A perioperative cardiovascular and cerebrovascular adverse event prediction method and prediction system
By extracting an initial parameter set from the actual physiological parameter set, performing data cleaning and confidence calculation, and combining a risk relationship comparison table and a basic pathological feature set to calculate correction coefficients, an early warning model is generated and a collaborative bias analysis is performed. This solves the problem of low prediction accuracy in perioperative cardiovascular and cerebrovascular adverse event prediction, achieving higher prediction accuracy and reliability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WEST CHINA HOSPITAL SICHUAN UNIV
- Filing Date
- 2026-05-25
- Publication Date
- 2026-06-19
AI Technical Summary
Existing technologies have low accuracy in predicting adverse cardiovascular and cerebrovascular events during the perioperative period, lack personalized correction mechanisms, and cannot accurately reflect the physiological changes of patients, resulting in large prediction errors and failing to support the accuracy of risk warnings.
By extracting an initial parameter set from the actual physiological parameter set, performing data cleaning and confidence calculation, and combining a risk relationship comparison table and a basic pathological feature set to calculate correction coefficients, an early warning model is generated, and collaborative bias analysis is performed to generate early warning results.
It improves the accuracy of predictions, eliminates errors caused by individual differences, ensures that the prediction model is deeply linked to the actual situation of patients, and enhances the precision and reliability of predictions, which is in line with clinical physiological laws.
Smart Images
Figure CN122245793A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical data processing technology, specifically to a method and system for predicting perioperative cardiovascular and cerebrovascular adverse events. Background Technology
[0002] Patients' physiological states fluctuate dramatically during the perioperative period (before, during, and after surgery). Abnormal changes in multiple physiological indicators such as heart rate, blood pressure, and blood oxygen saturation are key warning signals for adverse events such as postoperative complications, organ dysfunction, and even death. Therefore, real-time monitoring and accurate prediction of the physiological indicators of perioperative patients, as well as timely risk warnings, are crucial for improving the quality of clinical monitoring and reducing the incidence of adverse events. This is also one of the core needs in the current field of clinical monitoring.
[0003] Currently, perioperative physiological indicator monitoring and risk warning mainly rely on traditional monitoring equipment and manual judgment. Some solutions have introduced technologies such as virtual patient models, knowledge graphs, and data fitting to try to improve the accuracy and real-time performance of warnings. However, in practical applications, these methods often use general templates, and the parameters are fixed, lacking personalized correction mechanisms. At the same time, they do not dynamically adjust based on the patient's real-time physiological state and clinical risk factors, resulting in a low degree of fit between the predicted curve and the patient's actual physiological data. The predicted values of physiological indicators obtained based on this curve have large errors and cannot accurately reflect the individual physiological change trend of the patient, making it difficult to support the accuracy of subsequent risk warnings. Summary of the Invention
[0004] The main purpose of this application is to provide a method and system for predicting adverse cardiovascular and cerebrovascular events during the perioperative period, aiming to solve the problem of low prediction accuracy in the existing technology.
[0005] This application achieves the above objectives through the following technical solutions: A method for predicting perioperative adverse cardiovascular and cerebrovascular events includes the following steps: The initial parameter set is extracted from the actual physiological parameter set based on the movement time window; The initial parameter set is cleaned, and the confidence parameter set and target parameter set are calculated. Obtain the risk relationship comparison table and the basic pathological feature set, and calculate the first set of correction coefficients in combination with the confidence parameter set; A warning model is generated based on the first set of correction coefficients and the target parameter set; A collaborative deviation analysis is performed on the early warning model and the target parameter set to calculate the second correction coefficient; Early warning results are generated based on the second correction coefficient and the early warning model.
[0006] Optionally, an initial parameter set can be extracted from the actual physiological parameter set based on the moving time window, including the following steps: Obtain the actual set of physiological parameters; Set the move time window length and move step size, and generate the move time window; Based on the moving time window, the parameter closest to the current time is selected from the actual physiological parameter set as the initial parameter set.
[0007] Optionally, the initial parameter set is cleaned and the confidence parameter set and target parameter set are calculated, including the following steps: The initial parameter set is divided into several sub-parameter sets according to the types of physiological indicators; Obtain any sub-parameter set and calculate its mean and standard deviation; The sub-parameter set is cleaned based on the mean and the standard deviation. Interpolation is used to complete the abnormal parameters that have been cleaned and removed, generating a target parameter set; The confidence parameter is calculated based on the target parameter set, wherein the expression for the confidence parameter is as follows: Where i represents the data number, j represents the physiological indicator number, and x represents the data index number. i,j Let i represent the target parameter with index i for the j-th physiological indicator; n represents the number of data points. This represents the mean of the target parameter set. This represents the standard deviation of the target parameter set; Repeat the steps of obtaining any sub-parameter set and calculating its mean and standard deviation to obtain the confidence parameter set and the target parameter set.
