state transition matrix of the chronic disease patient is constructed through all state transition sequences. The probability distribution of the Markov order predicted state in the state transition matrix is ​​statistically analyzed, and the state transition probability of the Markov order in different traversal processes is calculated.

[0007] Preferably, the method for obtaining the Markov sample order is as follows: For any chronic disease patient, a validation matrix of the same order as the state transition matrix is ​​constructed using the validation sample sequence of the chronic disease patient's order. The difference between the Markov order prediction values ​​between the state transition matrix and the validation matrix is ​​compared to calculate the probability prediction accuracy of the Markov order in different traversals. Combining the state transition probability and the probability prediction accuracy, the personalized disease-aided evolution of different Markov orders is calculated. Based on the personalized disease-aided evolution, the Markov sample order of the chronic disease patient is marked from all Markov orders.

[0008] Preferably, the method for obtaining the accuracy of the probability prediction is as follows: For any Markov order in the traversal process, obtain several temporal traversal prediction values ​​of the Markov order from the state transition matrix, and obtain several temporal verification prediction values ​​of the Markov order from the verification matrix; the difference between all temporal traversal prediction values ​​and temporal verification prediction values ​​is taken as the probability prediction accuracy of the Markov order.

[0009] Preferably, the method for obtaining the disease time-series auxiliary vector is as follows: The length of the historical state window is determined by the Markov sample order of the target patient, and the historical state window is constructed. The historical state window is traversed over the historical monitoring data sequence of the target patient to obtain several characteristic symptom values ​​of the historical state window of the target patient in different traversal processes. The feature vector constructed based on the characteristic symptom values ​​is used as the symptom time-series auxiliary vector of the target patient in the historical state window in different traversal processes.

[0010] Preferably, the method for obtaining the characteristic symptom values ​​is as follows: For any historical state window in the traversal process, calculate the maximum value, minimum value, mean value, and variance of the monitoring data within the historical state window, and use the maximum value, minimum value, mean value, and variance value of the monitoring data as characteristic symptom values.

[0011] Preferably, the method for obtaining the time-series dynamically similar patients is as follows: The inverse proportional relationship between the symptom time-series auxiliary vectors of the target patient and other chronic disease patients is used as the order window similarity between the target patient and other chronic disease patients; and based on the order window similarity, several time-series dynamically similar patients to the target patient are selected from other chronic disease patients.

[0012] Preferably, the method for obtaining the timing state reference matrix is ​​as follows: Based on the data fluctuations contained in the historical status window of the target patient, the target ideal time interval is divided. Based on the monitoring data in different historical monitoring data sequences within the target ideal time interval, and combined with the Markov sample order of patients with different time-series dynamic similarities, several time-series status reference matrices of the target patient are constructed.

[0013] Preferably, the method for obtaining the time-series data monitoring trend matrix is ​​as follows: In the historical monitoring data sequence of the target patient, several trend label values ​​of the target patient are calculated; under the target ideal time interval, based on the sequence formed by the trend label values, a time series data monitoring trend matrix of the target patient is constructed according to a preset traversal step size.

[0014] Preferably, the method for obtaining the final monitoring data is as follows: Calculate the state candidate probability for each time-series state reference matrix and the trend candidate probability for the time-series data monitoring trend matrix; combine the state candidate probability and the trend candidate probability to calculate the predicted fusion probability of patients with similar dynamics at different time series; and based on the predicted fusion probability, select the final monitoring data of the target patient from all predicted values.

