Lupus nephritis condition monitoring system based on dynamic change of TWEAK and CD163

Abstract

The invention relates to the field of medical care informatics, and discloses a lupus nephritis condition monitoring system based on TWEAK and CD163 dynamic change, which comprises a data adaptation gateway module, a data rationality arbitration engine, a time sequence regularization engine module, a high-order feature extraction engine module and a risk layering module, the time sequence regularization engine module switches an interpolation or smoothing algorithm based on a qualitative result of the data rationality arbitration engine and a time domain where a data point is located so as to generate a continuous time function. Through the information processing architecture, the technical problem that sparse asynchronous original medical data is prone to being affected by outlier pollution and end point distortion is solved, and the accuracy of data processing is improved. The system reconstructs discrete data snapshots into continuous mathematical objects, provides a stable and analyzable technical basis for subsequent extraction of interpretable high-order dynamic features, and improves the credibility of clinical decision support.

Description

Technical Field

[0001] This invention relates to a lupus nephritis disease monitoring system based on the dynamic changes of TWEAK and CD163, belonging to the field of medical and health informatics technology. Background Technology

[0002] Current electronic health record systems and laboratory information systems are typically used in chronic disease management to store and display patients' biomarker test data long-term. For example, in the management of lupus nephritis, biomarkers such as TWEAK and CD163 are used for disease assessment. Currently, research and development for lupus nephritis focuses on two aspects: firstly, new therapeutic targets and drug development; for instance, Chinese invention patent CN117982510A discloses a new use for G-quadruplex DNA stabilizers, revealing that a G-quadruplex DNA stabilizer such as CX5461 can be used to prepare… Drugs for treating systemic lupus erythematosus or lupus nephritis are based on a drug development approach that uses a drug (G-quadruplex DNA stabilizer) to specifically downregulate the number of B cells, germinal center B cells, and plasma cells to alleviate kidney inflammation. However, this research and development path focuses on providing new treatments but fails to address the more prevalent and pressing informatics problem in clinical practice: regardless of the treatment regimen adopted by the patient, clinicians need a reliable information system to dynamically monitor and assess changes in the patient's condition, and existing information systems still have problems processing this monitoring data.

[0003] Existing information systems typically function as repositories for static data snapshots, passively recording and displaying past test results. In clinical practice, however, biomarker testing data often exhibits temporal sparsity, long testing cycles, and asynchronous testing dates for different biomarkers. Existing information processing methods present technical challenges: clinical decisions often rely on the dynamic trends of biomarkers over time, such as the rate or acceleration of ratio changes, rather than isolated data points. However, for raw data sequences composed of sparse, non-uniform time points, existing information systems struggle to obtain stable and effective dynamic derivatives through simple difference calculations. Direct calculations based on a few points are susceptible to distortion due to data noise. If complex artificial intelligence models are used for trend prediction, the internal calculation logic is opaque, and the output probabilistic results are difficult for clinicians to directly review and confirm.

[0004] Therefore, the technical problem to be solved by this invention is how to design an information processing system that can convert the input sparse, asynchronous detection data stream into a continuous time function that can be analyzed, extract high-order dynamic features with clear meaning from the function, and finally output risk indication information through an interpretable rule engine. Summary of the Invention

[0005] This invention provides a lupus nephritis disease monitoring system based on the dynamic changes of TWEAK and CD163. The main purpose is to solve the problem that existing information systems have difficulty converting sparse and asynchronous detection data into a continuous time function that can be analyzed, thus making it impossible to extract high-order dynamic features and conduct risk assessment through interpretable rules.

[0006] To achieve the above objectives, this invention provides a lupus nephritis disease monitoring system based on the dynamic changes of TWEAK and CD163. This system includes a data adaptation gateway module, a data rationality arbitration engine, a time-series regularization engine module, a high-order feature extraction engine module, and a risk stratification module. The data adaptation gateway module is used to obtain the patient's historical test data for subsequent evaluation from one or more laboratory information systems, which are sparse and asynchronous in time, including TWEAK and CD163 test data points. The data rationality arbitration engine is used to conduct a statistical rationality assessment of TWEAK and CD163 detection data points based on historical detection data before the time series regularization engine module processes them, and to classify them as reliable data or statistical outliers. The temporal regularization engine module is used to detect data points based on TWEAK and CD163, and generate corresponding continuous time functions for TWEAK and CD163 respectively. When generating any continuous time function, the temporal regularization engine module divides the TWEAK and CD163 detection data points into internal regions and endpoint regions; applies a first interpolation algorithm to the internal regions and a second low-order interpolation algorithm to the endpoint regions; and automatically switches to the smoothing algorithm in response to the data rationality arbitration engine's qualitative identification of statistical outliers of any TWEAK and CD163 detection data point. The high-order feature extraction engine module is used to compute high-order dynamic features based on continuous-time functions; The risk stratification module is used to generate risk indication information based on high-order dynamic features.