[0008] Optionally, obtain the risk relationship comparison table and the basic pathological feature set, and calculate the first correction coefficient set in combination with the confidence parameter set, including the following steps: Obtain a risk relationship comparison table; Generate corresponding basic pathological feature sets for each physiological indicator; Based on the basic pathological feature set, basic coefficients are extracted from the risk relationship comparison table, and corresponding basic coefficient sets are generated for each physiological indicator. Based on the confidence parameter set, each of the aforementioned basic coefficient sets is modified to generate the first modified coefficient set for each physiological indicator.
[0009] Optionally, the first modified coefficient set for each physiological indicator is generated by modifying each of the baseline coefficient sets according to the confidence parameter set, including the following steps: Retrieve the confidence parameters and baseline coefficient set of the same physiological indicator; A dynamic correction model is constructed based on the confidence parameters; wherein the calculation expression of the dynamic correction model is: : k represents the basic coefficient number, Indicates the base coefficient. This represents the adjustment factor, which is a constant ranging from 0.2 to 0.4; The basic coefficient set is modified according to the dynamic correction model to generate a dynamic basic coefficient set; The first correction coefficient is calculated based on the dynamic basic coefficient set, wherein the calculation expression for the first correction coefficient is: Q j Let represent the set of dynamic baseline coefficients for the j-th physiological indicator; Repeat the steps of retrieving the confidence parameters and baseline coefficient set of the same physiological indicator to obtain the first set of corrected coefficients for all physiological indicators.
[0010] Optionally, generating an early warning model based on the first set of correction coefficients and the target parameter set includes the following steps: Obtain the first correction coefficient and target parameter set for the same physiological indicator; The target parameter set is fitted and calculated to generate the corresponding fitting curve; The fitted curve is corrected according to the first correction coefficient to obtain a corrected fitted curve; wherein the calculation expression of the corrected fitted curve is as follows: , where X j (t) represents the fitted curve of the j-th physiological index; Repeat the steps of obtaining the first correction coefficient and target parameter set for the same physiological indicator, summarize the corrected fitting curves of each physiological indicator, and generate an early warning model.
[0011] Optionally, a collaborative deviation analysis is performed on the early warning model and the target parameter set to calculate the second correction coefficient, including the following steps: Based on the aforementioned early warning model, historical prediction parameter sets for each physiological indicator are generated; Based on physiological indicators, the target parameter sets and the historical prediction parameter sets are combined into several calculation groups; The average relative deviation rate of a single indicator is calculated based on each of the calculation groups, and the average relative deviation set of a single indicator is generated. Generate a collaborative deviation matrix based on the average relative deviation set of the single indicators; The second correction coefficient is calculated based on the cooperative deviation matrix.
[0012] Optionally, the formula for calculating the average relative deviation rate of a single indicator is as follows: ;where P i,j Represents historical prediction parameters; the calculation expression for the collaborative deviation matrix is as follows: Where m represents the number of physiological indicators, This represents the average relative deviation rate of the m-th physiological indicator. Let represent the correlation coefficient between the m-th physiological indicator and the first physiological indicator, which is a constant; the expression for calculating the second correction coefficient is: : Optionally, generating an early warning result based on the second correction coefficient and the early warning model includes the following steps: Obtain the predicted time and generate an initial set of prediction parameters for the time to be predicted based on the early warning model; The initial prediction parameter set is corrected according to the second correction coefficient to generate a corrected prediction parameter set; wherein the calculation expression for the corrected prediction parameter is as follows: ,in Indicates the predicted time; Obtain the single-index comparison model and the alarm determination model; wherein the expression of the single-index comparison model is: S j，min and S j,max Let these represent the lower and upper thresholds of the j-th physiological indicator, respectively. The calculation expression for the alarm judgment model is as follows: , where B represents the number of abnormal indicators; The number of abnormal indicators B is calculated based on the modified prediction parameter set and the single indicator comparison model. Early warning results are generated based on the number of abnormal indicators and the alarm determination model.
[0013] Accordingly, this application also discloses a prediction system based on the above prediction method, including: The parameter acquisition module is used to extract an initial parameter set from the actual physiological parameter set based on the movement time window; The data cleaning and calculation module is used to clean the initial parameter set and calculate the confidence parameter set and the target parameter set. The first calculation module is used to obtain the risk relationship comparison table and the basic pathological feature set, and calculate the first correction coefficient set in combination with the confidence parameter set. The model generation module is used to generate an early warning model based on the first set of correction coefficients and the target parameter set; The second calculation module is used to perform a collaborative deviation analysis on the early warning model and the target parameter set, and to calculate the second correction coefficient. The early warning module is used to generate early warning results based on the second correction coefficient and the early warning model.