[0015] The beneficial effects of the technical solution of this invention are as follows: This invention randomly divides historical monitoring data sequences into state transition sample sequences and order verification sample sequences; and combines the two to traverse the Markov order for evaluation and judgment, marking the Markov sample order of different chronic disease patients; by defining the Markov sample order of chronic disease patients already stored in the database, the Markov algorithm is endowed with intelligent response capability when facing different data samples, improving the utilization efficiency of the Markov algorithm for data samples; then, based on the Markov sample order, several patients with similar time-series dynamics are selected; the data source range for expanding the amount of training sample data is determined, making the subsequently expanded training samples more reasonable and accurate; then, a time-series state reference matrix and a time-series data monitoring trend matrix are constructed and synergistically fused to predict the final monitoring data of the target patient; eliminating the limitations and interference of the Markov algorithm and the training samples, effectively mitigating the conflict between the insufficient amount of data samples and the prediction effect of low-order Markov algorithms based on data samples. This invention intelligently applies different orders of Markov algorithms to historical monitoring data of different chronic disease patients in the database and expands the data sources of training samples. This eliminates the limitations and interference of the Markov algorithm and the training samples, effectively reduces the conflict between the insufficient amount of data samples and the prediction effect of low-order Markov algorithms based on data samples, improves the accuracy of data prediction, and enhances the auxiliary early warning effect for follow-up visits of chronic disease patients. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the steps in the time-series analysis-based auxiliary early warning data processing method for chronic disease patients' follow-up visits according to the present invention. Figure 2 This is a schematic diagram of the probability prediction accuracy curve of the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the time-series analysis-based auxiliary early warning data processing method for chronic disease patients proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the chronic disease patient follow-up visit auxiliary early warning data processing method provided by the present invention.

[0021] Please see Figure 1 The diagram illustrates a flowchart of a method for processing auxiliary early warning data for follow-up visits of chronic disease patients based on time-series analysis, according to an embodiment of the present invention. The method includes the following steps: Step S001: Obtain historical monitoring data sequences for patients with different chronic diseases.

[0022] It should be noted that existing technologies typically utilize neural networks to collect a large number of patient data samples as training samples. Then, leveraging the inherent randomness and regularity of low-order Markov algorithms, the training samples are optimized to establish an individualized early warning model. However, in real-world scenarios, acquiring a large number of patient data samples is costly and difficult, often resulting in insufficient training data to support the optimization processing of low-order Markov algorithms. This leads to increased data discrepancies between the data sample size and the algorithm's capabilities, reducing the accuracy of predictions and diminishing the effectiveness of early warning systems for chronic disease patients during follow-up visits.

[0023] In one specific implementation of this invention, the method for obtaining historical monitoring data sequences is as follows: randomly selecting from the patient sample database. Name patients with chronic diseases who have the same symptoms as the current patient, and derive the recent data for each chronic disease patient. The historical monitoring data recorded within a day is used to construct a sequence as the historical monitoring data sequence for each chronic disease patient. This embodiment uses... , This example is used for illustration; no specific limitations are set in this embodiment. , It depends on the specific implementation situation.

[0024] It should be noted that, in this embodiment, chronic diseases are described using diabetes as an example, and the recorded historical monitoring data are described using blood glucose levels as an example. The recording frequency of historical monitoring data in the patient sample database is... Taking 0.03 times / second as an example, the historical monitoring data sequence in this embodiment is as follows: {6.0, 8.0, 8.5, 6.0, 7.0, 9.0, 10.0, 6.5, 8.0, 9.0, 5.5, 7.0, 7.0, 9.5, 11.0, 6.0, 7.5, 8.0, 11.0, 7.0, 7.5, 9.0, 10.5, 6.0, 6.5, 8.0, 9.0, 10.0, 6.5, 7.0}, with units of mmol / L.

[0025] Thus, the historical monitoring data sequences of patients with different chronic diseases were obtained through the above methods.

[0026] Step S002: Randomly divide the historical monitoring data sequence into a state transition sample sequence and an order verification sample sequence; traverse the Markov order through the state transition sample sequence, construct a state transition matrix, and calculate the state transition probability of the Markov order in different traversal processes; evaluate and determine the state transition probability through the order verification sample sequence, and mark the Markov sample order of different chronic disease patients among all Markov orders.

[0027] It should be noted that in real-world scenarios, obtaining a large number of relevant patient data samples is costly and difficult, resulting in insufficient training data to support the optimization of low-order Markov algorithms. This leads to an increase in the data discrepancy between the amount of data in the training samples and the low-order Markov algorithm. To effectively solve this technical problem, it is necessary to address the issues of insufficient data in the training samples and the application of low-order Markov algorithms.

