Data management system and method for the medical community

Abstract

The application provides a data management system and method for the medical community, and relates to the technical field of data management. The method comprises the following steps: periodically collecting the running state log and the corresponding data collection task log of each institution server node. The server node running state log comprises node identification, heartbeat information, resource occupation rate and available state, and is used for monitoring whether the collection node is in a normal running state. The data collection task log comprises task identification, execution state and execution result, and is used for monitoring the execution process of multiple collection tasks on each node. The management terminal can display the running information of the server and the task in real time, and provide task start-stop, priority adjustment and retry strategy configuration functions, so as to realize server priority monitoring and task whole-process management, and guarantee the stability and controllability of medical community data collection and exchange.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data management, in particular to a data management system and method for medical community. BACKGROUND

[0002] The demand for data integration and sharing among multiple medical institutions in a region is growing. To meet the requirements of unified management and data interconnection, it is common to deploy data collection server nodes locally in each member unit, and regularly transmit structured data in the business system to the medical community data center through ETL (Extraction, Transformation, Loading) tasks. However, due to the large number of collection nodes, complex task types, and large differences in running environments, traditional manual supervision methods are difficult to timely grasp the running status of each node and its task, and are prone to problems such as task execution failure, node disconnection, and delayed data transmission, which seriously affect the completeness and timeliness of data collection.

[0003] In the prior art, only whether the task is completed is generally concerned, and there is a lack of systematic monitoring and linkage management of server running status and task execution process. Especially in actual medical community deployment, it is common to run multiple collection tasks on one server, and task state abnormalities are often highly related to node resource abnormalities, such as CPU overload, memory overflow, network interruption, etc. If node and task coordination monitoring and management cannot be achieved, it will lead to low fault troubleshooting efficiency, high task failure rate, and prolonged data interruption time. In addition, some systems still rely on manual methods to start and stop tasks and handle exceptions, and lack a unified task scheduling and management mechanism, making it difficult to meet the visualization, automation, and stabilization requirements of the data collection process in large-scale medical community scenarios.

[0004] Therefore, there is an urgent need for a data management method for medical community, which can realize whole-process monitoring of the data collection and exchange process under a unified architecture. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a data management system and method for medical community.

[0006] In a first aspect, the present application provides a data management method for medical community, comprising:

[0007] obtaining data collection task logs of each target medical institution and running state logs of corresponding server nodes according to a first preset period; the running state logs of the server nodes include: node identifier, node heartbeat information, node resource occupation rate, and node available state;

[0008] Based on the data collection task log and the running state log of the server node, data quality indicators of each target medical institution and running state indicators of the server node are respectively counted, and time series data of each data quality indicator and running state indicator are generated according to a second preset period; the running state indicators include: node failure rate and resource abnormal occupation rate;

[0009] For the same data quality indicator, the time series data corresponding to each target medical institution is stacked according to the institution dimension to construct a joint feature matrix, and a multivariate fitting algorithm is executed based on the joint feature matrix to obtain the trend change parameter of each target medical institution;

[0010] For the trend change parameter, combined with the running state indicators of the server node, a deterioration trend detection result is generated according to a preset judgment condition; based on the deterioration trend detection result, trend early warning information is generated and output to a management terminal.

[0011] As an optional implementation, the management terminal is further used for:

[0012] Real-time display of running state information of the server node corresponding to each target medical institution and execution state information of the data collection task;

[0013] The running state information includes: node available state, node resource occupation rate, node failure prompt, and the execution state information includes: task current state, task execution progress, and task execution result;

[0014] The management terminal is further used for: providing task management function according to the execution state information of the data collection task and the running state information of the server node; the task management function includes task start-stop control, task priority adjustment, and task retry strategy management.

[0015] As an optional implementation, the data quality indicators include: data missing rate, format error rate, and data collection delay; the data collection task log includes: collection task identifier, collection time, collection item quantity, collection result, and collection state;

[0016] For the trend change parameter, a deterioration trend detection result is generated according to a preset judgment condition, including:

[0017] According to the trend change parameter of each target medical institution, corresponding single-institution judgment result information is generated;

[0018] The single-institution judgment result information is input as multi-source evidence information, and evidence theory is used for synthesis processing to obtain a comprehensive confidence result for the overall data collection situation;

[0019] Calculate the conflict degree of the comprehensive confidence result, and perform weight adjustment on the conflict evidence information when the conflict degree exceeds a preset range to obtain a modified comprehensive confidence result;

[0020] Compare the modified comprehensive confidence result with a preset credibility threshold to determine an abnormal confidence level of the medical community level;

[0021] According to the abnormal confidence level and the single-institution determination result information of each target medical institution, generate the deterioration trend detection result.

[0022] As an optional implementation, the obtaining of the comprehensive confidence result for the overall data collection situation includes:

[0023] According to the single-institution determination result information of each target medical institution, construct an initial basic confidence allocation;

[0024] Calculate the conflict degree between the initial basic confidence allocations, and adjust the weight of the conflict evidence information to update the basic confidence allocation when the conflict degree exceeds a preset range;

[0025] Perform iterative synthesis processing on the updated basic confidence allocation to obtain a first synthesis confidence and a second synthesis confidence; wherein the first synthesis confidence is used to represent that the data collection quality is in a deterioration trend, and the second synthesis confidence is used to represent that the data collection quality is in a normal trend;

[0026] Determine an uncertainty entropy value based on the first synthesis confidence and the second synthesis confidence;

[0027] Generate a comprehensive confidence result based on the first synthesis confidence, the second synthesis confidence, and the uncertainty entropy value;

[0028] The uncertainty entropy value is used to represent the consistency degree of the overall evidence of the medical community.

[0029] As an optional implementation, it further includes:

[0030] Generate a geographical proximity matrix based on geographical position information between each target medical institution;

[0031] According to the geographical proximity matrix, perform proximity-based weighting processing on the single-institution determination result information of each target medical institution;

[0032] Use the weighted single-institution determination result information to generate the initial basic confidence allocation;

[0033] The generating of the geographical proximity matrix based on the geographical position information between each target medical institution includes:

[0034] Performing geographical analysis on the address information of each target medical institution to obtain corresponding institution location information parameters;

[0035] When the address of any target medical institution cannot be successfully resolved or the confidence of the corresponding institution location information parameter is lower than a preset threshold, the institution location information parameter is supplemented based on a predefined auxiliary location information generation rule;

[0036] Based on the institution location information parameter, the geographical proximity between any two target medical institutions is calculated, and the geographical proximity matrix is constructed.

[0037] As an optional implementation, after calculating the geographical proximity between any two target medical institutions based on the institution location information parameter, the following steps are further included:

[0038] In a third preset period, collect the business interaction log of each target medical institution, which at least records the interaction timestamp, the interaction counterpart identifier, the interaction type identifier, and the interaction data volume;

[0039] For any two target medical institutions, the number of interactions and the interaction data volume in the third preset period are counted;

[0040] According to the number of interactions, the interaction data volume, and the geographical proximity, the weight adjustment is performed on the corresponding proximity value to obtain the updated proximity value;

[0041] Based on the updated proximity value, the geographical proximity matrix is constructed.

[0042] As an optional implementation, the stacking of the time series data of each target medical institution corresponding to the same data quality indicator according to the institution dimension to construct the joint feature matrix includes:

[0043] Based on the geographical proximity matrix, target medical institutions with a geographical proximity higher than a preset grouping threshold are divided into the same proximity cluster according to a proximity aggregation rule;

[0044] For each proximity cluster, the number of abnormal counts of target medical institutions within the cluster is counted at each time series index, and when the number of abnormal counts reaches a preset cluster threshold, a regional synchronous abnormality marker is generated;

[0045] When the regional synchronous abnormality marker exists, the column-level merging processing is performed on the corresponding time series data in a cluster unit according to a preset merging rule to obtain a cluster-level representative data column;

[0046] Stack the cluster-level representative data column and the time series data column of target medical institutions outside the cluster according to the institution dimension, and fill it into the joint feature matrix.

[0047] As an optional implementation, the filling of the joint feature matrix by stacking the cluster-level representative data column and the time series data column of the target medical institution outside the cluster in the institution dimension comprises:

[0048] For each adjacent cluster, the time series data column of the member institution inside the cluster is subjected to difference processing using the cluster-level representative data column to generate a corresponding residual data column;

[0049] The cluster-level representative data column and each residual data column are stacked in the institution dimension and combined with the time series data column of the target medical institution outside the cluster to fill the joint feature matrix.

[0050] As an optional implementation, a multivariate fitting algorithm is performed based on the joint feature matrix to obtain the trend change parameter of each target medical institution, comprising:

[0051] The joint feature matrix is partitioned according to a preset sliding time window to obtain a plurality of window feature sub-matrices;

[0052] For each window feature sub-matrix, weighted regression fitting is performed to obtain a corresponding candidate trend parameter and a residual confidence value;

[0053] The candidate trend parameter and the residual confidence value of each window are used as multi-source evidence information to construct a window-level basic confidence distribution;

[0054] The window-level basic confidence distribution is subjected to evidence theory synthesis processing, and the weight of conflicting evidence information is dynamically adjusted when the conflict degree exceeds a preset range to generate a comprehensive trend change parameter of each target medical institution.

[0055] In a second aspect, the present application provides a data management system for the medical community, comprising:

[0056] A collection module is configured to obtain data collection task logs of each target medical institution and running state logs of corresponding server nodes according to a first preset period; the running state logs of the server nodes include node identification, node heartbeat information, node resource occupation rate, and node available state;

[0057] A first processing module is configured to respectively count data quality indicators of each target medical institution and running state indicators of server nodes based on the data collection task logs and the running state logs of the server nodes, and generate time series data of each data quality indicator and running state indicator according to a second preset period; the running state indicators include node failure rate and resource abnormal occupation rate;

[0058] The second processing module is configured to, for the same data quality index, stack the time-series data corresponding to each target medical institution in the institutional dimension, construct a joint feature matrix, and perform a multivariate fitting algorithm based on the joint feature matrix to obtain a trend change parameter of each target medical institution.