[0007] Preferably, the high-order feature extraction engine module calculates high-order dynamic features, including: calculating a ratio function based on the continuous-time function generated for TWEAK and the continuous-time function generated for CD163; calculating the first derivative of the ratio function to obtain the rate of change feature, and calculating the second derivative of the ratio function to obtain the acceleration feature. The risk stratification module is used to generate risk indication information based on the rate of change feature and the acceleration feature.

[0008] Preferably, the data rationality arbitration engine performs statistical rationality assessment, including: retrieving historical detection data and calculating the interquartile range of the historical detection data; constructing a dynamic rationality statistical window based on the interquartile range; comparing the values ​​of TWEAK and CD163 detection data points with the rationality statistical window and classifying them as reliable data or statistical outliers; wherein, the time series regularization engine module switches to execute a smoothing algorithm and performs smoothing spline operations.

[0009] Preferably, the temporal regularization engine module applies a first interpolation algorithm, including the application of a shape-preserving spline interpolation algorithm, to fit the data in the internal region; applies a second low-order interpolation algorithm, including the application of a local polynomial fitting algorithm, to fit the data in the endpoint region; and a high-order feature extraction engine module is used to calculate high-order dynamic features based on the function fragments of continuous-time functions in the endpoint regions generated by the second low-order interpolation algorithm.

[0010] Preferably, the data adaptation gateway module is also used to obtain the context information of medical order events associated with TWEAK and CD163 test data points from the hospital information system; the system also includes a context weight engine, which is used to assign information weights to TWEAK and CD163 test data points based on the context information of the medical order events, wherein the test data points corresponding to temporary medical orders are assigned higher information weights than the test data points corresponding to regular medical orders; and the time series regularization engine module incorporates information weights when applying the first interpolation algorithm, the second low-order interpolation algorithm, or the smoothing algorithm to perform weighted interpolation or weighted smoothing.

[0011] Preferably, the risk stratification module is also used to: monitor and calculate the data staleness time difference between the current system time and the timestamp of the most recent TWEAK and CD163 detection data points obtained by the data adaptation gateway module; when the data staleness time difference is less than a preset timeliness threshold, perform risk stratification based on high-order dynamic features; when the data staleness time difference exceeds the preset timeliness threshold, stop risk stratification and generate specific risk indication information indicating that the risk status is unknown and the data is stale.

[0012] Preferably, the high-order feature extraction engine module is also used to: calculate the cumulative inflammatory burden feature as the chronic injury feature by integrating the continuous time function within a preset long period time window while calculating the high-order dynamic feature as the acute risk feature; the risk stratification module is also used to: generate the first risk indication information based on the acute risk feature; and generate the second risk indication information based on the chronic injury feature.

[0013] Preferably, the risk stratification module includes a rule engine that compares the rate of change characteristics and acceleration of change characteristics with a set of clinical thresholds to determine risk indication information; wherein the rule engine includes at least a high-risk alert rule, which is defined as: when and When this occurs, a risk indication message is triggered; among which, Characterized by the rate of change. For changing acceleration characteristics, The preset rate threshold, This is the preset acceleration threshold.

[0014] Preferably, the data adaptation gateway module is used to: actively poll the laboratory information system at a preset frequency to pull TWEAK and CD163 test data points, convert the pulled TWEAK and CD163 test data points into internal standard events, and push them to the data rationality arbitration engine.

[0015] Preferably, the first interpolation algorithm used by the temporal regularization engine module is a form-preserving PCHIP interpolation algorithm.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. The information processing system provided by this invention, by setting a time-series regularization engine module, converts the time-sparse and asynchronous original detection data points obtained by the data adaptation gateway module into continuous time functions. This information processing method reconstructs discrete data snapshots into continuous mathematical objects in the healthcare informatics system, providing a stable and analyzable technical foundation for the subsequent high-order feature extraction engine module to perform dynamic feature calculations, enabling the system to extract dynamically changing features from the original non-uniform data.

[0017] 2. By linking and coordinating the temporal regularization engine, the high-order feature extraction engine, and the interpretable risk stratification module, an information processing pipeline with transparent computational logic is constructed. The risk indication information output by the system is generated through a path that includes continuous functions, rate of change, and comparison with clinical thresholds. All of these are based on explicit mathematical operations and expert rules, avoiding the uninterpretability caused by complex models. This enables medical personnel to understand and review the complete logical chain of risk judgment, thereby improving the credibility of the information system in clinical decision support.