[0028] It's important to further clarify that the application of low-order Markov algorithms primarily hinges on the Markov order within the algorithm itself. This is because Markov algorithms predict future states based on historical states; in terms of data, this means predicting the next possible moment by combining historical data. The Markov order determines the amount of historical data used. Since the Markov order and the relevant data samples input to the Markov algorithm directly form a strong correlation, the core issue in addressing the application of low-order Markov algorithms is the proper determination of the Markov order.

[0029] Taking any chronic disease patient as an example, the historical monitoring data sequence is... The data sequence of the day was used as a state transition sample sequence for this chronic disease patient; the historical monitoring data sequence was then... The data sequence of the day was used as the order verification sample sequence for this chronic patient. In this embodiment, the data sequence of the day was used as the order verification sample sequence for this chronic patient. , This example is used for illustration; no specific limitations are set in this embodiment. , It depends on the specific implementation situation and must meet the following requirements. , Restrictions.

[0030] Preferably, in some implementations of the present invention, the method for obtaining the state transition probability is as follows: For any chronic disease patient, a preset range of Markov order traversal is defined, and Markov order traversal is performed on the state transition sample sequence of the chronic disease patient to obtain the state transition sequence of the chronic disease patient in different traversal processes; a state transition matrix of the chronic disease patient is constructed through all state transition sequences; the probability distribution of the Markov order predicted state within the state transition matrix is ​​statistically analyzed, and the state transition probability of the Markov order in different traversal processes is calculated. The specific process is as follows: Presuppose a range of orders for a Markov order. ,Will Starting from a step size of 1, the Markov order is continuously decreased until it reaches 1, at which point the traversal stops. Several Markov orders are obtained during this traversal, and each Markov order is used as the Markov trial order for the patient with the chronic disease. This embodiment uses... This example is used for illustration; no specific limitations are set in this embodiment. It depends on the specific implementation situation and must meet the following requirements. The constraint is that the result must be a positive integer. Additionally, each chronic disease patient corresponds to multiple Markov trial orders.

[0031] Furthermore, taking any Markov trial order as an example, starting from the first historical monitoring data in the state transition sample sequence of the chronic disease patient, with a step size of 1, the traversal begins until the last historical monitoring data is reached. The predicted value is obtained through each traversal process using the Markov trial order, and the sequence constructed from the historical monitoring data and the predicted value in each traversal process is used as the state transition sequence for that Markov trial order. The state transition sequences of the chronic disease patient in different traversal processes are obtained. Each Markov trial order corresponds to multiple state transition sequences. The process of obtaining the predicted value based on the Markov trial order is a well-known aspect of Markov algorithms and will not be elaborated upon in this embodiment.

[0032] It should be noted that the last data in the state transition sequence in this embodiment is the predicted value. An example of the state transition sequence is as follows: There is a historical monitoring data sequence {5.0, 5.5, 6.0}, and a Markov trial order of 3. Using the Markov algorithm, the predicted value 6.5 can be obtained based on {5.0, 5.5, 6.0}, where 6.5 is the predicted value obtained through the Markov trial order, and the corresponding state transition sequence is {5.0, 5.5, 6.0, 6.5}.

[0033] Furthermore, each state transition sequence of the Markov trial order is treated as a row in a matrix, and these rows are sequentially placed into the matrix in chronological order. The resulting matrix is ​​then used as the state transition matrix for that Markov trial order. The frequency of occurrence of the same predicted value in the state transition matrix is ​​counted. The normalized value of the product of the frequency of all occurrences of the same predicted value with the eigenvalues ​​of the state transition matrix is ​​taken as the state transition probability for that Markov order.

[0034] It should be noted that, in this embodiment, the default is... Normalization is performed on the function as an example, and the implementer can choose the normalization function according to the actual situation.

[0035] It should be noted that, for a matrix, its eigenvalues ​​represent the inherent physical parameters of the system in independent mode. In the state transition matrix constructed in this embodiment, since the matrix components are relevant data that can represent the state of patients with chronic diseases, the corresponding eigenvalues ​​can, to a certain extent, represent the state transition probabilities of patients with chronic diseases under a fixed Markov order. By combining the analysis of eigenvalues ​​within the matrix, the physical representational significance of the Markov order can be effectively improved.

