Post-operative cardiovascular risk identification system based on multi-physiological parameter feedback
By constructing a pharmacokinetic mask window and dynamic baseline updates, combined with patient characteristic similarity, the problem of time alignment error and individual differences between physiological parameter changes and drug intervention behavior in postoperative cardiovascular monitoring was solved, thus achieving accuracy and stability in postoperative cardiovascular risk identification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEILONGJIANG UNIV OF CHINESE MEDICINE
- Filing Date
- 2026-05-21
- Publication Date
- 2026-07-21
AI Technical Summary
In current cardiovascular surgery monitoring, the timing misalignment between changes in physiological parameters and drug interventions, as well as individual differences, lead to inaccurate risk assessment and make it difficult to achieve accurate risk identification.
By constructing a pharmacokinetic-based mask window, combining patient baseline similarity, dynamically updating the baseline, and using a pre-trained classification model for risk identification, the accuracy of risk identification is improved by eliminating time alignment errors and individual differences.
It has achieved accuracy and stability in identifying postoperative cardiovascular risks, reduced errors caused by drug intervention and individual differences, and improved the individualization and medical rationality of the monitoring system.
Smart Images
Figure CN122229405B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of cardiovascular medical technology, and more specifically to a cardiovascular postoperative risk identification system based on multi-physiological parameter feedback. Background Technology
[0002] Post-cardiovascular surgery patients typically require continuous monitoring of vital signs using multi-parameter monitoring equipment, along with medication and supportive interventions based on clinical orders. With advancements in intensive care technology, bedside monitoring devices can collect real-time data on multidimensional physiological parameters such as heart rate, blood pressure, and blood oxygen saturation, enabling centralized management and retrieval of this data through hospital information systems. Building upon this foundation, and combining medication information recorded in electronic medical records with the patient's baseline characteristics, a comprehensive assessment of the post-operative patient's physiological status has become a crucial direction for clinical monitoring and risk assessment.
[0003] During post-cardiovascular surgery monitoring, changes in patients' physiological parameters are not only influenced by their own pathological state but also closely related to clinical medication and intervention behaviors. Because current data recording methods typically use the prescribed time as a reference for medication administration, they fail to accurately reflect the actual action of the drug in the body, leading to a time alignment error between changes in physiological parameters and intervention behaviors. Furthermore, individual differences exist in the physiological responses of different patients under the same medication conditions, making it difficult for risk assessment methods based on uniform standards to accurately characterize the true risk status. Summary of the Invention
[0004] This invention provides a cardiovascular postoperative risk identification system based on multi-physiological parameter feedback to solve the existing problem of how to effectively model multiple physiological parameters while taking into account the dynamic effects of medication and individual differences, so as to achieve more accurate cardiovascular postoperative risk identification.
[0005] The cardiovascular postoperative risk identification system based on multi-physiological parameter feedback of the present invention adopts the following technical solution:
[0006] One embodiment of the present invention provides a cardiovascular postoperative risk identification system based on multi-physiological parameter feedback, the system comprising the following modules:
[0007] The medical data module is used to acquire physiological status data, medication record data, and basic characteristic data for the current patient and several historical patients.
[0008] The influence analysis module is used to construct a mask window and perform pharmacokinetic characteristic analysis based on drug record data to determine the window size of the mask window. It also calculates the similarity by combining the basic characteristic data of the current patient and historical patients to obtain the reference influence coefficient of each physiological parameter of the current patient at any time of drug administration.
[0009] The dynamic baseline module is used to predict the trend of physiological parameters based on the reference influence coefficient, mask window and drug concentration changes, and obtain the baseline of the previous moment. Then, the baseline of the previous moment is iteratively updated by using the predicted values of physiological parameters to obtain the updated baseline.
[0010] The risk assessment module is used to determine the current patient's risk level based on the deviation of physiological parameters from the updated baseline, and to identify the corresponding risk type using a pre-trained classification model.
