Chronic disease patient health management system based on big data

Abstract

The invention relates to the technical field of medical health management, in particular to a chronic disease patient health management system based on big data, which comprises the following steps: constructing a full-dimensional personal health record containing static and dynamic information; and the data analysis module performs abnormal mode mining on the continuous physical sign data and associates the periodic test data with historical records to generate a trend graph. And the data association module constructs a risk association matrix for quantifying the influence contribution degree of the environmental behavior factors to the health indexes based on the environmental exposure data. And the risk modeling module integrates the information. And the prediction and intervention generation module performs calculation by using the chronic disease progress prediction model, and synchronously outputs an individualized disease risk early warning signal and a staged health intervention scheme matched with the individualized disease risk early warning signal. According to the invention, quantitative attribution of health risk factors and dynamic and staged accurate health management are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical health management, and particularly relates to a chronic disease patient health management system based on big data. BACKGROUND

[0002] The existing chronic disease patient health management system is mostly based on electronic health record integration and wearable device data access, and provides data visualization and simple early warning function based on fixed threshold. These technical solutions mainly realize centralized storage and trend display of data such as signs and tests, and trigger a reminder when the monitoring data exceeds the preset general safety range. The core defect of this kind of solution is that the analysis and early warning mechanism stays in the comparison of data appearance, and cannot explain the potential causes of sign fluctuation or index change. The alarm generated by the system often only indicates that “a certain index is abnormal”, but cannot answer “why the abnormality occurs at this time”. In addition, the health suggestions given by the system are usually static and general knowledge base entries, which cannot be closely combined with the current specific and dynamically changing risk status of the patient, resulting in insufficient individualization and timeliness of intervention.

[0003] The existing technology lacks the ability to systematically quantify and attribute analyze the multiple external factors affecting chronic diseases. The health status of patients is affected by multiple factors such as genetics, behavior and environment. Although the conventional system can record environmental or behavioral data, it does not establish a quantitative correlation model between these external factors and the changes in internal physiological indicators. This makes it impossible for health management to accurately identify and focus on key risk driving factors that can be changed. At the same time, due to the lack of predictive judgment of the dynamic trajectory of disease progression, the intervention suggestions of the system are often reactive and one-time, and cannot form a dynamic management plan that matches the evolution stage of disease risk and gradually progresses, making it difficult to achieve real effective long-term health outcome improvement. SUMMARY

[0004] The purpose of the present application is to solve the problems existing in the prior art, and to provide a chronic disease patient health management system based on big data.

[0005] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows: a chronic disease patient health management system based on big data, comprising:

[0006] An archive construction module constructs a full-dimension personal health archive of the chronic disease patient, the full-dimension personal health archive contains a static basic information set and a dynamic monitoring information set, the static basic information set includes genetic history records, past diagnosis records and long-term medication records, and the dynamic monitoring information set includes continuous sign monitoring data, periodic test data and environmental exposure data;

[0007] a data analysis module, which performs time series anomaly pattern mining on the continuous vital sign monitoring data, identifies vital sign fluctuation segments deviating from individual baseline patterns, and performs correlation mapping analysis on the periodic test data and the past diagnosis records to generate test index trend evolution maps,

[0008] a data correlation module, which constructs a risk correlation matrix of patient behavior activities and external environment based on the environmental exposure data, and quantifies the contribution of different environmental factors and behavior patterns to each vital sign fluctuation and test index change;

[0009] a risk modeling module, which integrates the vital sign fluctuation segments, the test index trend evolution maps, and the risk correlation matrix;

[0010] a prediction and intervention generation module, which uses a preset chronic disease progression prediction model to generate individualized disease risk warning signals and phased health intervention plans.

[0011] As a further scheme of the present application, the time series anomaly pattern mining on the continuous vital sign monitoring data to identify vital sign fluctuation segments deviating from individual baseline patterns comprises:

[0012] The vital sign fluctuation segments include heart rate variability abnormal interval, blood pressure circadian rhythm disorder period, and blood oxygen saturation periodic decline event.

[0013] A historical data time window for calculating individual baseline patterns is set, and historical monitoring sequences within the historical data time window are extracted from the continuous vital sign monitoring data;

[0014] The average, standard deviation, and periodic component of each vital sign index in the historical monitoring sequences are calculated to determine the normal fluctuation range and rhythm characteristics of individual vital sign indexes;

[0015] The newly acquired real-time monitoring sequences are compared point by point with the normal fluctuation range and rhythm characteristics of the vital sign indexes, and data points exceeding the fluctuation range or violating the rhythm characteristics are marked;

[0016] The marked abnormal data points are subjected to cluster analysis, and time-continuous or close abnormal data points of the same type are aggregated into the vital sign fluctuation segments;

[0017] Each vital sign fluctuation segment is assigned a type label and a severity score, and the type label includes heart rate variability abnormality, blood pressure circadian rhythm disorder, or blood oxygen saturation periodic decline.