[0036] Preferably, in some implementations of the present invention, the method for obtaining the Markov sample order is as follows: For any chronic disease patient, a verification matrix of the same order as the state transition matrix is ​​constructed using the order verification sample sequence of the chronic disease patient; the difference in the predicted Markov order values ​​between the state transition matrix and the verification matrix is ​​compared, and the probability prediction accuracy of the Markov order in different traversal processes is calculated; combining the state transition probability and the probability prediction accuracy, the personalized disease-aided evolution of different Markov orders is calculated; and based on the personalized disease-aided evolution, the Markov sample order of the chronic disease patient is marked from all Markov orders. The specific process is as follows: Referring to the method for obtaining the state transition matrix of the Markov trial order, the state transition sample sequence of the chronic disease patient is replaced with the order validation sample sequence. Following the same process, the state transition matrix of the Markov trial order is obtained, and this state transition matrix is ​​used as the validation matrix of the Markov trial order. The state transition matrix and the validation matrix have the same number of columns.

[0037] Preferably, in some implementations of the present invention, the method for obtaining the probability prediction accuracy is as follows: for any Markov order in the traversal process, obtain several temporal traversal prediction values ​​of the Markov order from the state transition matrix, and obtain several temporal verification prediction values ​​of the Markov order from the verification matrix; the difference between all temporal traversal prediction values ​​and temporal verification prediction values ​​is taken as the probability prediction accuracy of the Markov order. The specific process is as follows: As an example, the probability prediction accuracy of this Markov trial order can be calculated using the following formula. :

[0038] In the formula, This represents all observation types in the state transition matrix of the Markov trial order; The validation matrix represents all observation types in the order of the Markov trial; The state transition matrix representing the order of the Markov trial is the nth... The numerical values ​​for each observation category; The verification matrix representing the order of the Markov trial is the first... The numerical values ​​for each observation category; Indicates removing the absolute value; This represents the preset hyperparameters. In this embodiment, we use... For example, this is used to prevent the denominator from being 0.

[0039] It should be noted that in this embodiment, there are no identical observations among different observation types. The time-series verification prediction value refers to the numerical value of the observation type in the state transition matrix, and the time-series verification prediction value refers to the numerical value of the observation type in the verification matrix. Please refer to [link / reference]. Figure 2 It shows a schematic diagram of the probability prediction accuracy curve, where the horizontal axis represents the Markov trial order and the vertical axis represents the probability prediction accuracy.

[0040] Furthermore, the normalized value of the product between the probability prediction accuracy and the state transition probability of the Markov trial order is used as the personalized disease-aided variability of the Markov trial order; the personalized disease-aided variability of all Markov trial orders is obtained, and the Markov trial order with the highest personalized disease-aided variability is taken as the Markov sample order of the chronic disease patient; the Markov sample order of each chronic disease patient is obtained. Each sexually transmitted disease patient corresponds to one Markov sample order.

[0041] Thus, the Markov sample order for each chronic disease patient was obtained using the above method.

[0042] Step S003: Take any chronic disease patient as the target patient, traverse the historical monitoring data sequence of the target patient through the Markov sample order, construct the historical state window of the target patient, and calculate several symptom time-series auxiliary vectors of the target patient; by comparing the similarity of the symptom time-series auxiliary vectors between the target patient and other chronic disease patients, select several time-series dynamic similar patients of the target patient from other chronic disease patients.

