Risk monitoring and early warning method and system for rejection after kidney transplantation

Abstract

The invention relates to a renal transplantation post-operation rejection risk monitoring and early warning method and a renal transplantation post-operation rejection risk monitoring and early warning system. The method comprises the steps of collecting recipient nursing monitoring data, laboratory indexes and transplanted kidney ultrasonic blood flow parameters in a follow-up visit period, performing timestamp alignment, deletion processing and standardization on multi-source data, extracting features to construct a time sequence feature sequence, inputting the time sequence feature sequence into a pre-training risk prediction model, outputting the rejection reaction occurrence probability of the next period, and forming a risk trend. Calculating a nursing sensitive index contribution weight based on the model contribution information, and screening a target nursing monitoring index; and establishing an individualized baseline model to obtain a baseline value and an allowable fluctuation interval, extracting characteristics such as deviation amplitude, direction, rate, fluctuation and continuous deviation duration and the like, and performing individualized calibration on the occurrence probability to obtain a calibration risk score. When the threshold value is not reached and the trend is not triggered, generating a nursing monitoring suggestion of the next period; and pushing early warning and generating grading intervention suggestions when a threshold value is reached or a trend is triggered.

Description

Technical Field

[0001] This application relates to the field of postoperative monitoring technology, and in particular to a method and system for monitoring and early warning of rejection risk after kidney transplantation. Background Technology

[0002] Kidney transplantation is one of the main treatment methods for restoring kidney function in patients with end-stage renal disease. Recipients still face risks such as acute rejection and chronic rejection-related damage after kidney transplantation. If rejection is not identified and intervened in a timely manner in the early stages, it can easily lead to rapid deterioration of transplanted kidney function, thereby increasing the probability of adverse outcomes such as transplant failure and re-dialysis. Therefore, how to continuously monitor, accurately assess, and provide timely warnings of rejection risk during postoperative follow-up management is an important issue in long-term post-transplant management.

[0003] Current methods for monitoring rejection primarily rely on laboratory indicators (such as serum creatinine and urine protein) and imaging examinations (such as ultrasound blood flow parameters of the transplanted kidney), combined with clinical symptoms and signs for comprehensive judgment. However, these indicators have certain lag and nonspecificity: for example, serum creatinine is affected by multiple factors such as dehydration, infection, and drug toxicity, which may lead to misjudgment or missed diagnosis of rejection risk; ultrasound blood flow parameters are affected by operator experience, timing of examination, and individual differences, resulting in variability and insufficient repeatability. At the same time, a large amount of nursing monitoring data during the postoperative follow-up period (such as urine output, weight, body temperature, and blood pressure) is usually used for daily records and empirical observation, and a unified assessment framework for quantifying rejection risk has not yet been formed, making it difficult to fully utilize its value as an early sensitive signal.

[0004] With the development of medical informatization and intelligentization, some studies have attempted to use machine learning or deep learning models to predict post-transplant risks, but the following shortcomings still exist: First, most models use single test results or single time-point features as input, lacking time stamp alignment and time-series modeling of multi-source heterogeneous data such as nursing, laboratory, and ultrasound data, making it difficult to reflect the trend characteristics of risk evolution over time; Second, the model output is usually a probability or score, lacking interpretable analysis of the contribution of nursing monitoring indicators, making it difficult for nursing staff to clearly identify which nursing-sensitive indicators should be focused on, resulting in insufficient clinical usability and feasibility; Third, existing solutions mostly use population statistical thresholds for early warning, without fully considering the individualized baseline, allowable fluctuation range, and deviation behavior characteristics of the recipient, which easily leads to frequent false alarms or missed alarms when there are large individual differences; Fourth, after early warning, there is a lack of task-oriented and closed-loop intervention organization methods that match the risk level, making it difficult to form a traceable and verifiable graded treatment process. Summary of the Invention

[0005] To at least partially overcome the problems in related technologies, such as reliance on single or lagging indicators for rejection risk monitoring, lack of time-series fusion and trend assessment of multi-source follow-up data, lack of quantification of nursing sensitivity indicators leading to insufficient interpretability and feasibility of early warning, and the tendency to generate false alarms and missed alarms due to failure to combine individualized baseline and allowable fluctuation range of the recipient, this application provides a method and system for monitoring and early warning of rejection risk after kidney transplantation.

[0006] The proposed solution is as follows:

[0007] According to a first aspect of the embodiments of this application, a method for monitoring and early warning of rejection risk after kidney transplantation is provided, characterized in that it includes:

[0008] During the current follow-up period after kidney transplantation, nursing monitoring data, laboratory indicator data, and ultrasound blood flow parameters of the transplanted kidney are collected from the recipient.

[0009] The collected data is subjected to timestamp alignment, missing data handling, and standardization. Feature extraction is then performed to construct a temporal feature sequence.

[0010] The time-series feature sequence is input into a pre-trained rejection reaction risk prediction model, which outputs the probability of rejection reaction in the next period and combines it with the probability of rejection reaction in historical periods to form a risk trend.

[0011] Nursing monitoring indicators are determined. Based on the probability of rejection, a nursing sensitivity indicator weight analysis algorithm is executed to calculate the contribution weight of each nursing monitoring indicator to the probability of rejection. Nursing monitoring indicators with contribution weights higher than preset weight values ​​are used as target nursing monitoring indicators.

[0012] Based on the nursing monitoring data corresponding to the target nursing monitoring indicators, an individualized baseline model of the recipient is constructed to obtain the baseline value and allowable fluctuation range of the target nursing monitoring indicators, and the deviation features of the nursing monitoring data corresponding to the target nursing monitoring indicators relative to the baseline value and allowable fluctuation range are extracted.

[0013] The probability of the rejection reaction is individually calibrated based on the deviation characteristics to obtain a calibration risk score;

[0014] When the calibration risk score fails to reach the preset risk warning score threshold and the risk trend does not meet the preset trend warning conditions, nursing monitoring recommendations for the next cycle are generated based on the target nursing monitoring indicators.

[0015] When the calibrated risk score reaches the preset risk warning score threshold, or when the risk trend meets the preset trend warning conditions, a warning message is pushed to the nursing terminal, and a graded intervention recommendation for the next cycle is generated based on the risk level corresponding to the calibrated risk score and the target nursing monitoring indicators.

[0016] Preferably, nursing monitoring indicators are determined, and based on the probability of rejection, a nursing sensitivity indicator weighting analysis algorithm is executed to calculate the contribution weight of each nursing monitoring indicator to the probability of rejection, including:

[0017] Obtain the contribution information of each time-series feature corresponding to the rejection reaction risk prediction model when outputting the probability of rejection reaction occurrence;

[0018] The contribution degree corresponding to each nursing monitoring indicator is extracted from the contribution information of time-series features, and the contribution degree is aggregated in the time dimension to obtain the initial contribution weight of each nursing monitoring indicator.

[0019] The initial contribution weights were subjected to stability testing and normalization to obtain the contribution weights of each nursing monitoring indicator to the probability of the rejection reaction.

[0020] The stability test includes at least: determining whether the initial contribution weight meets the continuity threshold and / or volatility threshold within a preset sliding time window, so as to suppress weight distortion caused by occasional anomalies.

[0021] Preferably, the contribution information of each time-series feature corresponding to the rejection reaction risk prediction model when outputting the probability of rejection reaction occurrence includes:

[0022] The gradient information of the probability of the rejection reaction occurring relative to each feature value at each time step in the temporal feature sequence is calculated based on backpropagation.

[0023] The contribution of each temporal feature is determined based on the gradient information; the contribution includes: the product of the value of the temporal feature and the gradient information, and the integrated gradient value obtained by integrating the gradient information.