[0018] 3. The information processing architecture constructed in this invention introduces data processing logic before the time-series regularization engine, such as statistical rationality arbitration or medical order event context weighting, enabling the core pipeline to dynamically preprocess the input data points. The system can automatically switch the working mode of the interpolation algorithm based on the historical data distribution, such as switching from interpolation to smoothing to attenuate the influence of outliers or assigning higher weights to specific data points based on the type of medical order to perform weighted interpolation. This collaborative design of the information processing flow enables the system to maintain the clinical relevance of the analysis results even when facing unavoidable data contamination or data heterogeneity in real medical environments. Attached Figure Description

[0019] Figure 1 This is a diagram illustrating the information processing architecture and data flow of the monitoring system of this invention. Figure 2 This is a statistical chart showing the arbitration results of the rationality of the TWEAK and CD163 data points in this invention; Figure 3 This is a flowchart illustrating the system operation mode and data timeliness management process of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of this invention are for illustrative purposes only and are not intended to limit the scope of protection of this invention.

[0021] This invention provides a lupus nephritis disease monitoring system based on the dynamic changes of TWEAK and CD163. The system constructs a serial information processing pipeline at the healthcare informatics level to address the sparsity, asynchronicity, and data contamination issues of clinical biomarker data. The system architecture includes: a data adaptation gateway module, a data rationality arbitration engine, a time-series regularization engine module, a high-order feature extraction engine module, and a risk stratification module. The system's information processing flow is as follows: the data adaptation gateway module acquires raw data points from heterogeneous laboratory information systems (LIS); the data rationality arbitration engine performs statistical evaluation on the new data to identify and label outliers; the time-series regularization engine module receives the evaluation results and reconstructs the discrete data point sequence into a continuous-time function; the high-order feature extraction engine module extracts dynamic features such as rate and acceleration based on this function; finally, the risk stratification module generates risk indication information based on the dynamic features. The data adaptation gateway module serves as the interface between the system and external healthcare information systems, used to acquire patients' TWEAK and CD163 test data points from one or more laboratory information systems (LIS), given the sparsity and asynchronicity of such test data. This module employs proactive polling logic to actively pull new test results from the LIS database at a preset frequency, such as every 5 minutes. After acquiring new data, the module standardizes it, encapsulating it into an internal standard event containing the patient ID, test item, value, unit, and test timestamp, and pushes it to the system's internal data bus. Simultaneously, this module also retrieves the patient's historical test data sequence for use by the data validity arbitration engine. In another implementation, this data adaptation gateway module is also used to interface with the hospital information system (HIS) for medical order entry. The system (CPOE) obtains contextual information about medical orders associated with each test data point, particularly distinguishing whether the order is a routine or ad hoc order. The data rationality arbitration engine, located between the data adaptation gateway module and the time-series regularization engine module, performs statistical rationality assessments on newly received test data points, identifying data contamination or entry errors. One implementation approach uses the interquartile range (IQR) method: when a new TWEAK or CD163 data point is received, the engine retrieves the corresponding historical test data sequence for that patient and calculates the first quartile of that historical sequence. ) and the third quartile ( ); Calculate the interquartile range And construct a dynamic rationality statistics window, the range of which is defined as The engine compares the value of the new data point with this statistical window. If the data point is within the window, it is classified as reliable data. If it is outside the window, it is classified as a statistical outlier. This classification result is sent to the time series regularization engine module along with the data point.

[0022] The system may also include a context weighting engine, which receives the medical order event context information obtained by the data adaptation gateway module. This engine assigns information weights to the data point based on preset rules. For example, data points generated based on routine medical orders, with weights The data points generated based on temporary medical orders are assigned a baseline value of 1.0, and their weights are... Information weights assigned higher than the baseline value, such as 1.8; this information weight... The data is passed to the temporal regularization engine module for weighted calculation; the temporal regularization engine module is used to generate corresponding continuous-time functions for TWEAK and CD163 based on the detected data points, i.e. and In its implementation, the engine employs a hybrid interpolation logic based on temporal domain partitioning: the engine divides the data point sequence into internal and endpoint regions; for the internal region, the engine applies a first interpolation algorithm, such as shape-preserving spline interpolation or PCHIP interpolation, to fit historical trends; for the endpoint region, the engine switches to a second low-order interpolation algorithm, such as robust local polynomial fitting, to provide a robust estimate of the current trend; subsequent high-order feature extraction is calculated based on function fragments from this endpoint region; furthermore, the temporal regularization engine module responds to the qualitative results of the data rationality arbitration engine: when the received data point is qualitatively identified as reliable data, the engine performs standard interpolation operations, causing the function curve to pass through the point; while when the data point is qualitatively identified as a statistical outlier, the engine automatically switches to a smoothing algorithm, such as smoothing spline operations, allowing the generated function to pass near the outlier while attenuating its influence, rather than forcibly passing through it; in addition, the temporal regularization engine module incorporates information weights assigned by the context weight engine when applying the first interpolation algorithm, the second low-order interpolation algorithm, or the smoothing algorithm. This is used to perform weighted interpolation or weighted smoothing, and this weighted calculation results in a continuous function. Prioritize fitting data points with higher information weights; the high-order feature extraction engine module receives the continuous-time function output by the temporal regularization engine module. and In this implementation, the engine calculates the ratio function based on two functions. The engine's ratio function Perform differentiation operations and calculate its first derivative to obtain the rate of change characteristics. ), and calculate its second derivative to obtain the characteristics of the changing acceleration ( ). ).