[0043] It should be noted that step S002 primarily addressed the issue of applying low-order Markov algorithms; the next step is to analyze and address the problem of insufficient training sample data. While the historical monitoring data of each chronic disease patient may seem isolated, the aggregation of massive amounts of similar data reveals common patterns within the same chronic disease: for example, the definition of glycated hemoglobin control targets stems from a macroscopic analysis of the risk of complications within a group. More importantly, by matching characteristics such as age, disease duration, and pancreatic function, we can identify "patient groups" highly similar to the current patient. The treatment plans of these patients can provide direct reference for individualized medication, help predict rare reactions, and issue early warnings of abnormal fluctuations. Complex situations such as advanced age and multiple medications often require real-world data to compensate for the lack of clinical trials. Therefore, the historical monitoring data of different patients with the same chronic disease also has certain reference value. By analyzing the historical monitoring data of other chronic disease patients and using optimized Markov algorithms for prediction and transformation, we can, to some extent, supplement the current patient's required training sample data.

[0044] Preferably, in some implementations of the present invention, the method for obtaining the symptom time-series auxiliary vector is as follows: the length of the historical state window is determined by the Markov sample order of the target patient, and the historical state window is constructed; the historical state window is traversed over the historical monitoring data sequence of the target patient to obtain several characteristic symptom values ​​of the historical state window of the target patient in different traversal processes; the feature vector constructed based on the characteristic symptom values ​​is used as the symptom time-series auxiliary vector of the historical state window of the target patient in different traversal processes. The specific process is as follows: Preferably, in some implementations of the present invention, the method for obtaining the characteristic symptom values ​​is as follows: for any historical state window of the traversal process, calculate the maximum value, minimum value, mean value, and variance of the monitoring data within the historical state window, and use the maximum value, minimum value, mean value, and variance of the monitoring data as the characteristic symptom values. The specific process is as follows: Using any chronic disease patient as the target patient, a historical state window of the same length as the Markov sample order of the target patient is constructed. Starting from the first historical monitoring data in the target patient's historical monitoring data sequence, with a step size of 1, the historical state window is traversed. The maximum, minimum, mean, and variance of the monitoring data within the historical state window are calculated during each traversal. The maximum, minimum, mean, and variance of the monitoring data are all used as characteristic symptom values. The traversal stops when the historical state window begins to contain the last historical monitoring data in the historical monitoring data sequence. Several characteristic symptom values ​​of the target patient in the historical state window are obtained during each traversal.

[0045] Furthermore, in each traversal process, the feature vector constructed from the maximum value, minimum value, mean value, and variance of the monitoring data is used as the disease time-series auxiliary vector of the target patient in the historical state window during each traversal process.

[0046] Preferably, in some implementations of the present invention, the method for obtaining time-series dynamically similar patients is as follows: the inverse proportional relationship between the time-series auxiliary vectors of the target patient and other chronic disease patients is used as the order window similarity between the target patient and other chronic disease patients; and based on the order window similarity, several time-series dynamically similar patients of the target patient are selected from other chronic disease patients. The specific process is as follows: Patients with chronic diseases other than the target patient were used as control patients. The inverse proportional value of the Euclidean distance between the disease time-series auxiliary vectors of the target patient and each control patient was used as the order-window similarity between the target patient and each control patient. A preset order-window similarity threshold was set. The similarity of the order window is greater than The control patients are used as time-series dynamic similar patients to the target patients. All time-series dynamic similar patients of the target patients are obtained. Each target patient corresponds to multiple time-series dynamic similar patients. Furthermore, this embodiment uses... This example is used for illustration; no specific limitations are set in this embodiment. It depends on the specific implementation situation.

[0047] It should be noted that this embodiment uses the default setting. The model is used to represent the inverse proportional relationship. As input to the model, implementers can choose an inverse proportional function based on the actual situation.

[0048] Thus, several patients with similar temporal dynamics were obtained through the above methods.

[0049] Step S004: Based on the historical status window of the target patient, integrate the status transition matrices of all time-series dynamic similar patients within the same time range to construct several time-series status reference matrices for the target patient; based on the historical monitoring data sequence of the target patient, analyze the trend of the target patient's own monitoring data affecting the disease status to construct the time-series data monitoring trend matrix for the target patient; synergistically fuse the time-series status reference matrix and the time-series data monitoring trend matrix to predict the final monitoring data of the target patient.