[0018] As a further scheme of the present application, the point-by-point comparison of the newly acquired real-time monitoring sequence with the normal fluctuation range and rhythm characteristics of the physical indicators, and the marking of data points exceeding the fluctuation range or violating the rhythm characteristics, comprises:

[0019] Extracting a single physical indicator time sequence from the real-time monitoring sequence, and performing numerical comparison with the corresponding normal fluctuation range of the physical indicator; if a plurality of consecutive data points exceed the upper or lower limit of the normal fluctuation range, it is marked as continuous abnormality;

[0020] Analyzing the change pattern of the single physical indicator time sequence within 24 hours, and performing pattern similarity calculation with the typical diurnal change curve defined in the rhythm characteristics; if the similarity is lower than a preset threshold, it is marked as rhythm abnormality;

[0021] Detecting whether there is a sharply rising or falling segment in the single physical indicator time sequence within a short period of time; by calculating the difference value of adjacent data points and comparing it with the historical average change rate, if it exceeds the historical average change rate by several multiples, it is marked as mutation abnormality;

[0022] Recording data points meeting any of the conditions of continuous abnormality, rhythm abnormality or mutation abnormality as abnormal data points, and recording the specific abnormality type.

[0023] As a further scheme of the present application, the correlation mapping analysis of the periodic test data and the past diagnosis records to generate a test indicator trend evolution map comprises:

[0024] The test indicator trend evolution map reflects the change trajectory of the value of a specific test item over time and its corresponding relationship with historical diagnosis events;

[0025] Extracting multiple test results of the same test item from the periodic test data, and sorting them by test time to form a numerical sequence of a specific item;

[0026] Extracting the time points of diagnosis events and diagnosis conclusions from the past diagnosis records, the diagnosis conclusions including disease staging or severity grade;

[0027] Aligning the numerical sequence of the specific item with the time points of the diagnosis events on the time axis, and establishing the time correspondence between test values and diagnosis events;

[0028] Analyzing the change trend, change rate and peak or valley value reached by the numerical sequence of the specific item before and after each diagnosis event;

[0029] Based on the time correspondence and change characteristics, a graphical map is generated with time as the horizontal axis, test item value as the vertical axis, and diagnosis event points marked, i.e. the test indicator trend evolution map.

[0030] As a further aspect of the present invention, the analysis of the changing trend, rate of change, and peak or trough values ​​of the numerical sequence of the specific item before and after each diagnostic event includes:

[0031] Using the time of the diagnostic event as the dividing point, time windows of predetermined lengths are selected before and after the event, and the average value of the numerical sequence of the specific items within the window is calculated as the baseline value before diagnosis and the baseline value after diagnosis.

[0032] The difference between the baseline value after diagnosis and the baseline value before diagnosis is calculated as the net change in the impact of the diagnostic event on the test indicators.

[0033] Within the pre-diagnosis time window, the linear regression slope of the numerical series is calculated as the pre-event trend slope; within the post-diagnosis time window, the linear regression slope of the numerical series is calculated as the post-event trend slope.

[0034] Identify the global maximum and global minimum values ​​of the entire numerical sequence over the entire time range, and record them as peak values ​​and valley values, respectively;

[0035] The average daily change from the pre-diagnosis baseline value to the post-diagnosis baseline value was recorded as the average daily rate of change.

[0036] As a further aspect of the present invention, the step of constructing a risk correlation matrix between patient behavior and the external environment based on the environmental exposure data includes:

[0037] Multiple environmental exposure factors and patient behavioral activity records were extracted from the environmental exposure data. The environmental exposure factors included air quality index, temperature and humidity levels, and noise decibel levels. The patient behavioral activity records included physical activity intensity, sleep duration, and dietary content.

[0038] For each health outcome associated with a chronic disease, a set of relevant physical and laboratory indicators are defined as target variables for risk association.

[0039] Multivariate correlation analysis and regression analysis were used to calculate the statistical association strength coefficient between each environmental exposure factor and each patient behavioral activity record on each target variable.

[0040] All the calculated statistical association strength coefficients are organized into a multi-dimensional numerical table, namely the risk association matrix, according to the environmental exposure factor, the behavioral activity record rows, and the target variable as the column.

[0041] The coefficients in the risk correlation matrix are standardized to make the coefficients of different target variables comparable.

[0042] As a further scheme of the present application, the statistical correlation strength coefficient of each environmental exposure factor and each patient behavior activity record to each target variable is calculated respectively by using the multivariate correlation analysis and regression analysis method, including:

[0043] For each specific combination of environmental exposure factor and target variable, the historical data pairs of the environmental exposure factor and the target variable at the same time point are collected;

[0044] The correlation coefficient between the historical data sequence of the environmental exposure factor and the historical data sequence of the target variable is calculated;

[0045] A linear regression model is established with the environmental exposure factor as the independent variable and the target variable as the dependent variable, and the regression coefficient is fitted;

[0046] The correlation coefficient and the regression coefficient are comprehensively weighted to obtain the statistical correlation strength coefficient of the environmental exposure factor to the target variable;

[0047] For each combination of patient behavior activity record and target variable, the steps from data pair collection to comprehensive weighting are repeated.

[0048] As a further scheme of the present application, the sign fluctuation segment, the test index trend evolution graph and the risk correlation matrix are integrated, and a preset chronic disease progression prediction model is used for calculation to generate an individualized disease risk warning signal, including:

[0049] All the sign fluctuation segments identified in the current monitoring period are converted into input feature vectors of the prediction model according to their types and severity scores;

[0050] The recent trend slope and the deviation of the current value from the historical baseline are extracted from the latest test index trend evolution graph as another group of input features;

[0051] According to the environmental exposure data and behavior activity record recorded in the patient's recent record, the comprehensive risk exposure score under the current environmental and behavior combination is calculated by combining the risk correlation matrix as a third group of input features;

[0052] The input feature vectors, recent trend slope and deviation, and comprehensive risk exposure score are input into the chronic disease progression prediction model;

[0053] The chronic disease progression prediction model outputs the probability value of occurrence of a specific acute event or disease staging progression of a specific chronic disease within a specified period of time in the future. When the probability value exceeds a preset warning threshold, the corresponding disease risk warning signal is generated, and the signal contains the risk event type and the predicted occurrence time window.