[0011] Furthermore, the specific method for obtaining the mask window is as follows:
[0012] Based on the medication record data, the start and end times of drug administration are determined, and the pharmacokinetic parameters of the corresponding drugs are obtained.
[0013] Based on the elimination rate constant, steady-state concentration during continuous infusion, and preset minimum effective concentration threshold in the pharmacokinetic parameters, the decay time when the drug concentration drops below the minimum effective concentration threshold is calculated.
[0014] The time interval between the drug administration start time and the decay time is used as the mask window.
[0015] Furthermore, the specific method for calculating similarity by combining the basic characteristic data of the current patient and historical patients includes:
[0016] Standardize the basic feature data of the current patient and each historical patient to obtain the basic feature vector of the current patient and the basic feature vector of each historical patient.
[0017] Calculate the distance metric between the current patient's baseline feature vector and the baseline feature vectors of each historical patient;
[0018] The distance metric is transformed by a preset decreasing mapping function to obtain the similarity between the current patient and each historical patient.
[0019] Furthermore, the specific method for obtaining the reference influence coefficient is as follows:
[0020] Based on the similarity between the current patient and each historical patient, several target historical patients are selected from all historical patients. Then, through medication record data and physiological state data, the changes in physiological indicators and corresponding medication dosages of each target historical patient under the medication event in the current patient are obtained.
[0021] Calculate the ratio of changes in physiological indicators to drug dosage for each target patient's history to obtain the unit effect ratio;
[0022] Using the similarity as a weight, the unit influence ratios of several target historical patients are weighted and summed to obtain the reference influence coefficient of the current patient on each physiological parameter under the medication event.
[0023] Furthermore, the specific methods for predicting the changing trends of physiological parameters include:
[0024] For current drug use events, a piecewise time response function is constructed to characterize the drug's onset and decay processes;
[0025] The predicted impact of a single medication administration is obtained by multiplying the reference impact coefficient corresponding to the medication event, the current medication dosage, and the time response function.
[0026] The statistical mean of physiological state data that were not within the mask window before the first baseline update was obtained as the initial baseline. The predicted impact of all medication events within the mask window at the current moment was summed with the initial baseline to obtain the predicted value of the physiological parameter.
[0027] Furthermore, the specific method for obtaining the updated baseline is as follows:
[0028] Get the baseline updated in the previous time step;
[0029] According to a preset smoothing coefficient, the baseline updated at the previous moment and the predicted values of physiological parameters at the current moment are weighted and fused to obtain the baseline updated at the current moment; wherein, the smoothing coefficient is used to control the weight ratio of the baseline updated at the previous moment and the predicted values in the update process.
[0030] Furthermore, the specific methods for determining the current patient's risk level based on the deviation of physiological parameters from the updated baseline include:
[0031] Calculate the difference between each physiological parameter at the current moment and the updated baseline, and normalize the difference by combining the historical fluctuation range to obtain the degree of abnormality of each physiological parameter;
[0032] The degree of abnormality of each physiological parameter is integrated to obtain a comprehensive abnormality index;
[0033] A sliding calculation is performed on the time series of the comprehensive anomaly index using a sliding window of preset duration to obtain a steady-state risk measurement value.
[0034] When the steady-state risk metric is greater than a preset risk threshold and the duration reaches a preset condition, the corresponding risk level is determined.
[0035] Furthermore, the medical data module includes data preprocessing before acquiring data, including the following specific methods:
[0036] Heterogeneous data is received and timestamped via a medical information bus.
[0037] Physiological state data with different sampling frequencies are resampled to a uniform time scale;
[0038] For any missing physiological state data segments, interpolation methods are used to complete them.
[0039] Furthermore, the method for obtaining the pre-trained classification model is as follows:
[0040] Obtain training samples from historical patients whose real clinical events have been labeled. The characteristics of the training samples include combinations of different physiological parameter abnormalities and are labeled with specific risk types.
