A health data abnormal distribution identification method

Abstract

The application discloses a health and hygiene data abnormal distribution identification method, and relates to the technical field of data analysis, comprising obtaining health and hygiene data, performing multi-dimensional index processing of time dimension, space dimension, data category and value on the health and hygiene data, and generating a space-time data cube, wherein the data category comprises one or more of patient total number, medical treatment person-time, medical resource consumption and medical expenses. In the application, the health and hygiene data is converted into four-dimensional structured data of the space-time data cube, which provides a basis for subsequent fine-grained analysis of specific dimensions. Meanwhile, the cube data cooperates with covariates to construct a more detailed baseline model based on autoregression, integration and moving average. On this basis, potential outliers are subjected to abnormal cluster determination based on a control chart, so that seasonal disease prevalence, temperature change, holiday travel and the like can be reduced or even excluded from the identification of natural fluctuations in prescription amount.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method for identifying abnormal distributions in health data. Background Technology

[0002] With the development of digital health and intelligent regulatory technologies, the in-depth digital transformation of the health sector has gradually become an effective path to improve the efficiency of public health governance. In-depth analysis and pattern classification of medical and health data from different regions and time periods can provide a data foundation for specific areas such as optimizing medical resource allocation and early warning of public health events.

[0003] Traditional health data monitoring systems primarily rely on periodic summary reports from medical institutions and manual statistical analysis, making it difficult to track prescription behavior in a granular and timely manner. In particular, they lack a systematic characterization of the geographical distribution characteristics of health data and its evolution over time. Against this backdrop, identifying anomalous signals deviating from normal patterns from massive, multi-dimensional health data has become a core challenge in improving public health emergency early warning capabilities and the efficiency of medical resource allocation. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for identifying abnormal distributions in health data.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for identifying abnormal distributions in health data, comprising: Acquire health data, perform multi-dimensional indexing on the health data based on time dimension, spatial dimension, data category and value, and generate a spatiotemporal data cube, where the data categories include one or more of the following: total number of patients, number of outpatient visits, medical resource consumption and medical expenses. Read the spatiotemporal data cube within the first time period, and read one or more of the following as covariates: seasonal, climate, weekday / holiday identifiers, and epidemic monitoring data. Perform dynamic baseline construction processing based on an autoregressive integral moving average model on the spatiotemporal data cube and covariates within the first time period to generate a dynamic baseline model. The dynamic baseline model outputs the expected range, upper limit of outliers, and lower limit of outliers for specified data categories by region and time period. Read the spatiotemporal data cube within the second time period, perform spatiotemporal clustering processing on the spatiotemporal data cube within the second time period to generate clustering results, and perform potential outlier identification processing on the clustering results to generate potential outliers, wherein the second time period is later than the first time period; Using the mean of the expected range as the center line, the upper limit of outliers as the upper control limit, and the lower limit of outliers as the lower control limit, control chart construction and anomaly detection are performed on potential outliers to generate anomaly clusters. Read the anomaly clusters and geographical factors, construct a regression function of the anomaly cluster values ​​with respect to the geographical factors, and perform attribution processing on the anomaly clusters based on the coefficients of the geographical factors in the regression function to generate attribution results; Read the attribution results, perform spatial unit feature correspondence processing, similarity relationship matching processing, and intervention action organization processing on the attribution results, and generate intervention strategy suggestions.

[0006] As a further description of the above technical solution: The generation of spatiotemporal data cubes includes: Obtain health service records, perform timestamp reading and time slot aggregation on the health service records, and generate time-dimensional content; Read the spatial attribution content corresponding to the health service records, and perform administrative division correspondence processing, geographic grid correspondence processing, or medical service radius correspondence processing on the spatial attribution content to generate spatial dimension content; Read the data category content corresponding to the health service records, perform category merging and category indexing on the data category content, and generate data category content; Read one or more of the following: total number of patients, number of outpatient visits, medical resource consumption, and medical expenses; perform value sorting and processing on the read results to generate value content. Perform multidimensional index binding processing on time dimension content, spatial dimension content, data category content, and value content to generate a spatiotemporal data cube.

[0007] As a further description of the above technical solution: The generation of the dynamic baseline model includes: Read the spatiotemporal data cube within the first time period, and perform master variable organization processing on the spatiotemporal data cube within the first time period by region, time period, and specified data category to generate a master variable sequence; Read one or more of the following: seasonal, climate, weekday / holiday identifiers, and epidemiological surveillance data; perform covariate processing on the reading results to generate a covariate sequence. Perform time-correspondence processing and region-correspondence processing on the main variable sequence and covariate sequence to generate dynamic baseline input data; Baseline fitting processing based on an autoregressive integral moving average model is performed on the dynamic baseline input data to generate a dynamic baseline model; The dynamic baseline model performs expected range calculation, outlier upper limit calculation, and outlier lower limit calculation on specified data categories by region and time period, generating expected range, outlier upper limit, and outlier lower limit.

[0008] As a further description of the above technical solution: The generation of clustering results includes: Read the spatiotemporal data cube within the second time period, and read the median time interval between adjacent health events within a specified spatial cell, the area of ​​the specified spatial cell, and the value of the specified spatial cell within the second time period; An adaptive function is generated by performing adaptive function construction processing on the median, area, and values ​​of the time interval; The output of the adaptive function is used as the neighborhood radius to process the spatiotemporal data cube in the second time period using the improved ST-DBSCAN clustering algorithm, generating clustering results.

[0009] As a further description of the above technical solution: The generation of potential outliers includes: Read the minimum number of cluster points in the improved ST-DBSCAN clustering algorithm and determine the value of k based on the minimum number of cluster points; Based on the k-value, the k-nearest neighbor method is applied to the clustering results to generate the k-nearest neighbor clusters for each cluster; Read each cluster and its k nearest neighbor clusters, perform local reachability density calculation on each cluster, and perform average local reachability density calculation on the k nearest neighbor clusters to generate local reachability density and average local reachability density; Calculate the ratio of local reachability density to average local reachability density. Clusters whose ratios meet the set conditions are marked to generate potential outliers.

[0010] As a further description of the above technical solution: The generation of anomalous clusters includes: Read the specified spatial granularity according to the user's instructions, perform spatial unit partitioning on the spatiotemporal data cube in the second time period, and generate multiple spatial units; For each spatial unit, read the potential outliers, expected range, upper outlier limit, and lower outlier limit from the clustering results; Using the mean of the expected range as the center line, the upper limit of outliers as the upper control limit, and the lower limit of outliers as the lower control limit, control chart construction is performed on potential outliers in each spatial unit to generate a spatial unit control chart. Anomaly detection and processing are performed on potential outliers based on spatial unit control charts to generate anomaly clusters.