[0024] Preferably, the allowable fluctuation range is adaptively determined based on the historical distribution of the target nursing monitoring indicator, including:

[0025] Within the pre-defined baseline sample segment, the standard deviation is calculated based on the historical distribution of the target nursing monitoring indicators;

[0026] Centered on the baseline value, an allowable fluctuation range is formed by setting the baseline value ± k times the standard deviation;

[0027] Where k is a configurable parameter, and can be set in segments or dynamically adjusted according to the postoperative time stage.

[0028] Preferably, extracting deviation features includes:

[0029] Within a preset time window, calculate the difference or ratio between the current value and the baseline value as the deviation from the baseline;

[0030] Use the sign of the deviation magnitude as the deviation direction;

[0031] Calculate the slope of the deviation magnitude over the time dimension as the deviation rate;

[0032] Calculate the variance, standard deviation, or coefficient of variation of the deviation as the degree of volatility;

[0033] The cumulative duration or the cumulative number of sampling points during which the deviation exceeds the allowable fluctuation range is calculated as the continuous deviation duration.

[0034] Preferably, generating nursing monitoring recommendations for the next cycle based on the target nursing monitoring indicators includes:

[0035] Based on the baseline values ​​and allowable fluctuation ranges of the target nursing monitoring indicators, the recommended collection frequency, collection time points, and collection methods for each target nursing monitoring indicator in the next cycle are determined.

[0036] Based on the deviation characteristics of the target nursing monitoring indicator in the current period, key monitoring prompts are generated for the deviation direction, deviation magnitude, and continuous deviation duration, and the retest trigger conditions for the target nursing monitoring indicator are determined; the retest trigger conditions include at least one or more of the following: deviation magnitude exceeds a preset magnitude threshold, deviation rate exceeds a preset rate threshold, and continuous deviation duration reaches a preset duration threshold.

[0037] Preferably, based on the risk level corresponding to the calibrated risk score and the target care monitoring indicators, a tiered intervention recommendation for the next cycle is generated, including:

[0038] The calibration risk score is divided into corresponding risk levels according to a preset mapping rule;

[0039] Based on the risk level, an intervention template is obtained by matching from a pre-configured tiered intervention strategy library. The intervention template includes at least a set of intervention actions, a set of execution roles, and an execution time limit.

[0040] Based on the abnormality types and deviation characteristics of the target nursing monitoring indicators, intervention actions corresponding to the abnormality types are selected from the set of intervention actions to form an intervention task list for the next cycle. The intervention task list includes at least one or more of the following: communication and follow-up tasks, compliance verification tasks, follow-up assessment tasks, and diagnosis and treatment collaboration tasks.

[0041] Configure task priority, execution role and closed-loop verification conditions for each intervention task in the intervention task list, and configure task time limit for each intervention task in the intervention task list according to the execution time limit.

[0042] Preferably, the probability of the rejection reaction is individually calibrated based on the deviation characteristics to obtain a calibration risk score, including:

[0043] The deviation features are quantified according to the deviation direction, deviation magnitude, deviation rate, fluctuation degree, and continuous deviation duration to generate a deviation feature vector;

[0044] A personalized calibration factor is constructed based on the deviation feature vector; the personalized calibration factor is positively correlated with the deviation magnitude, deviation rate, and continuous deviation duration, and is also associated with the deviation direction;

[0045] The probability of the rejection reaction is corrected based on the individualized calibration factor, and a calibration risk score is output.

[0046] The calibration risk score is subject to boundary constraints and smoothing to suppress score jumps caused by occasional anomalies.

[0047] Preferably, the nursing monitoring data includes at least one or more of the following: daily urine output, weight change, body temperature, and blood pressure;

[0048] The laboratory indicator data shall include at least one or more of the following: serum creatinine, urine protein, and blood concentration of immunosuppressant drugs.

[0049] The ultrasound blood flow parameters of the transplanted kidney include at least: blood flow velocity parameters and / or resistance-related parameters;

[0050] The preset trend warning conditions include at least one or more of the following: the risk trend rises continuously for a preset number of cycles, the rate of increase of the risk trend exceeds a preset rate threshold, and the cumulative increase of the risk trend within a preset time window exceeds a preset magnitude threshold.

[0051] According to a second aspect of the embodiments of this application, a system for monitoring and early warning of rejection risk after kidney transplantation is provided, comprising:

[0052] Processor and memory;

[0053] The processor and memory are connected via a communication bus:

[0054] The processor is used to call and execute the program stored in the memory;

[0055] The memory is used to store a program, which is at least used to execute a method for monitoring and early warning of rejection risk after kidney transplantation as described in any of the above.

[0056] The technical solution provided in this application may include the following beneficial effects:

[0057] This application achieves continuous monitoring and quantitative assessment of post-kidney transplant rejection risk by integrating multi-source heterogeneous data and performing time-series modeling. By identifying nursing sensitivity indicators and constructing individualized baselines for recipients, this method enables precise risk warning, reducing false positives and false negatives. Furthermore, it generates targeted nursing monitoring and tiered intervention recommendations, improving clinical usability and intervention efficiency, thereby facilitating early identification and timely intervention for post-kidney transplant rejection and improving recipient prognosis.

[0058] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0059] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0060] Figure 1 This is a flowchart illustrating a method for monitoring and early warning of rejection risk after kidney transplantation, provided in one embodiment of this application.

[0061] Figure 2 This is a schematic diagram of a kidney transplant rejection risk monitoring and early warning system provided in one embodiment of this application.

[0062] Reference numerals: Processor-21; Memory-22. Detailed Implementation

[0063] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0064] Example 1

[0065] Figure 1 This is a flowchart illustrating a method for monitoring and early warning of rejection risk after kidney transplantation, provided in one embodiment of this application. (Refer to...) Figure 1 A method for monitoring and early warning of rejection risk after kidney transplantation, comprising:

[0066] S11. During the current follow-up period after kidney transplantation, collect the recipient's nursing monitoring data, laboratory indicator data, and ultrasound blood flow parameters of the transplanted kidney.

[0067] S12. Perform timestamp alignment, missing data handling, and standardization on the collected data, extract features, and construct a temporal feature sequence.

[0068] S13. Input the time series feature sequence into the pre-trained rejection reaction risk prediction model, output the probability of rejection reaction in the next period, and combine it with the probability of rejection reaction in the historical period to form a risk trend.

[0069] S14. Determine nursing monitoring indicators. Based on the probability of rejection, execute the nursing sensitivity indicator weight analysis algorithm to calculate the contribution weight of each nursing monitoring indicator to the probability of rejection. Nursing monitoring indicators with contribution weights higher than the preset weight values ​​are used as target nursing monitoring indicators.

[0070] S15. Based on the nursing monitoring data corresponding to the target nursing monitoring indicators, construct an individualized baseline model for the recipient, obtain the baseline value and allowable fluctuation range of the target nursing monitoring indicators, and extract the deviation characteristics of the nursing monitoring data corresponding to the target nursing monitoring indicators relative to the baseline value and allowable fluctuation range; the deviation characteristics include: the deviation magnitude, deviation direction, deviation rate, fluctuation degree and continuous deviation duration relative to the baseline;

[0071] S16. Based on the deviation characteristics, the probability of rejection reaction is individually calibrated to obtain a calibration risk score;

[0072] S17. When the calibration risk score does not reach the preset risk warning score threshold and the risk trend does not meet the preset trend warning conditions, generate nursing monitoring recommendations for the next cycle based on the target nursing monitoring indicators.

[0073] S18. When the calibrated risk score reaches the preset risk warning score threshold, or when the risk trend meets the preset trend warning conditions, push the warning information to the nursing terminal, and generate the next cycle's graded intervention recommendations based on the risk level corresponding to the calibrated risk score and the target nursing monitoring indicators.

[0074] It should be noted that:

[0075] Nursing monitoring data refers to physiological and behavioral data collected by nursing staff or through automated equipment during the follow-up process after kidney transplantation, such as daily urine output, weight changes, body temperature, and blood pressure. This type of data reflects the recipient's daily condition and is both real-time and continuous.