[0054] As a further scheme of the present application, the generating the periodic health intervention scheme comprises:

[0055] The health intervention scheme contains medication adjustment suggestions, behavior pattern optimization lists, and recheck plans;

[0056] According to the risk event type indicated in the disease risk early warning signal, a corresponding intervention measure set is matched from a preset intervention knowledge base;

[0057] Combined with the patient's current long-term medication record, the medication suggestions in the intervention measure set are subjected to compatibility check to screen out non-conflicting medication adjustment suggestions;

[0058] According to the patient's personal work and rest habit preferences and the behavior activity record, specific execution time, frequency and intensity are set for the behavior intervention suggestions in the intervention measure set to form a behavior pattern optimization list;

[0059] According to the key indicator monitoring requirements indicated in the test index trend evolution map and the urgency of the disease risk early warning signal, a recheck plan containing next test items and recommended recheck time is made;

[0060] The medication adjustment suggestions, behavior pattern optimization lists and recheck plans are integrated into a periodic health intervention scheme document with clear execution cycle.

[0061] As a further scheme of the present application, the construction steps of the chronic disease progression prediction model comprise:

[0062] Collecting complete health monitoring data sets of historical chronic disease patient groups, the health monitoring data sets containing continuous body sign monitoring data, periodic test data, environmental exposure data and corresponding disease progression event records;

[0063] Time series segmentation processing is performed on the continuous body sign monitoring data to extract body sign fluctuation pattern feature vectors of each patient in different time periods;

[0064] Trend feature extraction processing is performed on the periodic test data to generate test index change trajectory descriptors of each patient;

[0065] Feature encoding processing is performed on the environmental exposure data and patient behavior activity records to obtain environmental behavior feature matrix;

[0066] Multi-source feature fusion processing is performed on the body sign fluctuation pattern feature vectors, test index change trajectory descriptors and environmental behavior feature matrix to generate comprehensive feature representation;

[0067] The disease progression state labels of each patient sample are annotated based on the disease progression event records;

[0068] The trained time series deep learning network is trained by using the comprehensive feature representation and the disease progression state label, and the network parameters are optimized by a back propagation algorithm to obtain a trained chronic disease progression prediction model.

[0069] Compared with the prior art, the application has the advantages and positive effects that:

[0070] By structurally associating the behavior activities of patients with the external environment based on environmental exposure data, a risk association matrix is constructed, and statistical analysis or machine learning algorithms are used to calculate specific contribution values of different environment-behavior combinations on various health indicators. This scheme converts the originally general environment and behavior risks into quantifiable, sortable and comparable numerical indicators. This can clearly reveal the key external driving factors affecting the health fluctuations of a specific patient and their relative importance, enabling health management to move from focusing on whether the indicators are normal to understanding why the indicators change, providing direct quantitative decision-making basis for taking targeted and precise environmental adjustments and behavior interventions, and surpassing the traditional mode based on experience or qualitative suggestions.

[0071] By using the preset chronic disease progression prediction model, multi-source fusion data is integrated to generate a dynamic health intervention scheme directly bound to individualized disease risk warning signals and executed in stages. Instead of outputting a static recommendation list, the scheme outputs a dynamic action plan sequence containing different execution time nodes and different intervention focuses. This enables health intervention to actively match the predicted risk evolution rhythm and the specific health stage currently experienced by the patient. Intervention measures can be scientifically arranged in time according to risk priority and feasibility to form a complete management path combining immediate short-term measures and continuous long-term plans, thereby converting one-time reminders into an executable and adjustable continuous health management process. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 A timing diagram of the chronic disease patient health management system based on big data described in the application;

[0073] Figure 2 A flowchart for generating an index trend evolution graph. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solutions and advantages of the application clearer, the application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application and do not limit the application.

[0075] In the description of the present application, it needs to be understood that the terms "length", "width", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, in the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly and specifically limited.

[0076] Referring to Figure 1 The archive construction module is responsible for constructing a full-dimension personal health archive of a chronic disease patient, which contains a static basic information set and a dynamic monitoring information set. The static basic information set includes genetic history records, past diagnosis records and long-term medication records. The dynamic monitoring information set includes continuous sign monitoring data, periodic test data and environmental exposure data. The data analysis module performs time series anomaly pattern mining on the continuous sign monitoring data to identify sign fluctuation segments deviating from the personal baseline pattern. The data analysis module also performs correlation mapping analysis on the periodic test data and the past diagnosis records to generate a test index trend evolution graph. The data correlation module constructs a risk correlation matrix of patient behavior activities and external environment based on the environmental exposure data, which quantifies the contribution of different environmental factors and behavior patterns to the fluctuation of various signs and the change of test indexes. The risk modeling module integrates the sign fluctuation segments, the test index trend evolution graph and the risk correlation matrix. The prediction and intervention generation module uses a preset chronic disease progression prediction model to calculate and generate individualized disease risk warning signals and phased health intervention plans.

