An abnormal event automatic classification and recording method in a nephropathy follow-up system

Abstract

The present application relates to the technical field of medical information processing, in particular to an abnormal event automatic classification and recording method in a kidney disease follow-up system, comprising collecting kidney function parameters and symptom codes, identifying abnormal sequences, extracting time anchor points and characteristic identifiers, calculating flip frequency and amplitude, constructing a feature vector to evaluate risk and generating event classification records. In the present application, by constructing an abnormal trigger ordered sequence and extracting anchor points to determine composite features, combining symptom flip frequency to construct a persistence intensity coefficient, quantifying disease complexity from the dimensions of time evolution and etiology coupling, using a multi-dimensional feature vector to fuse physiological amplitude and statistical characteristics, obtaining disease risk probability values through weighted calculation, realizing the transition from qualitative classification to quantitative risk grading, solving the false alarm and missed alarm problems caused by single index determination, providing dynamic diagnosis basis containing risk level for clinical, significantly improving the recognition accuracy and intervention timeliness of the hidden deterioration trend in kidney disease follow-up.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and in particular to a method for automatic classification and recording of abnormal events in a kidney disease follow-up system. Background Technology

[0002] The field of medical information processing technology refers to a collection of technologies related to the collection, storage, transmission, organization, analysis and management of various types of information generated in medical activities, including patient basic information management, diagnosis and treatment process data recording, follow-up information maintenance, medical event annotation and structured management of long-term health data.

[0003] Among them, the automatic classification and recording method of abnormal events in the kidney disease follow-up system refers to a processing method that identifies and classifies abnormal situations that occur in the follow-up records during the long-term follow-up of kidney disease patients and forms corresponding records. This type of method usually compares the examination results, medication, symptom descriptions and follow-up time nodes collected during the follow-up process through pre-set abnormal judgment conditions. When the data meets the predetermined abnormal conditions, it is mapped to the predefined abnormal type and written into the follow-up file or abnormal event table according to a fixed data item structure to complete the classification and recording of abnormal events.

[0004] Current kidney disease follow-up management models rely solely on preset fixed conditions to discretely compare physiological indicators or symptoms at a single time point. This static and isolated judgment logic severs the evolutionary relationship of abnormal data over time, ignores the pathological coupling effect between parameter fluctuation frequency and concurrent complex symptoms, and lacks in-depth quantitative analysis of the degree of disease persistence and multidimensional risk characteristics. This makes it impossible to accurately assess the true probability of deterioration of complex diseases. It is difficult to distinguish between transient noise and persistent lesions based solely on a single threshold classification, resulting in a lack of dynamic data support for clinical grading intervention and a tendency to delay the early identification and treatment of occult high-risk diseases. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose an automatic classification and recording method for abnormal events in a kidney disease follow-up system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an automatic classification and recording method for abnormal events in a kidney disease follow-up system, comprising the following steps:

[0007] S1: Collect renal function physiological parameters and symptom status codes, determine abnormal trigger data items based on the changes of renal function physiological parameters and symptom status codes over time, and integrate the abnormal trigger ordered sequence;

[0008] S2: Extract the timestamp corresponding to the first abnormal trigger data item in the abnormal trigger ordered sequence as the starting anchor point, determine the composite feature identifier bit and assign a value to obtain the composite feature identifier assignment result;

[0009] S3: Calculate the symptom state reversal frequency in the symptom state code records of the abnormal trigger data items in the abnormal trigger ordered sequence within a preset specified period, and set the persistence intensity coefficient.

[0010] S4: Based on the abnormal triggering ordered sequence, calculate the relative amplitude value of renal function physiological parameters, and construct a multidimensional feature vector according to the assigned composite feature identifier bit in the composite feature identifier assignment result, the persistence intensity coefficient and the relative amplitude value of renal function physiological parameters, and calculate the disease risk probability value.

[0011] S5: Determine the risk level of the current renal function abnormality event based on the risk probability value of the condition, classify the event into category labels according to the risk level, and generate a classification record of the current renal function abnormality event.

[0012] As a further aspect of the present invention, the abnormal triggering ordered sequence includes renal function physiological parameters and symptom status codes that are determined to be abnormal triggering data items. The composite feature identifier assignment result is specifically an assigned value obtained after extracting the starting anchor point and determining the composite feature identifier bit. The persistence intensity coefficient is specifically a numerical coefficient determined based on the symptom status reversal frequency of the symptom status code within a preset specified period. The disease risk probability value is specifically a probability value representing the degree of risk calculated based on a multidimensional feature vector constructed from the composite feature identifier bit, persistence intensity coefficient, and relative amplitude values ​​of renal function physiological parameters. The current renal function abnormality event classification record includes the risk level of the current renal function abnormality event determined based on the disease risk probability value and the event classification category label divided according to the risk level.

[0013] As a further aspect of the present invention, the step of obtaining the abnormally triggered ordered sequence specifically comprises:

[0014] S111: By collecting renal function physiological parameters composed of blood pressure, heart rate and urine protein concentration in real time, and obtaining symptom status codes that identify edema and nausea symptoms, reading the timestamp when each renal function physiological parameter is generated, performing linear interval mapping processing on the renal function physiological parameters according to the preset parameter range, and performing association combination with symptom status codes and timestamps to generate follow-up data items.