[0023] In another parallel implementation, the higher-order feature extraction engine module is also used to perform integration operations, through a comparison value function. Integration is performed within a preset long-term time window to calculate the cumulative inflammatory burden characteristic. This cumulative inflammatory burden characteristic, as a chronic injury characteristic, is output in parallel to the rate of change and acceleration of change characteristics, which are acute risk characteristics, and is then sent to the risk stratification module. The risk stratification module is used to generate risk indication information based on higher-order dynamic features. In this implementation, the module performs information timeliness verification: monitoring and calculating the current system time. The timestamp of the most recent detection data point obtained by the data adaptation gateway module Data staleness time difference This module has a built-in preset timeliness threshold, such as 45 days; when When this timeliness threshold is exceeded, the module will suspend risk stratification based on higher-order dynamic features and generate specific risk indication information indicating that the risk status is unknown and the data is outdated; only when When the timeliness threshold is less than this, the module performs regular risk stratification. This module includes a rules engine that receives acute risk characteristics, i.e., rate of change characteristics. With changing acceleration characteristics and chronic injury characteristics, i.e., cumulative inflammatory burden characteristics. For acute risks, the rules engine will and Compared to a set of preset clinical thresholds, a high-risk alert rule can be defined as: when the rate of change characteristic (Right now (greater than the preset rate threshold) And the changing acceleration characteristics (Right now (greater than the preset acceleration threshold) When a risk indication message is triggered, for chronic injuries, the rule engine will process it in parallel. With chronic threshold In comparison, when At that time, a second risk indication is generated, indicating the cumulative risk of chronic injury.

[0024] Example 1: In a long-term monitoring case of a patient with lupus nephritis, the patient's TWEAK and CD163 test data fluctuated around the baseline for a long period. In week 20, the data adaptation gateway module retrieved a new TWEAK test data point triggered by a routine medical order from the Laboratory Information System (LIS), with a value of 1500 pg / ml. In week 21, the LIS entered a new TWEAK test data point with a value of 8500 pg / ml, which was generated by a temporary visit to the patient and a temporary medical order issued by the doctor. In week 22, the LIS entered a TWEAK test data point with a value of 25000 pg / ml, which was associated with a routine medical order. The system's information processing pipeline was activated sequentially. The 1500 pg / ml data point from week 20 entered the data rationality arbitration engine. This engine retrieved historical data to calculate the IQR statistical window, determined that the data point was within the window, and classified it as reliable data. The context weight engine identified its routine medical order attribute and assigned information weights. The data point of 8500 pg / ml from week 21 was entered. The data rationality arbitration engine determined that this value was outside the IQR window and was therefore classified as a statistical outlier. At the same time, the context weighting engine identified the attributes of the temporary medical order and assigned information weights. Next, the 25,000 pg / ml data point from week 22 was entered, and the data rationality arbitration engine also determined it to be a statistical outlier; the context weight engine identified its routine medical order attributes and assigned information weights. .

[0025] The time series regularization engine module receives three data points with different qualitative results and different information weights, and generates a continuous-time function. At that time, a collaborative adaptive algorithm is executed: for data points in week 20 that are reliable, Perform standard weighted interpolation; for outliers in the data points of week 21, The engine switches to execute a weighted smoothing algorithm, due to the information weights The smoothing algorithm generates a function curve that is guided to approximate the 8500 pg / ml point with higher fidelity; for the data points in week 22 (outliers), The engine also executes a weighted smoothing algorithm, but due to information weights The smoothing algorithm assigns a lower confidence level to this 25000 pg / ml outlier, resulting in a smaller impact on the generated function curve and thus mitigating the distortion of the overall function shape caused by this data contamination. The higher-order feature extraction engine module is based on this weighted smoothed, robust continuous-time function. and the corresponding Calculate the ratio function Its derivative; since the function shape is mainly dominated by the high-weight temporary medical order point (8500 pg / ml) in week 21, while the influence of the low-weight outlier point (25000 pg / ml) in week 22 is suppressed, the system calculates a change rate feature greater than zero. And the same positive zero acceleration characteristic The rule engine of the risk stratification module receives these two feature values ​​and compares them with a preset rate threshold. and acceleration threshold Comparison, final judgment and This triggers and outputs risk indication information for high-risk alerts.