[0041] The initial multilayer neural network model is trained under supervision using the training samples, enabling the multilayer neural network model to learn the mapping relationship between physiological abnormality combinations and risk types, thus obtaining the pre-trained classification model.
[0042] Furthermore, the specific method for identifying the corresponding risk type using a pre-trained classification model includes:
[0043] Extract the degree of abnormality of each physiological parameter at the current moment and construct a multidimensional abnormal feature vector;
[0044] The multidimensional anomaly feature vector is input into the pre-trained classification model, and the probability distribution of each candidate risk type is output through the inference of the classification model.
[0045] The candidate risk type corresponding to the highest probability value in the probability distribution is determined as the risk type of the current patient.
[0046] The beneficial effects of the technical solution of this invention are as follows: By introducing pharmacokinetic features to construct a dynamic mask window, the continuous time interval for the drug to produce real physiological effects in vivo can be determined, eliminating the time alignment error between physiological parameter fluctuations and drug intervention, and avoiding misjudging drug metabolic decay as patient pathological deterioration; when calculating drug effects, by introducing a historical patient weight screening mechanism based on basic feature similarity, the acquisition of reference effect coefficients no longer depends on a unified group standard, but is adapted to the individual physiological basis of the current patient, thereby improving the individualization and medical rationality of baseline correction; furthermore, by using prediction and dynamic baseline updates, drastic drift in baseline estimation is avoided, thereby effectively filtering out instantaneous noise interference caused by changes in body position, lead dropout, etc., solving the problems of medication time alignment error and individual differences, and improving the accuracy and stability of cardiovascular postoperative risk identification. Attached Figure Description
[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0048] Figure 1 This is a structural block diagram of the cardiovascular postoperative risk identification system based on multi-physiological parameter feedback of the present invention;
[0049] Figure 2 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present invention. Detailed Implementation
[0050] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the cardiovascular postoperative risk identification system based on multi-physiological parameter feedback proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0052] The following description, in conjunction with the accompanying drawings, details the specific scheme of the cardiovascular postoperative risk identification system based on multi-physiological parameter feedback provided by this invention.
[0053] Please see Figure 1 The diagram illustrates a structural block diagram of a cardiovascular postoperative risk identification system based on multi-physiological parameter feedback, according to an embodiment of the present invention. The system includes the following modules:
[0054] The medical data module 101 is used to acquire physiological status data, medication record data, and basic characteristic data corresponding to the current patient and several historical patients.
[0055] It should be noted that this step is used to construct the raw input data system for the cardiovascular postoperative risk identification system. Its core objective is to transform heterogeneous data from different devices and systems into structured input that can be used for subsequent modeling. Based on the temporal attributes and functional differences of the data, the collected data is divided into two categories: one is a sequence of physiological states that changes continuously over time. Another type is static feature vectors that reflect the basic characteristics of individual patients. These two processes serve the subsequent dynamic modeling and similarity analysis processes, respectively.
[0056] To implement the cardiovascular postoperative risk identification system based on multi-physiological parameter feedback proposed in this embodiment, it is first necessary to acquire physiological state data, medication record data, and basic characteristic data corresponding to the current patient and several historical patients, respectively. The specific process is as follows:
[0057] First, by using a multi-parameter bedside monitor and an electronic medical record system, physiological status data, medication record data, and basic characteristic data of the current patient and several historical patients are obtained.
[0058] As an optional embodiment, the specific method for acquiring the physiological state data is as follows: A three-lead electrocardiogram signal is acquired through the electrocardiogram monitoring module of a multi-parameter bedside monitor, and feature extraction is performed at the data access end to obtain a heart rate sequence updated every second; blood pressure data is acquired through an invasive arterial blood pressure sensor or an oscillometric non-invasive blood pressure module, and discrete measurement values are uniformly mapped to a 1Hz time scale by combining time interpolation and resampling strategies; photoplethysmography signals are acquired through a pulse oximeter probe, and blood oxygen saturation and perfusion index are calculated. The obtained heart rate sequence, blood pressure data, blood oxygen saturation, and perfusion index are collectively referred to as the patient's physiological state data.