[0015] S112: Based on the follow-up data items, perform sliding window truncation processing on the renal function physiological parameters according to the timestamp, filter out items whose renal function physiological parameters exceed the preset safety benchmark threshold range, and at the same time filter out items whose symptom status code value is equal to one, to obtain abnormal trigger data items.

[0016] S113: Gather all the aforementioned abnormal trigger data items, extract the timestamps from them as the sorting basis, perform a time-series rearrangement operation, link the sorted abnormal trigger data items according to the time dimension, and generate an ordered sequence of abnormal triggers.

[0017] As a further aspect of the present invention, the step of obtaining the composite feature identifier assignment result is specifically as follows:

[0018] S211: Extract the timestamp corresponding to the first abnormal trigger data item in the ordered sequence of abnormal triggers and set it as the starting anchor point. Traverse each remaining abnormal trigger data item in the ordered sequence of abnormal triggers, read the corresponding timestamp, and calculate the time interval between the timestamp of each subsequent abnormal trigger data item and the starting anchor point.

[0019] S212: If the time interval is less than the preset time threshold and the abnormality type is inconsistent, it is determined to be a composite abnormal symptom state; if the time interval is greater than the time threshold or the abnormal triggering ordered sequence has only one starting anchor point, it is determined to be a non-composite abnormal symptom state, and the composite abnormality determination result is obtained.

[0020] S213: Perform a binary assignment operation based on the composite anomaly determination result. If the determination result is a composite anomaly, set the flag bit to the value one. If the determination result is a non-composite anomaly, set the flag bit to the value zero. Quantify the complexity of the anomaly event and obtain the composite feature flag assignment result.

[0021] As a further aspect of the present invention, the step of obtaining the durability strength coefficient specifically includes:

[0022] S311: Retrieve the symptom status code of the abnormal trigger data item in the abnormal trigger ordered sequence, trace back all symptom status records of the symptom status code within the specified period, perform time-series symptom status change detection on the record sequence, identify the jump points of the symptom status code, accumulate the total number of jump points, and obtain the flip count value.

[0023] S312: Count the total number of times the symptom status code appears in the symptom status record, calculate the ratio of the flip count value to the total number of times the symptom appears, and obtain the symptom status flip frequency;

[0024] S313: Compare the frequency of symptom state reversal with a preset stability coefficient threshold, and quantify the degree of disease stubbornness based on the comparison difference to obtain the persistence intensity coefficient.

[0025] As a further aspect of the present invention, the step of obtaining the disease risk probability value specifically includes:

[0026] S411: Extract renal function physiological parameters from the abnormal trigger data items in the abnormal trigger ordered sequence and normalize them, calculate the deviation between the normalized renal function physiological parameters and the corresponding safety benchmark threshold, and use it as the relative amplitude value.

[0027] S412: Perform vector concatenation processing on the binary value in the composite feature identifier assignment result, the persistence intensity coefficient, and the relative amplitude value of the physiological parameter, and construct a multidimensional data structure according to the preset dimensional order to obtain a multidimensional feature vector;

[0028] S413: The multidimensional feature vector is matched with the preset risk weights using a machine learning algorithm. A linear weighted mapping calculation is performed based on the risk weights to obtain the risk projection value. The risk projection value is then input into the Sigmoid activation function to perform a nonlinear transformation calculation, quantifying the possibility of the current condition worsening and obtaining the condition risk probability value.

[0029] As a further aspect of the present invention, the steps for obtaining the current renal function abnormality event classification record are specifically as follows:

[0030] S511: Match the disease risk probability value with the preset risk grading standard interval to determine the risk interval to which the current disease risk probability value belongs, define the current risk severity according to the matched interval, and determine the risk level of the current renal function abnormality event;

[0031] S512: Based on the risk level of the current renal function abnormality event, retrieve the preset classification mapping table, match the corresponding descriptive text, and generate a unique index code to identify the current abnormal event, and divide the event into category labels;

[0032] S513: Based on the timestamp, perform trend rendering on the physiological parameters of renal function and the symptom status code to generate a visualization chart. Perform structured encapsulation on the risk level of the current renal function abnormal event, the event classification category label, the index code of the abnormal event, and the visualization chart data to generate a classification record of the current renal function abnormal event.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0034] In this invention, by constructing an abnormal triggering ordered sequence and extracting anchor point judgment composite features, and combining the symptom reversal frequency to construct a persistence intensity coefficient, the complexity of the disease is quantified from the dimensions of temporal evolution and etiological coupling. By using multi-dimensional feature vectors to fuse physiological amplitude and statistical features, the risk probability value of the disease is obtained through weighted calculation, realizing the transformation from qualitative classification to quantitative risk grading, solving the problem of false alarms and false negatives caused by single indicator judgment, providing dynamic diagnostic basis including risk level for clinical practice, and significantly improving the accuracy of identifying the occult deterioration trend in kidney disease follow-up and the timeliness of intervention. Attached Figure Description

[0035] Figure 1 This is the overall flowchart of the method for automatic classification and recording of abnormal events in the kidney disease follow-up system of the present invention;

[0036] Figure 2 This is a flowchart illustrating the construction of an ordered sequence triggered by an anomaly in this invention.

[0037] Figure 3 This is a flowchart illustrating the process of assigning values ​​to the composite feature identifier of the present invention.

[0038] Figure 4 This is a flowchart illustrating the calculation of the durability strength coefficient of this invention.