[0011] As a further description of the above technical solution: The generation of attribution results includes: Read the values ​​of the abnormal clusters and the spatial units corresponding to the abnormal clusters, and read the geographical factors; Spatial location correspondence processing is performed on the values, spatial units, and geographical factors of the outlier clusters to generate geographic regression input data; Based on the geographic regression input data, a regression function is constructed to determine the values ​​of the anomaly clusters with respect to geographic factors, and the regression function is generated. Read the coefficients of geographical factors in the regression function, perform influencing factor identification and outlier cluster attribution processing on the coefficients, and generate attribution results.

[0012] As a further description of the above technical solution: The generation of intervention strategy recommendations includes: Read the attribution results of the spatial units corresponding to the abnormal clusters, and read one or more of the following for each spatial unit: the sliding standard deviation of the time window of the number of visits, the distribution of patient age, the distribution of patient address, and the cost concentration coefficient; High-dimensional feature calculation processing is performed on one or more of the following factors: sliding standard deviation of outpatient visits during time window, patient age distribution, patient address distribution, and cost concentration coefficient, to generate high-dimensional features of spatial units; Similarity calculation is performed on the high-dimensional features of multiple spatial units to generate similarity relationships between spatial units. When multiple spatial units whose similarity of high-dimensional features meets the set conditions form a set, the attribution results corresponding to each spatial unit in the set are read, and the common influencing factors among multiple spatial units are determined based on the coefficients of geographical factors in the attribution results, generating intervention strategy suggestions for cross-regional collaborative anomaly patterns.

[0013] As a further description of the above technical solution: The generation of intervention strategy recommendations also includes: Read the attribution results of the spatial units corresponding to the anomaly clusters, and read the high-dimensional features of the spatial units and the high-dimensional features of the historical spatial units; Perform similarity calculation on the high-dimensional features of spatial units and the high-dimensional features of historical spatial units to generate historical similarity relationships; When the similarity between the high-dimensional features of a spatial unit and the high-dimensional features of any historical spatial unit meets the set conditions, the effective intervention actions of the historical spatial unit are read. Based on the coefficients of geographical factors in the attribution results, effective intervention actions for historical spatial units are matched and processed to generate intervention strategy recommendations.

[0014] As a further description of the above technical solution: A system for identifying abnormal distributions of health data, characterized in that it includes: The spatiotemporal data cube construction module is used to acquire health data, perform multidimensional indexing on the health data in terms of time dimension, spatial dimension, data category and value, and generate a spatiotemporal data cube, wherein the data categories include one or more of the following: total number of patients, number of outpatient visits, medical resource consumption and medical expenses. The dynamic baseline model module is used to read the spatiotemporal data cube within the first time period and read one or more of the following as covariates: seasonal, climate, weekday / holiday identifiers, and epidemic monitoring data. It performs dynamic baseline construction processing based on an autoregressive integral moving average model on the spatiotemporal data cube and covariates within the first time period to generate a dynamic baseline model. The dynamic baseline model outputs the expected range, upper limit of outliers, and lower limit of outliers for specified data categories by region and time period. The anomaly cluster module is used to read the spatiotemporal data cube within the second time period, perform spatiotemporal clustering processing on the spatiotemporal data cube within the second time period to generate clustering results, and perform potential outlier identification processing on the clustering results to generate potential outliers, wherein the second time period is later than the first time period; The anomaly cluster module is also used to perform control chart construction and anomaly detection processing on potential outliers, with the mean of the expected range as the center line, the upper limit of outliers as the upper control limit, and the lower limit of outliers as the lower control limit, to generate anomaly clusters. The geographic regression intervention module is used to read outlier clusters and geographic factors, construct a regression function of the values ​​of outlier clusters with respect to geographic factors, and perform attribution processing on outlier clusters based on the coefficients of geographic factors in the regression function to generate attribution results. The geographic regression intervention module is also used to read attribution results, perform spatial unit feature correspondence processing, similarity relationship matching processing, and intervention action organization processing on the attribution results, and generate intervention strategy suggestions.

[0015] The present invention has the following beneficial effects: In this invention, health data is converted into four-dimensional structured data in a spatiotemporal data cube, providing a foundation for subsequent fine-grained analysis of specific dimensions. Simultaneously, the cube data, combined with covariates, constructs a more detailed baseline model based on autoregression, integral, and moving average. Based on this, anomaly cluster identification based on control charts is performed on potential outliers, reducing or even eliminating the identification of natural fluctuations in prescription volume caused by seasonal disease outbreaks, temperature changes, and holiday travel as anomalies. Geographically weighted regression is used to attribute the anomaly clusters, providing a more reliable data foundation for generating intervention strategy recommendations. Attached Figure Description

[0016] Figure 1 This is a diagram illustrating the method steps of the present invention; Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Reference Figure 1-2 The present invention provides an embodiment of a method for identifying abnormal distributions in health data, comprising: Acquire health data, perform multi-dimensional indexing on the health data based on time dimension, spatial dimension, data category and value, and generate a spatiotemporal data cube, where the data categories include one or more of the following: total number of patients, number of outpatient visits, medical resource consumption and medical expenses. Read the spatiotemporal data cube within the first time period, and read one or more of the following as covariates: seasonal, climate, weekday / holiday identifiers, and epidemic monitoring data. Perform dynamic baseline construction processing based on an autoregressive integral moving average model on the spatiotemporal data cube and covariates within the first time period to generate a dynamic baseline model. The dynamic baseline model outputs the expected range, upper limit of outliers, and lower limit of outliers for specified data categories by region and time period. Read the spatiotemporal data cube within the second time period, perform spatiotemporal clustering processing on the spatiotemporal data cube within the second time period to generate clustering results, and perform potential outlier identification processing on the clustering results to generate potential outliers, wherein the second time period is later than the first time period; Using the mean of the expected range as the center line, the upper limit of outliers as the upper control limit, and the lower limit of outliers as the lower control limit, control chart construction and anomaly detection are performed on potential outliers to generate anomaly clusters. Read the anomaly clusters and geographical factors, construct a regression function of the anomaly cluster values ​​with respect to the geographical factors, and perform attribution processing on the anomaly clusters based on the coefficients of the geographical factors in the regression function to generate attribution results; Read the attribution results, perform spatial unit feature correspondence processing, similarity relationship matching processing, and intervention action organization processing on the attribution results, and generate intervention strategy suggestions.

[0019] This embodiment provides a method for identifying abnormal distributions in health data, which is used to perform spatiotemporal organization, dynamic baseline modeling, cluster identification, anomaly determination, geographic attribution, and intervention strategy generation on health data, so that the process of identifying abnormal distributions in health data forms a continuous processing chain from raw data to intervention strategy recommendations.