[0059] The detection module is configured to, for the trend change parameter, generate a degradation trend detection result according to a preset judgment condition in combination with a running state index of the server node, generate trend early warning information based on the degradation trend detection result, and output the trend early warning information to a management terminal.

[0060] Compared with the prior art, the present application can realize accurate quantitative evaluation of the data quality in the medical community by automatically obtaining the data collection task logs of each medical institution and further statistically analyzing multiple indexes such as data missing rate, format error rate and data collection delay of each institution. Meanwhile, with the aid of joint feature matrix construction and multivariate fitting algorithm, the method can accurately and efficiently capture and reflect the trend change of the data quality of the medical institutions over time, greatly improving the real-time performance and accuracy of abnormal trend detection. In addition, the present application can automatically generate trend early warning information according to the trend change parameter and push it to the management terminal in real time, thereby helping the managers to take timely measures and effectively avoiding the risk of medical decision deviation and service interruption caused by continuous degradation of data quality. Compared with the prior art, the present application significantly reduces the time delay and subjective error of manual analysis, and enhances the sensitivity and rapid response capability to data quality abnormalities in the region, and is more suitable for the collaborative management of the medical community with large scale and huge data volume. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 The flowchart of the data management method for the medical community provided by the present application;

[0062] Figure 2 The flowchart of the method for generating a degradation trend detection result provided by the present application;

[0063] Figure 3 The flowchart of the method for obtaining a comprehensive confidence result for the overall data collection situation provided by the present application;

[0064] Figure 4 The schematic diagram of the data management system for the medical community provided by the present application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments.

[0066] Refer toFigure 1 As shown, the flowchart of the data management method for the medical community provided in the present application, the method comprises steps S101-S104, wherein:

[0067] S101: Obtain the data collection task log of each target medical institution and the running state log of the corresponding server node according to the first preset period; the running state log of the server node includes: node identification, node heartbeat information, node resource occupation rate and node available state;

[0068] S102: Based on the data collection task log and the running state log of the server node, the data quality indicators of each target medical institution and the running state indicators of the server node are respectively counted, and the time series data of each data quality indicator and running state indicator is generated according to the second preset period; the running state indicators include: node failure rate and resource abnormal occupation rate;

[0069] S103: For the same data quality indicator, the time series data corresponding to each target medical institution is stacked according to the institution dimension to construct a joint feature matrix, and a multivariate fitting algorithm is executed based on the joint feature matrix to obtain the trend change parameter of each target medical institution;

[0070] S104: For the trend change parameter, combined with the running state indicators of the server node, the degradation trend detection result is generated according to the preset judgment condition; based on the degradation trend detection result, the trend early warning information is generated, and the trend early warning information is output to the management terminal.

[0071] As an optional implementation, the management terminal is also used for:

[0072] Real-time display of the running state information of the server node corresponding to each target medical institution and the execution state information of the data collection task;

[0073] The running state information includes: node available state, node resource occupation rate, node failure prompt, and the execution state information includes: task current state, task execution progress, task execution result;

[0074] The management terminal is also used for: providing task management function according to the execution state information of the data collection task and the running state information of the server node; the task management function includes task start-stop control, task priority adjustment and task retry strategy management.

[0075] As an optional implementation, the data quality indicators include: data missing rate, format error rate and data collection delay; the data collection task log includes: collection task identification, collection time, collection item quantity, collection result and collection state;

[0076] Wherein, the data collection task log refers to a structured record set automatically generated by the medical information system, and each field has the following meanings: the collection task identifier is used to uniquely identify a data extraction job; the collection time indicates the time when the extraction is triggered; the collection item quantity describes the total number of data tables or interfaces extracted this time; the collection result gives an enumeration state of successful, failed or partially successful extraction; and the collection state records the real-time state of the current job, such as queuing, execution or completion. The first preset period can be a five-minute rolling time window triggered by a unified scheduling strategy.

[0077] The present application continuously reads the task logs uploaded by each target medical institution to the message queue, writes the log entries into the time series database and maintains them in a zipped manner. Then, the index calculation process is executed: the log slices of each medical institution within the second preset period are traversed, and three indicators, the missing rate, the format error rate and the collection delay, are calculated. The missing rate is obtained by comparing the number of extraction items with the actual number of successful items, the format error rate is determined based on the number of parsing failure entries, and the collection delay is mapped to the delay interval label by the difference between the log reporting time and the collection time. The calculation results are written back to the original database and supplemented in chronological order, forming the index time series segment corresponding to each institution.

[0078] The system performs a stacking operation on the same indicator in the institution dimension, first reads the indicator columns of each institution with the missing part filled, and then performs horizontal splicing according to the unique code of the institution to generate a joint feature matrix. The row dimension of the matrix represents the time index, the column dimension is sorted by institution, and the matrix unit contains discrete values of the same indicator in the same time slice. The present application calls a multivariate fitting algorithm to implement fitting on the matrix column vector with a sliding time window as the basic unit. The fitting algorithm uses a column-by-column residual recursive method to achieve without formula: first, normalize all columns to zero mean and unit variance, then estimate the column trend slope using the iterative least squares method, and then update the latest trend change parameter by continuously smoothing the initial value of the early sample slope through a rolling window. The trend change parameter is divided into three intervals: positive, negative and approximately zero. Positive indicates that the indicator is deteriorating, negative indicates that the indicator is improving, and approximately zero indicates stability.

[0079] The system compares the trend change parameter with the predetermined judgment threshold, and the threshold is determined by the historical index variance through a percentile voting strategy. When the trend change parameter of any institution exceeds the corresponding threshold, the system immediately marks it as a single-institution deterioration. When the proportion of deteriorating institutions within the same time window is higher than fifteen percentage points, the present application aggregates the single-institution marking information to generate a deterioration trend detection result, and records the detection entries according to the institution code and the indicator type.

[0080] The detection entry enters the early warning generation process, and the system assembles trend early warning information according to the early warning strategy. The early warning information field includes: early warning identifier, trigger mechanism list, impact index, trigger time, trend parameter snapshot and recommended disposal action. The JSON structure generated by the present application is pushed to the management terminal through the REST interface, and is visualized in real time on the alarm large screen in red, orange and yellow blocks.

[0081] This embodiment ensures that the data quality trends of each institution in the medical community are quickly visible and can be automatically warned by unified zippered task log management, three-index time series calculation, joint feature matrix construction, multi-element slope recursion and threshold determination closed loop. The index degradation can be captured within five minutes, which is several hours earlier than the manual weekly or daily report method, significantly improving the timeliness of data governance; the joint feature matrix retains the institution granularity information and avoids analysis discontinuity caused by column gaps, laying a stable input foundation for subsequent evidence theory synthesis.

[0082] For the above S101:

[0083] In this embodiment, the present application deploys a log pushing service in each target medical institution, and the log is written in a unified JSON structure to a message queue topic. The scheduling module triggers a pull operation according to a first preset period. The first preset period can be set to five minutes; the example optional range is one minute to fifteen minutes, which is selected by the operation and maintenance personnel according to the data volume and monitoring real-time performance. The number of message entries pulled by the scheduling module at a time is limited by the batch size, and the batch size can be set to 500, with an example optional range of 100 to 1000, to balance network bandwidth occupation and processing delay.

[0084] After receiving the log, the system parses the five fields of collection task identifier, collection time, collection item quantity, collection result and collection state in turn. The collection task identifier needs to be unique within the current period; the collection time is converted to UTC representation through time zone conversion; the collection item quantity and the collection result must be verified by range verification and enumeration verification; the collection state is mapped to generate an internal state code. The example collection state can be set to three enumeration values: running, completed or failed.

[0085] After parsing and verification, the system writes the institution code and task identifier as the primary key into the time series database, and writes the timestamp, item number, result code, state code and institution identifier fields at the same time. The writing adopts a batch submission mode, and the batch size is consistent with the pull batch size. If the writing occurs abnormally, the present application automatically executes a retry mechanism, and the retry threshold can be set to two times, with an example optional range of one to five times. If the writing still cannot be written after reaching the threshold, the system stores the corresponding batch of logs to the error channel and generates an alarm event to remind the operation and maintenance personnel to intervene.

[0086] For medical institutions with small data volume or using traditional databases, the application can also enable active pulling based on direct database connection. The same field extraction and verification are completed by querying the log table by time interval, and then the results are written into the time series database. Time zone conversion, uniqueness verification, enumeration mapping and retry strategy remain consistent with the message queue path, and only the triggering mechanism changes from passive consumption to active polling.

[0087] Exemplarily, in an optional implementation, the data collection task can be executed by an ETL engine, and the running state log thereof is an ETL task execution log.

[0088] Through the above parameterized configuration and pipeline processing, the application can collect the data collection task logs of each target medical institution of the medical community in each first preset period in a complete and reliable manner, and provide a raw data basis with timeliness and consistency for subsequent data quality index statistics and trend analysis.

[0089] For S102 described above:

[0090] The application starts the index generation task after completing the log storage, and the task scheduler executes in a cycle with a second preset period as the triggering granularity. The second preset period can be set to thirty minutes; exemplarily, the optional range is ten minutes to one hour, which can be configured according to the balance between trend sensitivity and calculation resource occupation. The scheduler traverses the log records newly added in the current period window in the time series database according to the institution code, and calculates the data missing rate, format error rate and data collection delay of each target medical institution.