[0039] Figure 5 This is a flowchart illustrating the calculation of disease risk probability in this invention.

[0040] Figure 6 A flowchart for generating abnormal event classification records in this invention. Detailed Implementation

[0041] 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 embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] Please see Figure 1 This invention provides a technical solution: an automatic classification and recording method for abnormal events in a kidney disease follow-up system, comprising the following steps:

[0043] S1: Collect renal function physiological parameters and symptom status codes, determine abnormal trigger data items based on the changes of renal function physiological parameters and symptom status codes over time, and integrate the abnormal trigger ordered sequence;

[0044] S2: Extract the timestamp corresponding to the first abnormal trigger data item in the abnormal trigger ordered sequence as the starting anchor point, determine the composite feature identifier bit and assign a value to obtain the composite feature identifier assignment result;

[0045] S3: Calculate the symptom state reversal frequency and set the persistence intensity coefficient in the symptom state code of the abnormal trigger data item in the statistical abnormal trigger ordered sequence within a preset specified period.

[0046] S4: Based on the abnormally triggered ordered sequence, calculate the relative amplitude values ​​of renal function physiological parameters. Based on the assigned composite feature identifier bits, persistence intensity coefficient and relative amplitude values ​​of renal function physiological parameters in the composite feature identifier assignment results, construct a multidimensional feature vector and calculate the disease risk probability value.

[0047] S5: Determine the risk level of the current renal function abnormality event based on the probability value of the condition, classify the event into category labels according to the risk level, and generate a classification record of the current renal function abnormality event.

[0048] The abnormal trigger ordered sequence includes renal function physiological parameters and symptom status codes that are identified as abnormal trigger data items. The composite feature identifier assignment result is specifically the assigned value obtained after extracting the starting anchor point and determining the composite feature identifier bit. The persistence intensity coefficient is specifically a numerical coefficient determined based on the symptom status code's symptom status reversal frequency within a preset specified period. The disease risk probability value is specifically a probability value representing the degree of risk calculated based on a multidimensional feature vector constructed from the composite feature identifier bit, persistence intensity coefficient, and relative amplitude values ​​of renal function physiological parameters. The current renal function abnormal event classification record includes the risk level of the current renal function abnormal event determined based on the disease risk probability value and the event classification category label based on the risk level.

[0049] Please see Figure 2 The specific steps for obtaining an ordered sequence triggered by an anomaly are as follows:

[0050] S111: By collecting renal function physiological parameters composed of blood pressure, heart rate and urine protein concentration in real time, and obtaining symptom status codes that identify edema and nausea symptoms, reading the timestamp when each renal function physiological parameter is generated, performing linear interval mapping processing on the renal function physiological parameters according to the preset parameter range, and performing association combination with symptom status codes and timestamps to generate follow-up data items.

[0051] Using a piezoelectric pressure sensor, a photoplethysmography (PPG) sensor, and an electrochemical urinary protein biosensor integrated into a wearable monitoring device, continuous analog signal acquisition is performed on the user's systolic and diastolic blood pressure, heart rate per minute, and urinary protein content. The acquired analog electrical signals are converted into digital signals via an analog-to-digital converter. Simultaneously, the user's subjective symptom feedback interface is invoked. When the user confirms lower limb or eyelid edema, the edema symptom status code is marked with a value of 1; otherwise, it is marked with a value of 0. When the user confirms nausea or vomiting, the nausea symptom status code is marked with a value of 1; otherwise, it is marked with a value of 0. The Greenwich Mean Time (GMT) at the time of each data point generation is recorded synchronously as a timestamp. During the linear interval mapping process for the parameter range, preset sensor linear calibration parameters are retrieved. These linear calibration parameters are obtained by selecting the minimum pressure value corresponding to the current value (e.g., 4 mA) from the standard pressure source output as the low point and the maximum pressure value corresponding to the current value (e.g., 20 mA) as the high point during the device's factory calibration phase, and constructing a linear regression equation to determine the slope and intercept. Taking systolic blood pressure measurement as an example, the sensor output current range is set to 4 mA to 20 mA, corresponding to a physical measurement range of 0 mmHg to 300 mmHg. If the real-time measured blood pressure current value is 12.32 mA, the following linear calculation is performed: The current systolic blood pressure value was thus obtained as 156 mmHg. The converted renal function physiological parameter value of 156, the corresponding symptom status code value of 1, and the timestamp accurate to milliseconds, December 18, 09:00:00, were encapsulated in the form of key-value pairs to construct a follow-up data item with the structure of "parameter type-value-status-time".

[0052] S112: Based on follow-up data items, perform sliding window truncation processing on renal function physiological parameters according to timestamps, filter items whose renal function physiological parameters exceed the preset safety baseline threshold range, and at the same time filter items whose symptom status code value is equal to one, to obtain abnormal trigger data items.