[0020] After acquiring health data, the time dimension, spatial dimension, data category, and corresponding value content are first read. Then, the time dimension content is written to the time index position, the spatial dimension content is written to the spatial index position, the data category content is written to the category index position, and the value content is written to the value index position. After writing to each index position, multi-dimensional binding is performed according to the correspondence between the time dimension, spatial dimension, data category, and value to generate a spatiotemporal data cube. The data categories include one or more of the following: total number of patients, number of outpatient visits, medical resource consumption, and medical expenses.

[0021] After generating the spatiotemporal data cube, the spatiotemporal data cube within the first time period is read, and one or more of the following are read as covariates: seasonality, climate, weekday / holiday identification, and epidemiological monitoring data. During the reading process, the spatiotemporal data cube within the first time period is used as the source of the main variables, and the covariates are matched with the spatiotemporal data cube within the first time period according to the same region and the same time period. After the matching is completed, dynamic baseline construction processing based on the autoregressive integral moving average model is performed on the spatiotemporal data cube and covariates within the first time period to generate a dynamic baseline model. The dynamic baseline model outputs and organizes the specified data categories by region and time period to generate the expected range, upper limit of outliers, and lower limit of outliers.

[0022] After the dynamic baseline model is generated, the spatiotemporal data cubes in the second time period are read, while maintaining the temporal order of the second time period being later than the first time period. Spatiotemporal clustering is performed on the spatiotemporal data cubes in the second time period, so that the values ​​under the same or similar time dimensions, spatial dimensions and data categories in the second time period are included in the same clustering process to generate clustering results. Potential outlier identification processing is then performed on the clustering results to generate potential outliers, so that the spatiotemporal distribution changes in the second time period can be subsequently correlated with the output results of the dynamic baseline model formed in the first time period.

[0023] After generating potential outliers, the expected range, upper and lower outlier limits are read, and the mean of the expected range is used as the center line, the upper outlier limit as the upper control limit, and the lower outlier limit as the lower control limit. Potential outliers are written to the control chart construction location and arranged according to their corresponding time dimension, spatial dimension, data category, and value. After the control chart is constructed, anomaly detection processing is performed on the potential outliers, and potential outliers that meet the anomaly detection relationship are organized into anomaly clusters.

[0024] After the anomaly clusters are generated, the anomaly clusters and geographical factors are read, and the values ​​of the anomaly clusters and geographical factors are spatially mapped according to the spatial units corresponding to the anomaly clusters. After the spatial mapping is completed, a regression function of the values ​​of the anomaly clusters with respect to the geographical factors is constructed. The coefficients of the geographical factors in the regression function are read, and the anomaly clusters are identified and attributed based on the coefficients of the geographical factors to generate attribution results.

[0025] After the attribution results are generated, the attribution results are read and the corresponding content of the spatial units is read; spatial unit feature correspondence processing is performed on the attribution results to form a stable correspondence between the attribution results and the corresponding spatial units; similarity relationship matching processing is performed on the feature relationships between spatial units, and intervention action sorting processing is performed on the historical effective intervention actions or cross-regional collaborative abnormal patterns formed by the matching to generate intervention strategy suggestions.

[0026] Through the above steps, health data, spatiotemporal data cubes, dynamic baseline models, clustering results, potential outliers, abnormal clusters, attribution results, and intervention strategy recommendations are sequentially integrated to form a closed data processing chain.

[0027] This embodiment is used to illustrate the process of generating a spatiotemporal data cube.

[0028] After obtaining the health service records, the timestamp content in the health service records is read first, and then the records are aggregated according to the time slots in which the timestamp content is located. During the aggregation process, health service records belonging to the same time slot are written to the same time dimension position, and health service records belonging to different time slots are written to consecutive time dimension positions to generate time dimension content.

[0029] After generating the time dimension content, the spatial attribution content corresponding to the health service record is read, and administrative division correspondence processing, geographic grid correspondence processing, or medical service radius correspondence processing are performed according to the spatial attribution content. During the correspondence processing, the spatial attribution content is written to the corresponding spatial index position, and the correspondence between the spatial index position and the health service record is maintained to generate the spatial dimension content.

[0030] After generating spatial dimension content, the data category content corresponding to the health service records is read, and category merging processing is performed on the data category content. During the category merging process, one or more of the content belonging to the total number of patients, number of visits, medical resource consumption, and medical expenses are written into the corresponding category position. After the category merging is completed, category indexing processing is performed on the category position to generate data category content.

[0031] After generating the data category content, read one or more of the following: total number of patients, number of outpatient visits, medical resource consumption, and medical expenses, and perform value sorting and processing on the read results; during the value sorting and processing, align each value with the corresponding time dimension content, spatial dimension content, and data category content to generate value content.

[0032] After generating time-dimensional content, spatial-dimensional content, data category content, and value content, the four types of content are bound together using a multidimensional index based on the same source of health service records. During the binding process, the time-dimensional content is used to determine the data's location in time, the spatial-dimensional content is used to determine the data's location in space, the data category content is used to determine the data's category, and the value content is used to determine the data's measurement result. After the binding is completed, a spatiotemporal data cube is generated.

[0033] This embodiment is used to illustrate the process of generating a dynamic baseline model.

[0034] After reading the spatiotemporal data cube within the first time period, the corresponding values ​​in the spatiotemporal data cube within the first time period are read according to region, time period, and specified data category. Then, the read values ​​are arranged continuously according to region order, time period order, and specified data category order to generate the master variable sequence.

[0035] After generating the main variable sequence, one or more of the following data are read: seasonal, climate, weekday / holiday identifiers, and epidemiological monitoring data. The results are then processed to organize the covariates. During the covariate organization process, each covariate is arranged according to the same region and time period as the main variable sequence, so that the covariates can form a one-to-one correspondence with the values ​​in the main variable sequence, thus generating the covariate sequence.

[0036] After generating the main variable sequence and covariate sequence, time correspondence processing and region correspondence processing are performed on the main variable sequence and covariate sequence. In the time correspondence processing, the time period content in the main variable sequence is aligned with the time period content in the covariate sequence. In the region correspondence processing, the region content in the main variable sequence is aligned with the region content in the covariate sequence. After the time correspondence and region correspondence are completed, dynamic baseline input data is generated.

[0037] After generating dynamic baseline input data, baseline fitting processing based on an autoregressive integral moving average model is performed on the dynamic baseline input data. During the baseline fitting process, the main variable sequence is used to characterize the regional and time-segmented changes in health data within the first time period, and the covariate sequence is used to supplement the influence of seasonality, climate, weekday / holiday identification, and epidemiological monitoring data on the changes. After the baseline fitting is completed, a dynamic baseline model is generated.