[0091] To obtain the data missing rate, the application first reads the total record number and the successful record number of the collection task of the institution in the window. The missing task is defined as the collection result field marked as failed or partially successful. If the total record number is zero, the system marks the missing rate of the current period as invalid and attaches the coverage rate identifier; otherwise, the missing task proportion is calculated, and the result is mapped to the percentage format and retained to two decimal places.

[0092] Exemplarily, the interval can be judged as excellent for 0-5%, acceptable for 5-15%, and unqualified for more than 15%. The interval threshold is set by the medical community data governance specification in the deployment stage, and can be dynamically adjusted through the management terminal later.

[0093] The calculation of the format error rate is based on the number of format verification failures recorded when the log is parsed. The application counts the number of parsing failure entries of the institution in the window, and compares it with the total number of task records to obtain the format error rate. To prevent misjudgment when the number of collection items is too low, the system does not calculate the format error rate in the window with less than ten total records, and marks the index as insufficient coverage.

[0094] For example, the threshold is 0.5%, 1% and 3% respectively, corresponding to the three levels of excellent, warning and risk.

[0095] The data collection delay measures the delay degree of the log reporting time from the collection time. The application calculates the difference value for each log and maps it to the delay gear according to the delay length. For example, the gears can be set to 0-5 minutes, 5-15 minutes, 15-30 minutes, and more than 30 minutes. The application takes the median of the delay gears of all logs of the same window in the same institution as the window delay indicator. If more than one-third of the logs in the window are in the "more than 30 minutes" delay gear, the system will directly mark the delay indicator as the highest risk gear to ensure that the late batch report is not hidden by the average number.

[0096] After the three indicators are generated, the application writes the missing rate, the format error rate and the delay gear into the institution indicator time sequence table respectively. The corresponding fields include the window end timestamp, the institution code, the indicator type and the indicator value. When writing, if the timestamp and institution combination already has an old value, the system will overwrite it with time priority and record the overwrite log to prevent repeated statistics. For the case where there is no valid log in the window, the system writes a placeholder record and marks the null value in the indicator value field. The subsequent trend fitting stage can detect and handle the gap.

[0097] The indicator time sequence table triggers the data completion module after writing is completed, and fills the gap forward or linearly interpolates. The application defaults to the forward filling strategy; example optional strategies include backward filling or external mean filling, which can be switched through a configuration file. After completion, the system writes an update identifier to the task progress table to provide continuous, non-hollow, standardized time series data for the construction of the joint feature matrix in S103.

[0098] In this way, the application can generate traceable and comparable data quality indicator time sequences for each target medical institution in the medical community in each second preset period, laying an accurate and stable input foundation for subsequent trend analysis.

[0099] For the above S103:

[0100] After completing the index time sequence filling, the application enters the joint feature matrix construction phase. The system first reads the current index generation completion identifier in the task schedule and processes each index type one by one. Taking the "data missing rate" as an example, the system scans all the index time series of the institutions according to the unified time index, aligns the time index to the second level Unix timestamp and completes the time zone unification. For any time point still marked with null value, the system fills it forward with the last valid value; if the length of continuous null value exceeds thirty index intervals, an additional gap identifier column is inserted to facilitate subsequent model identification of coverage defects. After filling, the system horizontally splices each column in ascending order of institution code to generate a joint feature matrix with time index as row dimension and institution code as column dimension.

[0101] To reduce the influence of dimension difference between columns on fitting, the application performs column-level standardization processing on the matrix. The standardization method adopts the zero mean plus unit variance scheme, and the standardization parameters are calculated and persisted at the first run to ensure that the dimensions remain consistent between different rolling windows. If the variance of a certain institution in the historical sample is below the one percent threshold, the system automatically marks the column as a low variation column and skips it during fitting to prevent numerical instability caused by near-zero variance.

[0102] After the matrix is prepared, the application performs multivariate fitting on each window. The system performs rolling with a third preset window size, and the exemplary window length can be set to one hundred and twenty time indexes, and the specific range can be adjusted to thirty to two hundred and forty. The application calculates the linear slope in each window using the incremental least squares recursive algorithm for matrix column vectors, uses the slope of the previous window as the initial value, and updates the latest estimate with five micro-steps. To weaken the instantaneous impact of abnormal peaks on the slope, the system performs three quantile truncation on each row before recursion, replacing abnormal values exceeding one times the interquartile range of the upper or lower limit with adjacent non-abnormal values.

[0103] After obtaining the window slope, the system generates a trend change parameter time series for each column. To facilitate threshold determination, the application maps the slope value to the "accelerated deterioration", "slight deterioration", "stable", "slight improvement", and "accelerated improvement" five interval. The exemplary boundaries can be set according to twenty-five percent, fifty percent, zero, negative fifty percent, and negative twenty-five percent of the standard deviation, and can be automatically learned according to the historical fluctuation range during deployment. The system appends a trend level column below the same column in the matrix, synchronously stores the original slope and level label, and calls the next stage of deterioration detection.

[0104] In order to reduce the full-matrix persistence occupation, the application releases the row direction buffer of the previous window after completing each window fitting, and only retains four types of information, i.e., the archived sample index, the standardization parameter, the slope vector and the trend gear label. For the monthly or quarterly trend that needs to be compared for a long time, the system resamples and aggregates the slope sequence in the offline batch task. Through the construction-fitting-compression pipeline of iteration by window, the application not only maintains the rolling real-time of trend calculation, but also controls the storage and memory overhead.

[0105] For the above S104:

[0106] After obtaining the window slope and mapping it to the trend gear, the system immediately enters the deterioration detection stage, and maintains a dynamic state register for each medical institution. The register field includes: the trend gear sequence of the last five windows, the cumulative deterioration count, the continuous deterioration start time and the current deterioration level. The application performs the following actions at each window update.

[0107] Firstly, the system reads the latest trend gear; if the gear is "accelerated deterioration" or "slight deterioration", the cumulative deterioration count in the register is increased by one; otherwise, it is cleared. The cumulative count threshold can be set to three, and the example optional range is two to five, which is used to control the "continuous window deterioration" trigger sensitivity. When the cumulative count reaches the threshold, the system marks the institution as "single institution deterioration", and records the continuous deterioration start time. If the trend gear does not fall to "stable" or better in the next two windows, the system upgrades the deterioration level to "aggravation". The level enumeration includes four states: initial, deterioration, aggravation and mitigation. The application automatically pushes the "technical operation check" task to the institution that has not been mitigated for three windows in the aggravation state.

[0108] Subsequently, the system conducts horizontal aggregation of all institutions at the window granularity. The application counts the number of institutions in the "deterioration" or "aggravation" state in the current window, and divides it by the total number of institutions to obtain the deterioration proportion. The deterioration proportion threshold can be configured as fifteen percent, and the example adjustable range is ten percent to thirty percent. If the proportion exceeds the threshold and covers at least two data quality indicators, the system generates a "regional deterioration event". The event field includes: event identification (concatenated by indicator type + window end time), trigger institution list, deterioration indicator, trigger threshold, proportion value and deterioration level distribution.

[0109] After the event is generated, it enters the early warning assembly process. According to the event level, the application matches the alarm strategy: single mechanism degradation is mapped to yellow warning, regional degradation event is mapped to orange warning, and regional aggravation event is mapped to red warning. In addition to basic information, the assembly field also contains suggested operation text. Suggested operation examples can be "check VPN channel connectivity", "check interface field mapping rules", "reset ETL extraction node", etc., and the content is automatically matched by the operation and maintenance knowledge base according to the index type and mechanism level.

[0110] After the early warning information is generated, it is written into the alarm table and pushed to the management terminal through asynchronous messages. The application defaults to use REST interface; exemplary, optional channels also include WebSocket push stream and email notification. After the terminal receives the information, it presents it on the large screen in color blocks and a scrolling list simultaneously. If the same mechanism triggers the same level of warning repeatedly within ten minutes, the system combines it into one and updates the trigger times and the latest time in the message body.

[0111] The application archives all the early warnings of the previous day to the historical alarm library at zero o'clock every day, and the archive field includes closed-loop processing results and time consumption, which is used for subsequent performance evaluation and model threshold adaptive training. Through the above-mentioned register maintenance, horizontal proportion detection, hierarchical mapping and multi-channel pushing closed-loop process, the system can continuously analyze the data quality trend of the medical community and output early warning instructions that are easy to execute, ensuring early detection of abnormalities, early assignment of responsibilities, and traceable disposal.

[0112] Exemplarily, after the platform completes the storage of the server node running state log, it will periodically perform statistical analysis on the node running state. Specifically, the system can read the heartbeat information of each server node every 30 seconds and record the response of the node. If the heartbeat information of a node is missing or unresponsive for three consecutive times, it is considered as a node failure sample, and the system calculates the node failure rate based on this. The calculation method can be: the number of node failure samples in the statistical period divided by the total number of theoretical heartbeats, finally presented in percentage form, and rounded to two decimal places.

[0113] Among them, the resource abnormal occupation rate of the node is determined based on the use of CPU, memory and network bandwidth of the periodically sampled node. The specific sampling threshold can be exemplarily set as: CPU occupation rate exceeding 85%, memory occupation rate exceeding 80%, and bandwidth occupation exceeding 70%. In a sampling, if any indicator exceeds the above threshold, it is considered as an abnormal sampling. The system divides the number of abnormal samplings in a period by the total number of samplings at the end of each statistical period to obtain the resource abnormal occupation rate.