[0076] Laboratory parameters refer to biochemical or immunological parameters obtained through testing of samples such as blood and urine, including serum creatinine, urine protein, and blood concentrations of immunosuppressants. These data are commonly used to assess renal function and drug efficacy.

[0077] Transplant kidney ultrasound blood flow parameters refer to hemodynamic indicators of the transplant kidney obtained through ultrasound examination, such as blood flow velocity parameters and / or resistance-related parameters. These parameters can be used to assess the perfusion status of the transplant kidney and the presence of vascular lesions.

[0078] A temporal feature sequence refers to a set of features extracted from raw data of different sources and sampling frequencies after timestamp alignment, missing data handling, and standardization. This sequence can reflect the changing patterns of a recipient's physiological state over time.

[0079] A rejection risk prediction model is a machine learning or deep learning model that, after being trained on a large amount of historical data, can predict the probability of a recipient experiencing a rejection reaction within a future period based on the input time-series feature sequence. This model aims to provide a quantitative risk assessment and can use a long short-term memory network to learn the correlation patterns between the temporal evolution of various parameters and rejection events, outputting the probability of a rejection reaction occurring in the next 7 to 14 days.

[0080] The Nursing Sensitivity Indicator Weighting Analysis Algorithm is an algorithm used to assess the contribution of various nursing monitoring indicators to the probability of rejection. This algorithm helps identify nursing monitoring indicators with a high correlation to the risk of rejection by quantifying the importance of these indicators.

[0081] Targeted nursing monitoring indicators refer to nursing monitoring indicators that, after being screened using a nursing sensitivity indicator weighting analysis algorithm, are identified as having a high contribution weight to the probability of rejection. These indicators are the focus of subsequent individualized monitoring and intervention.

[0082] An individualized baseline model is a model built based on the recipient's own historical nursing monitoring data to determine the individualized baseline value and allowable fluctuation range of a specific target nursing monitoring indicator. This model takes into account individual differences and avoids using a uniform group threshold.

[0083] Deviation characteristics refer to the differences between the current value of a target nursing monitoring indicator and its individualized baseline value and allowable fluctuation range. These characteristics include the magnitude of deviation, direction of deviation, rate of deviation, degree of fluctuation, and duration of continuous deviation, and are used to finely describe the abnormal state of the indicator.

[0084] Extracting deviation features includes:

[0085] Within a preset time window, calculate the difference or ratio between the current value and the baseline value as the deviation from the baseline;

[0086] Use the sign of the deviation magnitude as the deviation direction;

[0087] Calculate the slope of the deviation magnitude over the time dimension as the deviation rate;

[0088] Calculate the variance, standard deviation, or coefficient of variation of the deviation as the degree of volatility;

[0089] The cumulative duration or the cumulative number of sampling points during which the deviation exceeds the allowable fluctuation range is calculated as the continuous deviation duration.

[0090] Deviation direction refers to whether the current value of a target nursing care monitoring indicator is higher or lower than its baseline value. This characteristic is determined by judging the sign of the deviation magnitude; for example, a positive difference indicates an increase in the indicator, while a negative difference indicates a decrease. For ratios, a value greater than 1 indicates an increase, and a value less than 1 indicates a decrease. Determining the deviation direction is crucial for understanding the nature of physiological changes; for example, an increase in body temperature and a decrease in body temperature may indicate different clinical problems.

[0091] The rate of deviation refers to the speed at which the deviation of a target nursing care monitoring indicator changes over time. This characteristic can be obtained by calculating the slope of the deviation over a preset time period. For example, the steepness of the trend can be calculated by comparing the deviation over multiple consecutive sampling points or time windows. The rate of deviation can reflect the urgency of changes in physiological state; for example, a rapidly rising creatinine level may indicate acute kidney injury, while a slow rise may indicate chronic progression.

[0092] Volatility refers to the stability or variability of the deviation of a target nursing monitoring indicator over a period of time. This characteristic can be quantified by calculating the variance, standard deviation, or coefficient of variation of the deviation. Variance and standard deviation directly reflect the dispersion of the data, while the coefficient of variation is the relative dispersion considering the mean. High volatility may indicate physiological instability, and even if the average value is within the normal range, there may be potential risks.

[0093] Continuous deviation duration refers to the length of time or number of sampling points during which the deviation of a target nursing monitoring indicator continuously exceeds its allowable fluctuation range. This characteristic is calculated by the cumulative duration or number of sampling points where the deviation exceeds the allowable fluctuation range. For example, if an indicator exceeds the allowable fluctuation range for three consecutive days, the continuous deviation duration is three days. Continuous deviation duration reflects the persistence of the abnormal state; prolonged continuous deviation often has higher clinical significance and risk than short-term deviation.

[0094] Through the refined extraction of deviation features described above, this application can comprehensively and deeply characterize the abnormal state of individualized nursing monitoring indicators for recipients from multiple dimensions. Specifically, the deviation magnitude relative to the baseline quantifies the degree of abnormality, the deviation direction clarifies the nature of the abnormality (increase or decrease), the deviation rate reveals the urgency of the abnormal change, the degree of fluctuation reflects the stability of the abnormal state, and the duration of continuous deviation indicates the persistence of the abnormal state. These multi-dimensional deviation features together constitute a richer and more accurate profile of the individualized physiological state, enabling subsequent individualized calibration of the probability of rejection to more accurately reflect the recipient's true risk. Compared to extracting only single or vague deviation features, this refined feature extraction method can identify potential rejection risk signals earlier and more accurately, avoiding misjudgments or omissions due to insufficient information. This provides nursing staff with more targeted and timely early warning information and graded intervention suggestions, significantly improving the effectiveness and individualization of post-kidney transplant rejection risk monitoring and early warning.

[0095] The calibrated risk score is a risk assessment score obtained by adjusting the probability of rejection by incorporating individualized deviation characteristics. This score comprehensively considers both group risk and individual specificity, providing a more accurate risk assessment.

[0096] The risk warning score threshold refers to a preset calibration risk score cutoff value used to trigger a risk warning. When the calibration risk score reaches or exceeds this threshold, the system will issue a warning.

[0097] A risk trend refers to the overall direction in which the probability of a rejection reaction occurs changes over time. This trend is formed by combining the probability of occurrence with historical cycles and is used to assess the dynamic evolution of risk.

[0098] Trend warning conditions refer to preset risk trend change patterns used to trigger risk warnings. For example, a risk trend may rise continuously for a preset number of periods, the rate of increase may exceed a preset threshold, or the cumulative increase may exceed a preset threshold.

[0099] Nursing monitoring recommendations refer to guidance provided to nursing staff regarding the frequency, timing, methods, and key concerns of the next monitoring cycle, based on the current status of the target nursing monitoring indicators, when the risk has not reached the warning level.

[0100] Early warning information refers to the prompt information pushed by the system to the nursing staff when the risk reaches the warning conditions, which aims to remind nursing staff to pay attention to the recipient's condition and take appropriate measures.

[0101] Tiered intervention recommendations refer to a list of specific and actionable interventions provided to nursing staff or healthcare teams based on the risk level corresponding to the calibrated risk score and any abnormalities in the target care monitoring indicators. This recommendation aims to achieve precise and closed-loop risk management.

[0102] The implementation of a method for monitoring and early warning of rejection risk after kidney transplantation may include the following steps:

[0103] During the current follow-up period after kidney transplantation, recipient nursing monitoring data, laboratory indicator data, and ultrasound blood flow parameters of the transplanted kidney are collected. For example, nursing staff can manually record daily urine output, weight, body temperature, blood pressure, and other nursing monitoring data; laboratories can periodically test the recipient's serum creatinine, urine protein, and other laboratory indicators; and the radiology department can perform ultrasound examinations of the recipient's transplanted kidney to obtain blood flow parameters. This data can be manually entered or exported from different information systems (such as electronic medical record systems, laboratory systems, and imaging systems).