[0038] After generating the dynamic baseline model, the expected range calculation, outlier upper limit calculation, and outlier lower limit calculation are performed on the specified data categories by region and time period using the dynamic baseline model. During the calculation process, the central variation interval output by the dynamic baseline model is organized into the expected range, the outlier boundary above the expected range is organized into the outlier upper limit, and the outlier boundary below the expected range is organized into the outlier lower limit. After all three types of results are generated, the dynamic baseline output result is formed for subsequent outlier cluster generation.

[0039] This embodiment is used to illustrate the process of generating clustering results.

[0040] After reading the spatiotemporal data cube within the second time period, first read the median time interval of adjacent health events within the specified spatial cell, the area of ​​the specified spatial cell, and the value of the specified spatial cell within the second time period; during the reading process, bind the median time interval, area, and value to the same specified spatial cell, so that the three types of content enter the same adaptive function construction process.

[0041] After the reading is completed, an adaptive function is constructed for the median, area, and value of the time interval. During the construction process, the median of the time interval is used to represent the temporal concentration relationship of adjacent health events within a specified spatial unit, the area is used to represent the spatial range of the specified spatial unit, and the value is used to represent the data intensity of the specified spatial unit in the second time period. After the median, area, and value of the time interval are sorted accordingly, the adaptive function is generated.

[0042] After generating the adaptive function, its output is used as the neighborhood radius. Based on the neighborhood radius, an improved ST-DBSCAN clustering algorithm is applied to the spatiotemporal data cubes in the second time period. When calculating the distance between two spatiotemporal points, a weighted Euclidean distance is used, the mathematical expression of which is as follows:

[0043] in, , For the first , A spatial coordinate, , For the first , A timestamp, , These are the spatial and temporal weighting coefficients, respectively. By dynamically adjusting the weights, the model can flexibly handle different types of anomaly patterns, such as "spatial concentration and temporal dispersion" or "temporal concentration and spatial dispersion." Based on the above spatiotemporal weighted distance calculation results, the distance between each spatiotemporal point and its adjacent spatiotemporal points is compared with the neighborhood radius, and clustering results are generated based on the comparison results. During the clustering process, the spatiotemporal data cubes in the second time period enter the neighborhood judgment process according to the time dimension, spatial dimension, data category, and value, and the data content that meets the spatiotemporal proximity relationship is assigned to the corresponding cluster according to the neighborhood radius; after the clustering process is completed, clustering results are generated.

[0044] This embodiment is used to illustrate the process of generating potential outliers.

[0045] After reading the minimum cluster point count value in the improved ST-DBSCAN clustering algorithm, the k value is determined based on the minimum cluster point count value. In the process of determining the k value, the minimum cluster point count value is used as the source of the number of nearest neighbors required for the k-nearest neighbor method, so that the clustering results and the subsequent k-nearest neighbor method processing form a parameter connection relationship.

[0046] After determining the value of k, the k-nearest neighbor method is applied to the clustering results based on the value of k. During the processing, each cluster in the clustering results is read, and the k-nearest neighbor clusters are determined for each cluster according to the proximity relationship between the clusters. After the nearest neighbor correspondence is completed for each cluster, the k-nearest neighbor clusters of each cluster are generated.

[0047] After generating the k nearest neighbor clusters for each cluster, each cluster and its k nearest neighbor clusters are read; local reachability density calculation is performed on each cluster to generate local reachability density; average local reachability density calculation is performed on the k nearest neighbor clusters to generate average local reachability density; during the calculation process, the local reachability density and average local reachability density maintain the correspondence with the same cluster.

[0048] After generating the local reachability density and the average local reachability density, the ratio of the local reachability density to the average local reachability density is calculated; clusters whose ratios meet the set conditions are marked and sorted into potential outliers; after sorting, potential outliers are generated and passed to the anomalous cluster generation process.

[0049] This embodiment is used to illustrate the process of generating abnormal clusters.

[0050] After reading the specified spatial granularity according to the user's instructions, the spatiotemporal data cube in the second time period is divided into spatial units. During the spatial unit division process, the spatiotemporal data cube in the second time period is written into multiple spatial unit positions according to the specified spatial granularity, and the binding relationship between each spatial unit and the corresponding time dimension, data category and value is maintained, thereby generating multiple spatial units.

[0051] After generating multiple spatial units, the potential outliers, expected range, upper outlier limit, and lower outlier limit in the clustering results are read for each spatial unit. During the reading process, the potential outliers are matched with the corresponding expected range, upper outlier limit, and lower outlier limit within the same spatial unit, so that each spatial unit has input content that can be used for control chart construction.

[0052] After the reading is completed, the mean of the expected range is used as the center line, the upper limit of outliers is used as the upper control limit, and the lower limit of outliers is used as the lower control limit. Control chart construction is performed on the potential outliers in each spatial unit. During the construction process, the potential outliers are arranged in the control chart according to the time dimension and the order of values. The center line, upper control limit and lower control limit are written into the same control chart judgment position to generate the spatial unit control chart.

[0053] After generating the spatial unit control chart, anomaly detection processing is performed on potential outliers based on the spatial unit control chart. In the anomaly detection processing, the values ​​of potential outliers are compared with the center line, upper control limit, and lower control limit, and potential outliers that meet the anomaly detection relationship are merged into anomaly clusters. After merging, anomaly clusters are generated.

[0054] This embodiment is used to illustrate the process of generating attribution results.

[0055] After reading the values ​​of the outlier clusters and the spatial units corresponding to the outlier clusters, the geographical factors are then read. During the reading process, the values ​​of the outlier clusters are used as the explanatory content of the regression function, the geographical factors are used as the explanatory content of the regression function, and the spatial units corresponding to the outlier clusters are used as the basis for the spatial correspondence between the two.

[0056] After the reading is completed, spatial location correspondence processing is performed on the values ​​of the anomaly clusters, spatial units, and geographic factors. During the processing, the values ​​of the anomaly clusters are written to the value positions of the corresponding spatial units, and the geographic factors are written to the factor positions of the same spatial unit, while maintaining a one-to-one correspondence between the value positions and the factor positions, thus generating geographic regression input data.

[0057] After generating the geographic regression input data, a regression function is constructed based on the geographic regression input data to determine the values ​​of outlier clusters with respect to geographic factors; the regression function is as follows:

[0058] in, For the first Health data values ​​for each spatial unit. For the first The geographic center coordinates of each spatial unit It is the first One geographical factor variable, It is the first that changes with spatial location Regression coefficient function of geographical factor variables, yes The base value, This is the part that cannot be taken into account by geographical factors, ideally it is zero (or, in other words, The first priority goal of regression is to If the constraint is zero, and the regression result is unsatisfactory, an interpretable constraint is introduced. Perform parameter tuning and regression.