[0053] A sliding window with a time span of 30 minutes is set, and the data is truncated on the timeline of the follow-up data items in 5-minute increments. Within each sliding window, renal function physiological parameters are read one by one and compared with a preset safety baseline threshold range. The safety baseline threshold range is set based on: collecting all historical physiological parameter records of the patient during the past 6 months when the condition was stable (no hospitalization record, no medication adjustment), calculating the mean and standard deviation of the dataset, and selecting the range of values ​​between the mean and twice the standard deviation as the personalized safety baseline threshold range. For example, the calculated upper limit of the safety baseline threshold for systolic blood pressure is set to 140 mmHg. When executing the screening logic, physiological parameters and symptom status codes are checked simultaneously: the following possibilities exist: 1) physiological parameters are not exceeded and the status code is 0; 2) physiological parameters are not exceeded but the status code is 1; 3) physiological parameters exceed the limit but the status code is 0; 4) physiological parameters exceed the limit and the status code is 1. This step only retains the fourth case. Referring to the example in S111 above, the systolic blood pressure reading in the current follow-up data item is 156 mmHg, and the associated symptom status code has a value of 1 (indicating the presence of edema). Perform a numerical comparison: If the result is greater than 0, the physiological parameter is determined to be out of limit; simultaneously, the status code is checked to see if it equals 1. Since both conditions are met, the entry is determined to be a valid anomaly, retained, and marked as an anomaly trigger data item. If only one condition is met, it is considered noise data and discarded.

[0054] S113: Gather all abnormal trigger data items, extract the timestamps from them as the sorting basis to perform time-series rearrangement, link the sorted abnormal trigger data items according to the time dimension, and generate an ordered sequence of abnormal triggers.

[0055] A temporary doubly linked list data structure is established to store the filtered anomaly trigger data items. All anomaly trigger data items are traversed, and the timestamp field at the head of each data item is extracted. A quicksort algorithm is used, with the timestamp value as the comparison key, to sort all anomaly trigger data items in ascending order. During the sorting process, if two data items have completely identical timestamps, a secondary sort is performed based on the priority of the parameter type (setting the priority value of blood pressure to 3, heart rate to 2, and urine protein to 1). Assuming the filtered anomaly data item set contains three records with timestamps of 09:00, 10:30, and 09:45, after sorting, the data items are linked sequentially in the order of 09:00, 09:45, and 10:30. The sorted data item nodes are then sequentially added to the linked list, and pointer indexes are established for the predecessor and successor nodes to ensure the continuity and traceability of the data in the time dimension. The resulting linked list structure is an ordered sequence of exception triggers, which contains a complete record of the disease progression from the earliest time point when the exception occurred to the latest time point when the exception occurred.

[0056] Please see Figure 3 The specific steps for obtaining the composite feature identifier assignment result are as follows:

[0057] S211: Extract the timestamp corresponding to the first abnormal trigger data item in the abnormal trigger ordered sequence and set it as the starting anchor point. Iterate through each remaining abnormal trigger data item in the abnormal trigger ordered sequence, read the corresponding timestamp, and calculate the time interval between the timestamp of each subsequent abnormal trigger data item and the starting anchor point.

[0058] Access the head node of the ordered sequence of exception-triggered data, read its timestamp information, and define it as the starting anchor time. In the previous example, the starting anchor time is 09:00 on December 18th. Then, initialize an iteration pointer to point to the second node in the sequence and iterate backwards. During the traversal, read the timestamp of each subsequent exception-triggered data item. Perform a time difference calculation, that is, subtract the timestamp of the starting anchor from the timestamp of the currently traversed data item, and convert the difference to hours. Assuming the timestamp of the second data item in the sequence is 10:30 on December 18th, perform the calculation: That is, the time interval is 1.5 hours. Assuming the timestamp of the third data item in the sequence is 15:00 on December 18th, perform the calculation: This means the time interval is 6.0 hours. This subtraction operation is repeated for each non-first data item in the sequence to obtain a set of time interval values ​​relative to the starting anchor point.

[0059] S212: If there is a time interval less than the preset time threshold and the abnormality type is inconsistent, it is determined as a complex abnormal symptom state. If the time interval is greater than the time threshold or the abnormal triggering ordered sequence has only one starting anchor point, it is determined as a non-complex abnormal symptom state. Obtain the complex abnormality determination result.

[0060] A preset time threshold of 24 hours is set. This threshold is obtained by statistically analyzing the "optimal time window for early intervention" defined in the clinical guidelines for acute kidney injury (AKI) and using its recommended observation duration as the threshold. During the traversal examination, the abnormality type of the current data item is compared with the abnormality type corresponding to the starting anchor point. The judgment logic includes two judgment paths: Path 1, the calculated time interval value is less than the preset time threshold (24 hours), and the abnormality type of the current data item is inconsistent with the abnormality type of the starting anchor point (e.g., the starting point is abnormal blood pressure, and the current point is abnormal proteinuria); Path 2, the time interval value is greater than the time threshold, or although it is within the threshold, the abnormality type is consistent, or the sequence contains only one node. If Path 1 is satisfied, it is judged as a complex abnormal symptom state; if Path 2 is satisfied, it is judged as a non-complex abnormal symptom state. For a 1.5-hour time interval node (less than 24 hours), if the starting point is abnormal blood pressure and the node is abnormal proteinuria, then Path 1 is satisfied, and the judgment result is a complex abnormal symptom state. For subsequent nodes, if this condition is not met, they are judged separately. Finally, the complex abnormality judgment result for the entire sequence is obtained.

[0061] S213: Perform a binary assignment operation based on the composite anomaly determination result. If the determination result is a composite anomaly, set the flag bit to the value one. If the determination result is a non-composite anomaly, set the flag bit to the value zero. Quantify the complexity of the anomaly event and obtain the composite feature flag assignment result.