[0050] It should be noted that steps S002 and S003 above address the issues of insufficient training data in low-order Markov algorithms. Since these problems are the result of separate, relatively independent analysis and optimization, while the core technical problem this invention aims to solve is the contradiction between the two when they are integrated, this method integrates and analyzes them. This eliminates the limitations of their individual optimization operations to a certain extent, ultimately obtaining more accurate prediction data and achieving more precise prediction results.

[0051] Preferably, in some implementations of the present invention, the method for obtaining the time-series state reference matrix is ​​as follows: based on the data fluctuations contained in the historical state window of the target patient, the target ideal time interval of the target patient is divided; based on the monitoring data occupied by the target ideal time interval in different historical monitoring data sequences, and combined with the Markov sample order of patients with different time-series dynamic similarities, several time-series state reference matrices of the target patient are constructed. The specific process is as follows: Within the historical state window of the target patient across all traversal processes, the longest time interval occupied by the historical state window with the largest variance in the monitored data is taken as the target ideal time interval for the target patient. The sequence of historical monitoring data for each time-series dynamically similar patient within the target ideal time interval is taken as the concurrent monitoring data sequence for each time-series dynamically similar patient. Referring to the method for obtaining the state transition matrix of the Markov trial order, the Markov trial order is replaced with the Markov sample order for each time-series dynamically similar patient, and the state transition sample sequence is replaced with the concurrent monitoring data sequence for each time-series dynamically similar patient. Similarly, the state transition matrix for each time-series dynamically similar patient is obtained and used as the temporal state reference matrix for each time-series dynamically similar patient. The target patient corresponds to multiple time-series dynamically similar patients, and each time-series dynamically similar patient corresponds to one temporal state reference matrix; that is, the target patient corresponds to multiple time-series state reference matrices.

[0052] Preferably, in some implementations of the present invention, the method for obtaining the time-series data monitoring trend matrix is ​​as follows: In the historical monitoring data sequence of the target patient, calculate several trend label values ​​for the target patient; within the target ideal time interval, construct the time-series data monitoring trend matrix of the target patient based on the sequence formed by the trend label values ​​and according to a preset traversal step size. The specific process is as follows: In the historical monitoring data sequence of the target patient, the slope of each historical monitoring data point within the target ideal time interval is used as the trend label value of the target patient; the sequence composed of all trend label values ​​is used as the ideal trend label sequence of the target patient. Referring to the method for obtaining the state transition matrix of the Markov trial order, the Markov trial order is replaced with the Markov sample order of the target patient, and the state transition sample sequence is replaced with the ideal trend label sequence of the target patient. Similarly, the state transition matrix of the target patient is obtained and used as the time-series data monitoring trend matrix of the target patient.

[0053] It should be noted that when the trend label value is greater than 0, the target patient's blood glucose level is in a short-term upward state; when the trend label value is equal to 0, the target patient's blood glucose level is in a short-term stable state; and when the trend label value is less than 0, the target patient's blood glucose level is in a short-term downward state.

[0054] Preferably, in some implementations of the present invention, the method for obtaining the final monitoring data is as follows: calculating the state candidate probability of each time-series state reference matrix, calculating the trend candidate probability of the time-series data monitoring trend matrix; combining the state candidate probability and the trend candidate probability to calculate the prediction fusion occurrence probability; and based on the prediction fusion occurrence probability, selecting the final monitoring data of the target patient from all predicted values. The specific process is as follows: Within each time-series state reference matrix, the ratio of the number of predicted values ​​to the number of predicted values ​​is used as the state candidate probability for each time-series state reference matrix. Within the time-series data monitoring trend matrix, the ratio of the types of predicted values ​​to the number of predicted values ​​is used as the state candidate probability of the time-series data monitoring trend matrix. As an example, the predicted fusion probability for each time-series dynamically similar patient can be calculated using the following formula. :

[0055] In the formula, This represents the state candidate probability of each temporal state reference matrix; This represents the state candidate probability of the time series data monitoring trend matrix; Represents the eigenvalues ​​of each time-series state reference matrix; The eigenvalues ​​represent the trend matrix of time-series data monitoring. This represents the normalization function.