[0014] Compared with the prior art, this application has the following beneficial effects: This application first extracts an initial parameter set from the actual physiological parameter set based on the moving time window, then cleans the initial parameter set and calculates the confidence parameter set and the target parameter set; subsequently, it obtains a risk relationship comparison table and a basic pathological feature set, calculates a first correction coefficient set in combination with the confidence parameter set, generates an early warning model based on the first correction coefficient set and the target parameter set, performs a co-variance analysis on the early warning model and the target parameter set, calculates a second correction coefficient, and finally generates an early warning result based on the second correction coefficient and the early warning model. Compared with the prior art, this application uses the measured parameters of the subject as the basic parameters for calculation, thereby deeply binding the entire prediction result and calculation process with the actual situation of the subject, thereby eliminating prediction errors caused by individual differences as much as possible, making the prediction model reflect the actual situation of the patient as realistically as possible, and effectively improving the accuracy of prediction. Secondly, the risk relationship comparison table comprehensively lists various risks and their influencing factors, and also details the influence coefficients of different influencing factors on each risk. The above parameters are generated based on medical knowledge, and the basic pathological feature set is derived from the test subject, reflecting the influencing factors they possess. That is, for a certain risk, the test subject may have one or more corresponding influencing factors. Through the risk relationship comparison table and the basic pathological feature set, the influencing factors of a certain risk on the test subject's own factors can be determined from a medical perspective, thereby constraining the parameter prediction. Meanwhile, by cleaning the data, we can ensure the accuracy of the data and correct each early warning model by using the confidence parameter set. The confidence level reflects the overall credibility of the parameters after cleaning. Based on the first correction coefficient obtained by the influence factor and the confidence level, the early warning model can be constrained by both medical professional knowledge and the authenticity of measured data, which can effectively improve the model's prediction accuracy. Finally, no matter how the early warning model is improved, its calculation bias will always exist. At the same time, physiological indicators are not isolated, but are interconnected and mutually influential (such as the linkage between heart rate and blood pressure, and blood oxygen saturation and respiratory rate). The bias of a single indicator is easily affected by data noise. It is necessary to combine the correlation between indicators to conduct a comprehensive bias assessment in order to better reflect the actual physiological laws in clinical practice and ensure the comprehensiveness and objectivity of the bias assessment. In the technical solution described in this application, the above purpose is achieved by conducting a collaborative bias analysis between the calculated value of the early warning model and the target parameter set, thereby minimizing the impact of the above bias on the final result and improving the accuracy of the prediction. Attached Figure Description
[0015] Figure 1 A flowchart of a method for predicting perioperative cardiovascular and cerebrovascular adverse events provided in Embodiment 1 of this application; Figure 2This is a schematic diagram of the structure of a prediction system provided in Embodiment 2 of this application; The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0017] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0018] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.
[0019] Implementation Method 1 Reference Figure 1 This embodiment, as an optional implementation of this application, discloses a method for predicting perioperative adverse cardiovascular and cerebrovascular events, including the following steps: S1. Extract the initial parameter set from the actual physiological parameter set according to the moving time window; S11. Obtain the actual physiological parameter set; First, set the target physiological indicator, then retrieve all parameters of the corresponding physiological indicator from the database, and generate a unique timestamp number for each parameter. At the same time, generate a corresponding number according to the different physiological indicators, thereby generating the actual set of physiological parameters. It should be noted that the target physiological indicators refer to vital signs that are suitable for perioperative clinical monitoring scenarios, can be collected continuously in real time, have a clear clinical correlation with the occurrence, development and risk evolution of perioperative adverse cardiovascular and cerebrovascular events, and can be used for cardiovascular and cerebrovascular risk assessment and trend prediction. The above indicators are selected by medical staff in this field from the preset standardized candidate physiological indicators based on clinical experience, perioperative monitoring standards and evidence-based diagnosis and treatment guidelines for cardiovascular and cerebrovascular diseases. That is, the target physiological indicators are composed of several or all of the indicators selected by medical staff from a number of alternative physiological indicators based on the actual situation; Furthermore, a duration can be set to filter parameters, removing outdated historical parameters and reducing computational load.
[0020] S12. Set the length of the moving time window and the moving step size, and generate the moving time window; Based on the perioperative monitoring sampling frequency and clinical monitoring needs, the length of the moving time window and the moving step size are determined, and the corresponding moving time window is generated; S13. Based on the moving time window, extract the parameter closest to the current time from the actual physiological parameter set as the initial parameter set.
[0021] Starting with the real-time physiological parameters closest to the current moment, the system works backwards according to the length of the moving time window to extract all real-time physiological parameters closest to the current moment, thereby generating an initial parameter set. By extracting the parameters mentioned above, we can reduce the number of parameters to be calculated, which is beneficial to improving computational efficiency. Secondly, in the case of a short time span, the changes in physiological state are relatively stable, and the data can truly reflect the current state. However, in the case of a long time span, the changes in physiological state are more obvious. By extracting the parameters that are closest to the current moment as the initial parameter set, we can effectively eliminate the influence of historical parameters on the current state and improve the accuracy of prediction.