[0104] The collected nursing monitoring data, laboratory indicator data, and transplant kidney ultrasound blood flow parameters were time-stamp aligned, missing data were handled, and standardized within the current kidney transplant follow-up period. Subsequently, feature extraction was performed to construct a temporal feature sequence. For example, for data collected at different time points, time-stamp alignment could be performed using nearest neighbor interpolation or linear interpolation; for missing data, mean imputation or deletion of records containing missing values ​​could be used; for data with different dimensions, maximum / minimum normalization or Z-score standardization could be used. Feature extraction could involve directly using the original indicators as features and arranging them in chronological order to form a sequence.

[0105] The time-series feature sequence is input into a pre-trained rejection reaction risk prediction model to output the probability of rejection reaction occurring in the next period. This probability is then combined with the rejection reaction probabilities from historical periods to form a risk trend. For example, the prediction model could be a simple regression model built on statistical principles, whose output is a value between 0 and 1, representing the likelihood of rejection reaction occurring. The risk trend can be determined by simply comparing the probability of occurrence over several consecutive periods to determine whether it is increasing, decreasing, or remaining stable.

[0106] Once nursing monitoring indicators are identified, a nursing sensitivity indicator weighting analysis algorithm is executed based on the probability of rejection. This algorithm calculates the contribution weight of each nursing monitoring indicator to the probability of rejection, and indicators with contribution weights higher than preset weight values ​​are designated as target nursing monitoring indicators. For example, clinical experts can directly designate certain nursing monitoring indicators (such as urine output and weight) as sensitivity indicators based on experience, assigning them fixed weight values. When the weights of these indicators exceed preset thresholds, they are identified as target nursing monitoring indicators.

[0107] Based on the nursing monitoring data corresponding to the target nursing monitoring indicator, an individualized baseline model is constructed for the recipient to obtain the baseline value and allowable fluctuation range of the target nursing monitoring indicator. Subsequently, the deviation characteristics of the nursing monitoring data corresponding to the target nursing monitoring indicator relative to the baseline value and allowable fluctuation range are extracted. These deviation characteristics may include: the magnitude of deviation relative to the baseline, the direction of deviation, the rate of deviation, the degree of fluctuation, and the duration of continuous deviation. For example, the individualized baseline model can simply calculate the average value of a target nursing monitoring indicator for the recipient during the stable period as the baseline value, and set a fixed percentage range (e.g., ±10%) as the allowable fluctuation range. The extraction of deviation characteristics can simply calculate the difference between the current value and the baseline value as the magnitude of deviation, and determine the direction of deviation based on the sign of the difference.

[0108] The calibration risk score is obtained by individually calibrating the probability of rejection based on the deviation characteristics. For example, individualized calibration can be achieved by multiplying a simple quantitative value of the deviation characteristics (such as the deviation magnitude) by a preset calibration coefficient, and then adding or subtracting it from the original probability of rejection.

[0109] When the calibration risk score fails to reach the preset risk warning score threshold, and the risk trend does not meet the preset trend warning conditions, nursing monitoring recommendations for the next cycle are generated based on the target nursing monitoring indicator. For example, when the risk is low, the system can recommend that nursing staff monitor all target nursing monitoring indicators at the usual frequency based on a preset general template, without adjusting for specific deviations.

[0110] When the calibrated risk score reaches a preset risk warning score threshold, or when the risk trend meets preset trend warning conditions, a warning message is pushed to the nursing staff. Simultaneously, based on the risk level corresponding to the calibrated risk score and the target nursing monitoring indicator, a tiered intervention recommendation for the next cycle is generated. For example, when the calibrated risk score exceeds a certain fixed threshold, the system can send a simple text warning notification to the nursing staff. Tiered intervention recommendations can be based on preset simple rules; for example, when the risk level is "high," a uniform recommendation is made to immediately contact the attending physician, regardless of the specific type of abnormal indicator.

[0111] This application achieves continuous monitoring and quantitative assessment of post-kidney transplant rejection risk by integrating multi-source heterogeneous data and performing time-series modeling. By identifying nursing sensitivity indicators and constructing individualized baselines for recipients, this method enables precise risk warning, reducing false positives and false negatives. Furthermore, it generates targeted nursing monitoring and tiered intervention recommendations, improving clinical usability and intervention efficiency, thereby facilitating early identification and timely intervention for post-kidney transplant rejection and improving recipient prognosis.

[0112] Example 2

[0113] It should be noted that, based on the probability of rejection, the nursing monitoring indicators are determined using a nursing sensitivity indicator weighting analysis algorithm to calculate the contribution weight of each nursing monitoring indicator to the probability of rejection, including:

[0114] Obtain the contribution information of each time-series feature when the rejection reaction risk prediction model outputs the probability of rejection reaction occurrence;

[0115] The contribution degree corresponding to each nursing monitoring indicator is extracted from the contribution information of time-series features, and the contribution degree is aggregated in the time dimension to obtain the initial contribution weight of each nursing monitoring indicator.

[0116] The initial contribution weights were subjected to stability testing and normalization to obtain the contribution weights of each nursing monitoring indicator to the probability of rejection.

[0117] The stability test includes at least the following: determining whether the initial contribution weights meet the continuity threshold and / or volatility threshold within a preset sliding time window, so as to suppress weight distortion caused by occasional anomalies.

[0118] Specifically, the contribution information of each time-series feature corresponding to the output probability of rejection reaction in the rejection reaction risk prediction model is obtained, including:

[0119] The gradient information of the probability of rejection occurring relative to each feature value at each time step in the time-series feature sequence is calculated based on backpropagation.

[0120] The contribution of each temporal feature is determined based on the gradient information; the contribution includes the product of the temporal feature value and the gradient information, as well as the integrated gradient value obtained by integrating the gradient information.

[0121] This application calculates the gradient information of the probability of rejection with respect to each feature value at each time step in the time-series feature sequence based on backpropagation. Backpropagation is a widely used algorithm in neural networks. Its core idea is to use the chain rule to calculate the gradient of the loss function (which can be understood here as the probability of rejection) with respect to the model input (each feature value at each time step in the time-series feature sequence). In specific implementation, after the rejection risk prediction model completes forward propagation to calculate the probability of rejection, there is no need to retrain the model. Instead, automatic differentiation is used to trace back along the computational graph of the model to calculate the partial derivative of the output probability with respect to each input feature. These partial derivatives constitute the gradient information, which quantifies the degree and direction of the influence of small changes in each input feature on the probability of rejection. For example, if the gradient of a certain feature is positive and large, it indicates that an increase in the value of that feature will significantly increase the probability of rejection.

[0122] After acquiring gradient information, this application further determines the contribution of each temporal feature based on the gradient information. The contribution is a quantitative indicator measuring the degree of influence of each temporal feature on the probability of a rejection response. One way to determine the contribution is to calculate the product of the temporal feature value and the gradient information. This method intuitively combines the actual value of the feature with its local sensitivity to the output to estimate its contribution. Another more robust way to determine the contribution is to integrate the gradient information to obtain an integrated gradient value. The integrated gradient method accumulates gradient information along the way by defining a path in the input space from the baseline (e.g., an input where all features are zero or average) to the actual input and integrating the gradient along that path. This integration process effectively solves the gradient saturation problem that may occur in nonlinear models using simple gradient methods, more comprehensively captures the global contribution of features, and makes the contribution assessment more accurate and stable.

[0123] Through the above technical solution, this application can accurately obtain the contribution information of each time-series feature in the rejection risk prediction model to the probability of rejection. Specifically, by calculating the gradient information of the probability of rejection relative to the feature values ​​at each time step in the time-series feature sequence, the direction and intensity of the influence of each input feature on the model output can be quantified. Furthermore, by multiplying the value of the time-series feature with the gradient information, or by integrating the gradient information to obtain the integrated gradient value, the contribution of each time-series feature can be determined more comprehensively and accurately. This method overcomes the limitation of insufficient interpretability of traditional models, making the sensitivity analysis of various nursing monitoring indicators more reliable, providing a solid foundation for subsequent contribution weight calculation, thereby improving the accuracy and interpretability of rejection risk monitoring and early warning.