[0059] During the construction process, the values ​​of the anomaly clusters are used to form the target input of the regression function, the geographical factors are used to form the factor input of the regression function, and the spatial units are used to maintain the spatial correspondence between the target input and the factor input; after the regression function is constructed, the regression function is generated.

[0060] After generating the regression function, the coefficients of the geographical factors in the regression function are read; the influencing factor identification processing is performed on the coefficients to obtain the influence relationship of geographical factors on the values ​​of outlier clusters; the outlier cluster attribution processing is then performed on the influence relationship to establish an attribution correspondence between outlier clusters and corresponding geographical factors, and attribution results are generated.

[0061] This implementation method is used to illustrate the process of generating intervention strategy recommendations for cross-regional collaborative abnormal patterns.

[0062] After reading the attribution results of the spatial units corresponding to the abnormal clusters, read one or more of the following for each spatial unit: the sliding standard deviation of the time window for the number of visits, the distribution of patient age, the distribution of patient address, and the cost concentration coefficient; during the reading process, bind each read content to the same spatial unit so that the attribution results correspond to the characteristic sources of the spatial unit.

[0063] After the reading is completed, high-dimensional feature calculation processing is performed on one or more of the following: the sliding standard deviation of the time window for the number of visits, the distribution of patient age, the distribution of patient address, and the cost concentration coefficient. During the calculation process, the various contents corresponding to the same spatial unit are uniformly organized, and the organized results are written into the high-dimensional feature position of the spatial unit to generate the high-dimensional feature of the spatial unit.

[0064] After generating the high-dimensional features of spatial units, similarity calculation is performed on the high-dimensional features of multiple spatial units. During the calculation, the high-dimensional features of different spatial units are compared in the same feature order, and the similarity relationship between spatial units is formed based on the comparison results, thus generating spatial unit similarity relationship.

[0065] After generating spatial unit similarity relationships, when multiple spatial units whose high-dimensional feature similarity meets the set conditions form a set, the attribution results corresponding to each spatial unit in the set are read; the common influencing factors among multiple spatial units are determined based on the coefficients of geographical factors in the attribution results; after the common influencing factors are determined, intervention actions are performed on the common influencing factors and spatial unit similarity relationships to generate intervention strategy suggestions for cross-regional collaborative anomaly patterns.

[0066] This embodiment is used to illustrate the process of generating intervention strategy suggestions based on effective intervention actions of historical spatial units.

[0067] After reading the attribution results of the spatial units corresponding to the anomaly clusters, the high-dimensional features of the spatial units and the high-dimensional features of the historical spatial units are read. During the reading process, the high-dimensional features of the current spatial unit and the high-dimensional features of the historical spatial units are arranged in the same feature order so that they can enter the same similarity calculation process.

[0068] After the reading is completed, similarity calculation is performed on the high-dimensional features of the spatial unit and the high-dimensional features of the historical spatial units. During the calculation, the high-dimensional features of the current spatial unit are matched with the high-dimensional features of each historical spatial unit, and historical similarity relationships are generated based on the comparison results.

[0069] After generating historical similarity relationships, when the similarity between the high-dimensional features of a spatial unit and the high-dimensional features of any historical spatial unit meets the set conditions, the effective intervention actions of the historical spatial unit are read. During the reading process, the effective intervention actions are bound to the corresponding historical spatial unit and its high-dimensional features, so that the effective intervention actions can be subsequently matched with the attribution results of the current spatial unit.

[0070] After reading the effective intervention actions of historical spatial units, matching processing is performed on the effective intervention actions of historical spatial units according to the coefficients of geographical factors in the attribution results; during the matching process, the attribution results of the current spatial unit and the effective intervention actions of historical spatial units are organized according to the correspondence of geographical factor coefficients to generate intervention strategy suggestions.

[0071] This embodiment provides a device for identifying abnormal distributions of health data. The device includes a spatiotemporal data cube construction module, a dynamic baseline model module, an anomaly cluster module, and a geographic regression intervention module.

[0072] After acquiring health data, the spatiotemporal data cube construction module performs multidimensional indexing on the health data, including time dimension, spatial dimension, data category, and value. During the processing, the time dimension, spatial dimension, data category, and value of the health data are written into the corresponding index positions, and a spatiotemporal data cube is generated according to the correspondence between the index positions. The data categories include one or more of the following: total number of patients, number of outpatient visits, medical resource consumption, and medical expenses.

[0073] The dynamic baseline model module reads the spatiotemporal data cube within the first time period and reads one or more of the following as covariates: seasonality, climate, weekday / holiday identification, and epidemic monitoring data. After reading, it performs dynamic baseline construction processing based on an autoregressive integral moving average model on the spatiotemporal data cube and covariates within the first time period to generate a dynamic baseline model. The dynamic baseline model outputs the expected range, upper limit of outliers, and lower limit of outliers for specified data categories by region and time period.

[0074] The anomaly cluster module reads the spatiotemporal data cube within the second time period, performs spatiotemporal clustering processing on the spatiotemporal data cube within the second time period, and generates clustering results. After the clustering results are generated, the anomaly cluster module performs potential outlier identification processing on the clustering results, generating potential outliers, where the second time period is later than the first time period. The anomaly cluster module continues to use the mean of the expected range as the center line, the upper limit of outliers as the upper control limit, and the lower limit of outliers as the lower control limit, and performs control chart construction processing and anomaly judgment processing on the potential outliers to generate anomaly clusters.

[0075] The geographic regression intervention module reads outlier clusters and geographic factors, and constructs a regression function for the values ​​of outlier clusters with respect to geographic factors. After the regression function is generated, the geographic regression intervention module reads the coefficients of geographic factors in the regression function and performs attribution processing on the outlier clusters based on the coefficients to generate attribution results. The geographic regression intervention module continues to read the attribution results and performs spatial unit feature correspondence processing, similarity relationship matching processing, and intervention action organization processing on the attribution results to generate intervention strategy suggestions.

[0076] This embodiment uses the identification of abnormal distributions in public health data as an application scenario to illustrate the complete execution process of the method for identifying abnormal distributions in health data. In this embodiment, the health data originates from health service records generated by medical institutions, disease control and prevention centers, and community health centers. After entering the spatiotemporal data cube generation process, the health service records are organized into multidimensional correspondences between time dimensions, spatial dimensions, data categories, and values.

[0077] In this embodiment, health data is first acquired, and then multidimensional indexing processing is performed on the health data based on time dimension, spatial dimension, data category, and value. The time dimension is organized according to the timestamp content in the health service records, the spatial dimension is organized according to the spatial affiliation content corresponding to the health service records, the data categories are organized according to one or more of the following: total number of patients, number of visits, medical resource consumption, and medical expenses, and the values ​​are organized according to the statistical results of each data category under the corresponding time and spatial dimensions. After the multidimensional index binding is completed, a spatiotemporal data cube is generated.