[0022] S2. Perform data cleaning on the initial parameter set and calculate the confidence parameter set and the target parameter set; S21. Divide the initial parameter set into several sub-parameter sets according to the types of physiological indicators; First, the initial parameter set is retrieved, and then all parameters with the same physiological index number are aggregated into the same set and sorted according to the number generated by the timestamp. That is, its expression is {x 1,1 x 2,1 x 3,1 , ..., x i,1 , ..., x n,j}, {x 1,2 x 2,2 x3,2 , ..., x i,2 , ..., x n,2}, {x 1,3 x 2,3 x 3,3 , ..., x i,3 , ..., x n,3},...,{x 1,j x 2,j x 3,j , ..., x i,j , ..., x n,j},...,{x 1,m x 2,m x 3,m , ..., x i,m , ..., x n,m}; where n represents the number of data points and m represents the number of physiological indicators; S22. Obtain any sub-parameter set and calculate its mean and standard deviation; Retrieve any sub-parameter set obtained in step S21, and calculate its mean and standard deviation respectively; S23. Perform data cleaning on the sub-parameter set based on the mean and the standard deviation; Based on the 3σ principle, the mean and standard deviation calculated in step S22 are retrieved to clean the sub-parameter set, so as to eliminate parameters with abnormal deviations and ensure the accuracy of the data. S24. Use interpolation to complete the abnormal parameters that have been cleaned and removed, and generate the target parameter set; For the parameters that have been cleaned up, retrieve the two adjacent parameters and calculate their theoretical values using interpolation. Replace the cleaned-up parameters with the theoretical values to generate the target parameter set. The above cleaning and interpolation completion calculations can effectively ensure the accuracy of the data while maintaining the number of parameters, thereby improving the accuracy of the final prediction results. S25. Calculate the confidence parameter based on the target parameter set; Retrieve the target parameter set obtained in step S24, and then first calculate its mean and standard deviation; The confidence level parameter is calculated by combining the calculated mean and standard deviation, wherein the expression for the confidence level parameter is as follows: Where i represents the data number, j represents the physiological indicator number, and x represents the data index number. i,j Let i represent the target parameter with index i for the j-th physiological indicator; n represents the number of data points. This represents the mean of the target parameter set. This represents the standard deviation of the target parameter set; The calculation of confidence parameters enables quantitative assessment and comprehensive evaluation of the quality of parameters within the sub-parameter set, thus providing basic parameters for subsequent corrections. S26. Repeat the steps of obtaining any sub-parameter set and calculating its mean and standard deviation to obtain the confidence parameter set and the target parameter set; Repeating steps S22-S25 will generate a confidence parameter for each sub-parameter set. Since each sub-parameter set corresponds to a physiological indicator, this step will generate a unique confidence parameter for each physiological indicator. Combined with step S21, the target parameter set corresponding to each physiological indicator can be obtained.
[0023] S3. Obtain the risk relationship comparison table and the basic pathological feature set, and calculate the first correction coefficient set in combination with the confidence parameter set; S31. Obtain a risk relationship comparison table; This application pre-constructs and locally stores a perioperative cardiovascular and cerebrovascular risk relationship comparison table, which is uniformly calibrated and generated based on clinical evidence-based medicine data and the association patterns of perioperative complications. In the table: physiological indicators are the core vital signs monitored in real time during surgery; risk factors are independent influencing factors that induce adverse cardiovascular and cerebrovascular events; and the baseline coefficient is the quantitative correlation value of risk factors on the fluctuation of corresponding physiological indicators. Among them, a positive coefficient indicates that the risk factor aggravates the abnormal fluctuation of physiological indicators, and a negative coefficient indicates that the risk factor inhibits the abnormal fluctuation of physiological indicators. The risk relationship comparison is shown below: physiological indicators Risk factors base coefficient MAP (Mean Arterial Pressure) hypertension +0.22 MAP Cerebral aneurysm -0.25 MAP Anticoagulants -0.2 MAP Intraoperative hypotension -0.18 SBP (systolic blood pressure) hypertension +0.24 SBP Cerebral aneurysm -0.26 SBP Anticoagulants -0.21 HR (Heart Rate) advanced age -0.12 HR Atrial fibrillation +0.18 HR Heart failure +0.15 It should be noted that the table above is just an example; in the actual compilation process, all physiological indicators and risk factors need to be listed. The risk relationship lookup table is a pre-configured structured data table, which is stored in the local terminal of the prediction system in the form of a database form. It supports operations such as adding indicators, iterating risk factors, batch updating basic coefficients, and offline calling. S32, generate the corresponding basic pathological feature sets for each physiological indicator; Obtain the basic pathological feature set of the subject to be tested, wherein the basic pathological feature set is all the risk factors that the subject to be tested possesses; S33. Extract basic coefficients from the risk relationship comparison table based on the basic pathological feature set, and generate corresponding basic coefficient sets for each physiological indicator. Based on the basic pathological feature set obtained in step S32, basic coefficients are extracted from the risk relationship comparison table, and all basic coefficients are summarized to generate a basic coefficient set. Specifically, if the set physiological indicators are MAP, SBP, and HR, and the basic pathological feature set is {hypertension, cerebral aneurysm, and heart failure}, then the basic coefficient set extracted from the example table above is {+0.22, -0.25, +0.24, -0.26, +0.15}.