[0124] Subsequently, the contribution values ​​corresponding to each nursing monitoring indicator are extracted from the contribution information of the time-series features, and these contribution values ​​are aggregated along the time dimension to obtain the initial contribution weights of each nursing monitoring indicator. The purpose of this step is to map and summarize the contribution information obtained from the model for fine-grained time-series features onto more macroscopically meaningful nursing monitoring indicators. A nursing monitoring indicator (e.g., daily urine output) may correspond to multiple time-series features (e.g., urine output values ​​at different time points, urine output change rate, etc.). Therefore, it is necessary to identify all time-series features associated with a specific nursing monitoring indicator and integrate their respective contribution values. Aggregation along the time dimension can employ methods such as summation, averaging, or weighted averaging to combine the contribution values ​​of the same nursing monitoring indicator at different time steps into a single initial contribution weight, thereby reflecting the comprehensive impact of the indicator on the probability of rejection throughout the entire follow-up period.

[0125] The initial contribution weights are subjected to stability testing and normalization to obtain the contribution weights of each nursing monitoring indicator to the probability of rejection. The stability test is a crucial step in ensuring the reliability of the calculated weights, aiming to suppress weight distortion caused by occasional abnormal data or instantaneous model fluctuations. Specifically, the stability test includes at least determining whether the initial contribution weights meet continuity and / or volatility thresholds within a preset sliding time window. For example, the system can maintain a historical sliding time window covering the most recent N follow-up periods or time points. For the initial contribution weight of each nursing monitoring indicator, its mean, standard deviation, or coefficient of variation can be calculated within the sliding time window. If the initial contribution weight is below a preset continuity threshold for X consecutive periods or time points within the sliding time window, or its volatility (e.g., standard deviation) exceeds a preset volatility threshold, the weight is considered unstable and requires adjustment or smoothing. In this way, weight fluctuations caused by short-term noise or random events can be effectively filtered out, ensuring that the final contribution weights reflect the long-term, stable importance of the indicators. Normalization is usually performed after stability testing, aiming to adjust the contribution weights of the stability-tested indicators to a uniform scale. For example, the sum of the contribution weights of all nursing monitoring indicators is made to 1, so as to facilitate subsequent comparisons and threshold setting.

[0126] Through the aforementioned technical solution, this application achieves a deep understanding of the decision-making mechanism within the rejection risk prediction model, thereby accurately quantifying the impact of various time-series features on the prediction results. Furthermore, by aggregating these contribution information onto specific nursing monitoring indicators and performing stability tests and normalization, interference from occasional abnormal data in the contribution weight calculation is effectively avoided, ensuring that the determined contribution weights of each nursing monitoring indicator are more stable, reliable, and practically instructive. This enables the system to more accurately identify key nursing monitoring indicators that significantly influence rejection risk, providing a solid foundation for subsequent individualized baseline model construction, deviation feature extraction, and the generation of nursing monitoring recommendations and tiered intervention recommendations, significantly improving the accuracy and effectiveness of risk monitoring and early warning.

[0127] Example 3

[0128] It should be noted that the allowable fluctuation range is adaptively determined based on the historical distribution of the target nursing monitoring indicators, including:

[0129] Within the pre-defined baseline sample segment, the standard deviation is calculated based on the historical distribution of the target nursing monitoring indicators;

[0130] Centered on the baseline value, an allowable fluctuation range is formed by setting the baseline value ± k times the standard deviation;

[0131] Where k is a configurable parameter, and can be set in segments or dynamically adjusted according to the postoperative time stage.

[0132] The allowable fluctuation range is adaptively determined based on the historical distribution of the target nursing monitoring indicators. This aims to enable the monitoring system to define the normal physiological fluctuation range based on the individual's physiological data characteristics, rather than a uniform fixed threshold. By analyzing individual historical data, the system can learn and adapt to the individual's unique physiological patterns, thereby more accurately identifying true abnormal deviations and avoiding false alarms or missed alarms due to individual differences. This adaptability ensures personalized and accurate monitoring.

[0133] Within a pre-defined baseline sample segment, the standard deviation is calculated based on the historical distribution of the target nursing monitoring indicators. This pre-defined baseline sample segment typically refers to nursing monitoring data collected from the recipient during a specific period after kidney transplantation, when their physiological state is relatively stable and there are no significant rejection reactions or other complications. For example, it could be data from several consecutive days or weeks during a stable period post-surgery. Selecting a stable baseline sample segment ensures that the calculated historical distribution accurately reflects the physiological fluctuations of the recipient in a healthy or stable state, providing a reliable reference for subsequent abnormality assessment. The historical distribution of the target nursing monitoring indicators refers to the set of values ​​and statistical characteristics of specific target nursing monitoring indicators (such as daily urine output, weight change, body temperature, blood pressure, etc.) over time within the baseline sample segment. Analyzing this historical data allows us to understand the average level, fluctuation range, and degree of variability of the indicator under stable conditions. Calculating the standard deviation is a statistic that measures the dispersion of data. Here, calculating the standard deviation of the target nursing monitoring indicator within the baseline sample segment quantifies the natural fluctuation range of the indicator under stable conditions. A larger standard deviation indicates a wider fluctuation range of the indicator under stable conditions; conversely, a smaller standard deviation indicates a narrower fluctuation range. This value is a key parameter for constructing an individualized allowable fluctuation range, reflecting the inherent characteristics of the individual's physiological fluctuations.

[0134] Centered on the baseline value, an allowable fluctuation range is formed by setting the baseline value ± k times the standard deviation. The baseline value refers to the reference value obtained through an individualized baseline model, representing the target nursing monitoring indicator under stable conditions. The "k" in the k times the standard deviation is a configurable multiplier parameter used to adjust the width of the allowable fluctuation range. By adding or subtracting k times the standard deviation from the baseline value, an interval centered on the baseline value and with a width of 2k times the standard deviation can be constructed. This interval defines the range of physiological fluctuations that are statistically considered normal. For example, when k=1, the interval covers approximately 68% of normal fluctuations; when k=2, it covers approximately 95%; and when k=3, it covers approximately 99.7%. By adjusting the k value, the sensitivity and specificity of monitoring can be flexibly controlled to adapt to different clinical needs and risk preferences.

[0135] Here, 'k' is a configurable parameter, which can be set in segments or dynamically adjusted according to the postoperative time stage. The fact that 'k' is a configurable parameter means that it can be flexibly set and optimized based on clinical experience, expert consensus, or actual monitoring results. This flexibility allows the system to adapt to the needs of different medical institutions, different patient groups, or different monitoring purposes. Segmenting or dynamically adjusting according to the postoperative time stage is because the recipient's physiological state and indicator fluctuation characteristics may differ significantly at different stages after kidney transplantation. For example, fluctuations may be larger in the early postoperative period, while they are relatively stable during the stable period. By setting different 'k' values ​​at different postoperative time stages (such as within 1 week, 1 week to 1 month, 1 month to 3 months, and after 3 months), or dynamically adjusting the 'k' value according to the recipient's actual recovery, it can be ensured that the allowable fluctuation range always matches the recipient's current physiological state. This helps improve monitoring accuracy, avoids excessive false alarms in the early stage with large fluctuations, and can promptly capture minor abnormal changes during the stable period.