[0077] Referring to Figure 2 In an embodiment of the present application, time series anomaly pattern mining is performed on continuous sign monitoring data to identify sign fluctuation segments deviating from the personal baseline pattern, including heart rate variability abnormal interval, blood pressure circadian rhythm disorder period and blood oxygen saturation periodic decline event. A historical data time window for calculating the personal baseline pattern is set, for example, the past thirty days, the historical monitoring sequence within the historical data time window is extracted from the continuous sign monitoring data, which contains hourly recorded heart rate, systolic pressure, diastolic pressure and blood oxygen saturation values. The average, standard deviation and periodic component of different sign indicators in the historical monitoring sequence are calculated respectively to determine the personalized sign indicator normal fluctuation range and rhythm characteristics. The personalized sign indicator normal fluctuation range is defined by the average value plus or minus twice the standard deviation, and the rhythm characteristics are extracted by Fourier analysis to obtain the typical waveform within twenty-four hours. In some embodiments, the length of the historical data time window can be dynamically adjusted according to the data acquisition frequency.

[0078] In specific implementation, the newly acquired real-time monitoring sequence is compared with the normal fluctuation range and rhythm characteristics of the physical indicators point by point, and the data points exceeding the fluctuation range or violating the rhythm characteristics are marked. The single physical indicator time sequence in the real-time monitoring sequence, such as the single-day heart rate time sequence, is extracted and compared with the corresponding normal fluctuation range of the heart rate indicator in numerical value. If the consecutive three data points exceed the upper limit or lower limit of the normal fluctuation range, it is marked as continuous abnormality. The change pattern of the single physical indicator time sequence within twenty-four hours is analyzed, and the pattern similarity is calculated with the typical diurnal change curve defined in the rhythm characteristics. The pattern similarity is calculated by dynamic time warping algorithm. If the calculated dynamic time warping distance is higher than the preset threshold, it is marked as rhythm abnormality. It is detected whether there is a sharp rise or fall segment in the single physical indicator time sequence within a short period of time. The difference between adjacent data points is calculated and compared with the historical average change rate. If it exceeds three times of the historical average change rate, it is marked as mutation abnormality. It can be understood that the marking rules can be configured. The data points meeting any of the continuous abnormality, rhythm abnormality or mutation abnormality are recorded as abnormal data points, and the specific abnormal type is recorded.

[0079] In specific implementation, the marked abnormal data points are subjected to cluster analysis, and the marked points that are continuous or close in time and have consistent abnormal types are aggregated into physical fluctuation segments. It can be understood that the aggregation process is based on the time stamp and type label of the abnormal data points, and is realized by using a density-based clustering algorithm. Each physical fluctuation segment is given a type label and a severity score. The type label includes heart rate variation abnormality, blood pressure diurnal rhythm disorder or blood oxygen saturation periodic decline. The severity score is calculated by a formula, which is:

[0080]

[0081] Wherein: represents the severity score, represents the number of abnormal data points contained in the physical fluctuation segment, represents the actual monitoring value of the th abnormal data point, represents the expected value of the physical indicator in the personal baseline mode at the corresponding time, represents the standard deviation of the physical indicator in the personal baseline mode at the corresponding time, represents the weight coefficient preset according to the abnormal type.

[0082] In one embodiment of the present application, the periodic test data is correlated and mapped with the past diagnosis records to generate a test index trend evolution map, which reflects the changing trajectory of a specific test item value over time and its corresponding relationship with historical diagnosis events. The multiple test results of the same test item are extracted from the periodic test data, such as the glycated hemoglobin test values of a diabetic patient, and sorted by test time to form a numerical sequence of a specific item, which contains multiple time stamps and corresponding test values. The time points of diagnosis events and diagnosis conclusions are extracted from the past diagnosis records, and the diagnosis conclusions include disease stages or severity levels, such as the diagnosis conclusion of "diabetic nephropathy stage III". The numerical sequence of a specific item and the time points of diagnosis events are aligned on the time axis to establish the time correspondence between test values and diagnosis events, so that each glycated hemoglobin test value can be associated with the nearest diagnosis event before or after it on the time axis.

[0083] In some embodiments, the change trend, change rate, and peak or valley value exhibited by the numerical sequence of a specific item before and after each diagnosis event are analyzed. With the diagnosis event time as the dividing point, a predetermined length of time window is selected forward and backward, respectively, such as six months before diagnosis and six months after diagnosis as the time window, and the average value of the numerical sequence of a specific item in the window is calculated as the baseline value before diagnosis and the baseline value after diagnosis. The difference between the baseline value after diagnosis and the baseline value before diagnosis is calculated as the net change of the influence of the diagnosis event on the test index. In the pre-diagnosis time window, the linear regression slope of the numerical sequence is calculated as the pre-event trend slope; in the post-diagnosis time window, the linear regression slope of the numerical sequence is calculated as the post-event trend slope. It can be understood that the length of the time window can be adjusted according to the type of disease and the clinical significance of the test item. The global maximum and global minimum of the entire numerical sequence in the entire time range are identified and recorded as the peak value and the valley value, respectively. The average daily change from the baseline value before diagnosis to the baseline value after diagnosis is recorded as the average daily change rate, and its calculation formula is:

[0084]

[0085] wherein: represents the average daily change rate, represents the baseline value before diagnosis, represents the baseline value after diagnosis, represents the actual number of days experienced from the time point corresponding to the pre-baseline value to the time point corresponding to the post-baseline value. Optionally, when calculating the baseline values before and after diagnosis, a moving average of all values in the time window can be used.