[0026] Example 2: To objectively verify the robustness of the system of the present invention in information processing under typical data challenges in healthcare informatics, this example constructs a semi-synthetic benchmark test. This test defines a known mathematical function for the true trajectory of a patient's condition. Starting from this function, a set of TWEAK and CD163 detection data point sequences, simulating common clinical practice data, are generated, exhibiting sparsity, asynchronicity, and data contamination. Finally, this contaminated data sequence is input into the system of the present invention and two control system systems, respectively. The performance of the system is quantitatively evaluated by comparing the error between the higher-order dynamic features calculated by each system and the true derivative value of the true trajectory function of the patient's condition. The experimental platform and data generation procedure are as follows: The ratio function of the true trajectory of the patient's condition is set. ,in The function represents a constant positive acceleration (time unit: weeks). The process involves generating a sparse and asynchronous set of TWEAK and CD163 sampling points over a time span of 0 to 10 weeks, based on this function, and calculating the corresponding ratio. Meanwhile, in week 8 ( When artificially introducing deviations from the true value ( Statistical outliers Simulated detection or input errors; the final generated sparse contamination data sequence ( , The values ​​are: (0, 1.02), (3, 1.32), (5, 1.55), (8, 4.50), (10, 2.51). The objective of the experiment is to evaluate the performance of each system at the current time. During the period, the characteristics of the changing acceleration ( Right now The accuracy of the calculation is 0.02, and the actual value is 0.02.

[0027] The experiment consisted of three treatment groups: Control Group A (existing technology simulation): This system did not include a data rationality arbitration engine or endpoint region algorithm, and used a standard spline interpolation algorithm to attempt to fit all five data points, including the outlier in week 8; Control Group B (partial features): This system included a data rationality arbitration engine but did not include a second low-order interpolation algorithm for endpoint regions; The IQR evaluation was used to assess... Points identified as statistical outliers are removed from the data sequence. Then, standard spline interpolation is applied to the remaining four data points, and calculations are performed. The second derivative at the point; Sample of this invention (complete solution): This system adopts a complete information processing pipeline, and the data rationality arbitration engine will... The point was identified as a statistical outlier. Upon receiving this qualitative result, the time-series regularization engine module automatically switches to a smoothing algorithm to attenuate its impact when processing the point. Simultaneously, during the calculation... When the endpoint region is located, the robust local polynomial fitting (LOESS) algorithm (i.e., the second low-order interpolation algorithm) is used to calculate the derivative. The processing results of each processing group on the same set of sparse contaminated data sequences are shown in Table 1.

[0028] Table 1: Characteristics of each system in response to changing acceleration ( Comparison table of calculation results

[0029] Experimental data showed that control group A, due to forced fitting, This statistical outlier distorts the shape of the generated continuous function. The calculated acceleration value was -0.118, which is consistent with the actual trajectory of the illness. The trend is opposite to that of the control group B; the control group B removes outliers through the data rationality arbitration engine to avoid trend errors, but the standard spline interpolation used shows instability in the endpoint region of sparse data, and the calculated acceleration value is 0.093 with a relative error of 365%, indicating a calculation bias; the sample group of this invention, through the collaboration of the data rationality arbitration engine and the time series regularization engine module, applies a smoothing algorithm to attenuate the influence of outliers on the one hand, and applies a second low-order interpolation algorithm that is mathematically more robust to endpoint data on the other hand. Its calculated acceleration value is 0.024 with a relative error of only 20%. At the information processing level, the calculation result is closest to the second derivative value of the actual trajectory of the disease.

[0030] Example 3: This example combines Figures 1 to 3 A description of the lupus nephritis disease monitoring system based on TWEAK and CD163 dynamic changes, such as... Figure 1As shown, this architecture obtains test data points and medical order event contexts from external Laboratory Information System (LIS) and Hospital Information System (HIS). The data adaptation gateway module is responsible for acquiring sparse, asynchronous TWEAK and CD163 data points and medical order contexts, and passing the test data points and historical data to the data rationality arbitration engine. This engine classifies the test data points as reliable data or statistical outliers based on historical data. At the same time, it passes the medical order event context to the context weight engine, which assigns information weights based on temporary or regular medical orders. The time series regularization engine module receives the test data points, qualitative results, and information weights to reconstruct a continuous time function and adaptively switches between interpolation or smoothing algorithms. The generated continuous time function is input to the high-order feature extraction engine module, which calculates high-order dynamic features, such as rate of change and acceleration of change. Finally, the risk stratification module generates risk indication information, such as high-risk alarms or unknown status, based on the calculated high-order dynamic features and rules, combined with data staleness verification.