[0078] After generating the spatiotemporal data cube, the spatiotemporal data cube within the first time period is read as the source of the main variable, and one or more of the following are read as covariates: seasonality, climate, weekday / holiday identification, and epidemic monitoring data. After reading, the spatiotemporal data cube and covariates within the first time period are organized according to the same region and the same time period to generate dynamic baseline input data. Dynamic baseline construction processing based on the autoregressive integral moving average model is performed on the dynamic baseline input data to generate a dynamic baseline model. The dynamic baseline model outputs the expected range, upper limit of outliers, and lower limit of outliers for specified data categories by region and time period.

[0079] After the dynamic baseline model is output, the spatiotemporal data cube for the second time period is read. For the spatiotemporal data cube for the second time period, the median time interval of adjacent health events within a specified spatial cell, the area of ​​the specified spatial cell, and the value of the specified spatial cell within the second time period are read. An adaptive function is constructed by performing adaptive function construction on the median time interval, area, and value. The output of the adaptive function is used as the neighborhood radius, and the improved ST-DBSCAN clustering algorithm is performed on the spatiotemporal data cube for the second time period to generate clustering results.

[0080] After the clustering results are generated, the minimum number of cluster points in the improved ST-DBSCAN clustering algorithm is read, and the k value is determined based on the minimum number of cluster points. The k-nearest neighbor method is applied to the clustering results based on the k value to generate the k-nearest neighbor clusters for each cluster. Each cluster and its k-nearest neighbor clusters are read, and the local reachability density is calculated for each cluster. The average local reachability density is calculated for the k-nearest neighbor clusters to generate the local reachability density and the average local reachability density. The ratio of the local reachability density to the average local reachability density is calculated. Clusters whose ratios meet the set conditions are marked to generate potential outliers.

[0081] After generating potential outliers, the system reads the specified spatial granularity according to user instructions and performs spatial unit partitioning on the spatiotemporal data cube within the second time period to generate multiple spatial units. For each spatial unit, the system reads the potential outliers, expected range, upper outlier limit, and lower outlier limit from the clustering results. Using the mean of the expected range as the center line, the upper outlier limit as the upper control limit, and the lower outlier limit as the lower control limit, the system performs control chart construction on the potential outliers within each spatial unit to generate a spatial unit control chart. Based on the spatial unit control chart, the system performs anomaly detection processing on the potential outliers to generate anomaly clusters.

[0082] After the abnormal clusters are generated, the values ​​of the abnormal clusters and the spatial units corresponding to the abnormal clusters are read, and the geographical factors are read; spatial location correspondence processing is performed on the values ​​of the abnormal clusters, spatial units and geographical factors to generate geographical regression input data; a regression function of the values ​​of the abnormal clusters with respect to geographical factors is constructed based on the geographical regression input data to generate the regression function; the coefficients of geographical factors in the regression function are read, and influencing factor identification processing and abnormal cluster attribution processing are performed on the coefficients to generate attribution results.

[0083] After the attribution results are generated, the attribution results of the spatial units corresponding to the anomaly clusters are read, and one or more of the following are read for each spatial unit: the sliding standard deviation of the time window for the number of visits, the patient age distribution, the patient address distribution, and the cost concentration coefficient. High-dimensional feature calculation processing is performed on the read content to generate high-dimensional features of the spatial units. Similarity calculation processing is performed on the high-dimensional features of multiple spatial units to generate spatial unit similarity relationships. When multiple spatial units whose high-dimensional feature similarity meets the set conditions form a set, the attribution results corresponding to each spatial unit in the set are read, and the common influencing factors among multiple spatial units are determined based on the coefficients of geographical factors in the attribution results, generating intervention strategy suggestions for cross-regional collaborative anomaly patterns.

[0084] In this embodiment, health data is generated through spatiotemporal data cube generation, dynamic baseline model construction, clustering result generation, potential outlier generation, abnormal cluster generation, attribution result generation, and intervention strategy recommendation generation. The input and output data of each stage are sequentially connected, which can support the identification, attribution, and collaborative intervention recommendation generation of abnormal distributions in public health.

[0085] This embodiment uses the identification of abnormal medical resource consumption and medical expenses as an application scenario to illustrate another implementation process of the method for identifying abnormal distributions in health data. In this embodiment, the data categories in the health data include one or more of medical resource consumption and medical expenses, and the values ​​are used to characterize the data change results within the corresponding spatial unit and the corresponding time period.

[0086] In this embodiment, health data is first acquired, and the time dimension, spatial dimension, data category, and values ​​are read from the health data. The time dimension is processed by timestamp reading and time slot aggregation to generate time dimension content. The spatial dimension is processed by administrative division correspondence, geographic grid correspondence, or medical service radius correspondence to generate spatial dimension content. The data categories are processed by category merging and category indexing to generate data category content. The values ​​are processed by sorting to generate value content. Finally, the time dimension content, spatial dimension content, data category content, and value content are bound by a multidimensional index to generate a spatiotemporal data cube.

[0087] After the spatiotemporal data cube is generated, the spatiotemporal data cube within the first time period is read. The main variables within the first time period are processed by region, time period, and specified data category to generate a main variable sequence. One or more of the following data are read: seasonal, climate, weekday / holiday identifiers, and epidemiological monitoring data. The covariate processing is performed on the read results to generate a covariate sequence. Time-correlation and region-correlation processing is performed on the main variable sequence and covariate sequence to generate dynamic baseline input data. Baseline fitting processing based on an autoregressive integral moving average model is performed on the dynamic baseline input data to generate a dynamic baseline model. The expected range, upper outlier limit, and lower outlier limit are generated using the dynamic baseline model.

[0088] After the dynamic baseline model is generated, the spatiotemporal data cube within the second time period is read, and the median time interval of adjacent health events within a specified spatial cell, the area of ​​the specified spatial cell, and the value of the specified spatial cell within the second time period are read. An adaptive function is constructed by performing adaptive function construction on the median time interval, area, and value. The output of the adaptive function is used as the neighborhood radius, and the improved ST-DBSCAN clustering algorithm is performed on the spatiotemporal data cube within the second time period to generate clustering results.

[0089] After the clustering results are generated, the minimum number of cluster points in the improved ST-DBSCAN clustering algorithm is read, and the k value is determined based on the minimum number of cluster points. The k-nearest neighbor method is applied to the clustering results based on the k value to generate the k-nearest neighbor clusters for each cluster. Each cluster and its k-nearest neighbor clusters are read, and the local reachability density is calculated for each cluster. The average local reachability density is calculated for the k-nearest neighbor clusters to generate the local reachability density and the average local reachability density. The ratio of the local reachability density to the average local reachability density is calculated. Clusters whose ratios meet the set conditions are marked to generate potential outliers.