[0024] S34. Based on the confidence parameter set, each of the aforementioned basic coefficient sets is modified to generate the first modified coefficient set for each physiological indicator; S341. Retrieve the confidence parameters and baseline coefficient set of the same physiological indicator; S342. Construct a dynamic correction model based on the confidence parameters; wherein the calculation expression of the dynamic correction model is: : k represents the basic coefficient number, Indicates the base coefficient. This represents the adjustment factor, which is a constant ranging from 0.2 to 0.4; First, obtain the general expression of the dynamic correction model. Then, call the confidence parameters calculated in step S25 and import them into the general expression to generate the corresponding correction model for each physiological indicator. By summing up all the correction models, the dynamic correction model can be obtained. S343. Based on the dynamic correction model, the basic coefficient set is corrected to generate a dynamic basic coefficient set; Obtain a dynamic correction model, and then dynamically correct each basic coefficient using the dynamic correction model to obtain a dynamic basic coefficient set; The essence of the baseline coefficient is the quantification of the real clinical correlation pattern. It needs to be set based on a large amount of real clinical data. Therefore, the authenticity and accuracy of clinical parameters are highly related to the setting of the baseline coefficient. That is, the confidence parameter corresponding to the default setting of the baseline coefficient is 1. When the confidence level is closer to 1, it means that the target parameter set can more accurately reflect the current physiological state, and the corresponding dynamic correction coefficient is closer to the corresponding baseline coefficient, meaning that the baseline coefficient does not need to be adjusted significantly. Conversely, it means that the measured parameters cannot reflect the true physiological state and need to be adjusted accordingly to reduce the impact of data unreliability and avoid prediction bias. This application calibrates the reliability of the correlation between risk factors and measured physiological parameters by confidence level, so that each basic coefficient is more in line with the real-time data quality, thereby ensuring the individualized adaptation accuracy of the early warning model and avoiding misjudgment of early warning due to data quality issues. Compared with traditional fixed values, this application offers greater flexibility and effectively guarantees data accuracy, thereby improving the final prediction accuracy. S344. Calculate the first correction coefficient based on the dynamic basic coefficient set; The first correction coefficient can be calculated using the dynamic coefficient set, wherein the expression for the first correction coefficient is as follows: Q j Let represent the set of dynamic baseline coefficients for the j-th physiological indicator; The first correction coefficient can comprehensively weight the dynamic base coefficients corresponding to multiple risk factors, that is, measure the combined impact of multiple factors. Compared with the calculation of the impact of a single indicator, its calculation is more comprehensive and effectively ensures the accuracy and reliability of the early warning results. S345. Repeat the steps of retrieving the confidence parameters and baseline coefficient set of the same physiological indicator to obtain the first modified coefficient set of all physiological indicators.
[0025] S4. Generate an early warning model based on the first set of correction coefficients and the target parameter set; S41. Obtain the first correction coefficient and target parameter set for the same physiological indicator; S42. Perform fitting calculations on the target parameter set to generate the corresponding fitting curve; First, a two-dimensional standard coordinate system is generated. The horizontal axis of the standard two-dimensional coordinate system represents the time parameter, and the vertical axis represents the target parameter. Then, according to the target parameter set, several calibration points are calibrated in the standard coordinate system. Finally, the corresponding fitted curve is obtained through computer fitting. S43. Correct the fitted curve according to the first correction coefficient to obtain a corrected fitted curve; Then, the corresponding first correction coefficient obtained in step S3 is used to correct the fitted curve in step S42, resulting in a corrected fitted curve. The calculation expression for the corrected fitted curve is as follows: , where X j (t) represents the fitted curve of the j-th physiological index; The first correction coefficient can be used to correct the fitted curve based on clinical medical principles, so that the fitted curve can more accurately fit the patient's own physiological state, while constraining its prediction, thereby making parameter predictions as based as possible on the patient's real physiological state and improving the accuracy of prediction. S44. Repeat the steps of obtaining the first correction coefficient and target parameter set of the same physiological index, summarize the corrected fitting curves of each physiological index, and generate an early warning model. Repeating steps S41 to S43 will yield fitting models for all physiological indicators, which will then generate an early warning model.