[0062] If the judgment result is "complex abnormal symptom state", the preset composite feature identifier variable is assigned the value 1; if the judgment result is "non-complex abnormal symptom state", the identifier variable is assigned the value 0. This assignment operation aims to transform the qualitative clinical complication logic into quantitative numerical features. In the aforementioned example, since the inconsistent types of blood pressure and urinary protein appeared within 1.5 hours (within 24 hours), the composite judgment was triggered, so the composite feature identifier was set to the value 1. This value 1, as a quantitative indicator, directly participates in the subsequent multidimensional vector construction, representing that the current abnormal event has complex characteristics of multiple etiological coupling, while the value 0 represents a single etiology or discrete features.

[0063] Please see Figure 4 The specific steps for obtaining the persistence strength coefficient are as follows:

[0064] S311: Retrieve the symptom status code of the abnormal trigger data item in the abnormal trigger ordered sequence, trace back all symptom status records of the symptom status code within the specified period, perform time-series symptom status change detection on the record sequence, identify the jump points of the symptom status code, accumulate the total number of jump points, and obtain the flip count value.

[0065] The process of performing temporal symptom state change detection on the recorded sequence, identifying transition points in symptom state codes, and accumulating the total number of transition points is as follows:

[0066] Extract the symptom status codes from the symptom status records one by one in ascending order of time as the current detection target;

[0067] Obtain the symptom status code of the previous time point of the adjacent currently detected object as the preceding comparison object;

[0068] Perform a numerical equality comparison between the value of the currently detected object and the value of the previously compared object;

[0069] If the comparison results indicate that the values ​​are not equal, a jump point is identified and the count value is incremented by one.

[0070] Initialize the flip count to 0. Set the backtracking period to the past 7 days. Extract all symptom status codes from the abnormal trigger ordered sequence, arranging them chronologically to form a status sequence of 0s and 1s. Assume the symptom status code sequence fragment extracted within the 7-day period is "0, 0, 1, 1, 0, 1". Traverse the sequence starting from the second element in ascending chronological order. Use the current element as the current detection object and the previous element as the preceding comparison object. Perform numerical equality comparison: First, compare the current object 0 with the preceding object 0; if the values ​​are equal, the count remains unchanged. Second, compare the current object 1 with the preceding object 0; if the values ​​are not equal, a jump point is identified, and the count is incremented by one. Third step: Compare the current object 1 with the previous object 1. If they are equal, leave it unchanged. Fourth step: Compare the current object 0 with the previous object 1. If they are not equal, increment by one. Fifth step: Compare the current object 1 with the previous object 0. If they are not equal, increment by one. After the traversal is complete, the flip count is 3.

[0071] S312: Count the total number of times the symptom status code appears in the symptom status record, calculate the ratio of the flip count value to the total number of times the symptom appears, and obtain the symptom status flip frequency;

[0072] The total number of symptom status codes collected during the 7-day follow-up period was counted. This total number was obtained directly by querying the database for the number of entries in all follow-up records within that time period. Assume that 30 status records were collected during the monitoring period. Combining this with the reversal count value of 3 obtained in S311, a division operation was performed to calculate the symptom status reversal frequency. The calculation formula is: The value of 0.1 represents the symptom state reversal frequency. This frequency can have several possibilities: the closer the value is to 0, the more stable the symptom state (persistent symptoms or persistent asymptomaticity); the closer the value is to 0.5 (under a uniform 0 / 1 distribution), the more frequently symptoms appear and disappear, indicating a highly unstable condition. In this example, 0.1 represents a certain degree of fluctuation.

[0073] S313: Compare the symptom state reversal frequency with a preset stability coefficient threshold, and quantify the degree of disease stubbornness based on the comparison difference to obtain the persistence intensity coefficient.

[0074] Retrieve the preset stability coefficient threshold. This threshold is obtained by selecting a sample of chronic kidney disease patients diagnosed by doctors as having "well-controlled disease," collecting their symptom status code sequences over three months, calculating the average symptom reversal frequency for this group, and setting this average value as the stability coefficient threshold, for example, 0.05. Compare the symptom reversal frequency calculated in S312 with the stability coefficient threshold. To more smoothly quantify the degree of disease refractoryness, a nonlinear mapping algorithm using the hyperbolic tangent function (Tanh) is constructed to calculate the persistence intensity coefficient. The calculation formula for this algorithm is: ,in, The value represents the persistence intensity coefficient, which is used to characterize the severity of fluctuations in the condition; the larger the value, the more severe the fluctuations. This represents the gain sensitivity coefficient, used to adjust the sensitivity to frequency deviations. It is set to 10. The larger the value, the more sensitive it is to slight frequency deviations, thus pushing the coefficient towards the limit value of 1 more quickly. This represents the flip count value obtained in S311, which is 3 in this scenario; This represents the total number of times the symptoms obtained in S312 occurred, which is 30 in this scenario; This represents the preset stability coefficient threshold, which is 0.05 in this scenario. This value represents the allowable background noise of normal physiological fluctuations. Based on the values ​​in the previous example, the parameters are directly substituted into the formula to perform the calculation: If the calculation result If the result is negative, the formula must be zeroed out. In this example, the result is positive, therefore the continuity strength coefficient... The final value was determined to be 0.462. This value quantifies the nonlinear intensity of the current disease fluctuations exceeding the normal stable state; the larger the value, the more severe the recurrence and difficulty in controlling the disease.