[0136] Through the above technical solution, this application can adaptively determine the allowable fluctuation range based on the historical distribution of the recipient's individual target nursing monitoring indicators, rather than using a uniform fixed threshold. Specifically, the standard deviation is calculated within a preset baseline sample segment, and the allowable fluctuation range is formed by setting the baseline value ± k times the standard deviation, centered on the baseline value. Simultaneously, the allowable k value can be segmented or dynamically adjusted according to the postoperative time stage. This approach allows the allowable fluctuation range to accurately reflect the individual physiological fluctuation characteristics of the recipient at different postoperative stages, effectively distinguishing between normal physiological fluctuations and potential abnormal deviations. Compared to a fixed threshold, this scheme significantly improves the accuracy and sensitivity of deviation feature extraction, reducing false positives and false negatives caused by individual differences or dynamic changes in physiological state. Therefore, subsequent individualized calibration of the probability of rejection based on deviation features will be more reliable, resulting in more accurate and timely risk warnings and tiered intervention recommendations. This helps clinicians and nurses to identify and intervene in potential rejection risks earlier, improving the management quality and prognosis of kidney transplant recipients.

[0137] Example 4

[0138] It should be noted that nursing monitoring recommendations for the next cycle are generated based on the target nursing monitoring indicators, including:

[0139] Based on the baseline values ​​and allowable fluctuation ranges of the target nursing monitoring indicators, the recommended collection frequency, collection time points, and collection methods for each target nursing monitoring indicator in the next cycle are determined.

[0140] Based on the deviation characteristics of the target nursing monitoring indicators in the current period, key monitoring prompts are generated for the direction, magnitude, and duration of deviation, and the retest trigger conditions for the target nursing monitoring indicators are determined. The retest trigger conditions include at least one or more of the following: deviation magnitude exceeds a preset magnitude threshold, deviation rate exceeds a preset rate threshold, and duration of deviation reaches a preset duration threshold.

[0141] The recommended collection frequency, collection time, and collection method for each target nursing monitoring indicator in the next cycle aim to provide recipients with personalized and prospective nursing monitoring guidance based on the individualized baseline values ​​and allowable fluctuation ranges of the target nursing monitoring indicators. Recommended collection frequency refers to the number of times measurements are suggested for a specific target nursing monitoring indicator (such as daily urine output, weight change, body temperature, blood pressure, etc.) within the next cycle, such as once daily, twice daily, or every other day. Collection time refers to the specific time of measurement suggested, such as 8:00 AM daily, before bedtime, or before and after specific medications. Collection method refers to the specific method or tool used for measurement, such as patient self-testing, nurse measurement, or automatic collection using specific intelligent monitoring equipment. By comprehensively considering the stability, volatility, and position of the indicator within the allowable fluctuation range, the system can intelligently adjust the intensity and method of monitoring, ensuring that unnecessary monitoring burden is reduced when the risk is low, and monitoring is strengthened when the indicator tends to be unstable.

[0142] The generation of key monitoring prompts targeting the direction, magnitude, and duration of deviation aims to provide nurses or recipients with targeted, high-priority monitoring guidance based on the deviation characteristics exhibited by the target nursing monitoring indicators within the current period. Key monitoring prompts are warning messages automatically generated by the system based on abnormal indicator performance, such as "Pay attention to continued weight gain," "Be alert to persistently high blood pressure," and "Pay attention to increased fluctuations in urine output." The direction of deviation indicates whether the indicator deviates upwards or downwards from the baseline, the magnitude of deviation quantifies the degree of deviation, and the duration of continuous deviation reflects the persistence of the abnormal state. These prompts help nurses quickly focus on the recipient's most pressing physiological indicators and their abnormal trends, improving monitoring efficiency and accuracy, and avoiding missing crucial risk signals.

[0143] The purpose of defining the retest trigger conditions for the target nursing monitoring indicators is to establish clear and operable conditions for immediate retesting or further evaluation of these indicators. The retest trigger conditions are a set of preset rules. When any deviation characteristic of the target nursing monitoring indicator meets any of these conditions, a retest is immediately triggered or medical staff are notified to intervene. For example, when the deviation magnitude (e.g., weight change) exceeds a preset magnitude threshold (e.g., a daily weight gain exceeding 2 kg), or the deviation rate (e.g., blood pressure change rate) exceeds a preset rate threshold (e.g., an hourly systolic blood pressure increase exceeding 20 mmHg), or the continuous deviation duration (e.g., body temperature remaining above 37.5℃ for more than 24 hours) reaches a preset duration threshold, the system will automatically issue a retest instruction. These conditions ensure that when indicators show significant abnormalities or a rapid deterioration trend, a secondary confirmation or emergency treatment process can be initiated promptly, gaining valuable time for early intervention.

[0144] Through the aforementioned technical solution, this application can dynamically generate nursing monitoring recommendations for the next cycle based on the baseline values, allowable fluctuation ranges, and deviation characteristics of the recipient's individualized target nursing monitoring indicators. Specifically, the recommended collection frequency, collection time points, and collection methods are determined based on the baseline values ​​and allowable fluctuation ranges, making monitoring activities more accurate and efficient, avoiding over- or under-monitoring, and reducing the burden on recipients and nursing staff. Simultaneously, generating key monitoring prompts based on the direction, magnitude, and duration of deviation can guide nursing staff's attention to key abnormal indicators and their trends, significantly improving the sensitivity and timeliness of risk identification. Furthermore, by setting clear retesting trigger conditions, such as deviation magnitude, deviation rate, or duration of continuous deviation exceeding preset thresholds, it ensures that retesting or intervention procedures can be quickly initiated when indicators show significant abnormalities. This effectively shortens the time interval from the appearance of risk signals to action, providing more refined and intelligent guidance for early warning and intervention of post-kidney transplant rejection, further improving the accuracy and clinical applicability of monitoring and early warning.

[0145] It should be noted that, based on the risk level corresponding to the calibrated risk score and the target care monitoring indicators, tiered intervention recommendations for the next cycle are generated, including:

[0146] The calibration risk score is divided into corresponding risk levels according to a preset mapping rule;

[0147] Intervention templates are matched from a pre-configured tiered intervention strategy library based on risk level. Each intervention template includes at least a set of intervention actions, a set of execution roles, and an execution time limit.

[0148] Based on the abnormality types and deviation characteristics of the target nursing monitoring indicators, intervention actions corresponding to the abnormality types are selected from the set of intervention actions to form the intervention task list for the next cycle. The intervention task list includes at least one or more of the following: communication and follow-up tasks, compliance verification tasks, follow-up assessment tasks, and diagnosis and treatment collaboration tasks.

[0149] Configure task priorities, execution roles, and closed-loop verification conditions for each intervention task in the intervention task list, and configure task time limits for each intervention task in the intervention task list according to the execution time limit.

[0150] First, the calibration risk scores are divided into corresponding risk levels according to a pre-defined mapping rule. This pre-defined mapping rule can transform continuous calibration risk scores into discrete, easily understood, and manageable risk levels. For example, by setting multiple scoring thresholds, the score range can be divided into multiple levels such as low risk, medium risk, high risk, and very high risk. These thresholds can be determined based on clinical experience, expert consensus, or historical data analysis to ensure the clinical effectiveness of the risk level classification.

[0151] Based on this, intervention templates are matched from a pre-configured tiered intervention strategy library according to the risk level. This library pre-stores standard intervention templates for different risk levels. Each template includes at least a set of intervention actions, a set of execution roles, and an execution time limit. For example, for a "high-risk" level, the intervention template might include intervention actions such as "immediately notify the attending physician," "arrange an urgent follow-up test for serum creatinine," and "strengthen patient health education." The set of execution roles specifies which medical staff (such as attending physicians, responsible nurses, pharmacists, etc.) are responsible for performing which intervention actions. The execution time limit specifies how long each intervention action should be completed to ensure timely intervention. The matching process can involve directly indexing the corresponding template based on the risk level.