[0026] S5. Perform a collaborative deviation analysis on the early warning model and the target parameter set, and calculate the second correction coefficient; S51. Generate historical prediction parameter sets for each physiological indicator based on the aforementioned early warning model; By retrieving the early warning model and target parameter set, retrieving the time parameters corresponding to each parameter in the target parameter set, and importing the time parameters into the early warning model, corresponding historical prediction parameter sets can be generated for each physiological indicator. S52. Based on physiological indicators, combine each set of target parameters and each set of historical prediction parameters into several calculation groups; First, the parameters are classified according to physiological indicators, that is, the historical prediction parameter set and the target parameter set under the same physiological indicator are summarized together. Then, the historical prediction parameters and target parameters at the same time are paired one by one to generate several calculation groups. In the same calculation group, there are historical prediction parameters and target parameters at the same time. The historical prediction parameters are the predicted values of the prediction model, while the target parameters are the actual measured values at the same time. S53. Calculate the average relative deviation rate of a single indicator according to each of the calculation groups, and collect them to generate a set of average relative deviations of a single indicator. Retrieve each calculation group in step S52 and calculate the difference between each calculation group, that is, the difference between the actual measured value and the predicted parameter, which accurately measures the calculation deviation of the early warning model. The average relative deviation rate of a single physiological indicator can be calculated by summing up the differences of all the same physiological indicator. The formula for calculating the average relative deviation rate of a single indicator is as follows: ;x i,j P represents the target parameter with index i for the j-th physiological indicator. i,j This represents the historical prediction parameters, and n represents the number of parameters. Repeating the above steps will generate the corresponding single-indicator average relative deviation rate for each physiological indicator. Collecting all the single-indicator average deviation rates into the same set will generate the single-indicator average relative deviation set. The average relative deviation rate of a single indicator can accurately quantify the calculation deviation of a certain physiological indicator, providing a basic parameter for further adjustment. S54. Generate a collaborative deviation matrix based on the single-index average relative deviation set; Retrieve the average relative deviation rate of each individual indicator calculated in step S53, and then construct a coordinated deviation matrix covering all physiological indicators based on the above average relative deviation rates of individual indicators. The calculation expression of the coordinated deviation matrix is as follows: Where m represents the number of physiological indicators, This represents the average relative deviation of the m-th physiological indicator. Let represent the correlation coefficient between the m-th physiological indicator and the first physiological indicator, which is a constant; When the average relative deviation of two physiological indicators deviates from the true value, the aforementioned collaborative deviation matrix can amplify the deviation. When one is larger than the other, the geometric mean can be used to take the middle value, avoiding the dominance of a single indicator deviation in the overall collaborative judgment. That is, the geometric mean of the average relative deviation rates of two single indicators is used to characterize the degree of collaborative deviation among multiple indicators, highlighting the common trend of the multi-indicator synchronous deviation early warning model, avoiding excessive influence of abnormal deviation of a single indicator on the overall collaborative judgment, and making the collaborative deviation assessment more stable. Based on this, a collaborative deviation matrix is constructed. The diagonal elements of the matrix are the average relative deviation rates of each individual indicator, while the off-diagonal elements are the collaborative deviation strengths of the average relative deviation rates of the individual indicators of the two corresponding physiological indicators. This enables a comprehensive quantitative assessment of the overall fitting consistency and linkage deviation characteristics of multiple indicators. Compared with existing technologies, by associating all physiological indicators through the above-mentioned synergistic deviation matrix and using it for parameter deviation assessment, the synergistic effect between different physiological indicators is strengthened, enabling the prediction process to more objectively reflect the true state and thus improve the accuracy of prediction. S55. Calculate the second correction coefficient based on the cooperative deviation matrix; The second correction factor is then calculated based on the cooperative deviation matrix, wherein the expression for the calculation of the second correction factor is as follows: ; The above calculations ensure that the second correction coefficient is obtained by averaging all elements of the multi-indicator co-variance deviation matrix. This includes both the average relative deviation rate of each physiological indicator and the degree of co-variance deviation of the average relative deviation rate of each indicator, thus achieving a comprehensive evaluation of the overall model fitting effect.
[0027] S6. Generate an early warning result based on the second correction coefficient and the early warning model; S61. Obtain the predicted time and generate an initial prediction parameter set for the time to be predicted based on the early warning model; Set the prediction time according to the actual situation. For example, if the prediction time is set to 15 minutes, it means that the current time will be the starting point and the next 15 minutes will be the prediction time node. Once the prediction time is set, the time parameters can be substituted into the early warning model to generate the corresponding parameters. The initial prediction parameter set can be obtained by summarizing the predicted values of all physiological indicators. S62. Correct the initial prediction parameter set according to the second correction coefficient to generate a corrected prediction parameter set; Then, the second correction parameter calculated in step S5 is retrieved, and each element in the initial prediction parameter set is corrected using the second correction parameter to obtain the corrected prediction parameter set. The calculation expression for the corrected prediction parameter is as follows: ,in Indicates the predicted time; The second correction coefficient is based on the difference between the predicted parameters and the measured parameters. It reflects the inherent deviation that the early warning model cannot overcome. The above correction can further eliminate the inherent deviation, thereby further improving the accuracy of the calculation.
[0028] S63. Obtain the single-index comparison model and alarm judgment model; wherein the expression of the single-index comparison model is: S j，min and S j,max Let these represent the lower and upper thresholds of the j-th physiological indicator, respectively. The calculation expression for the alarm judgment model is as follows: , where B represents the number of abnormal indicators; A lower threshold and an upper threshold are set for each physiological indicator, and then a single-indicator comparison model is generated based on the upper threshold and the lower threshold. S64. Calculate the number of abnormal indicators B based on the modified prediction parameter set and the single indicator comparison model; Substitute the predicted values of each physiological indicator calculated in step S62 into the corresponding single indicator comparison model. If the output result is 1, it indicates that the physiological indicator is unqualified at the predicted time point; otherwise, it is qualified. At the same time, the number of unqualified physiological indicators is counted, and then the number of abnormal indicators B is generated; S65. Generate an early warning result based on the number of abnormal indicators and the alarm judgment model; The calculated abnormal index numbers are substituted into the alarm judgment model, and the corresponding early warning results are output. In the above calculation, an alarm will be triggered as soon as any physiological indicator fails to meet the standard, which can effectively lower the alarm threshold and issue an early warning as soon as an abnormality is detected.