[0075] Please see Figure 5 The specific steps for obtaining the probability value of disease risk are as follows:

[0076] S411: Extract renal function physiological parameters from the abnormal trigger data items in the abnormal trigger ordered sequence and normalize them. Calculate the deviation between the normalized renal function physiological parameters and the corresponding safety benchmark threshold as the relative amplitude value.

[0077] First, normalization is performed. The normalization parameters are obtained based on the medical survival limits of the physiological indicator; for example, the systolic blood pressure limit is set to 0 to 300. For the systolic blood pressure of 156 obtained in S111, normalization calculation is performed: Simultaneously, the corresponding safety baseline threshold (140 set in S112) is retrieved, and the same normalization calculation is performed: The deviation of normalized renal function physiological parameters from their corresponding safety baseline thresholds is calculated and used as the relative amplitude value. The calculation formula is: The value of 0.053 is the relative amplitude of the physiological parameter, used to eliminate the influence of dimensions and purely measure the degree of deviation from safety.

[0078] S412: Perform vector concatenation processing on the binary values, persistence intensity coefficients, and relative amplitude values ​​of physiological parameters in the composite feature identifier assignment results, and construct a multidimensional data structure according to the preset dimensional order to obtain a multidimensional feature vector;

[0079] A multidimensional data structure is constructed according to a preset dimensional order. The first dimension is filled with the composite feature identifier assignment result obtained in S213 (value 1); the second dimension is filled with the persistence intensity coefficient obtained in S313 (value 0.462); and the third dimension is filled with the relative amplitude value of the physiological parameter obtained in S411 (value 0.053). Vector concatenation is performed to generate a multidimensional feature vector. This vector integrates discrete categorical features (whether they are composite), statistical features (fluctuation frequency), and continuous numerical features (deviation magnitude) into a unified mathematical expression.

[0080] S413: The multidimensional feature vector is matched with the preset risk weights through machine learning algorithms. A linear weighted mapping calculation is performed based on the risk weights to obtain the risk projection value. The risk projection value is input into the Sigmoid activation function to perform nonlinear transformation calculation, quantify the possibility of the current condition deteriorating, and obtain the condition risk probability value.

[0081] The preset risk weights and intercept terms are obtained as follows: a training set containing multidimensional feature data of historical kidney disease patients and their final diagnosis (worsening / stabilizing) is constructed. Supervised learning is performed using logistic regression. After training convergence, the coefficients of each dimension of the model are extracted as risk weights, and the bias is extracted as the intercept term. Assume the weight vector obtained during training... Where 0.5 is the weight of the composite feature, and the larger the value, the greater the weight of the feature's influence on risk (indicating that the concurrent symptoms are the main source of risk); 0.3 is the weight of the persistence intensity coefficient; 0.2 is the weight of the relative amplitude value; and the intercept term... The Sigmoid activation function is used to perform a nonlinear transformation to calculate the probability value of disease risk. The calculation formula is as follows: ,in, This represents the probability value of disease risk, ranging from 0 to 1; Represents the first in the weight vector Each weight value represents the importance of the corresponding feature; Represents the first in a multidimensional feature vector One eigenvalue; This represents the intercept term, used to adjust the offset position of the classification plane; This represents the feature dimension, which is 3 in this example. The multidimensional feature vector obtained in S412 above... And the calculation is performed by substituting preset weights into the formula: The final disease risk probability value was 0.2057 (i.e., 20.57%), which quantifies the possibility of the current condition worsening.

[0082] Please see Figure 6 The specific steps for obtaining the current renal function abnormality event classification record are as follows:

[0083] S511: Match the disease risk probability value with the preset risk grading standard range to determine the risk range to which the current disease risk probability value belongs, define the current risk severity based on the matched range, and determine the risk level of the current renal function abnormality event;

[0084] The process of matching the disease risk probability value with a preset risk grading standard range to determine the risk range to which the current disease risk probability value belongs is as follows:

[0085] Construct a pre-defined risk classification standard interval that includes a low-risk monitoring interval, a medium-risk early warning interval, and a high-risk intervention interval. The low-risk monitoring interval, the medium-risk early warning interval, and the high-risk intervention interval are continuously distributed on the numerical axis and do not overlap with each other.

[0086] The position of the disease risk probability value within the preset risk grading standard range is determined.

[0087] If the probability value of the disease risk falls into the low-risk monitoring range, the current risk severity is defined as the observation level and the risk level of the current renal function abnormality event is set as Level 1 risk.

[0088] If the probability value of the disease risk falls into the medium risk warning range, the current risk severity will be defined as the alert level and the risk level of the current abnormal renal function event will be set as level two risk.

[0089] If the probability value of the disease risk falls into the high-risk intervention range, the current risk severity is defined as the emergency level and the risk level of the current renal function abnormality event is set as level three risk.

[0090] The preset risk grading standard range includes a low-risk monitoring range. Medium-risk warning zone and high-risk intervention zones The boundary values ​​of these intervals are determined based on the balance point between specificity and sensitivity, based on the maximum point of the Youden exponent on the receiver operating characteristic (ROC) curve. The interval in which the probability value of disease risk of 0.2057 falls is determined by: Therefore, it was determined to fall within the low-risk monitoring zone. According to the matched zone definitions: if falling within the low-risk monitoring zone, it is defined as the observation level, with a risk level set to Level 1; if falling within the medium-risk warning zone, it is defined as the alert level, with a risk level set to Level 2; if falling within the high-risk intervention zone, it is defined as the emergency response level, with a risk level set to Level 3. In this example, the risk level of the current renal function abnormality event was ultimately determined to be Level 1.