[0090] After generating potential outliers, the system reads the specified spatial granularity according to user instructions and performs spatial unit partitioning on the spatiotemporal data cube within the second time period to generate multiple spatial units. For each spatial unit, the system reads the potential outliers, expected range, upper outlier limit, and lower outlier limit from the clustering results. Using the mean of the expected range as the center line, the upper outlier limit as the upper control limit, and the lower outlier limit as the lower control limit, the system performs control chart construction on the potential outliers within each spatial unit to generate a spatial unit control chart. Based on the spatial unit control chart, the system performs anomaly detection processing on the potential outliers to generate anomaly clusters.

[0091] After the abnormal clusters are generated, the values ​​of the abnormal clusters and the spatial units corresponding to the abnormal clusters are read, and the geographical factors are read; spatial location correspondence processing is performed on the values ​​of the abnormal clusters, spatial units and geographical factors to generate geographical regression input data; a regression function of the values ​​of the abnormal clusters with respect to geographical factors is constructed based on the geographical regression input data to generate the regression function; the coefficients of geographical factors in the regression function are read, and influencing factor identification processing and abnormal cluster attribution processing are performed on the coefficients to generate attribution results.

[0092] After the attribution results are generated, the attribution results of the spatial units corresponding to the anomaly clusters are read, and the high-dimensional features of the spatial units and the historical spatial units are read. Similarity calculation is performed on the high-dimensional features of the spatial units and the historical spatial units to generate historical similarity relationships. When the similarity between the high-dimensional features of the spatial units and the high-dimensional features of any historical spatial units meets the set conditions, the effective intervention actions of the historical spatial units are read. Matching processing is performed on the effective intervention actions of the historical spatial units according to the coefficients of geographical factors in the attribution results to generate intervention strategy suggestions.

[0093] In this embodiment, medical resource consumption and medical expenses are included as data categories in the spatiotemporal data cube in the same dynamic baseline, cluster identification, anomaly determination and geographic attribution processing chain; by matching the attribution results with the effective intervention actions of historical spatial units, intervention strategy suggestions corresponding to the anomaly clusters can be generated.

[0094] This embodiment provides a device for identifying abnormal distributions of health data. The device includes a spatiotemporal data cube construction module, a dynamic baseline model module, an anomaly cluster module, and a geographic regression intervention module.

[0095] The spatiotemporal data cube construction module acquires health data and performs multidimensional indexing processing on the health data based on time dimension, spatial dimension, data category, and value to generate a spatiotemporal data cube. During the generation process, the spatiotemporal data cube construction module binds the time dimension, spatial dimension, data category, and value, so that one or more of the following can be processed in the corresponding time and spatial dimensions: total number of patients, number of visits, medical resource consumption, and medical expenses.

[0096] The dynamic baseline model module reads the spatiotemporal data cube within the first time period and reads one or more of the following as covariates: seasonality, climate, weekday / holiday identification, and epidemic monitoring data. The dynamic baseline model module performs dynamic baseline construction processing based on the autoregressive integral moving average model on the spatiotemporal data cube and covariates within the first time period to generate the dynamic baseline model. The dynamic baseline model outputs the expected range, upper limit of outliers, and lower limit of outliers for specified data categories by region and time period.

[0097] The anomaly cluster module reads the spatiotemporal data cube within the second time period, performs spatiotemporal clustering processing on the spatiotemporal data cube within the second time period, and generates clustering results; the anomaly cluster module performs potential outlier identification processing on the clustering results, and generates potential outliers; the anomaly cluster module uses the mean of the expected range as the center line, the upper limit of outliers as the upper control limit, and the lower limit of outliers as the lower control limit, and performs control chart construction processing and anomaly judgment processing on the potential outliers to generate anomaly clusters.

[0098] The geographic regression intervention module reads outlier clusters and geographic factors, and constructs a regression function of the values ​​of outlier clusters with respect to geographic factors. The geographic regression intervention module performs attribution processing on outlier clusters based on the coefficients of geographic factors in the regression function, and generates attribution results. The geographic regression intervention module reads the attribution results, performs spatial unit feature correspondence processing, similarity relationship matching processing, and intervention action sorting processing on the attribution results, and generates intervention strategy suggestions.

[0099] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying abnormal distributions in health data, characterized in that: include: Acquire health data, perform multi-dimensional indexing on the health data based on time dimension, spatial dimension, data category and value, and generate a spatiotemporal data cube, where the data categories include one or more of the following: total number of patients, number of outpatient visits, medical resource consumption and medical expenses. Read the spatiotemporal data cube within the first time period, and read one or more of the following as covariates: seasonal, climate, weekday / holiday identifiers, and epidemic monitoring data. Perform dynamic baseline construction processing based on an autoregressive integral moving average model on the spatiotemporal data cube and covariates within the first time period to generate a dynamic baseline model. The dynamic baseline model outputs the expected range, upper limit of outliers, and lower limit of outliers for specified data categories by region and time period. Read the spatiotemporal data cube within the second time period, perform spatiotemporal clustering processing on the spatiotemporal data cube within the second time period to generate clustering results, and perform potential outlier identification processing on the clustering results to generate potential outliers, wherein the second time period is later than the first time period; Using the mean of the expected range as the center line, the upper limit of outliers as the upper control limit, and the lower limit of outliers as the lower control limit, control chart construction and anomaly detection are performed on potential outliers to generate anomaly clusters. Read the anomaly clusters and geographical factors, construct a regression function of the anomaly cluster values ​​with respect to the geographical factors, and perform attribution processing on the anomaly clusters based on the coefficients of the geographical factors in the regression function to generate attribution results; Read the attribution results, perform spatial unit feature correspondence processing, similarity relationship matching processing, and intervention action organization processing on the attribution results, and generate intervention strategy suggestions.

2. The method for identifying abnormal distributions of health data according to claim 1, characterized in that: The generation of spatiotemporal data cubes includes: Obtain health service records, perform timestamp reading and time slot aggregation on the health service records, and generate time-dimensional content; Read the spatial attribution content corresponding to the health service records, and perform administrative division correspondence processing, geographic grid correspondence processing, or medical service radius correspondence processing on the spatial attribution content to generate spatial dimension content; Read the data category content corresponding to the health service records, perform category merging and category indexing on the data category content, and generate data category content; Read one or more of the following: total number of patients, number of outpatient visits, medical resource consumption, and medical expenses; perform value sorting and processing on the read results to generate value content. Perform multidimensional index binding processing on time dimension content, spatial dimension content, data category content, and value content to generate a spatiotemporal data cube.