[0056] Furthermore, a preset probability threshold for predicted fusion is established. The probability of predicted fusion is greater than Patients with similar temporal dynamics are selected as the target patient group in a time-series sample. Among the last predicted value in the historical monitoring data sequence of all patients in the time-series sample, the observation with the highest frequency is used as the final monitoring data for the target patient. This embodiment uses... This example is used for illustration; no specific limitations are set in this embodiment. It depends on the specific implementation situation.

[0057] It should be noted that the final monitoring data is the predicted data after the last historical monitoring data of the target patient, and is not actually recorded in the database.

[0058] Thus, the final monitoring data is obtained through the above methods.

[0059] Step S005: Based on the final monitoring data, assess the current patient's follow-up visit level and intelligently generate a follow-up visit auxiliary early warning plan.

[0060] In one specific implementation of this invention, the final monitoring data of each chronic disease patient is obtained, and the average of the final monitoring data of all chronic disease patients is used as the final monitoring data of the current patient. A preset low-risk range is established. Medium risk range High-risk areas If the patient's final monitoring data is within the low-risk range. The electronic report indicates that the patient's condition remains stable with a high probability of low risk; however, the final monitoring data for the patient may fall within the medium-risk range. The electronic report indicates a high probability that the patient's condition is progressing to a medium-risk level, and a planned follow-up visit is recommended; however, if the patient's final monitoring data falls within the high-risk range... The electronic report indicates that the patient's condition has a high probability of progressing to a high-risk stage, and recommends immediately scheduling a follow-up appointment, completing the process of obtaining auxiliary early warning data for chronic disease patient follow-up visits. This embodiment uses... [4.0 mmol / L, 10.0 mmol / L], =[10.0mmol / L, 14mmol / L), =[14.0mmol / L,+ The example described uses (0, 4 mmol / L) as an example. This embodiment does not impose specific limitations. , , It depends on the specific implementation; || represents logic or running algorithm.

[0061] This concludes the embodiment.

[0062] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for processing time-series analysis-based auxiliary early warning data for follow-up visits of chronic disease patients, characterized in that, The method includes the following steps: Obtain historical monitoring data sequences for patients with different chronic diseases; The historical monitoring data sequence is randomly divided into a state transition sample sequence and an order verification sample sequence. The state transition matrix is ​​constructed by traversing the Markov order through the state transition sample sequence, and the state transition probability of the Markov order is calculated in different traversal processes. The state transition probability is evaluated and determined by the order verification sample sequence, and the Markov sample order of different chronic disease patients is marked in all Markov orders. Taking any chronic disease patient as the target patient, the historical monitoring data sequence of the target patient is traversed through the Markov sample order to construct the historical state window of the target patient and calculate several symptom time-series auxiliary vectors of the target patient; by comparing the similarity of the symptom time-series auxiliary vectors of the target patient with those of other chronic disease patients, several time-series dynamic similar patients of the target patient are selected from other chronic disease patients. Based on the historical status window of the target patient, the status transition matrices of all time-series dynamic similar patients within the same time range are integrated to construct several time-series status reference matrices for the target patient; based on the historical monitoring data sequence of the target patient, the trend of the target patient's own monitoring data affecting the disease status is analyzed to construct the time-series data monitoring trend matrix for the target patient; the time-series status reference matrices and the time-series data monitoring trend matrices are synergistically fused to predict the final monitoring data of the target patient; Based on the final monitoring data, the current patient's follow-up visit level is assessed, and an intelligent follow-up visit auxiliary early warning plan is generated.

2. The method for processing auxiliary early warning data for follow-up visits of chronic disease patients based on time-series analysis according to claim 1, characterized in that, The method for obtaining the state transition probability is as follows: For any chronic disease patient, a predefined range of Markov order traversal is used to perform Markov order traversal on the state transition sample sequence of the chronic disease patient to obtain the state transition sequence of the chronic disease patient in different traversal processes. Construct a state transition matrix for patients with chronic diseases using all state transition sequences; The probability distribution of Markov order predicted states within the statistical state transition matrix is ​​calculated, and the state transition probabilities of Markov order are calculated for different traversals.