[0086] In a specific implementation, based on the time correspondence and the change characteristics, a graphical atlas is generated, with time as the horizontal axis, the test item value as the vertical axis, and the diagnostic event points marked, i.e. the test index trend evolution atlas. It can be understood that in the graphical atlas, all historical test values will be plotted in the form of scattered points, and the trend of the value sequence will be displayed with a connecting line, and at the same time, the occurrence time and conclusion of each diagnostic event are clearly marked with a vertical line and a label at the corresponding position of the time axis. In some embodiments, the pre-event trend line and the post-event trend line calculated can be superimposed and displayed in the atlas. Optionally, the peak value, the valley value, and the positions of the pre-diagnosis baseline value and the post-diagnosis baseline value calculated can also be marked in the atlas.

[0087] In an embodiment of the present application, a risk association matrix of patient behavior activities and external environment is constructed based on environmental exposure data. A plurality of environmental exposure factors and patient behavior activity records are parsed from the environmental exposure data, the environmental exposure factors including air quality index, temperature and humidity level, and noise decibel level, and the patient behavior activity records including physical activity intensity, sleep duration, and diet content. For each health outcome related to a chronic disease, such as an acute cardiovascular event of hypertension, a set of related sign indexes and test indexes are defined as target variables of risk association, which can include systolic pressure, diastolic pressure, and C-reactive protein concentration in blood. In some embodiments, the selection of target variables follows clinical guidelines.

[0088] In a specific implementation, multivariate correlation analysis and regression analysis methods are used to calculate the statistical association strength coefficient of each environmental exposure factor and each patient behavior activity record to each target variable. For each specific pair of environmental exposure factor and target variable, such as outdoor PM2.5 concentration and systolic pressure, pairs of historical data of the environmental exposure factor and the target variable at the same time point are collected, and the time point can be accurate to the day. The correlation coefficient between the historical data sequence of the environmental exposure factor and the historical data sequence of the target variable is calculated, and the correlation coefficient is the Pearson correlation coefficient. A linear regression model is established with the environmental exposure factor as the independent variable and the target variable as the dependent variable, and the regression coefficient is fitted. It can be understood that the regression coefficient reflects the expected change amount of the target variable caused by the unit change of the environmental exposure factor. The correlation coefficient and the regression coefficient are comprehensively weighted to obtain the statistical association strength coefficient of the environmental exposure factor to the target variable.

[0089] In some embodiments, all the calculated statistical association strength coefficients are organized into a multi-dimensional numerical table, i.e. a risk association matrix, according to the environmental exposure factor, the behavior activity record row, and the target variable column. The row title of the risk association matrix is the specific environmental exposure factor name and the patient behavior activity record name, the column title is the target variable name, and each cell in the matrix is filled with the corresponding statistical association strength coefficient The coefficients in the risk correlation matrix are standardized to make them comparable across different target variables. The standardization method used is Z-score standardization, which involves subtracting the mean of all coefficients in each column (corresponding to a target variable) and then dividing by its standard deviation. It can be understood that the standardized coefficients directly reflect the relative strength of the influence of different environmental or behavioral factors under the same target variable dimension.

[0090] In one embodiment of the present invention, vital sign fluctuation segments, laboratory indicator trend evolution maps, and risk correlation matrices are integrated, and a pre-defined chronic disease progression prediction model is used to calculate and generate a personalized disease risk warning signal. All vital sign fluctuation segments identified within the current monitoring period are scored according to their type and severity, and converted into input feature vectors for the prediction model. The input feature vector is a multi-dimensional numerical array, where each dimension corresponds to a type of vital sign fluctuation segment, and its value is aggregated from the severity scores of that type of segment. The aggregation method can be summation or taking the maximum value. From the latest laboratory indicator trend evolution map, the recent trend slope and the offset of the current value relative to the historical baseline are extracted as another set of input features. The recent trend slope is usually taken from the post-event trend slope, and the historical baseline value can be the post-diagnosis baseline value or the average value of a longer historical window. Based on the patient's recently recorded environmental exposure data and behavioral activity records, combined with the risk correlation matrix, a comprehensive risk exposure score under the current environmental and behavioral combination is calculated as a third set of input features. It is understandable that calculating the comprehensive risk exposure score requires traversing the rows in the risk association matrix corresponding to the patient's current environment and behavioral records, and then weighted summing their coefficients with respect to a specific target variable. The input feature vector, recent trend slope and offset, along with the comprehensive risk exposure score, are input into the chronic disease progression prediction model. The chronic disease progression prediction model outputs the probability value of a specific acute event or disease stage progression occurring within a specified future time period. When the probability value exceeds a preset warning threshold, a corresponding disease risk warning signal is generated, which includes the risk event type and the expected occurrence time window. See Table 1 for a specific example of converting a segment of vital sign fluctuations into an input feature vector.