[0059] In the process of acquiring three-lead electrocardiogram signals, the specific sampling frequency should not be lower than 100Hz. In one specific embodiment of the present invention, the sampling frequency is set to 100Hz. In other embodiments, it can be adjusted according to the actual situation.
[0060] As an optional embodiment, the specific method for obtaining the medication record data is as follows: extract the patient's medication records from the electronic medical record system. The medication records include various types of drugs, drug administration start time, end time, route of administration, and drug dosage rate. Each medication record is treated as a medication event, and all of the patient's medication events are used as the patient's medication record data.
[0061] In a specific embodiment, the drug type specifically includes: anticoagulants (such as heparin and warfarin), antiplatelet drugs (such as aspirin and clopidogrel), vasoactive drugs (such as norepinephrine and dopamine), and antiarrhythmic drugs (such as amiodarone); the route of administration specifically includes: intravenous bolus injection, continuous intravenous infusion, and oral administration.
[0062] As an optional embodiment, the specific method for obtaining the basic characteristic data is as follows: collecting the patient's demographic information, medical history, and postoperative key time, and using the patient's demographic information, medical history, and postoperative key time as the patient's basic characteristic data.
[0063] In one specific embodiment, the demographic information includes, but is not limited to: age, sex, and weight; the medical history includes, but is not limited to: hypertension, diabetes, and renal insufficiency; and the postoperative critical times include, but are not limited to: extubation time and first time out of bed.
[0064] Then, all raw data is transmitted to the central data server via a medical information bus (such as the HL7 protocol), and stored in the time series database after being timestamped. For data missing or outliers (such as flat segments caused by signal dropout), linear interpolation or neighboring effective value filling methods are used for preprocessing, and the preprocessed multi-physiological parameters and medication event sequences are used as input data for step two.
[0065] Thus, the physiological status data, medication record data, and basic characteristic data corresponding to the current patient and several historical patients have been obtained through the above methods.
[0066] The influence analysis module 102 is used to construct a mask window and perform pharmacokinetic characteristic analysis based on drug record data to determine the window size of the mask window. It also calculates the similarity by combining the basic characteristic data of the current patient and historical patients to obtain the reference influence coefficient of each physiological parameter of the current patient at any time of drug administration.
[0067] It is important to note that after a drug enters the body intravenously or orally, it undergoes a distribution, metabolism, and elimination half-life process. However, the dosing start time recorded in electronic medical records is usually the moment the nurse begins executing the doctor's order, and the end time is often assumed to be the time specified in the order to stop or the start time of the next dose. This recording method masks the continuous changes in drug concentration from onset to peak effect to decay. For example, after an intravenous bolus injection of norepinephrine, blood pressure begins to rise within 30 seconds, reaches its peak in 2-5 minutes, and the effect subsides within 5-10 minutes after discontinuation; while after continuous infusion of heparin, the prolonged effect of activated partial thromboplastin time takes 2-6 hours to reach steady state. Directly using the doctor's prescribed time as the drug's effective range can lead to misattribution of physiological parameter changes, misjudging fluctuations outside the drug-affected period as drug effects, or incorrectly attributing changes during the drug-effect period to the patient's own pathological evolution.
[0068] As a preferred embodiment, the specific method for obtaining the mask window is as follows: determining the dosing start time and dosing end time based on the medication record data, and obtaining the pharmacokinetic parameters of the corresponding drug; calculating the decay time when the drug concentration drops below the minimum effective concentration threshold based on the elimination rate constant, steady-state concentration during continuous infusion, and preset minimum effective concentration threshold in the pharmacokinetic parameters; and using the time interval between the dosing start time and the decay time as the mask window.