[0029] Compared with the prior art, this application uses the measured parameters of the subject as the basic parameters for calculation, thereby deeply binding the entire prediction result and calculation process with the actual situation of the subject, thereby eliminating prediction errors caused by individual differences as much as possible, making the prediction model reflect the actual situation of the patient as realistically as possible, and effectively improving the accuracy of prediction. Secondly, the risk relationship comparison table comprehensively lists various risks and their influencing factors, and also details the influence coefficients of different influencing factors on each risk. The above parameters are generated based on medical knowledge, and the basic pathological feature set comes from the test subject, reflecting the influencing factors they have. That is, for a certain risk, the test subject may have one or more corresponding influencing factors. Through the risk relationship comparison table and the basic pathological feature set, the influencing factors of a certain risk on the test subject's own factors can be determined from a medical perspective, thereby constraining the parameter prediction at the medical level. Meanwhile, by cleaning the data, we can ensure the accuracy of the data and also correct each early warning model through the confidence parameter set. That is, the confidence level reflects the overall credibility of the parameters after cleaning. The first correction coefficient generated by the above-mentioned influencing factors and confidence levels can simultaneously take into account both medical common sense and data authenticity to constrain the early warning model and improve the accuracy of prediction. Finally, no matter how the early warning model is improved, its calculation bias will always exist. At the same time, physiological indicators are not isolated, but are interconnected and mutually influential (such as the linkage between heart rate and blood pressure, and blood oxygen saturation and respiratory rate). The bias of a single indicator is easily affected by data noise. It is necessary to combine the correlation between indicators to conduct a comprehensive bias assessment in order to better reflect the actual physiological laws in clinical practice and ensure the comprehensiveness and objectivity of the bias assessment. In the technical solution described in this application, the above purpose is achieved by conducting a collaborative bias analysis between the calculated value of the early warning model and the target parameter set, thereby minimizing the impact of the above bias on the final result and improving the accuracy of the prediction.
[0030] Furthermore, refer to Figure 2 As another embodiment of this application, a prediction system based on the above prediction method is disclosed, including a parameter acquisition module and a data cleaning and calculation module. The parameter acquisition module extracts an initial parameter set from the actual physiological parameter set according to the moving time window. Its output is communicatively connected to the data cleaning and calculation module. The output of the data cleaning and calculation module is communicatively connected to a first calculation module. The output of the first calculation module is sequentially communicatively connected to a model generation module and a second calculation module. The output of the second calculation module is communicatively connected to an early warning module.
[0031] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.
Claims
1. A perioperative cardiovascular and cerebrovascular adverse event prediction method, characterized by, Includes the following steps: The initial parameter set is extracted from the actual physiological parameter set based on the movement time window; The initial parameter set is cleaned, and the confidence parameter set and target parameter set are calculated. Obtain the risk relationship comparison table and the basic pathological feature set, and calculate the first set of correction coefficients in combination with the confidence parameter set; A warning model is generated based on the first set of correction coefficients and the target parameter set; A collaborative deviation analysis is performed on the early warning model and the target parameter set to calculate the second correction coefficient; Early warning results are generated based on the second correction coefficient and the early warning model.
2. The perioperative cardiovascular and cerebrovascular adverse event prediction method of claim 1, wherein, The step of extracting an initial parameter set from the actual physiological parameter set based on the moving time window includes the following steps: Obtain the actual set of physiological parameters; Set the move time window length and move step size, and generate the move time window; Based on the moving time window, the parameter closest to the current time is selected from the actual physiological parameter set as the initial parameter set.
3. The perioperative cardiovascular and cerebrovascular adverse event prediction method of claim 1, wherein, The process of cleaning the initial parameter set and calculating the confidence parameter set and target parameter set includes the following steps: The initial parameter set is divided into several sub-parameter sets according to the types of physiological indicators; Obtain any sub-parameter set and calculate its mean and standard deviation; The sub-parameter set is cleaned based on the mean and the standard deviation. Interpolation is used to complete the abnormal parameters that have been cleaned and removed, generating a target parameter set; The confidence parameter is calculated based on the target parameter set, wherein the expression for the confidence parameter is as follows: Where i represents the data number, j represents the physiological indicator number, and x represents the data index number. i,j Let i represent the target parameter with index i for the j-th physiological indicator; n represents the number of data points. This represents the mean of the target parameter set. This represents the standard deviation of the target parameter set; Repeat the steps of obtaining any sub-parameter set and calculating its mean and standard deviation to obtain the confidence parameter set and the target parameter set.