[0114] For example, the system comprehensively considers the health status of the node and the execution of the data collection task to accurately determine the degradation trend of the data quality. First, the platform divides the health status of the node into three levels of normal, warning and failure according to the node failure rate and resource abnormal occupancy rate. When the health status of a server node is marked as the failure level and there is a failure record of the data collection task carried by the node in the same period, the system will directly determine that the degradation trend of the task belongs to “node failure trigger” and does not need to additionally determine the slope of the trend change parameter.

[0115] If the health status of the node is at the warning level and the data collection task on the node has a positive slope in the same period, the system will automatically increase the weight of the task in the degradation trend detection, for example, by 1.5 times. In this way, the abnormal fluctuation of the node status can be more obviously reflected in the trend determination process of the task data quality, and the correlation between the node running status and the task quality is strengthened.

[0116] For example, the management terminal has detailed task management functions, including task start-stop control, task priority adjustment and task retry strategy management. The management terminal realizes the start-stop of the task through a predefined REST interface, for example, by calling the interface / task / {id} / control and setting the action as “pause” or “resume” in the request parameter, so as to realize the real-time start-stop control of the specified data collection task.

[0117] For task priority adjustment, the management terminal adjusts the task priority in real time through the interface / task / {id} / priority, and the priority range accepted by the system is exemplarily 0 to 10. After each adjustment, the new priority value is immediately written into the metadata table of the task scheduling center, triggering the real-time rearrangement of the task scheduling queue, so that the high-priority task is preferentially executed.

[0118] In addition, the retry strategy of the task is also dynamically configured by the management terminal. The system supports setting the maximum number of retries (maxRetry) and the retry backoff interval (backoffSeconds) of the task. When the task running state is converted to failure, the scheduler will automatically re-schedule the task to the standby server node or the adjacent server node of the same medical institution according to the configured backoff strategy in the form of exponential backoff to continue execution until the task is successfully completed or the maximum number of retries is reached. When the task finally fails, the system will automatically generate a corresponding failure alarm on the management terminal for the management personnel to manually intervene and further process.

[0119] As an optional implementation, see Figure 2A flowchart of a method for generating a deterioration trend detection result is provided, including steps S201-S205, wherein:

[0120] S201: generating single-institution determination result information corresponding to each target medical institution according to the trend change parameters of the target medical institutions;

[0121] S202: inputting the single-institution determination result information as multi-source evidence information, performing synthesis processing using evidence theory to obtain a comprehensive confidence result for the overall data collection situation;

[0122] S203: calculating the conflict degree of the comprehensive confidence result, and performing weight adjustment on the conflicting evidence information when the conflict degree exceeds a preset range to obtain a modified comprehensive confidence result;

[0123] S204: comparing the modified comprehensive confidence result with a preset confidence threshold to determine an abnormal confidence level at the medical community level;

[0124] S205: generating the deterioration trend detection result according to the abnormal confidence level and the single-institution determination result information of each target medical institution.

[0125] The single-institution determination result information refers to a determination item formed based on the trend change parameters of a single medical institution within a specific time window, and each determination item includes fields such as institution code, index type, trend level, trend confidence, and evidence weight. The evidence theory synthesis processing refers to synthesizing multiple source determination items according to the improved Dempster rule or similar methods to obtain the comprehensive confidence distribution of the three propositions of overall deterioration, normality, and uncertainty in the current time window. The conflict degree represents the accumulated contradiction value when different determination items are opposite in proposition attribution, and when the contradiction value exceeds the range set in advance, the relevant evidence information needs to be down-weighted. The abnormal confidence level is determined by comparing the deterioration confidence with the preset confidence threshold, and finally the overall risk is classified into high, medium, or low levels for subsequent early warning decision-making.

[0126] In a specific implementation, the present application first constructs a basic belief assignment according to each single-institution decision result information, in which the trend confidence is combined with the evidence weight to map to the "deterioration" or "normal" proposition, and the remaining part not assigned is attributed to the "uncertain" proposition to explicitly show the case of incomplete information or ambiguous trend. Multiple basic belief assignments are sequentially input into the evidence theory synthesis module and the conflict degree is calculated in real time. If the conflict degree between the new decision item and the current synthesis result exceeds the predefined limit value, the system will reduce the evidence weight of the item to prevent extreme or low-coverage institutions from exerting unreasonable influence on the overall judgment. If the high conflict degree cannot be eliminated, the remaining conflict part will be suspended in the uncertain proposition for subsequent synthesis, thereby preserving local conflict information without destroying the consensus of most institutions. The final synthesis result contains three synthesis confidences of deterioration, normal and uncertain, and the system will further compare the deterioration confidence with the confidence threshold to determine the abnormal confidence level of the medical community as high, medium or low.

[0127] For example, the system generates three single-institution decision result information after collecting the trend change parameters of three medical institutions. If the trend gear of one of the institutions is determined to be accelerated deterioration, the trend confidence is high, and the coverage rate is more than 90%, which will bring strong support to the "deterioration" proposition during synthesis. The trend gear of another institution is determined to be slightly improved, but the coverage rate is low, so the evidence weight will be reduced, thereby the support for the "normal" proposition is weak during the synthesis process. Once the conflict degree accumulated by the two opposite evidence information exceeds the set limit value, the system will again reduce the evidence information with relatively low confidence and poor coverage. After several reductions, if the conflict degree is still higher than the limit value, the remaining conflict part will be attributed to the uncertain proposition to show the preservation and isolation of this local information and avoid interfering with the overall judgment. After the synthesis is completed, the present application obtains a relatively high comprehensive confidence of the "deterioration" proposition, and concludes that the abnormal confidence level is high. At the same time, the conflict reduction process is recorded in the system log to further investigate the reason and reliability of this potential extreme evidence item by the operation and maintenance personnel.

[0128] Thus, it has significant advantages for the scenario of multiple agencies and differentiated data sources coexisting in the medical community environment. By adopting the strategy of dynamic weight adjustment and uncertain proposition allocation in the synthesis stage, it can avoid the extreme misjudgment brought by simple voting or weighted average method, and also automatically weaken the evidence entries with low coverage or insufficient confidence when the multi-source evidence information conflicts. The synthesis result finally provides a quantitative index of whether the overall data collection quality has significantly deteriorated, and presents it with high, medium and low abnormal confidence levels, which is convenient for medical community managers to take different emergency measures according to the different risk levels. Compared with the method of relying only on a single threshold or simple statistics, this synthesis method based on evidence theory can more accurately identify the overall trend, reduce noise interference, and improve the accuracy and reliability of medical community data governance.

[0129] As an optional implementation, refer to Figure 3 The flowchart provided by the present application for obtaining a comprehensive confidence result for the overall data collection situation includes steps S301-S304, wherein:

[0130] S301: Construct an initial basic confidence distribution according to the single-agency determination result information of each target medical institution;

[0131] S302: Calculate the conflict degree between the initial basic confidence distributions, and adjust the weight of the conflicting evidence information to update the basic confidence distribution when the conflict degree exceeds the preset range;

[0132] S303: Perform iterative synthesis processing on the updated basic confidence distribution to obtain a first synthesis confidence and a second synthesis confidence; wherein the first synthesis confidence is used to represent that the data collection quality is in a deteriorating trend, and the second synthesis confidence is used to represent that the data collection quality is in a normal trend;

[0133] S304: Determine an uncertainty entropy value based on the first synthesis confidence and the second synthesis confidence; generate a comprehensive confidence result based on the first synthesis confidence, the second synthesis confidence, and the uncertainty entropy value; wherein the uncertainty entropy value is used to represent the consistency degree of the overall evidence of the medical community.

[0134] In the global synthesis of single-institution determination result information of each target medical institution, the application introduces a multi-iteration comprehensive processing means. In order to allow multi-source evidence to be corrected in time when the conflict degree exceeds the limit, while maintaining the compatibility of uncertain information, the application first allocates an initial basic confidence assignment to each single-institution determination result. The basic confidence assignment is used to characterize the support degree of the item to the two propositions of "data acquisition quality is in a deteriorating trend" and "data acquisition quality is in a normal trend", and reserves a part of the probability space for the uncertain proposition, mainly reflecting the information gap when the medical institution coverage is insufficient or the trend direction is ambiguous.

[0135] After obtaining all the initial basic confidence assignments, the application calculates the conflict degree for any two basic confidence assignments. In order to identify high-conflict situations caused by institution determination results with significant differences in coverage or extreme direction, the system accumulates the parts of the two assignments that are completely opposite on the deterioration and normal propositions into the conflict value. If the conflict degree exceeds the threshold range set in the deployment stage, the system will lock the item that is recently included in the conflict degree iteration calculation process and is obviously opposite to other evidence, and adjust the weight corresponding thereto.

[0136] For example, if the weight of an institution is originally high, but the coverage is insufficient to support its high credibility, then in the conflict degree iteration and merging, this item will be preferentially down-weighted or partially transferred to the uncertain proposition. After one or more rounds of adjustment, when the conflict degree falls within an acceptable range, the system takes the merging result as the output of the current iteration and enters the next round of synthesis.

[0137] In the final output of multiple iteration synthesis, the application focuses on two key confidence degrees, namely the first synthesis confidence degree and the second synthesis confidence degree. The first synthesis confidence degree is used to represent the support strength of the proposition that the medical community as a whole is in a "deteriorating trend" after the synthesis of all institution information. If this value is higher than a certain credibility threshold, it means that the system determines that the overall deterioration state is relatively certain. The second synthesis confidence degree is used to quantify the degree of certainty of the system to the proposition of "normal trend" under the comprehensive effect of the same batch of institution determination results.

[0138] By introducing a dynamic weight adjustment mechanism in each iteration, extreme conflict evidence can be gradually weakened or removed from the two-pole attribution, thereby presenting a more reasonable overall trend evaluation in the synthesis output.