[0091] S512: Based on the risk level of the current renal function abnormality event, retrieve the preset classification mapping table, match the corresponding descriptive text, and generate a unique index code to identify the current abnormal event, and divide the event into category labels;

[0092] The classification mapping table is constructed as follows: based on the consensus of nephrology experts, standard clinical management recommendations corresponding to different risk levels are compiled into a text library. For Level 1 risk, the matched descriptive text is "minor fluctuations in physiological indicators; it is recommended to maintain the current treatment plan and continue observation." Simultaneously, a unique index code is generated to identify the current abnormal event. The generation logic for the index code is: combining "event type prefix" + "current date" + "risk level code" + "random hash checksum". For example, the generated code is "EVT-20251218-L1-X9Y". Based on the descriptive text and risk level, event classification labels are assigned: Level 1 risk corresponds to "routine monitoring," Level 2 to "early warning and attention," and Level 3 to "emergency intervention." In this example, it is classified under the "routine monitoring" label.

[0093] S513: Based on timestamps, perform trend rendering on renal function physiological parameters and symptom status codes to generate visualization charts. Perform structured encapsulation on the risk level, event classification category label, index code of abnormal events, and visualization chart data of the current renal function abnormality event to generate the current renal function abnormality event classification record.

[0094] The chart generation logic is as follows: Using the timestamp as the horizontal axis, and physiological parameter values ​​(e.g., 156 mmHg) and symptom status codes (0 or 1) as the left and right vertical axes respectively, line trend charts and bar charts are drawn. The risk level determined in S511 (Level 1 risk), the event category label determined in S512 (routine monitoring category), the index code of the abnormal event (EVT-20251218-L1-X9Y), and the visualization chart data (the Base64 encoded string or data point set of the chart) are structurally encapsulated. The encapsulation format uses JSON object format, for example, {"id": "EVT-20251218-L1-X9Y", "level": 1, "category": "routine monitoring category", "chart_data": [...]}. Finally, a current renal function abnormality event classification record containing complete diagnostic information is generated and stored in the patient's electronic health record.

[0095] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An abnormal event automatic classification and recording method in a kidney disease follow-up system, characterized by, Includes the following steps: S1: Collect renal function physiological parameters and symptom status codes, determine abnormal trigger data items based on the changes of renal function physiological parameters and symptom status codes over time, and integrate the abnormal trigger ordered sequence; S2: Extract the timestamp corresponding to the first abnormal trigger data item in the abnormal trigger ordered sequence as the starting anchor point, determine the composite feature identifier bit and assign a value to obtain the composite feature identifier assignment result; S3: Calculate the symptom state reversal frequency in the symptom state code records of the abnormal trigger data items in the abnormal trigger ordered sequence within a preset specified period, and set the persistence intensity coefficient. S4: Based on the abnormal triggering ordered sequence, calculate the relative amplitude value of renal function physiological parameters, and construct a multidimensional feature vector according to the assigned composite feature identifier bit in the composite feature identifier assignment result, the persistence intensity coefficient and the relative amplitude value of renal function physiological parameters, and calculate the disease risk probability value. S5: Determine the risk level of the current renal function abnormality event based on the aforementioned disease risk probability value, classify the event category label according to the risk level, and generate a current renal function abnormality event classification record; The specific steps for obtaining the composite feature identifier assignment result are as follows: S211: Extract the timestamp corresponding to the first abnormal trigger data item in the ordered sequence of abnormal triggers and set it as the starting anchor point. Traverse each remaining abnormal trigger data item in the ordered sequence of abnormal triggers, read the corresponding timestamp, and calculate the time interval between the timestamp of each subsequent abnormal trigger data item and the starting anchor point. S212: If the time interval is less than the preset time threshold and the abnormality type is inconsistent, it is determined to be a composite abnormal symptom state; if the time interval is greater than the time threshold or the abnormal triggering ordered sequence has only one starting anchor point, it is determined to be a non-composite abnormal symptom state, and the composite abnormality determination result is obtained. S213: Perform a binary assignment operation based on the composite anomaly determination result. If the determination result is a composite anomaly, set the flag bit to the value one. If the determination result is a non-composite anomaly, set the flag bit to the value zero. Quantify the complexity of the anomaly event and obtain the composite feature flag assignment result.

2. The method of claim 1, wherein the method further comprises: The abnormal triggering ordered sequence includes renal function physiological parameters and symptom status codes that are determined to be abnormal triggering data items. The composite feature identifier assignment result is specifically the assigned value obtained after extracting the starting anchor point and determining the composite feature identifier bit. The persistence intensity coefficient is specifically a numerical coefficient determined based on the symptom status code's symptom status reversal frequency within a preset specified period. The disease risk probability value is specifically a probability value representing the degree of risk calculated based on a multidimensional feature vector constructed from the composite feature identifier bit, persistence intensity coefficient, and relative amplitude values ​​of renal function physiological parameters. The current renal function abnormality event classification record includes the risk level of the current renal function abnormality event determined based on the disease risk probability value and the event classification category label divided according to the risk level.