3. The method for identifying abnormal distributions of health data according to claim 1, characterized in that: The generation of the dynamic baseline model includes: Read the spatiotemporal data cube within the first time period, and perform master variable organization processing on the spatiotemporal data cube within the first time period by region, time period, and specified data category to generate a master variable sequence; Read one or more of the following: seasonal, climate, weekday / holiday identifiers, and epidemiological surveillance data; perform covariate processing on the reading results to generate a covariate sequence. Perform time-correspondence processing and region-correspondence processing on the main variable sequence and covariate sequence to generate dynamic baseline input data; Baseline fitting processing based on an autoregressive integral moving average model is performed on the dynamic baseline input data to generate a dynamic baseline model; The dynamic baseline model performs expected range calculation, outlier upper limit calculation, and outlier lower limit calculation on specified data categories by region and time period, generating expected range, outlier upper limit, and outlier lower limit.

4. The method for identifying abnormal distributions of health data according to claim 1, characterized in that: The generation of clustering results includes: Read the spatiotemporal data cube within the second time period, and read the median time interval between adjacent health events within a specified spatial cell, the area of ​​the specified spatial cell, and the value of the specified spatial cell within the second time period; An adaptive function is generated by performing adaptive function construction processing on the median, area, and values ​​of the time interval; The output of the adaptive function is used as the neighborhood radius to process the spatiotemporal data cube in the second time period using the improved ST-DBSCAN clustering algorithm, generating clustering results.

5. The method for identifying abnormal distributions of health data according to claim 1, characterized in that: The generation of potential outliers includes: Read the minimum number of cluster points in the improved ST-DBSCAN clustering algorithm and determine the value of k based on the minimum number of cluster points; Based on the k-value, the k-nearest neighbor method is applied to the clustering results to generate the k-nearest neighbor clusters for each cluster; Read each cluster and its k nearest neighbor clusters, perform local reachability density calculation on each cluster, and perform average local reachability density calculation on the k nearest neighbor clusters to generate local reachability density and average local reachability density; Calculate the ratio of local reachability density to average local reachability density. Clusters whose ratios meet the set conditions are marked to generate potential outliers.

6. The method for identifying abnormal distributions of health data according to claim 1, characterized in that: The generation of anomalous clusters includes: Read the specified spatial granularity according to the user's instructions, perform spatial unit partitioning on the spatiotemporal data cube in the second time period, and generate multiple spatial units; For each spatial unit, read the potential outliers, expected range, upper outlier limit, and lower outlier limit from the clustering results; Using the mean of the expected range as the center line, the upper limit of outliers as the upper control limit, and the lower limit of outliers as the lower control limit, control chart construction is performed on potential outliers in each spatial unit to generate a spatial unit control chart. Anomaly detection and processing are performed on potential outliers based on spatial unit control charts to generate anomaly clusters.

7. The method for identifying abnormal distributions of health data according to claim 1, characterized in that: The generation of attribution results includes: Read the values ​​of the abnormal clusters and the spatial units corresponding to the abnormal clusters, and read the geographical factors; Spatial location correspondence processing is performed on the values, spatial units, and geographical factors of the outlier clusters to generate geographic regression input data; Based on the geographic regression input data, a regression function is constructed to determine the values ​​of the anomaly clusters with respect to geographic factors, and the regression function is generated. Read the coefficients of geographical factors in the regression function, perform influencing factor identification and outlier cluster attribution processing on the coefficients, and generate attribution results.

8. The method for identifying abnormal distributions of health data according to claim 1, characterized in that: The generation of intervention strategy recommendations includes: Read the attribution results of the spatial units corresponding to the abnormal clusters, and read one or more of the following for each spatial unit: the sliding standard deviation of the time window of the number of visits, the distribution of patient age, the distribution of patient address, and the cost concentration coefficient; High-dimensional feature calculation processing is performed on one or more of the following factors: sliding standard deviation of outpatient visits during time window, patient age distribution, patient address distribution, and cost concentration coefficient, to generate high-dimensional features of spatial units; Similarity calculation is performed on the high-dimensional features of multiple spatial units to generate similarity relationships between spatial units. When multiple spatial units whose similarity of high-dimensional features meets the set conditions form a set, the attribution results corresponding to each spatial unit in the set are read, and the common influencing factors among multiple spatial units are determined based on the coefficients of geographical factors in the attribution results, generating intervention strategy suggestions for cross-regional collaborative anomaly patterns.

9. The method for identifying abnormal distributions of health data according to claim 1, characterized in that: The generation of intervention strategy recommendations also includes: Read the attribution results of the spatial units corresponding to the anomaly clusters, and read the high-dimensional features of the spatial units and the high-dimensional features of the historical spatial units; Perform similarity calculation on the high-dimensional features of spatial units and the high-dimensional features of historical spatial units to generate historical similarity relationships; When the similarity between the high-dimensional features of a spatial unit and the high-dimensional features of any historical spatial unit meets the set conditions, the effective intervention actions of the historical spatial unit are read. Based on the coefficients of geographical factors in the attribution results, effective intervention actions for historical spatial units are matched and processed to generate intervention strategy recommendations.

10. A system for identifying abnormal distributions of health data, characterized in that, include: The spatiotemporal data cube construction module is used to acquire health data, perform multidimensional indexing on the health data in terms of time dimension, spatial dimension, data category and value, and generate a spatiotemporal data cube, where the data categories include one or more of the following: total number of patients, number of outpatient visits, medical resource consumption and medical expenses. The dynamic baseline model module is used to read the spatiotemporal data cube within the first time period and read one or more of the following as covariates: seasonal, climate, weekday / holiday identifiers, and epidemic monitoring data. It performs dynamic baseline construction processing based on an autoregressive integral moving average model on the spatiotemporal data cube and covariates within the first time period to generate a dynamic baseline model. The dynamic baseline model outputs the expected range, upper limit of outliers, and lower limit of outliers for specified data categories by region and time period. The anomaly cluster module is used to read the spatiotemporal data cube within the second time period, perform spatiotemporal clustering processing on the spatiotemporal data cube within the second time period to generate clustering results, and perform potential outlier identification processing on the clustering results to generate potential outliers, wherein the second time period is later than the first time period; The anomaly cluster module is also used to perform control chart construction and anomaly detection processing on potential outliers, with the mean of the expected range as the center line, the upper limit of outliers as the upper control limit, and the lower limit of outliers as the lower control limit, to generate anomaly clusters. The geographic regression intervention module is used to read outlier clusters and geographic factors, construct a regression function of the values ​​of outlier clusters with respect to geographic factors, and perform attribution processing on outlier clusters based on the coefficients of geographic factors in the regression function to generate attribution results. The geographic regression intervention module is also used to read attribution results, perform spatial unit feature correspondence processing, similarity relationship matching processing, and intervention action organization processing on the attribution results, and generate intervention strategy suggestions.

Images