[0139] After obtaining the first synthesis confidence degree and the second synthesis confidence degree, the application derives an uncertainty entropy value from the degree of inconsistency between the two and the amount of information not assigned to any explicit trend proposition. A high entropy value means that there is still a large range of viewpoint dispersion or conflict residue among multi-source evidence; a low entropy value means that most of the evidence can be relatively consistent in pointing to a certain trend state, and there is little residual evidence after dynamic conflict processing.

[0140] Based on the first synthetic confidence, the second synthetic confidence and the uncertainty entropy value, the present application integrates them together as the final comprehensive confidence result. The result can be a three-element combination, such as presenting the values of "degradation confidence", "normal confidence" and "uncertainty entropy", or further converted into a form of grade, color or alert level in the back end, for direct reference by the management terminal or visual panel.

[0141] In this way, the present application can more robustly digest the noise brought by extreme or low-coverage evidence when facing heterogeneous and partially opposite medical institution trend determination results. Relying on the design of uncertain propositions and entropy values, medical community managers can also clearly understand the divergence degree of the overall trend determination, providing more reasonable decision-making reference for subsequent operation and maintenance intervention. Thus, the present application not only realizes the global support degree calculation of the "degradation" and "normal" propositions, but also greatly reduces the interference risk of conflicting evidence on the final result while preserving uncertain information, thereby improving the accuracy and interpretability of medical community data quality trend detection.

[0142] As an optional implementation, it further comprises:

[0143] Generating a geographical proximity matrix based on the geographical position information between the target medical institutions;

[0144] Performing proximity-based weighting processing on the single-institution determination result information of each target medical institution according to the geographical proximity matrix;

[0145] Using the weighted single-institution determination result information to generate the initial basic confidence distribution.

[0146] In specific implementation, first, address data of each target medical institution is collected in the deployment phase, and latitude and longitude coordinate conversion is performed on the address by using third-party geographic analysis services, such as offline API of Gaode Map, Baidu Map or OpenStreetMap. The present application writes the latitude and longitude results together with the confidence score returned by the analysis into the institution position information database, and simultaneously registers auxiliary fields such as hospital code, level, administrative division code in the entry.

[0147] In the running phase, when the system needs to calculate the geographical proximity of two medical institutions, the corresponding coordinates are searched from the position information database, and the spatial distance between the two is calculated by a predetermined engineering distance algorithm, such as spherical distance method or city-level road network query. If the distance is obviously less than a certain kilometer threshold and they belong to the same city-level administrative area, the system will determine them as high proximity.

[0148] Multiple distance bins can be set in the configuration file, for example, proximity within 5 km is determined as 0.9 or above, 5-30 km can be between 0.8 and 0.5, and 30 km or above is reduced to 0.4 or below. For institutions that have failed address resolution or have too low resolution confidence, the system will invoke a set of predefined fuzzy positioning schemes, such as using the coordinates of the city or county center to replace the true location, and automatically reducing the offset score by 0.1-0.2 when calculating proximity to reflect the positioning uncertainty.

[0149] After obtaining the proximity score for all pairs of institutions, the application generates a geographic proximity matrix, with rows and columns corresponding to the unique identifiers of the institutions, and the matrix elements recording the proximity of the two institutions. In order to support fast query in subsequent steps, the application can save this matrix in a distributed cache or a relational database, indexed by "institution pair" to "proximity score", to ensure that the proximity of any two institutions can be quickly retrieved even in a large medical community.

[0150] Before the application synthesizes the evidence of the single-institution determination result information of each target medical institution, it reads the proximity of the institutions involved in each determination from the proximity matrix and adjusts the weight according to the actual location relationship between the institutions. The specific approach includes: if the proximity between a certain institution and its neighbor institutions in the matrix is in the high segment, for example, 0.85 or above, the weight corresponding to the determination entry can be multiplied by an amplification coefficient, for example, it can be increased from 1.0 to 1.2; if the proximity is in the medium segment, for example, 0.6-0.7, the original weight is kept unchanged; if the proximity is lower than 0.4, the weight of the corresponding entry is set to 0.8, and if the proximity is further reduced, the weight can be further reduced, for example, to 0.6 or lower. The application updates these calculated weights in the single-institution determination result information, and then generates the initial basic confidence allocation based on this. In implementation, a field can be added to the determination entry structure to store the maximum weight after proximity weighting.

[0151] In actual scenarios, if a township health center often shares data with several health centers in the same county through the same VPN and a central hospital, the proximity between them should be evaluated as high. If these institutions collectively expose high missing rates or delay anomalies within the same time window, the application will make their determination entries have higher influence in the subsequent synthesis stage, thereby improving the ability to identify regional failures or network bottlenecks. Conversely, if a specialized hospital is too far away from other institutions or does not use a unified network, it will not have a strong impact on the evidence information of these institutions.

[0152] Through this clear engineering process, the application can flexibly incorporate geographical location differences into the weight calculation of single institution judgment results, providing more realistic distribution characteristic-based data support for subsequent evidence synthesis and trend early warning.

[0153] As an optional implementation, the generating a geographical proximity matrix based on geographical location information between each target medical institution comprises:

[0154] Performing geographical analysis processing on the address information of each target medical institution to obtain corresponding institution location information parameters;

[0155] When the address of any target medical institution cannot be successfully parsed or the confidence of the corresponding institution location information parameter is lower than a preset threshold, the institution location information parameter is supplemented based on a predefined auxiliary location information generation rule;

[0156] Based on the institution location information parameter, the geographical proximity between any two target medical institutions is calculated, and the geographical proximity matrix is constructed.

[0157] When generating the geographical proximity matrix, the application first performs geographical analysis processing on the address information of each target medical institution to obtain the institution location information parameter. Common address parsing services on the market can be used, such as submitting the registered address of a hospital, clinic or health center to a third-party geographic coding API to obtain latitude and longitude coordinates, administrative division level information and additional data such as parsing confidence identification. The parsing confidence is usually expressed in numerical percentage or grading level, reflecting the reliability of the address reverse solution by the geographic service.

[0158] In some cases, the address of the target medical institution may not be standardized or may not be successfully identified by the geographic service, or the returned confidence may be lower than the threshold set when the application is deployed. To deal with such situations where the accurate latitude and longitude cannot be directly obtained, the application sets predefined auxiliary location information generation rules for supplementation.

[0159] For example, the auxiliary information can be a pre-manually entered administrative division correspondence table, or a hospital level and county code mapping relationship maintained by the medical community management system. If the system returns a "no hit" or "confidence too low" prompt at the geographic service interface, it will directly refer to the correspondence table or code mapping file to perform fuzzy positioning on the address, such as assigning the unparsable township health center to the center coordinates of the county-level administrative area it is in, or combining the hospital level with the city ID it is in to estimate a substitute coordinate for subsequent proximity calculation.

[0160] After generating or completing the institution location information parameters, the application calculates the geographic proximity between any two target medical institutions based on these latitude and longitude or alternative coordinates. To better reflect the regional distribution characteristics in the medical community, the system usually adopts several combination strategies. The latitude and longitude of the two medical institutions can be input into the spherical distance algorithm or the apparent distance library of the urban road network. Exemplary methods include: if the institutions are located in the same city, the city partition weight is used to reduce the impact of the distance; if the institutions are across provinces or more than tens of kilometers apart, the proximity is set to a relatively low value.

[0161] The grading rules can also be defined, for example, within five kilometers is marked as "high proximity", corresponding to a proximity of 0.8 to 1.0, and more than thirty kilometers is marked as "low proximity", corresponding to a proximity of 0.2 to 0.4. The specific interval can be configured by the operation and maintenance personnel during the deployment phase.

[0162] The final geographic proximity matrix corresponds to the medical institution identifier, and the matrix element value represents the proximity score of the two institutions. If the address of some institutions is completed by auxiliary rules, the row and column in the matrix will have a special mark or weight adjustment, for example, an uncertainty coefficient can be added to the completed coordinates to slightly reduce the proximity score, so as to avoid overestimating the impact of the guessed location.

[0163] After construction, the geographic proximity matrix can be further used for proximity weighting or regional aggregation analysis. When several institutions are closely distributed in geographic space and have high missing rate or format error rate at the same time, the system will first suspect whether they belong to the same network node or the same operator line, and share similar fault risks; if the location of an individual institution in the matrix is significantly different from the main cluster, the evidence coupling strength between it and the main cluster can be reduced in subsequent evidence synthesis. Through this engineered address resolution, auxiliary completion, and graded proximity calculation, the application can more accurately reflect the real relationship of each medical institution in the region, providing a stable and reproducible basis for subsequent regional anomaly recognition and evidence weight adjustment.

[0164] As an optional implementation, based on the institution location information parameters, calculating the geographic proximity between any two target medical institutions further comprises:

[0165] In a third predetermined period, collecting business interaction logs of each target medical institution, the business interaction logs at least recording interaction timestamp, interaction counterpart identifier, interaction type identifier, and interaction data volume;

[0166] For any two target medical institutions, statistics the number of interactions and interaction data volume in the third predetermined period;

[0167] According to the interaction times, the interaction data amount and the geographical proximity, weight adjustment is performed on the corresponding proximity value to obtain an updated proximity value;

[0168] Based on the updated proximity value, the geographical proximity matrix is constructed.

[0169] The present application not only calculates geographical proximity based on the location information of the target medical institutions, but also collects and analyzes business interaction logs within a third preset period, superimposes the interaction times and the interaction data amount on the original geographical proximity, and presents a more practical application scenario in the final geographical proximity matrix.