[0152] Simultaneously, based on the abnormality types and deviation characteristics of the target nursing monitoring indicators, intervention actions corresponding to the abnormality types are selected from the set of intervention actions to form the intervention task list for the next cycle. This intervention task list includes at least one or more of the following: communication and follow-up tasks, compliance verification tasks, follow-up visit assessment tasks, and collaborative diagnosis and treatment tasks. For example, if the target nursing monitoring indicator shows a persistent decrease in urine output with a significant deviation, the system will select intervention actions related to "abnormal urine output" from the intervention template's set of intervention actions, such as "assessing the patient's fluid intake," "checking diuretic use," and "recommending the attending physician to adjust the fluid replacement regimen," and include them in the intervention task list. This selection mechanism ensures the individualization and precision of intervention measures, avoids unnecessary interventions, and improves intervention efficiency.

[0153] To ensure the orderly execution, clear responsibilities, and effective evaluation of intervention tasks, each intervention task in the intervention task list is assigned a task priority, execution role, and closed-loop verification conditions. A task time limit is also assigned to each intervention task in the intervention task list based on the execution time limit. Task priorities can be set according to risk level, severity of abnormal indicators, and clinical urgency. For example, a follow-up assessment task corresponding to a sharp increase in serum creatinine has the highest priority. The execution role clarifies which specific role is responsible for each task. Closed-loop verification conditions define the standards and verification methods for task completion. For example, the closed-loop condition for a follow-up communication task can be that patient feedback information has been recorded; the closed-loop condition for a follow-up assessment task can be that new test results have been entered and reviewed by the doctor. The task time limit sets a specific completion deadline for each specific task based on the execution time limit and task priority in the intervention template.

[0154] Through the aforementioned technical solutions, this application further provides systematic and individualized graded intervention recommendations based on the existing risk monitoring and early warning system for post-kidney transplant rejection. Transforming calibrated risk scores into clear risk levels makes clinical decision-making more intuitive. Matching intervention templates to risk levels ensures the standardization and timeliness of intervention measures. More importantly, by combining the abnormal types and deviation characteristics of target nursing monitoring indicators, the most suitable intervention tasks for individual patients can be accurately selected from the set of intervention actions, avoiding blind intervention and improving the targeting and effectiveness of interventions. By configuring priorities, execution roles, task time limits, and closed-loop verification conditions for intervention tasks, the smooth execution of the intervention process, clear division of responsibilities, and effective evaluation of intervention effects are ensured, thereby significantly improving the early intervention capability for post-kidney transplant rejection and the level of patient management.

[0155] Example 5

[0156] It should be noted that the probability of rejection is individually calibrated based on deviation characteristics to obtain a calibration risk score, including:

[0157] The deviation features are quantified according to the deviation direction, deviation magnitude, deviation rate, fluctuation degree, and continuous deviation duration to generate a deviation feature vector;

[0158] Individualized calibration factors are constructed based on deviation feature vectors; the individualized calibration factors are positively correlated with deviation magnitude, deviation rate, and duration of continuous deviation, and are also associated with the direction of deviation;

[0159] The probability of rejection is corrected based on the individualized calibration factor, and a calibration risk score is output.

[0160] Boundary constraints and smoothing processes are applied to the calibration risk score to suppress score jumps caused by occasional anomalies.

[0161] In some embodiments described above in this application, an individualized calibration of the probability of rejection based on deviation characteristics is proposed to obtain a calibrated risk score. However, in practical applications, how to effectively and stably transform multi-dimensional deviation characteristics into corrections to the probability of rejection to accurately reflect the individualized risk status of the recipient and avoid drastic fluctuations in the risk score due to occasional abnormal data is a technical problem that needs to be solved.

[0162] To address this, this application further proposes a specific method for individualizing the probability of the rejection response based on deviation characteristics to obtain a calibrated risk score. This method first quantifies the deviation characteristics according to deviation direction, deviation magnitude, deviation rate, volatility, and duration of continuous deviation, generating a deviation feature vector. Specifically, the deviation direction can be quantified as a discrete value representing deterioration or improvement; for example, upward deviation (deterioration trend) is assigned a value of +1, downward deviation (improvement trend) is assigned a value of -1, and no deviation is assigned a value of 0. The deviation magnitude, deviation rate, volatility, and duration of continuous deviation are directly calculated values. These quantified values ​​together constitute a multi-dimensional deviation feature vector, providing a structured input for subsequent risk correction.

[0163] Based on this, an individualized calibration factor is constructed using the aforementioned deviation feature vector. This individualized calibration factor is a crucial bridge connecting the individualized physiological deviation of the recipient with the correction of the probability of rejection. It aims to synthesize complex deviation patterns and generate a correction coefficient that reflects the current level of risk. For example, a weighted summation method can be used to assign different weights to each component in the deviation feature vector, performing linear or nonlinear combinations. These weights can be set through expert experience or learned from historical data to optimize the predictive ability of the calibration factor.

[0164] Furthermore, the individualized calibration factor is positively correlated with the magnitude of deviation, the rate of deviation, and the duration of continuous deviation, and is also associated with the direction of deviation. This means that when constructing the calibration factor, it is ensured that the magnitude of deviation, the rate of deviation, and the duration of continuous deviation are positive contributors in its calculation formula. When these characteristics increase, the calibration factor should also increase accordingly, resulting in a correction in the probability of rejection occurring towards an increased risk. Simultaneously, the calibration factor is adjusted based on whether the deviation is in a worsening or improving direction. For example, if the deviation indicates worsening, the calibration factor increases; if it indicates improvement, the calibration factor decreases or remains unchanged, ensuring that the calibration factor accurately reflects the degree and trend of physiological deterioration.

[0165] Subsequently, the probability of rejection is corrected based on the individualized calibration factor, and a calibration risk score is output. This step is the core of integrating individualized deviation information into the model prediction results. By using addition, multiplication, or nonlinear functions, the calibration factor is applied to the original probability of rejection, resulting in a more individualized and accurate risk assessment. For example, the calibration risk score can be corrected using the formula: Calibration Risk Score = Probability of Rejection + Individualized Calibration Factor, or Calibration Risk Score = Probability of Rejection * (1 + Individualized Calibration Factor).

[0166] Finally, the calibration risk scores are subject to boundary constraints and smoothing to suppress score jumps caused by occasional anomalies. Boundary constraints ensure that the calibrated risk scores always remain within a valid and meaningful range. For example, scores exceeding preset upper and lower limits are truncated to boundary values ​​using a truncation method, or the Sigmoid function or other mapping methods are used to limit them to between 0 and 1. Smoothing is achieved by processing calibration risk scores over multiple consecutive periods using methods such as moving averages, low-pass filtering, or time series models. This reduces drastic changes in risk scores caused by short-term, occasional data fluctuations or measurement errors, thereby improving the stability of risk assessment.

[0167] Through the aforementioned technical solution, this application can effectively quantify multi-dimensional deviation characteristics and construct an individualized calibration factor closely related to the recipient's physiological state change trend. This calibration factor is positively correlated with the deviation magnitude, deviation rate, and duration of continuous deviation, and is also correlated with the deviation direction, enabling the correction of the probability of rejection to accurately capture the individualized physiological deterioration or improvement trend of the recipient. Furthermore, the boundary constraints and smoothing processing applied to the calibration risk score effectively suppress drastic score jumps caused by occasional abnormal data, significantly improving the stability and reliability of risk assessment. This allows the system to provide more accurate risk warnings that better reflect the individual's actual situation, avoiding excessive or insufficient intervention, thereby improving the effectiveness and clinical applicability of post-kidney transplant rejection risk monitoring and warning.

[0168] Example 6

[0169] Figure 2 This is a schematic diagram of a kidney transplant rejection risk monitoring and early warning system according to one embodiment of this application, referring to... Figure 2 A kidney transplant rejection risk monitoring and early warning system, comprising:

[0170] Processor 21 and memory 22;

[0171] Processor 21 and memory 22 are connected via a communication bus:

[0172] The processor 21 is used to call and execute the program stored in the memory 22;

[0173] The memory 22 is used to store a program, which is used to execute at least one of the kidney transplant rejection risk monitoring and early warning methods in any of the above embodiments.