[0091] Table 1: Examples of converting vital sign fluctuation segments into input feature vectors

[0092] Vital sign fluctuation segment type label Severity score Corresponding input feature vector dimension Aggregate feature value Heart rate variability abnormality 8.5 Dimension 1 8.5 Heart rate variability abnormality 6.2 Dimension 1 8.5 (take max) Blood pressure circadian rhythm disorder 12.1 Dimension 2 12.1 Oxygen saturation periodic drop 3.0 Dimension 3 3.0

[0093] In some embodiments, the comprehensive risk exposure score can be calculated using the following formula:

[0094]

[0095] in: Representative regarding the first a comprehensive risk exposure score of the target variable, represents the total number of entries in the risk association matrix that are related to the environmental exposure factors and the behavioral activity records, is an indicator variable that takes the value 1 when the environmental or behavioral factor is observed in the recent records, and 0 otherwise, represents the normalized statistical association strength coefficient of the factor to the target variable in the risk association matrix. Optionally, the indicator variable can be replaced by the specific numerical value or rating of the factor in the recent records.

[0096] The steps of constructing the chronic disease progression prediction model include: collecting a complete health monitoring data set of a historical chronic disease patient population, the health monitoring data set including continuous vital sign monitoring data, periodic examination data, environmental exposure data, and corresponding disease progression event records. The continuous vital sign monitoring data is subjected to time series segmentation processing, and the vital sign fluctuation pattern feature vector of each patient in different time periods is extracted, which can include fluctuation frequency, average duration, average severity, etc. The periodic examination data is subjected to trend feature extraction processing to generate examination index change trajectory descriptors of each patient, which can include long-term slope, periodic amplitude, coefficient of variation, etc. The environmental exposure data and the patient behavior activity records are subjected to feature encoding processing to obtain an environmental behavior feature matrix, which can use one-hot encoding or discrete binning based on clinical knowledge. The vital sign fluctuation pattern feature vector, the examination index change trajectory descriptor, and the environmental behavior feature matrix are subjected to multi-source feature fusion processing to generate a comprehensive feature representation, which can be simple vector splicing or weighted fusion through an attention mechanism. The disease progression state label of each patient sample is labeled based on the disease progression event records, which can be binary progression or multi-classification progression type. The comprehensive feature representation and the disease progression state label are used to train a time series deep learning network, the network parameters are optimized through a back propagation algorithm, and a trained chronic disease progression prediction model is obtained, which can select a long short-term memory network or a time series convolution network. In some embodiments, the training process divides the training set, the validation set, and the test set. Optionally, the model training is completed to save its network structure and parameters for the prediction and intervention generation module to call.

[0097] In an embodiment of the present application, a periodic health intervention plan is generated, which contains medication adjustment suggestions, behavior pattern optimization list and re-visit plan. According to the type of risk event indicated in the disease risk early warning signal, for example, the early warning signal indicates an increased risk of acute exacerbation of heart failure within the next month, a corresponding intervention measure set is matched from the pre-set intervention knowledge base, which stores standardized intervention items corresponding to different risk event types. The matched intervention measure set may include "increase diuretic dose", "limit daily sodium intake" and "recommend cardiac ultrasound examination within a week". Combined with the patient's current long-term medication record, the medication adjustment suggestions in the intervention measure set are subjected to compatibility check to screen out non-conflicting medication adjustment suggestions. The compatibility check will check whether there is a known serious interaction between the drugs in the long-term medication record and the drugs to be added or adjusted, for example, the patient takes warfarin long-term, and the intervention measure suggests adding aspirin. After checking, it is found that there is an interaction of increased risk of bleeding, so this medication suggestion will be marked as conflicting and screened out. It can be understood that the compatibility check relies on a continuously updated drug interaction database.

[0098] In some embodiments, according to the patient's personal habit preference and behavior activity record, the specific execution time, frequency and intensity of the behavior intervention suggestions in the intervention measure set are set to form a behavior pattern optimization list. For example, for a behavior intervention suggestion of "perform moderate-intensity aerobic exercise", the system reads the patient's personal habit preference and finds that the patient usually has free time after 7:00 pm on weekdays, and combines the behavior activity history record to show that the patient can tolerate 30 minutes of walking each time, so the suggestion is specified as "walk 30 minutes each time at 19:30 on Monday, Wednesday and Friday every week", and this specific item is listed in the behavior pattern optimization list. Optionally, if the behavior activity record shows that the patient has knee discomfort recently, the system may automatically adjust the exercise intensity to "low intensity". According to the key indicator monitoring requirements indicated in the test indicator trend evolution map and the urgency of the disease risk early warning signal, a re-visit plan containing the next test item and recommended re-visit time is made. For example, the test indicator trend evolution map shows that the level of B-type natriuretic peptide has shown an upward trend recently, and the probability value of the disease risk early warning signal is high, so the re-visit plan will clearly list "B-type natriuretic peptide detection" as the next test item, and use a formula to calculate the recommended re-visit time:

[0099]

[0100] wherein: represents the recommended re-visit time, represents the current time, represents the start point of the predicted time window in the disease risk early warning signal, is a safety factor determined according to the risk level, represents rounding up. It can be understood that such calculation aims to ensure that the re-visit evaluation is completed before the risk event is expected to occur. The medication adjustment suggestion, the behavior pattern optimization list and the re-visit plan are integrated into a phased health intervention scheme document with a clear execution cycle, which will clearly indicate the execution cycle of the current phase of intervention. In some embodiments, the integrated document is presented to the patient and the medical staff in the form of a structured list and a calendar view. Optionally, at the end of the execution cycle, the system automatically triggers a new round of health status evaluation and scheme generation process.