[0031] like Figure 2 As shown in the figure, there are two types of data: reliable data and statistical outliers. The Y-axis represents the number of data points, with a scale range from 0 to 80. The X-axis distinguishes between TWEAK data points and CD163 data points. In the TWEAK data point category, the chart shows that there are 72 reliable data points and 8 statistical outliers. For the CD163 data point category, there are 68 reliable data points and 6 statistical outliers. From the visual comparison of the bar heights, it is clear that the number of reliable data points is higher than the number of statistical outliers in both TWEAK and CD163, indicating that outliers are a minority in the dataset.

[0032] like Figure 3 As shown, the system enters the initial running state when monitoring new patients to process the first N data points. After acquiring the N+1th data point, the system migrates to the steady-state monitoring state to perform routine risk stratification. In the steady-state monitoring state, if the timeliness of the acquired new data point is less than the threshold, the system remains in this state. If the data staleness is greater than the threshold, the system migrates to the unknown risk state. At this time, stratification will be stopped and an staleness prompt will be generated. When a new data point is acquired from the unknown risk state, the system returns to the steady-state monitoring state. In addition, the system can also migrate from the steady-state monitoring state to the stopped monitoring state.

[0033] Example 4: After the disease monitoring system of the present invention is deployed to the medical institution's information system in the information architecture, the key operating parameters in the risk stratification module can be calibrated offline to adapt it to the historical data distribution and clinical workflow of a specific institution. This calibration process utilizes the institution's existing anonymized patient cohort containing historical TWEAK and CD163 test data. The calibration process targets thresholds for acute risk characteristics. and The system selected two subsets of patients from the historical anonymized queue: one was the acute relapse group, which was identified at a specific time point. Within 7 days, patients were clinically diagnosed with acute exacerbation of lupus nephritis. Another group served as a stable control group, with no records of acute exacerbation during the same period. The system's information processing pipeline retrospectively processed all patients in the acute relapse group. Within the previous time window, such as Heavenly The sparse asynchronous detection data during the period is used, and a temporal regularization engine and a high-order feature extraction engine module are applied to calculate its value in... Rate of change characteristics within adjacent windows With changing acceleration characteristics The distribution; the same calculations were performed on the stable control group, and then, by comparing the two groups of patients... and Statistical analysis of the characteristic distribution can be performed using receiver operating characteristic (ROC) curve analysis to determine the threshold that balances the predictive sensitivity and specificity for the acute relapse group. In the calibration of this embodiment, the rate threshold... The acceleration threshold was determined to be 0.085 units per cycle. The value was set at 0.012 units / week². During the same calibration process, the timestamps of the test data of patients in the acute relapse group were analyzed, and the time of their most recent test was calculated. Time of acute recurrence Data staleness time difference Statistics show that the 95th percentile The value is 45 days; therefore, the preset timeliness threshold in the risk stratification module is [not specified]. It was marked as 45 days.

[0034] Finally, regarding the information weights in the context weight engine To perform optimization; the system uses Based on this, the temporary medical orders are iteratively adjusted in steps of 0.2 within the range of 1.2 to 2.2. Values ​​are assigned, and new weights are applied after each iteration. The entire historical anonymized cohort was reprocessed, and its predictive accuracy for the acute relapse group was calculated and evaluated. The available evaluation metric was the F1-score. The calibration results showed that when… At that time, the F1-score was 0.869. At that time, the F1-score was 0.874. At that time, the F1-score was 0.871; based on this optimization result, The information weight determined as a temporary medical order is used by the data rationality arbitration engine to construct the dynamic rationality statistics window. When, the multiplier The value is determined in the offline calibration procedure, which retrieves a historical anonymized queue. The test data points in this queue have been manually labeled as clinically reliable data or known data entry errors. The system iterates within a specific interval of 1.0 to 3.0 with a preset step size of 0.1. The value is calculated, and for each iteration, the F1-score that distinguishes known input errors is calculated. Finally, the value corresponding to the maximum F1-score is generated. The value is fixed as the operating parameter of the data rationality arbitration engine, and the preset long-period time window on which the high-order feature extraction engine module calculates the cumulative inflammatory burden feature, the specific duration of which is... The offline calibration procedure was defined to retrieve a historical anonymized cohort containing long-term follow-up results, including the two-year eGFR change rate or the chronic renal pathology index. The system used a set of candidate duration values ​​(3 months, 6 months, 9 months, and 12 months) to calculate the cumulative inflammatory burden characteristics of all patients in the cohort. The Pearson correlation coefficient between each set of characteristics and the long-term follow-up results was calculated, and the candidate duration value that resulted in the largest absolute value of the correlation coefficient was selected as the [specific value]. The fixed operating parameters are determined by executing the above offline calibration procedure. The information processing pipeline of the monitoring system, including the data rationality arbitration engine, the time series regularization engine module and the risk layering module, are monitored. All key parameters are obtained from the specific historical data distribution, which are used to complete the initialization of the system in the deployment environment.