3. The method for processing auxiliary early warning data for follow-up visits of chronic disease patients based on time-series analysis according to claim 1, characterized in that, The method for obtaining the Markov sample order is as follows: For any chronic disease patient, a validation matrix of the same order as the state transition matrix is ​​constructed using the validation sample sequence of the chronic disease patient's order. The difference between the Markov order prediction values ​​between the state transition matrix and the validation matrix is ​​compared to calculate the probability prediction accuracy of the Markov order in different traversals. Combining the state transition probability and the probability prediction accuracy, the personalized disease-aided evolution of different Markov orders is calculated. Based on the individualized symptom-assisted evolution, the Markov sample order of chronic disease patients was identified from all Markov orders.

4. The method for processing auxiliary early warning data for follow-up visits of chronic disease patients based on time-series analysis according to claim 3, characterized in that, The method for obtaining the accuracy of the probability prediction is as follows: For any Markov order in the traversal process, obtain several temporal traversal prediction values ​​of the Markov order from the state transition matrix, and obtain several temporal verification prediction values ​​of the Markov order from the verification matrix. The difference between all time-series ergodic predictions and time-series verification predictions is used as the probability prediction accuracy of the Markov order.

5. The method for processing auxiliary early warning data for follow-up visits of chronic disease patients based on time-series analysis according to claim 1, characterized in that, The method for obtaining the disease time-series auxiliary vector is as follows: The length of the historical state window is determined by the Markov sample order of the target patient, and the historical state window is constructed. The historical state window is traversed over the historical monitoring data sequence of the target patient to obtain several characteristic symptom values ​​of the historical state window of the target patient in different traversal processes. The feature vector constructed based on the characteristic symptom values ​​is used as the symptom time-series auxiliary vector of the target patient in the historical state window in different traversal processes.

6. The method for processing auxiliary early warning data for follow-up visits of chronic disease patients based on time-series analysis according to claim 5, characterized in that, The method for obtaining the characteristic symptom values ​​is as follows: For any historical state window in the traversal process, calculate the maximum value, minimum value, mean value, and variance of the monitoring data within the historical state window, and use the maximum value, minimum value, mean value, and variance value of the monitoring data as characteristic symptom values.

7. The method for processing auxiliary early warning data for follow-up visits of chronic disease patients based on time-series analysis according to claim 1, characterized in that, The method for obtaining the time-series dynamically similar patients is as follows: The inverse proportional relationship between the symptom time-series auxiliary vectors of the target patient and other chronic disease patients is used as the order window similarity between the target patient and other chronic disease patients. Based on the similarity of the order window, several time-series dynamic similar patients of the target patient are selected from other chronic disease patients.

8. The method for processing auxiliary early warning data for follow-up visits of chronic disease patients based on time-series analysis according to claim 1, characterized in that, The method for obtaining the time-series state reference matrix is ​​as follows: Based on the data fluctuations contained in the historical status window of the target patient, the target ideal time interval is divided. Based on the monitoring data in different historical monitoring data sequences within the target ideal time interval, and combined with the Markov sample order of patients with different time-series dynamic similarities, several time-series status reference matrices of the target patient are constructed.

9. The method for processing auxiliary early warning data for follow-up visits of chronic disease patients based on time-series analysis according to claim 8, characterized in that, The method for obtaining the time-series data monitoring trend matrix is ​​as follows: In the historical monitoring data sequence of the target patient, several trend label values ​​of the target patient are calculated; under the target ideal time interval, based on the sequence formed by the trend label values, a time series data monitoring trend matrix of the target patient is constructed according to a preset traversal step size.

10. The method for processing auxiliary early warning data for follow-up visits of chronic disease patients based on time-series analysis according to claim 1, characterized in that, The method for obtaining the final monitoring data is as follows: Calculate the state candidate probability for each time-series state reference matrix, and calculate the trend candidate probability for the time-series data monitoring trend matrix; combine the state candidate probability and the trend candidate probability to calculate the probability of prediction fusion for patients with similar dynamics at different time series. Based on the predicted probability of fusion, the final monitoring data of the target patient is selected from all predicted values.

Images