[0170] The present application first configures a third preset period during deployment, which is used to periodically pull the business interaction logs between the target medical institutions. An exemplary third preset period can be daily or weekly, depending on the evaluation of the cross-institution business collaboration speed by the operation and maintenance personnel. The business interaction logs at least contain interaction timestamps, interaction counterpart identifiers, interaction type identifiers and interaction data amounts, which are used to depict the data transmission or service call relationship between different medical institutions within the period. Common big data or log collection tools can be used for collection, such as recording traffic statistics at the edge nodes or VPN gateways of each institution, and then pushing the logs to the central node in batches through a log agent.

[0171] After receiving the business interaction logs, the present application performs aggregation and statistics for any two target medical institutions, such as accumulating the interaction times and the total interaction data amount within the period in the form of connection pairs or traffic pairs. If the interaction between two hospitals or health centers is significantly more frequent or has a larger data amount, the system can consider it as having a higher coupling degree in terms of business and network, and thus give additional weight in the proximity. To prevent extreme cases from interfering, the operation and maintenance personnel can configure the interaction amount threshold and the number of times threshold upper and lower limits during the deployment stage, such as setting less than ten times of interaction or less than 100MB of interaction data within each period as low interaction, and more than several GB or hundreds of connections as high-intensity interaction, and mapping these intervals and weight rules in the system configuration file.

[0172] After the statistics are completed, the system first compares the interaction results in parallel according to the previously calculated geographical proximity. If the geographical proximity is high but the actual interaction frequency is low, it means that the two institutions are physically close, but they may not be on the same medical service link or network topology; if the geographical proximity is moderate but the interaction times and data amount are significantly higher than the normal level, it means that the two institutions are geographically relatively dispersed, but they are closely cooperated in data, such as sharing a testing center or an image.

[0173] According to the above differences, the present application adjusts the proximity by weighting. The specific implementation can use simple multiplication coefficients or segment translation methods.

[0174] For example, if the original geographical proximity is 0.7, but the two institutions have a large amount of interaction in the third preset period, the final proximity can be increased from 0.7 to 0.85; if the physical proximity is 0.9, but the business interaction is almost zero, the proximity can be reduced to 0.75. In order to allow the operation and maintenance personnel to configure flexibly, these weighting rules can be stored in the form of "(interaction intensity interval, geographical proximity interval) to adjustment coefficient" in the database or configuration file, and dynamically loaded during program running.

[0175] Finally, the system uses these proximity values adjusted by weight to reconstruct or update the geographical proximity matrix. Compared with the proximity based on address resolution alone, the matrix at this time can better reflect the real distance of the medical community in the aspects of network bandwidth sharing, business cooperation and service calling, making the subsequent detection of regional failures or local anomalies more effective. For example, if a group of medical institutions with relatively close geographical positions present a high interaction due to long-term business cooperation, they will be evaluated as closer in the final proximity matrix. The system can identify that "these institutions constitute a business chain group" before synthesis determination, and it is easier to capture the high loss rate or delay phenomenon that occurs synchronously in the group.

[0176] This optional implementation mainly solves the "one-size-fits-all" problem that may be caused by relying only on geographical position information. Some institutions with relatively close geographical distances do not necessarily have frequent data interaction, and some institutions with relatively far distances have closer business interaction due to specialty cooperation or unified image cloud. By pulling the business interaction log in the third preset period and combining it with the geographical proximity for weighting, the present application realizes a more real and dynamic similarity evaluation between institutions at the engineering level, further improving the detection accuracy of subsequent trend anomalies and the recognition ability of cross-regional cooperation anomalies.

[0177] For example, first, a set of log collection or traffic probe tools is deployed to record connection handshake and data packet statistics on the VPN tunnel or message bus between institutions.

[0178] Then, the third preset period scheduler runs the summary script once an hour or daily, and stores the summary results to the interaction statistics table. The interaction statistics table can include fields such as "source institution identifier, target institution identifier, interaction type, total number, total data volume, sampling time range".

[0179] Next, the system references the original geographical proximity value for each pair of institutions, reads the interaction statistics table, and adjusts the proximity by combining preset intervals or function formulas, for example, floating up and down in the range of 0.6-1.0.

[0180] Finally, the updated proximity values are written back to the proximity matrix, and the query interface is opened to subsequent evidence synthesis or clustering detection services. This flexible and dynamic weight adjustment mechanism ensures that the medical community can maintain high regional anomaly recognition accuracy and trend warning effect when the network deployment frequently changes or new collaborative hospitals join.

[0181] As an optional implementation, the stacking of the time series data of each target medical institution according to the institution dimension to construct a joint feature matrix includes:

[0182] Based on the geographical proximity matrix, target medical institutions with a geographical proximity higher than a preset grouping threshold are divided into the same proximity cluster according to a proximity aggregation rule;

[0183] For each proximity cluster, the anomaly count of the target medical institutions in the cluster is counted at each time series index, and a regional synchronous anomaly label is generated when the anomaly count reaches a preset cluster threshold;

[0184] When the regional synchronous anomaly label exists, the corresponding time series data is processed by column-level merging according to a preset merging rule in a cluster unit to obtain a cluster-level representative data column;

[0185] The cluster-level representative data column and the time series data column of the target medical institutions outside the cluster are stacked according to the institution dimension and filled into the joint feature matrix.

[0186] When stacking multi-institution data for the same data quality indicator, the present application can further divide proximity clusters according to the geographical proximity matrix and perform column-level merging processing on the data of each target medical institution. In this way, "regional synchronous anomalies" can be identified before merging, thereby avoiding fitting bias or matrix rank deficiency problems caused by completely similar multi-columns, and providing more accurate and controllable input for subsequent trend analysis in the "joint feature matrix". This optional implementation can effectively solve the problem of "completely linearly correlated multi-columns" caused by simply stacking when several geographically close medical institutions with similar network environments synchronously occur anomalies in the same time window.

[0187] In a specific implementation, after reading the geographical proximity matrix, the present application performs grouping operation according to the proximity aggregation rule.

[0188] Exemplarily, if the proximity is higher than 0.8, it is considered as the same proximity cluster; if it is between 0.5 and 0.8, it is processed separately; and if it is lower than 0.5, it is considered as basically no direct correlation. Multiple proximity threshold values can be set in the configuration file or database for adapting to medical communities of different scales and geographical distribution. The system centrally processes the target medical institutions in the same cluster to facilitate the identification of the possibility of "regional cluster abnormality". If there is a deployment requirement, common clustering algorithms such as hierarchical clustering or DBSCAN can be used to automatically cluster the geographical proximity matrix, and these algorithms can be used in tools such as Python Scikit-learn or SparkMLlib.

[0189] For each divided proximity cluster, the application counts the abnormality count of the target medical institutions in the cluster at each time index. The abnormality count can be implemented according to predefined judgment conditions, such as a missing rate exceeding 15%, a format error rate higher than 3%, or a collection delay exceeding 30 minutes. If multiple institutions in the cluster trigger these thresholds at the same time point or in the same time slice, the abnormality is counted once. The system generates a "regional synchronous abnormality marker" when the cumulative value reaches a preset cluster threshold, indicating that the proximity cluster is in a synchronous deterioration state within the time index range. The cluster threshold can be set to 50% or 60% or more of the total number of institutions in the cluster, and the specific value is determined by the actual cluster size in the medical community, which can be optimized by the operation and maintenance personnel through the configuration file during the deployment phase.

[0190] When the system detects that a proximity cluster has triggered a regional synchronous abnormality marker, it invokes a preset merging rule to perform further column-level merging processing on the time series data column corresponding to the cluster. For example, in column-level merging, the average or median of multiple columns in the cluster can be calculated, or the maximum or minimum value can be selected as the "cluster-level representative", depending on the needs of the business operation and maintenance scenario for abnormal sensitivity. Exemplarily, if the maximum value is selected to highlight the conservative risk, it can more sensitively capture the deterioration situation; if the average or median value is selected to estimate the stability, it can weaken the influence of extreme values on the global. After the merging is implemented, a new "cluster-level representative data column" is formed, which replaces the original multiple columns or adds a new column in the matrix to represent the data of the cluster, avoiding redundant calculation of multiple highly overlapping columns in subsequent multivariate fitting.

[0191] Finally, the cluster-level representative data column and the time series data column of the target medical institutions outside the cluster are stacked according to the institution dimension and filled into the joint feature matrix. For those institutions that are not in the same cluster or have not reached the synchronous abnormality marker, they still maintain their independent columns.

[0192] The advantages of this approach are that it can improve the ability to identify regional synchronization anomalies and avoid the formation of too many row and column approximations, reducing the redundancy in the fitting process. This approach can achieve good efficiency and accuracy in practical deployment. If there are dozens of institutions in a medical community that are closely adjacent and share the same network, and all of them have synchronization error rate increases at the same time, if not combined, there will be many almost identical columns in the joint feature matrix. Through this method, the strongly related columns can be aggregated and the representative values can be written into a single column. Subsequent multivariate fitting performs a trend operation on the combined column, which can accurately identify regional fluctuations and significantly reduce algorithmic computation and storage overhead.

[0193] For example, the time series data can be clustered and counted using Python scripts or Spark SQL: find all pairs of institutions with a proximity of ≥0.8 in the proximity table and construct an "adjacent cluster" index. For each institution in the adjacent cluster, traverse the time series data table and count the number of abnormal institutions in the cluster at any time t. If the number exceeds a certain percentage threshold of the cluster size, such as 50%, a "regional synchronization anomaly flag" is generated at time t. If a synchronization anomaly flag is present, the columns in the cluster are combined: the "column average" or "column maximum" strategy can be used to write the result to a new cluster representative column, and the original multiple columns can be discarded or kept as reference columns. The final stacking is performed with other institution columns that do not trigger synchronization anomalies to form a new joint feature matrix.