[0035] Example 5: In the information processing environment of the disease monitoring system of the present invention, when establishing a monitoring file for a newly enrolled lupus nephritis patient, the patient's historical test data sequence is empty. At this time, the system's information processing pipeline automatically enters the initialization operation procedure to process the patient's first batch (taking N=5 points as an example) TWEAK or CD163 test data points. Under this procedure, the IQR statistical window construction logic of the data rationality arbitration engine is temporarily bypassed. The engine classifies these N=5 initial data points as reliable data and transmits them to the time series regularization engine module. Correspondingly, in this procedure, the time series regularization engine module does not perform the algorithmic division of the internal region and the endpoint region. Instead, it uniformly applies the robust local polynomial fitting or linear regression in the second low-order interpolation algorithm to the received sparse sequence of N<5 to generate a preliminary continuous time function. This procedure also defines the steady-state operating parameters after the initialization operation is completed, wherein the endpoint region is determined as the timestamp from the second to last test data point. The initialization process starts from the current time and runs in the time domain. When the system's data adaptation gateway module acquires the N+1th (i.e., the 6th) detection data point, the initialization process automatically terminates. At this time, the data rationality arbitration engine calls the stored N=5 historical detection data, calculates the first valid IQR rationality statistical window, and begins to perform statistical rationality evaluation on the 6th data point. At the same time, the time series regularization engine module also switches to steady-state operation mode and begins to apply the first interpolation algorithm and the second low-order interpolation algorithm to process its internal region and endpoint region respectively, so that the entire information processing pipeline can complete the transition from cold start to steady-state monitoring without manual intervention.

[0036] Example 6: In the information processing pipeline of the disease monitoring system of the present invention, the core algorithm module adopts a set of determined engineering parameters and implementation paths to ensure the reproducibility of information processing results. In the specific implementation of the time-series regularization engine module, the first interpolation algorithm applied to the internal region is the morphology-preserving piecewise cubic Hermitian interpolation polynomial (PCHIP) algorithm; the second low-order interpolation algorithm applied to the endpoint region is the robust local polynomial fitting (LOESS) algorithm. The engineering parameters of the LOESS algorithm are determined to use a second-order local polynomial, and the data point window (Span) used for fitting is set to cover 75% of the total number of data points. When the engine switches to execute a smoothing algorithm in response to the qualitative analysis of statistical outliers, a smoothing spline operation is performed. The smoothing parameters of this operation are... The value is determined within the range of 0 to 1. Taking 0.5 as an example, this value is optimized and fixed during the system development phase by minimizing the generalized cross-validation (GCV) score to balance data fit and curve smoothness. The high-order feature extraction engine module directly applies the computation procedure to the piecewise polynomial functions generated by PCHIP, LOESS, or smoothing splines output by the time series regularization engine module. This engine calculates the rate of change features. (Right now ) and characteristics of changing acceleration (Right now The path is for continuous-time functions. At a specific point in time On the corresponding piecewise polynomial, analytical differentiation, i.e., symbolic differentiation, is performed to obtain the exact function values ​​of its first and second derivatives; the engine calculates the cumulative inflammatory burden characteristics (i.e. The path is to construct within a preset long-period time window. All piecewise polynomial functions are subjected to analytical integration and summation piecewise; this information processing method based on analytical calculus ensures that the calculation results of high-order dynamic features are deterministic and avoids the approximation errors that may be introduced by numerical difference or numerical integration methods.

[0037] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A lupus nephritis disease monitoring system based on TWEAK and CD163 dynamic changes, characterized in that, The system includes a data adaptation gateway module, a data rationality arbitration engine, a time series regularization engine module, a high-order feature extraction engine module, and a risk stratification module. The data adaptation gateway module is used to obtain the patient's historical test data for subsequent evaluation from one or more laboratory information systems, which are sparse and asynchronous in time, including TWEAK and CD163 test data points. The data rationality arbitration engine is used to conduct a statistical rationality assessment of TWEAK and CD163 detection data points based on historical detection data before the time series regularization engine module processes them, and to classify them as reliable data or statistical outliers. The temporal regularization engine module is used to detect data points based on TWEAK and CD163, and generate corresponding continuous time functions for TWEAK and CD163 respectively. When generating any continuous time function, the temporal regularization engine module divides the TWEAK and CD163 detection data points into internal regions and endpoint regions; applies a first interpolation algorithm to the internal regions and a second low-order interpolation algorithm to the endpoint regions; and automatically switches to the smoothing algorithm in response to the data rationality arbitration engine's qualitative identification of statistical outliers of any TWEAK and CD163 detection data point. The high-order feature extraction engine module is used to compute high-order dynamic features based on continuous-time functions; The risk stratification module is used to generate risk indication information based on high-order dynamic features.