4. The method for predicting adverse cardiovascular and cerebrovascular events during the perioperative period according to claim 1, characterized in that, The process of obtaining the risk relationship comparison table and the basic pathological feature set, and calculating the first correction coefficient set in combination with the confidence parameter set, includes the following steps: Obtain a risk relationship comparison table; Generate corresponding basic pathological feature sets for each physiological indicator; Based on the basic pathological feature set, basic coefficients are extracted from the risk relationship comparison table, and corresponding basic coefficient sets are generated for each physiological indicator. Based on the confidence parameter set, each of the aforementioned basic coefficient sets is modified to generate the first modified coefficient set for each physiological indicator.
5. The method for predicting adverse cardiovascular and cerebrovascular events during the perioperative period according to claim 4, characterized in that, The step of generating a first set of corrected coefficients for each physiological indicator by correcting each of the baseline coefficient sets according to the confidence parameter set includes the following steps: Retrieve the confidence parameters and baseline coefficient set of the same physiological indicator; A dynamic correction model is constructed based on the confidence parameters; wherein the calculation expression of the dynamic correction model is: : k represents the basic coefficient number, Indicates the base coefficient. This represents the adjustment factor, which is a constant ranging from 0.2 to 0.4; The basic coefficient set is modified according to the dynamic correction model to generate a dynamic basic coefficient set; The first correction coefficient is calculated based on the dynamic basic coefficient set, wherein the calculation expression for the first correction coefficient is: Q j Let represent the set of dynamic baseline coefficients for the j-th physiological indicator; Repeat the steps of retrieving the confidence parameters and baseline coefficient set of the same physiological indicator to obtain the first set of corrected coefficients for all physiological indicators.
6. The method for predicting adverse cardiovascular and cerebrovascular events in the perioperative period according to claim 1, characterized in that, The step of generating an early warning model based on the first set of correction coefficients and the target parameter set includes the following steps: Obtain the first correction coefficient and target parameter set for the same physiological indicator; The target parameter set is fitted and calculated to generate the corresponding fitting curve; The fitted curve is corrected according to the first correction coefficient to obtain a corrected fitted curve; wherein the calculation expression of the corrected fitted curve is as follows: , where X j (t) represents the fitted curve of the j-th physiological index; Repeat the steps of obtaining the first correction coefficient and target parameter set for the same physiological indicator, summarize the corrected fitting curves of each physiological indicator, and generate an early warning model.
7. The method for predicting adverse cardiovascular and cerebrovascular events during the perioperative period according to claim 1, characterized in that, The step of performing a collaborative deviation analysis on the early warning model and the target parameter set, and calculating the second correction coefficient, includes the following steps: Based on the aforementioned early warning model, historical prediction parameter sets for each physiological indicator are generated; Based on physiological indicators, the target parameter sets and the historical prediction parameter sets are combined into several calculation groups; The average relative deviation rate of a single indicator is calculated based on each of the calculation groups, and the average relative deviation set of a single indicator is generated. Generate a collaborative deviation matrix based on the average relative deviation set of the single indicators; The second correction coefficient is calculated based on the cooperative deviation matrix.
8. The method for predicting adverse cardiovascular and cerebrovascular events during the perioperative period according to claim 7, characterized in that, The formula for calculating the average relative deviation rate of the single index is as follows: P i,j Represents historical prediction parameters; the calculation expression for the collaborative deviation matrix is as follows: Where m represents the number of physiological indicators, This represents the average relative deviation rate of the m-th physiological indicator. Let represent the correlation coefficient between the m-th physiological indicator and the first physiological indicator, which is a constant; the expression for calculating the second correction coefficient is: .
9. The method for predicting adverse cardiovascular and cerebrovascular events in the perioperative period according to claim 1, characterized in that, The step of generating an early warning result based on the second correction coefficient and the early warning model includes the following steps: Obtain the predicted time and generate an initial set of prediction parameters for the time to be predicted based on the early warning model; The initial prediction parameter set is corrected according to the second correction coefficient to generate a corrected prediction parameter set; wherein the calculation expression for the corrected prediction parameter is as follows: ,in Indicates the predicted time; Obtain the single-index comparison model and the alarm determination model; wherein the expression of the single-index comparison model is: S j，min and S j,max Let these represent the lower and upper thresholds of the j-th physiological indicator, respectively. The calculation expression for the alarm judgment model is as follows: , where B represents the number of abnormal indicators; The number of abnormal indicators B is calculated based on the modified prediction parameter set and the single indicator comparison model. Early warning results are generated based on the number of abnormal indicators and the alarm determination model.
10. A prediction system based on the method for predicting perioperative cardiovascular and cerebrovascular adverse events according to any one of claims 1-9, characterized in that, include: The parameter acquisition module is used to extract an initial parameter set from the actual physiological parameter set based on the movement time window; The data cleaning and calculation module is used to clean the initial parameter set and calculate the confidence parameter set and the target parameter set. The first calculation module is used to obtain the risk relationship comparison table and the basic pathological feature set, and calculate the first correction coefficient set in combination with the confidence parameter set. The model generation module is used to generate an early warning model based on the first set of correction coefficients and the target parameter set; The second calculation module is used to perform a collaborative deviation analysis on the early warning model and the target parameter set, and to calculate the second correction coefficient. The early warning module is used to generate early warning results based on the second correction coefficient and the early warning model.