[0194] This cluster-based merging process mainly solves the problem of severe multicollinearity caused by simultaneous anomalies in clustered institutions in a large-scale medical community, and can also enhance the analysis capability of the regional dimension. If it is indeed necessary to retain fine-grained information, the original columns and representative columns can be kept in parallel, and the fitting stage can be set to reduce the weight or screen the flag, with flexibility relying on system configuration file control. In summary, this optional implementation can enable the medical community to focus on regional aggregation failures in monitoring, reduce resource waste from repeated analysis of highly homogeneous data, and improve the timeliness and accuracy of trend analysis.

[0195] As an optional implementation, stacking the cluster-level representative data column and the time series data column of the target medical institution outside the cluster according to the institution dimension to fill the joint feature matrix includes:

[0196] For each adjacent cluster, perform difference processing on the time series data column of the member institutions in the cluster using the cluster-level representative data column to generate a corresponding residual data column;

[0197] Stack the cluster-level representative data column and each residual data column according to the institution dimension, and combine with the time series data column of the target medical institution outside the cluster to fill the joint feature matrix.

[0198] In an optional embodiment of this application, before stacking the cluster-level representative data column with the time-series data column of the target medical institution outside the cluster, the time-series data column of the member institutions within the cluster is first differentially processed to generate a residual data column. This can effectively solve the matrix rank loss problem caused by high correlation of multiple columns, while retaining the independent deviation information of each member institution relative to the cluster-level representative.

[0199] This application first obtains a cluster-level representative data column for each neighboring cluster. This representative column can be a "cluster-level" time series trajectory generated by the mean, maximum, or other summarization methods, used to summarize the average or extreme value trend of the cluster as a whole. For the time series data column of each member institution within the cluster, this application subtracts the corresponding time value of the representative column point by point, obtaining a residual column. For example, if the representative column records the average missing rate of the cluster under each time index, and the missing rate of a certain institution is significantly higher than the average at the same index, the residual is positive, and its value reflects the specific magnitude by which the missing rate of that institution is higher than the cluster-level representative value at that time. Through this differencing process, the complete linear dependency between the member column and the representative column can be eliminated, reducing the impact of collinearity problems in subsequent multivariate fitting.

[0200] In practical implementation, the system can use distributed data processing tools, such as Spark DataFrame or batch update scripts in traditional databases, to perform row-by-row differencing operations on each column within a cluster. If the representative column is stored in a "cluster_col" field and the member columns are stored in an "institution_col" field, this application can calculate "residual_col = institution_col – cluster_col" and write the result to the same table or a newly created table. To avoid unnecessary growth of long columns, the system can perform the differencing only once within a short window, store the differencing result and the representative column after the calculation is complete, and then release the temporary table to save storage space.

[0201] Subsequently, this application stacks the cluster-level representative data column and each residual data column along the institution dimension, and then merges them with the time-series data columns of other institutions outside the cluster to form a new joint feature matrix. In this matrix, the cluster-level representative column reflects the overall regional trend, while each residual column shows the degree of independent deviation of each institution relative to this trend. Since the linear correlation coefficient between the residual column and the representative column is theoretically significantly reduced, the matrix has higher discernibility, and the fitting algorithm can avoid secondary convergence on the repeated column information of institutions within the same cluster when capturing overall fluctuations.

[0202] This method can solve the problem of "multiple geographical proximity and almost synchronous trend" in large-scale medical community. If the multiple columns of extremely similar data are directly maintained, it is easy to cause problems such as unidentifiable coefficients and unstable model values in subsequent multiple regression or iterative fitting. By generating cluster-level representative columns first and then generating residual columns by difference, not only the "overall trend of the region" and "individual agency difference" information is retained, but also the model will not cause rank deficiency due to multiple columns of the same value information during training or fitting. In other words, this difference-stacking strategy takes into account the detection requirements of global and local anomalies in engineering, which helps to quickly locate whether the "overall abnormality of the region" or the "individual agency deviation".

[0203] For example, for a neighboring cluster containing five community health centers, the present application first uses their average missing rate columns to form "cluster-level representative data columns", and then subtracts each agency column in turn to obtain five residual data. If the missing rate of a certain agency is often 2-3 percentage points higher than the average value, the residual column will show "positive deviation" at multiple times.

[0204] In the subsequent joint feature matrix, researchers or fitting algorithms can consider the representative column as the overall level of the cluster, or can understand the persistent error of a single agency compared to the cluster average through the residual column. In this way, it can be found more quickly whether the agency follows the market trend or presents individual problems during subsequent feature modeling or anomaly diagnosis.

[0205] In this way, the embodiment not only maintains the ability to summarize the regional-level trend, but also reduces the modeling difficulty caused by multiple columns of homogeneous information through difference, so that the system can more efficiently and accurately identify local and overall trend problems in the medical community based on the joint feature matrix.

[0206] As an optional embodiment, a multivariate fitting algorithm is executed based on the joint feature matrix to obtain trend change parameters of each target medical agency, including:

[0207] The joint feature matrix is partitioned according to a preset sliding time window to obtain a plurality of window feature sub-matrices;

[0208] For each window feature sub-matrix, a weighted regression fitting is performed to obtain a corresponding candidate trend parameter and a residual confidence value;

[0209] The candidate trend parameters and residual confidence values of each window are used as multi-source evidence information to construct a window-level basic confidence distribution;

[0210] The window-level basic confidence distribution is subjected to evidence theory synthesis processing, and the weight of conflicting evidence information is dynamically adjusted when the conflict degree exceeds a preset range to generate comprehensive trend change parameters of each target medical agency.

[0211] Further, the application partitions the joint feature matrix by a preset sliding time window to capture dynamic evolution features more finely, and performs weighted regression fitting to obtain candidate trend parameters and residual confidence in each time window. Then, the application inputs these window results as multi-source evidence information into the evidence theory model, dynamically adjusts the weight of evidence information with excessively high conflict degree, and finally outputs the comprehensive trend change parameters of each target medical institution. This method is compatible with various data processing and machine learning frameworks in engineering, such as using pandas DataFrame or Spark RDD partition operation of Python to perform sliding window segmentation on the matrix, or using third-party regression modules such as statsmodels or scikit-learn to perform weighted fitting in each window.

[0212] In a specific implementation, the application first partitions the joint feature matrix by a preset sliding time window. An exemplary approach can set a window length and a step in the configuration file, such as stepping 10 time series data from 30 time series data to achieve sliding, thereby obtaining multiple “window feature sub-matrices”. If the medical community business cycle is long, it can also be divided by timestamp: the interval of each hour, day or week can be flexibly configured to adapt to different granularity of trend observation. In the medical community scene with large processing scale, the system can use a distributed computing engine to distribute the matrix to the worker nodes for parallel regression calculation after segmentation.

[0213] In each window feature sub-matrix, the application performs weighted regression fitting to obtain candidate trend parameters and residual confidence values. The weighting method can be based on the number of valid samples, coverage or abnormal proportion in the window, or can refer to the geographical proximity weighting result, which is integrated into the regression loss function. When the regression model converges, the application outputs a candidate trend parameter vector (for all medical institutions or all columns) and the corresponding residual confidence value. The residual confidence value can be used to measure to what extent the fitting effect of the window meets the real data. The higher the value, the better the model explains the current window data; if the residual confidence is low, there are more abnormal fluctuations or potential conflicts in the window.

[0214] After completing the regression of multiple windows, the application regards the candidate trend parameters and residual confidence generated by each window as multi-source evidence information, and constructs a window-level basic belief assignment. The specific approach includes: splitting the trend parameters into the support of “deterioration” and “normal” propositions, or mapping the parameter interval into different propositions according to the level, such as high slope, low slope or near zero, and then amplifying or reducing the confidence value of the corresponding proposition according to the residual confidence, and reserving the remaining amount that cannot be mapped in the uncertain proposition. In this way, a basic belief assignment set covering all windows is formed.

[0215] To fuse all window results, the application will multiply and accumulate the conflict degree of each basic belief assignment of the window level in the evidence theory synthesis processing. If the window belief assignment is significantly opposed to the mainstream result, resulting in a conflict degree exceeding the preset range, the system will automatically reduce the weight corresponding to the window.

[0216] For example, under the condition that the initial weight is 1, the weight can be reduced to 0.5 for a second synthesis when a conflict occurs, and if the conflict is still too large, it can be further reduced to 0.3. If the conflict remains high after multiple rounds of reduction, the system will regard the window result as a serious anomaly and allocate most of the probability to the uncertain proposition. After synthesis is completed, each target medical institution can obtain a comprehensive trend change parameter vector, which represents the overall trend estimation after aggregation across multiple windows.

[0217] This optional implementation mainly solves the problem that a single global fitting may be excessively disturbed by an abnormal window. In actual medical community operation, some time windows may be greatly different from the remaining windows due to network jitter, holiday business volume drop, or sudden equipment failure. If regression is directly performed on the full data, the extreme window will exert an illogical bias on the overall model. By synthesizing the candidate trend parameters and residual confidence of multiple windows through evidence theory, the application can dynamically adjust the window-level outlier results in engineering, retaining the time locality and suppressing the pollution of extreme windows to the global model. In this way, the final comprehensive trend change parameter can better represent the true status of most windows in the medical community, and the visual conflict degree or uncertain proposition allocation for individual abnormal windows is provided for subsequent auditing or operation and maintenance to be targeted for in-depth investigation.

[0218] Based on the same inventive concept, the disclosure embodiments also provide a medical community-oriented data management system corresponding to the medical community-oriented data management method. Since the system in the disclosure embodiments solves problems by similar principles to the above-mentioned medical community-oriented data management method in the disclosure embodiments, the implementation of the system can be referred to the implementation of the method, and the repeated parts will not be described again.