[0101] The above merely describes the preferred embodiments of the present application, but does not limit the present application in other forms. Any person skilled in the art can use the disclosed technical content to make changes or modifications into equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application to the above embodiments without departing from the technical solution content of the present application still falls within the protection scope of the present application.

Claims

1. A big data-based chronic disease patient health management system, characterized by, The system comprises: an archive construction module, which constructs a full-dimension personal health archive of a chronic disease patient, the full-dimension personal health archive comprising a static basic information set and a dynamic monitoring information set, the static basic information set including genetic history records, past diagnosis records and long-term medication records, and the dynamic monitoring information set including continuous sign monitoring data, periodic test data and environmental exposure data; a data analysis module, which performs time series anomaly pattern mining on the continuous sign monitoring data, identifies sign fluctuation segments deviating from personal baseline patterns, and performs correlation mapping analysis on the periodic test data and the past diagnosis records to generate test index trend evolution maps, a data correlation module, which constructs a risk correlation matrix of patient behavior activities and external environment based on the environmental exposure data, the risk correlation matrix quantifying the contribution degrees of different environmental factors and behavior patterns to each sign fluctuation and test index change; a risk modeling module, which integrates the sign fluctuation segments, the test index trend evolution maps and the risk correlation matrix; a prediction and intervention generation module, which performs calculation using a preset chronic disease progression prediction model to generate individualized disease risk warning signals and phased health intervention schemes. 2.The big data-based chronic disease patient health management system according to claim 1, wherein, The time series anomaly pattern mining on the continuous sign monitoring data to identify sign fluctuation segments deviating from personal baseline patterns comprises: the sign fluctuation segments include heart rate variability abnormal interval, blood pressure circadian rhythm disorder period and blood oxygen saturation periodic decline event; a historical data time window for calculating personal baseline patterns is set, and historical monitoring sequences within the historical data time window are extracted from the continuous sign monitoring data; the average value, standard deviation and periodic component of each sign index in the historical monitoring sequences are calculated to determine the normal fluctuation range and rhythm characteristics of the sign index; the newly acquired real-time monitoring sequence is compared with the normal fluctuation range and rhythm characteristics of the sign index point by point, and data points exceeding the fluctuation range or violating the rhythm characteristics are marked; the marked abnormal data points are subjected to clustering analysis, and time-continuous or close abnormal data points of the same type are aggregated into the sign fluctuation segments; each sign fluctuation segment is given a type label and a severity score, and the type label includes heart rate variability abnormality, blood pressure circadian rhythm disorder or blood oxygen saturation periodic decline. 3.The big data-based chronic disease patient health management system according to claim 2, characterized in that, The comparison of the newly acquired real-time monitoring sequence with the normal fluctuation range and rhythm characteristics of the sign index point by point to mark data points exceeding the fluctuation range or violating the rhythm characteristics comprises: a single sign index time sequence in the real-time monitoring sequence is extracted, and a numerical comparison is performed between the single sign index time sequence and the normal fluctuation range of the corresponding sign index, and if a plurality of continuous data points exceed the upper limit or lower limit of the normal fluctuation range, the single sign index time sequence is marked as continuous abnormality; the change pattern of the single sign index time sequence within twenty-four hours is analyzed, and a shape similarity calculation is performed between the change pattern and a typical circadian change curve defined in the rhythm characteristics, and if the similarity is lower than a preset threshold, the single sign index time sequence is marked as rhythm abnormality. Detecting whether there is a sharp rise or fall in the short term in the single sign time series, by calculating the difference between adjacent data points and comparing it with the historical average change rate, if it exceeds several times of the historical average change rate, it is marked as mutation anomaly; Recording the data points that meet any of the conditions of continuous anomaly, rhythm anomaly or mutation anomaly as abnormal data points, and recording the specific anomaly type. 4.The big data-based chronic disease patient health management system according to claim 1, wherein, The association mapping analysis of the periodic test data and the past diagnosis records generates a test index trend evolution map, which includes: The test index trend evolution map reflects the change trajectory of the specific test item value over time and its corresponding relationship with the historical diagnosis events; Extracting multiple test results of the same test item from the periodic test data, sorting them by test time to form a numerical sequence of a specific item; Extracting the time points of diagnosis events and diagnosis conclusions from the past diagnosis records, the diagnosis conclusions include disease staging or severity level; Aligning the numerical sequence of the specific item with the time points of the diagnosis events on the time axis to establish the time correspondence between the test values and the diagnosis events; Analyzing the change trend, change rate and peak or valley value of the numerical sequence of the specific item before and after each diagnosis event; Based on the time correspondence and change characteristics, a graphical map is generated with time as the horizontal axis, test item value as the vertical axis, and diagnosis event points labeled, which is the test index trend evolution map. 5.The big data-based chronic disease patient health management system according to claim 4, wherein, The analysis of the change trend, change rate and peak or valley value of the numerical sequence of the specific item before and after each diagnosis event includes: Taking the diagnosis event time as the dividing point, selecting a predetermined length of time window forward and backward respectively, calculating the average value of the numerical sequence of the specific item in the window as the pre-diagnosis baseline value and the post-diagnosis baseline value; Calculating the difference between the post-diagnosis baseline value and the pre-diagnosis baseline value as the net change of the influence of the diagnosis event on the test index; In the pre-diagnosis time window, the linear regression slope of the numerical sequence is calculated as the pre-event trend slope; in the post-diagnosis time window, the linear regression slope of