2. The lupus nephritis disease monitoring system based on dynamic changes of TWEAK and CD163 according to claim 1, characterized in that, The high-order feature extraction engine module calculates high-order dynamic features, including: calculating a ratio function based on the continuous-time function generated for TWEAK and the continuous-time function generated for CD163; calculating the first derivative of the ratio function to obtain the rate of change feature; and calculating the second derivative of the ratio function to obtain the acceleration feature. The risk stratification module is used to generate risk indication information based on the rate of change feature and the acceleration feature.

3. The lupus nephritis disease monitoring system based on dynamic changes of TWEAK and CD163 according to claim 1, characterized in that, The data rationality arbitration engine performs statistical rationality assessment, including: retrieving historical detection data and calculating the interquartile range of the historical detection data; constructing a dynamic rationality statistical window based on the interquartile range; comparing the values ​​of TWEAK and CD163 detection data points with the rationality statistical window and classifying them as reliable data or statistical outliers; among these, the time series regularization engine module switches to execute a smoothing algorithm to perform smooth spline operations.

4. The lupus nephritis disease monitoring system based on dynamic changes of TWEAK and CD163 according to claim 1, characterized in that, Temporal regularization engine module: applies a first interpolation algorithm, including a shape-preserving spline interpolation algorithm, to fit the data in the internal region; applies a second low-order interpolation algorithm, including a local polynomial fitting algorithm, to fit the data in the endpoint region. The high-order feature extraction engine module is used to compute high-order dynamic features based on function fragments of continuous-time functions in the endpoint regions generated by the second low-order interpolation algorithm.

5. The lupus nephritis disease monitoring system based on dynamic changes of TWEAK and CD163 according to claim 1, characterized in that, The data adaptation gateway module is also used to obtain the context information of medical order events associated with TWEAK and CD163 test data points from the hospital information system. The system also includes a context weight engine, which is used to assign information weights to TWEAK and CD163 test data points based on the context information of medical order events. Among them, the test data points corresponding to temporary medical orders are assigned higher information weights than the test data points corresponding to regular medical orders. Furthermore, the time series regularization engine module incorporates information weights when applying the first interpolation algorithm, the second low-order interpolation algorithm, or the smoothing algorithm to perform weighted interpolation or weighted smoothing.

6. The lupus nephritis disease monitoring system based on dynamic changes of TWEAK and CD163 according to claim 1, characterized in that, The risk stratification module is also used to: monitor and calculate the data staleness time difference between the current system time and the timestamp of the most recent TWEAK and CD163 detection data points obtained by the data adaptation gateway module; when the data staleness time difference is less than the preset timeliness threshold, risk stratification based on high-order dynamic features is performed. When the time difference of data staleness exceeds the preset timeliness threshold, risk stratification is terminated, and specific risk indication information indicating that the risk status is unknown and the data is outdated is generated.

7. The lupus nephritis disease monitoring system based on dynamic changes in TWEAK and CD163 according to claim 1, characterized in that, The high-order feature extraction engine module is also used to: calculate the cumulative inflammatory burden feature as the chronic injury feature by integrating the continuous time function within a preset long period time window while calculating the high-order dynamic features as acute risk features; the risk stratification module is also used to: generate first risk indication information based on acute risk features; and generate second risk indication information based on chronic injury features.

8. A lupus nephritis disease monitoring system based on dynamic changes in TWEAK and CD163 according to claim 2, characterized in that, The risk stratification module includes a rule engine that compares rate of change characteristics and acceleration of change characteristics with a set of clinical thresholds to determine risk indication information. The rule engine includes at least a high-risk alert rule, which is defined as: when and When this occurs, a risk indication message is triggered; among which, Characterized by the rate of change. For changing acceleration characteristics, The preset rate threshold, This is the preset acceleration threshold.

9. A lupus nephritis disease monitoring system based on dynamic changes of TWEAK and CD163 according to claim 1, characterized in that, The data adaptation gateway module is used to: actively poll the laboratory information system at a preset frequency to pull TWEAK and CD163 test data points, convert the pulled TWEAK and CD163 test data points into internal standard events, and push them to the data rationality arbitration engine.

10. A lupus nephritis disease monitoring system based on dynamic changes in TWEAK and CD163 according to claim 1, characterized in that, The first interpolation algorithm used by the temporal regularization engine module is a form-preserving PCHIP interpolation algorithm.

Images