[0219] Referring to Figure 4 FIG. 1 shows a schematic diagram of a medical community-oriented data management system provided by an embodiment of the application, which includes:

[0220] The acquisition module 10 is configured to acquire the data acquisition task log of each target medical institution according to a first preset period. The data acquisition task log includes an acquisition task identifier, an acquisition time, an acquisition item quantity, an acquisition result, and an acquisition state.

[0221] The first processing module 20 is configured to count data quality indexes of each target medical institution based on the data collection task log, and generate time sequence data of each data quality index according to a second preset period, wherein the data quality indexes include a data missing rate, a format error rate and a data collection delay;

[0222] The second processing module 30 is configured to stack the time sequence data corresponding to each target medical institution according to an institution dimension for the same data quality index, construct a joint feature matrix, and perform a multivariate fitting algorithm based on the joint feature matrix to obtain a trend change parameter of each target medical institution.

[0223] The detection module 40 is configured to generate a degradation trend detection result according to a preset judgment condition for the trend change parameter, generate trend early warning information based on the degradation trend detection result, and output the trend early warning information to a management terminal.

[0224] Those skilled in the art can understand that the units and algorithm steps of the examples described in combination with the embodiments disclosed in the present application can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software manner depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

Claims

1. A data management method for medical consortia, characterized in that, include: According to the first preset cycle, acquire the data collection task logs of each target medical institution and the corresponding server node operation status logs; The server node's running status log includes: node identifier, node heartbeat information, node resource utilization, and node availability status; Based on the data acquisition task log and the server node's operation status log, the data quality indicators and server node operation status indicators of each target medical institution are statistically analyzed, and time-series data of each data quality indicator and operation status indicator are generated according to a second preset period; the operation status indicators include: node failure rate and abnormal resource occupancy rate. For the same data quality indicator, the time series data corresponding to each target medical institution are stacked according to the institution dimension to construct a joint feature matrix, and a multivariate fitting algorithm is executed based on the joint feature matrix to obtain the trend change parameters of each target medical institution. Based on the trend change parameters and the operating status indicators of the server nodes, a degradation trend detection result is generated according to preset judgment conditions; based on the degradation trend detection result, a trend warning message is generated and the trend warning message is output to the management terminal. It also includes: generating a geographic proximity matrix based on the geographic location information between each target medical institution; The construction of the joint feature matrix includes: based on the geographic proximity matrix, classifying target medical institutions with geographic proximity higher than a preset grouping threshold into the same neighborhood cluster according to the proximity aggregation rule; For each neighboring cluster, the anomaly count of the target medical institution within the cluster is counted at each time-series index, and a regional synchronization anomaly marker is generated when the anomaly count reaches a preset cluster threshold. When the regional synchronization anomaly flag exists, the corresponding time-series data is merged at the cluster level according to the preset merging rules to obtain a cluster-level representative data column. For each neighboring cluster, the cluster-level representative data column is used to perform differential processing on the time-series data column of the member institutions within the cluster to generate the corresponding residual data column; The cluster-level representative data column and each of the residual data columns are stacked according to the institution dimension, and merged with the time-series data column of the target medical institution outside the cluster to fill the joint feature matrix.

2. The data management method for medical consortia according to claim 1, characterized in that, The management terminal is also used for: It displays the real-time running status information of the corresponding server nodes of each target medical institution and the execution status information of the data collection tasks; The operational status information includes: node availability status, node resource utilization rate, and node fault indication; the execution status information includes the current task status, task execution progress, and task execution result. The management terminal is also used to: provide task management functions based on the execution status information of the data acquisition task and the running status information of the server node; the task management functions include task start / stop control, task priority adjustment and task retry strategy management.

3. The data management method for medical consortia according to claim 1, characterized in that, The data quality indicators include: data missing rate, format error rate, and data acquisition delay; the data acquisition task log includes: acquisition task identifier, acquisition time, number of acquisition items, acquisition results, and acquisition status. Based on the aforementioned trend change parameters and preset judgment conditions, a degradation trend detection result is generated, including: Based on the trend change parameters of each target medical institution, generate corresponding single-institution judgment result information; The single-institution judgment result information is used as multi-source evidence information input, and evidence theory is used for synthesis processing to obtain a comprehensive confidence result for the overall data collection situation; The degree of conflict of the comprehensive confidence result is calculated, and when the degree of conflict exceeds a preset range, the weight of the conflicting evidence information is adjusted to obtain the corrected comprehensive confidence result. By comparing the corrected overall confidence result with the preset confidence threshold, the abnormal confidence level at the medical consortium level is determined; The degradation trend detection results are generated based on the abnormal confidence level and the single-institution judgment results of each target medical institution.

4. The data management method for medical consortia according to claim 3, characterized in that, The comprehensive confidence result obtained for the overall data collection situation includes: Based on the individual assessment results of each target medical institution, an initial basic confidence allocation is constructed; Calculate the degree of conflict between the initial basic confidence assignments, and when the degree of conflict exceeds a preset range, adjust the weight of conflict evidence information to update the basic confidence assignments; An iterative synthesis process is performed on the updated basic confidence assignment to obtain a first synthesized confidence level and a second synthesized confidence level; wherein, the first synthesized confidence level is used to characterize that the data acquisition quality is in a deteriorating trend, and the second synthesized confidence level is used to characterize that the data acquisition quality is in a normal trend. Based on the first composite confidence level and the second composite confidence level, the uncertainty entropy value is determined; Based on the first composite confidence level, the second composite confidence level, and the uncertainty entropy value, a comprehensive confidence result is generated; The uncertainty entropy value is used to characterize the degree of consistency of the overall evidence of the medical consortium.

5. The data management method for medical consortia according to claim 4, characterized in that, Also includes: Based on the geographic proximity matrix, the single-institution determination results of each target medical institution are subjected to proximity-based weighted processing. The weighted single-institution determination results are used to generate the initial basic confidence assignment. The generation of the geographic proximity matrix based on the geographic location information between each target medical institution includes: Geographic parsing is performed on the address information of each target medical institution to obtain the corresponding institution location information parameters; When the address of any target medical institution cannot be successfully resolved or the confidence level of the corresponding institution location information parameter is lower than a preset threshold, the institution location information parameter is supplemented based on a predefined auxiliary location information generation rule. Based on the location information parameters of the institutions, the geographical proximity between any two target medical institutions is calculated, and the geographical proximity matrix is ​​constructed.

6. The data management method for medical consortia according to claim 5, characterized in that, After calculating the geographical proximity between any two target medical institutions based on the aforementioned institutional location information parameters, the method further includes: Within the third preset period, business interaction logs of each target medical institution are collected. The business interaction logs record at least the interaction timestamp, the identifier of the other party, the interaction type identifier, and the amount of interaction data. For any two target medical institutions, count the number of interactions and the amount of interaction data within the third preset period; Based on the number of interactions, the amount of interaction data, and the geographical proximity, the corresponding proximity value is weighted and adjusted to obtain the updated proximity value. Based on the updated proximity values, the geographic proximity matrix is ​​constructed.

7. The data management method for medical consortia according to claim 1, characterized in that, Based on the joint feature matrix, a multivariate fitting algorithm is executed to obtain the trend change parameters of each target medical institution, including: The joint feature matrix is ​​partitioned according to a preset sliding time window to obtain multiple window feature sub-matrices; For each of the aforementioned window feature submatrices, a weighted regression fitting is performed to obtain the corresponding candidate trend parameters and residual confidence values. The candidate trend parameters and residual confidence values ​​of each window are used as multi-source evidence information to construct a window-level basic confidence assignment. The window-level basic confidence assignment is processed by evidence theory synthesis, and the weight of conflicting evidence information is dynamically adjusted when the degree of conflict exceeds a preset range, generating comprehensive trend change parameters for each target medical institution.

8. A data management system for medical consortia, characterized in that: include: The data acquisition module is used to acquire data acquisition task logs and corresponding server node operation status logs for each target medical institution according to a first preset cycle. The server node's running status log includes: node identifier, node heartbeat information, node resource utilization, and node availability status; The first processing module is used to calculate the data quality indicators and server node operation status indicators of each target medical institution based on the data acquisition task log and the server node operation status log, and generate time-series data of each data quality indicator and operation status indicator according to a second preset period; the operation status indicators include: node failure rate and abnormal resource occupancy rate. The second processing module is used to stack the time series data corresponding to each target medical institution according to the institution dimension for the same data quality index, construct a joint feature matrix, and perform a multivariate fitting algorithm based on the joint feature matrix to obtain the trend change parameters of each target medical institution. The detection module is used to generate a degradation trend detection result based on the trend change parameters and the operating status indicators of the server node, according to preset judgment conditions; based on the degradation trend detection result, generate trend warning information, and output the trend warning information to the management terminal. It also includes: generating a geographic proximity matrix based on the geographic location information between each target medical institution; The construction of the joint feature matrix includes: based on the geographic proximity matrix, classifying target medical institutions with geographic proximity higher than a preset grouping threshold into the same neighborhood cluster according to the proximity aggregation rule; For each neighboring cluster, the anomaly count of the target medical institution within the cluster is counted at each time-series index, and a regional synchronization anomaly marker is generated when the anomaly count reaches a preset cluster threshold. When the regional synchronization anomaly flag exists, the corresponding time-series data is merged at the cluster level according to the preset merging rules to obtain a cluster-level representative data column. For each neighboring cluster, the cluster-level representative data column is used to perform differential processing on the time-series data column of the member institutions within the cluster to generate the corresponding residual data column; The cluster-level representative data column and each of the residual data columns are stacked according to the institution dimension, and merged with the time-series data column of the target medical institution outside the cluster to fill the joint feature matrix.

Images