3. The method of claim 1, wherein the method further comprises: The specific steps for obtaining the abnormally triggered ordered sequence are as follows: S111: By collecting renal function physiological parameters composed of blood pressure, heart rate and urine protein concentration in real time, and obtaining symptom status codes that identify edema and nausea symptoms, reading the timestamp when each renal function physiological parameter is generated, performing linear interval mapping processing on the renal function physiological parameters according to the preset parameter range, and performing association combination with symptom status codes and timestamps to generate follow-up data items. S112: Based on the follow-up data items, perform sliding window truncation processing on the renal function physiological parameters according to the timestamp, filter out items whose renal function physiological parameters exceed the preset safety benchmark threshold range, and at the same time filter out items whose symptom status code value is equal to one, to obtain abnormal trigger data items. S113: Gather all the aforementioned abnormal trigger data items, extract the timestamps from them as the sorting basis, perform a time-series rearrangement operation, link the sorted abnormal trigger data items according to the time dimension, and generate an ordered sequence of abnormal triggers.

4. The method of claim 1, wherein the method further comprises: The specific steps for obtaining the durability strength coefficient are as follows: S311: Retrieve the symptom status code of the abnormal trigger data item in the abnormal trigger ordered sequence, trace back all symptom status records of the symptom status code within the specified period, perform time-series symptom status change detection on the record sequence, identify the jump points of the symptom status code, accumulate the total number of jump points, and obtain the flip count value. S312: Count the total number of times the symptom status code appears in the symptom status record, calculate the ratio of the flip count value to the total number of times the symptom appears, and obtain the symptom status flip frequency; S313: Compare the frequency of symptom state reversal with a preset stability coefficient threshold, and quantify the degree of disease stubbornness based on the comparison difference to obtain the persistence intensity coefficient.

5. The method of claim 4, wherein the method further comprises: The process of performing temporal symptom state change detection on the recorded sequence, identifying transition points in symptom state codes, and accumulating the total number of transition points is as follows: Extract the symptom status codes from the symptom status records one by one in ascending order of time as the current detection target; Obtain the symptom status code of the previous time point of the adjacent currently detected object as the preceding comparison object; Perform a numerical equality comparison between the value of the currently detected object and the value of the previously compared object; If the comparison results indicate that the values ​​are not equal, a jump point is identified and the count value is incremented by one.

6. The method of claim 4, wherein the method further comprises: The specific steps for obtaining the disease risk probability value are as follows: S411: Extract renal function physiological parameters from the abnormal trigger data items in the abnormal trigger ordered sequence and normalize them, calculate the deviation between the normalized renal function physiological parameters and the corresponding safety benchmark threshold, and use it as the relative amplitude value. S412: Perform vector concatenation processing on the binary value in the composite feature identifier assignment result, the persistence intensity coefficient, and the relative amplitude value of the physiological parameter, and construct a multidimensional data structure according to the preset dimensional order to obtain a multidimensional feature vector; S413: The multidimensional feature vector is matched with the preset risk weights using a machine learning algorithm. A linear weighted mapping calculation is performed based on the risk weights to obtain the risk projection value. The risk projection value is then input into the Sigmoid activation function to perform a nonlinear transformation calculation, quantifying the possibility of the current condition worsening and obtaining the condition risk probability value.

7. The method for automatic classification and recording of abnormal events in a kidney disease follow-up system according to claim 6, characterized in that, For the probability value of disease risk The formula used is: ； in, Represents the first in the weight vector Each weight value Represents the first in a multidimensional feature vector 1 eigenvalue, Represents the intercept term. Represents the feature dimension.

8. The method for automatic classification and recording of abnormal events in a kidney disease follow-up system according to claim 6, characterized in that, The specific steps for obtaining the current renal function abnormality event classification record are as follows: S511: Match the disease risk probability value with the preset risk grading standard interval to determine the risk interval to which the current disease risk probability value belongs, define the current risk severity according to the matched interval, and determine the risk level of the current renal function abnormality event; S512: Based on the risk level of the current renal function abnormality event, retrieve the preset classification mapping table, match the corresponding descriptive text, and generate a unique index code to identify the current abnormal event, and divide the event into category labels; S513: Based on the timestamp, perform trend rendering on the physiological parameters of renal function and the symptom status code to generate a visualization chart. Perform structured encapsulation on the risk level of the current renal function abnormal event, the event classification category label, the index code of the abnormal event, and the visualization chart data to generate a classification record of the current renal function abnormal event.

9. The method for automatic classification and recording of abnormal events in a kidney disease follow-up system according to claim 8, characterized in that, The process of matching the disease risk probability value with a preset risk grading standard range to determine the risk range to which the current disease risk probability value belongs is as follows: Construct a pre-defined risk classification standard interval that includes a low-risk monitoring interval, a medium-risk early warning interval, and a high-risk intervention interval. The low-risk monitoring interval, the medium-risk early warning interval, and the high-risk intervention interval are continuously distributed on the numerical axis and do not overlap with each other. The position of the disease risk probability value within the preset risk grading standard range is determined. If the probability value of the disease risk falls into the low-risk monitoring range, the current risk severity is defined as the observation level and the risk level of the current renal function abnormality event is set as Level 1 risk. If the probability value of the disease risk falls into the medium risk warning range, the current risk severity will be defined as the alert level and the risk level of the current abnormal renal function event will be set as level two risk. If the probability value of the disease risk falls into the high-risk intervention range, the current risk severity is defined as the emergency level and the risk level of the current renal function abnormality event is set as level three risk.

Images