[0174] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0175] It should be noted that in the description of this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means at least two.

[0176] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.

[0177] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0178] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0179] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0180] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0181] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0182] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for monitoring and early warning of rejection risk after kidney transplantation, characterized in that, include: During the current follow-up period after kidney transplantation, nursing monitoring data, laboratory indicator data, and ultrasound blood flow parameters of the transplanted kidney are collected from the recipient. The collected data is subjected to timestamp alignment, missing data handling, and standardization. Feature extraction is then performed to construct a temporal feature sequence. The time-series feature sequence is input into a pre-trained rejection reaction risk prediction model, which outputs the probability of rejection reaction in the next period and combines it with the probability of rejection reaction in historical periods to form a risk trend. Nursing monitoring indicators are determined. Based on the probability of rejection, a nursing sensitivity indicator weight analysis algorithm is executed to calculate the contribution weight of each nursing monitoring indicator to the probability of rejection. Nursing monitoring indicators with contribution weights higher than preset weight values ​​are used as target nursing monitoring indicators. Based on the nursing monitoring data corresponding to the target nursing monitoring indicators, an individualized baseline model of the recipient is constructed to obtain the baseline value and allowable fluctuation range of the target nursing monitoring indicators, and the deviation characteristics of the nursing monitoring data corresponding to the target nursing monitoring indicators relative to the baseline value and allowable fluctuation range are extracted. The probability of the rejection reaction is individually calibrated based on the deviation characteristics to obtain a calibration risk score; When the calibration risk score fails to reach the preset risk warning score threshold and the risk trend does not meet the preset trend warning conditions, nursing monitoring recommendations for the next cycle are generated based on the target nursing monitoring indicators. When the calibrated risk score reaches the preset risk warning score threshold, or when the risk trend meets the preset trend warning conditions, a warning message is pushed to the nursing terminal, and a graded intervention recommendation for the next cycle is generated based on the risk level corresponding to the calibrated risk score and the target nursing monitoring indicators.

2. The method according to claim 1, characterized in that, Nursing monitoring indicators are determined. Based on the probability of rejection, a nursing sensitivity indicator weighting analysis algorithm is executed to calculate the contribution weight of each nursing monitoring indicator to the probability of rejection, including: Obtain the contribution information of each time-series feature corresponding to the rejection reaction risk prediction model when outputting the probability of rejection reaction occurrence; The contribution degree corresponding to each nursing monitoring indicator is extracted from the contribution information of time-series features, and the contribution degree is aggregated in the time dimension to obtain the initial contribution weight of each nursing monitoring indicator. The initial contribution weights were subjected to stability testing and normalization to obtain the contribution weights of each nursing monitoring indicator to the probability of the rejection reaction. The stability test includes at least: determining whether the initial contribution weight meets the continuity threshold and / or volatility threshold within a preset sliding time window, so as to suppress weight distortion caused by occasional anomalies.

3. The method according to claim 2, characterized in that, Obtain the contribution information of each time-series feature corresponding to the rejection reaction risk prediction model when outputting the probability of rejection reaction occurrence, including: The gradient information of the probability of the rejection reaction occurring relative to each feature value at each time step in the temporal feature sequence is calculated based on backpropagation. The contribution of each temporal feature is determined based on the gradient information; the contribution includes: the product of the value of the temporal feature and the gradient information, and the integrated gradient value obtained by integrating the gradient information.

4. The method according to claim 1, characterized in that, The allowable fluctuation range is adaptively determined based on the historical distribution of the target nursing monitoring indicators, including: Within the pre-defined baseline sample segment, the standard deviation is calculated based on the historical distribution of the target nursing monitoring indicators; Centered on the baseline value, an allowable fluctuation range is formed by setting the baseline value ± k times the standard deviation; Where k is a configurable parameter, and can be set in segments or dynamically adjusted according to the postoperative time stage.

5. The method according to claim 1, characterized in that, The deviation characteristics include: the magnitude of deviation from the baseline, the direction of deviation, the rate of deviation, the degree of fluctuation, and the duration of continuous deviation; Extracting deviation features includes: Within a preset time window, calculate the difference or ratio between the current value and the baseline value as the deviation from the baseline; Use the sign of the deviation magnitude as the deviation direction; Calculate the slope of the deviation magnitude over the time dimension as the deviation rate; Calculate the variance, standard deviation, or coefficient of variation of the deviation as the degree of volatility; The cumulative duration or the cumulative number of sampling points during which the deviation exceeds the allowable fluctuation range is calculated as the continuous deviation duration.

6. The method according to claim 1, characterized in that, Based on the target nursing monitoring indicators, nursing monitoring recommendations for the next cycle are generated, including: Based on the baseline values ​​and allowable fluctuation ranges of the target nursing monitoring indicators, the recommended collection frequency, collection time points, and collection methods for each target nursing monitoring indicator in the next cycle are determined. Based on the deviation characteristics of the target nursing monitoring indicator in the current period, key monitoring prompts are generated for the deviation direction, deviation magnitude, and continuous deviation duration, and the retest trigger conditions for the target nursing monitoring indicator are determined; the retest trigger conditions include at least one or more of the following: deviation magnitude exceeds a preset magnitude threshold, deviation rate exceeds a preset rate threshold, and continuous deviation duration reaches a preset duration threshold.

7. The method according to claim 6, characterized in that, Based on the risk level corresponding to the calibrated risk score and the target care monitoring indicators, a tiered intervention recommendation for the next cycle is generated, including: The calibration risk score is divided into corresponding risk levels according to a preset mapping rule; An intervention template is obtained by matching the risk level from a pre-configured tiered intervention strategy library. The intervention template includes at least a set of intervention actions, a set of execution roles, and an execution time limit. Based on the abnormality types and deviation characteristics of the target nursing monitoring indicators, intervention actions corresponding to the abnormality types are selected from the set of intervention actions to form an intervention task list for the next cycle. The intervention task list includes at least one or more of the following: communication and follow-up tasks, compliance verification tasks, follow-up assessment tasks, and diagnosis and treatment collaboration tasks. Configure task priority, execution role and closed-loop verification conditions for each intervention task in the intervention task list, and configure task time limits for each intervention task in the intervention task list according to the execution time limit.

8. The method according to claim 1, characterized in that, The probability of rejection is individually calibrated based on deviation characteristics to obtain a calibration risk score, including: The deviation features are quantified according to the deviation direction, deviation magnitude, deviation rate, fluctuation degree, and continuous deviation duration to generate a deviation feature vector; A personalized calibration factor is constructed based on the deviation feature vector; the personalized calibration factor is positively correlated with the deviation magnitude, deviation rate, and continuous deviation duration, and is also associated with the deviation direction; The probability of the rejection reaction is corrected based on the individualized calibration factor, and a calibration risk score is output. The calibration risk score is subject to boundary constraints and smoothing to suppress score jumps caused by occasional anomalies.

9. The method according to claim 1, characterized in that, The nursing monitoring data shall include at least one or more of the following: daily urine output, weight change, body temperature, and blood pressure; The laboratory indicator data shall include at least one or more of the following: serum creatinine, urine protein, and blood concentration of immunosuppressant drugs. The ultrasound blood flow parameters of the transplanted kidney include at least: blood flow velocity parameters and / or resistance-related parameters; The preset trend warning conditions include at least one or more of the following: the risk trend rises continuously for a preset number of cycles, the rate of increase of the risk trend exceeds a preset rate threshold, and the cumulative increase of the risk trend within a preset time window exceeds a preset magnitude threshold.

10. A system for monitoring and early warning of rejection risk after kidney transplantation, characterized in that, include: Processor and memory; The processor and memory are connected via a communication bus: The processor is used to call and execute the program stored in the memory; The memory is used to store a program, which is at least used to execute the method for monitoring and early warning of rejection risk after kidney transplantation as described in any one of claims 1-9.

Images