the numerical sequence is calculated as the post-event trend slope; Identifying the global maximum and minimum values of the entire numerical sequence in the entire time range, respectively recording as the peak value and the valley value; Recording the average daily change from the pre-diagnosis baseline value to the post-diagnosis baseline value as the average daily change rate. 6.The big data-based chronic disease patient health management system according to claim 1, wherein, The construction of the risk association matrix of patient behavior activities and external environment based on the environmental exposure data includes: Parsing multiple environmental exposure factors and patient behavior activity records from the environmental exposure data, the environmental exposure factors include air quality index, temperature and humidity level and noise decibel level, the patient behavior activity records include physical activity intensity, sleep duration and diet content; For each chronic disease related health outcome, define a set of related physical indicators and test indicators as target variables for risk association; Adopting multivariate correlation analysis and regression analysis method, the statistical correlation strength coefficient of each environmental exposure factor and each patient behavior activity record to each target variable is calculated respectively; All the statistical correlation strength coefficients calculated are organized into a multi-dimensional numerical table according to environmental exposure factors, behavior activity records and target variables, that is, the risk correlation matrix; The coefficients in the risk correlation matrix are standardized to make the coefficients between different target variables comparable. 7.The big data-based chronic disease patient health management system according to claim 6, wherein, The statistical correlation strength coefficient of each environmental exposure factor and each patient behavior activity record to each target variable is calculated respectively by adopting multivariate correlation analysis and regression analysis method, including: For each specific pair of environmental exposure factor and target variable, the historical data pairs of environmental exposure factor and target variable at the same time point are collected; The correlation coefficient between the historical data sequence of the environmental exposure factor and the historical data sequence of the target variable is calculated; A linear regression model is established with the environmental exposure factor as the independent variable and the target variable as the dependent variable, and the regression coefficient is fitted; The correlation coefficient and the regression coefficient are comprehensively weighted to obtain the statistical correlation strength coefficient of the environmental exposure factor to the target variable; For each combination of patient behavior activity record and target variable, the steps from data pair collection to comprehensive weighting are repeated. 8.The big data-based chronic disease patient health management system of claim 1, wherein, The integration of the sign fluctuation segment, the test index trend evolution graph and the risk correlation matrix is calculated by using a preset chronic disease progression prediction model to generate an individualized disease risk warning signal, including: All the sign fluctuation segments identified in the current monitoring period are converted into input feature vectors of the prediction model according to their types and severity scores; The recent trend slope and the offset of the current value relative to the historical baseline are extracted from the latest test index trend evolution graph as another group of input features; According to the environmental exposure data and behavior activity record recorded by the patient recently, combined with the risk correlation matrix, the comprehensive risk exposure score under the current environmental and behavior combination is calculated as the third group of input features; The input feature vectors, recent trend slope and offset and comprehensive risk exposure score are input into the chronic disease progression prediction model together; The chronic disease progression prediction model outputs the probability value of occurrence of a specific acute event or disease staging progression of a specific chronic disease within a specified period of time in the future. When the probability value exceeds a preset warning threshold, the corresponding disease risk warning signal is generated. The signal contains the risk event type and the predicted occurrence time window. 9.The big data-based chronic disease patient health management system according to claim 8, wherein, Generating the phased health intervention scheme includes: The health intervention scheme includes medication adjustment suggestions, behavior pattern optimization lists and reconsultation plans; According to the risk event type indicated in the disease risk warning signal, the corresponding intervention measure set is matched from the preset intervention knowledge base; Combined with the patient's current long-term medication record, the medication adjustment suggestions in the intervention measure set are checked for compatibility to screen out medication adjustment suggestions without conflicts; According to the personal work and rest habits of the patient and the behavior activity record, specific execution time, frequency and intensity of the behavior intervention suggestions in the intervention measure set are set to form a behavior pattern optimization list; According to the key indicator monitoring requirements indicated in the test index trend evolution map and the urgency of the disease risk early warning signal, a re-visit plan including next test items and recommended re-visit time is developed; The medication adjustment suggestions, behavior pattern optimization list and re-visit plan are integrated into a phased health intervention scheme document with a clear execution cycle. 10.The big data-based chronic patient health management system of claim 1, wherein, The steps for constructing the chronic disease progression prediction model include: Collecting complete health monitoring data sets of historical chronic disease patient groups, the health monitoring data sets including continuous body sign monitoring data, periodic test data, environmental exposure data and corresponding disease progression event records; Performing time series segmentation processing on the continuous body sign monitoring data to extract body sign fluctuation pattern feature vectors of each patient in different time periods; Performing trend feature extraction processing on the periodic test data to generate test index change trajectory descriptors of each patient; Performing feature encoding processing on the environmental exposure data and patient behavior activity records to obtain environmental behavior feature matrices; Performing multi-source feature fusion processing on the body sign fluctuation pattern feature vectors, test index change trajectory descriptors and environmental behavior feature matrices to generate comprehensive feature representations; Labeling the disease progression state labels of each patient sample based on the disease progression event records; Training a time series deep learning network based on the comprehensive feature representations and disease progression state labels, optimizing network parameters through a back propagation algorithm to obtain a trained chronic disease progression prediction model.

Images