[0069] For example, the specific value of the minimum effective concentration threshold can be obtained from a pharmacology database, and the unit is mg / L.
[0070] It should be noted that fixed windows cannot accommodate the differences in half-lives among different drugs and administration methods. For example, using an excessively long window for intravenously administered drugs can introduce irrelevant time-related noise, while using an excessively short window for continuously infused drugs can truncate the drug's steady-state period, leading to baseline offset estimation bias. Therefore, by constructing a dynamic mask window, the duration during which drug intervention significantly affects physiological parameters can be precisely defined, providing an accurate temporal interval for subsequent baseline reconstruction. This addresses the time alignment error between multiple physiological parameters and drug intervention caused by the coarse temporal granularity of clinical medication records, such as the inconsistency between the start and end times of drug administration recorded in the doctor's order and the actual time period during which the drug produces its physiological effect in the body.
[0071] For example, under any medication event, the specific method for calculating the decay time can be:
[0072]
[0073] in, Indicates the decay time; This is the end time of drug administration. This represents the average drug concentration (mg / L) from the start to the end of the dosing period. This is the decay rate parameter; This represents the minimum effective concentration threshold of the drug. This represents the logarithmic function with the natural constant as the base.
[0074] For example, the specific calculation method for the decay rate parameter is as follows: obtain the elimination half-life of the drug, denoted as... ,Will As a parameter of decay rate.
[0075] By acquiring a masked window, after a patient's cardiovascular surgery, the actual impact window corresponding to each medication administration is formed based on the time interval of each administration. This represents the period during which the drug has a significant impact on physiological parameters. Utilizing the exponential decay characteristics of pharmacokinetics, discrete prescription time points are transformed into continuous probabilistic impact intervals. Simultaneously, by introducing a minimum effective concentration threshold, continuous concentration values are binarized into "influenced" and "no-influence" intervals, avoiding errors from directly judging the impact time window based on the patient's physiological indicators. Furthermore, the method for calculating the decay time includes an average drug concentration from the start to the end of administration, which reflects the steady-state concentration during the administration period.
[0076] As a preferred embodiment, the method for calculating similarity by combining the basic feature data of the current patient and historical patients includes: standardizing the basic feature data of the current patient and each historical patient to obtain the basic feature vector of the current patient and the basic feature vector of each historical patient; calculating the distance metric between the basic feature vector of the current patient and the basic feature vector of each historical patient; and converting the distance metric through a preset decreasing mapping function to obtain the similarity between the current patient and each historical patient.
[0077] As a preferred embodiment, the specific method for obtaining the reference influence coefficient is as follows: based on the similarity between the current patient and each historical patient, several target historical patients are selected from all historical patients, and through medication record data and physiological state data, the changes in physiological indicators and corresponding medication dosages of each target historical patient under the medication event in which the current patient is located are obtained; the ratio of the changes in physiological indicators to the medication dosage of each target historical patient is calculated to obtain the unit influence ratio; using the similarity as a weight, the unit influence ratios of several target historical patients are weighted and summed to obtain the reference influence coefficient of the current patient on each physiological parameter under the medication event.
[0078] It is important to note that after obtaining the actual time window (i.e., mask window) for each drug, baseline correction is needed for the physiological parameters within this time window to eliminate the direct impact of drug intervention on the original physiological data. By converting physiological changes within the drug's action interval from "observed values" to "expected reference values," a stable benchmark can be provided for subsequent calculations of the degree of abnormality. Furthermore, when performing baseline correction, relying solely on the current patient's historical data is insufficient to accurately depict the true trend of change under the intervention. Therefore, historical patients' medication records and corresponding physiological indicator changes are introduced as a reference. By utilizing the response patterns of historical patients under similar medication conditions, the problem of insufficient short-term observational data for the current patient can be compensated for, making the baseline estimation more stable and medically interpretable, while avoiding baseline shifts caused by individual patient fluctuations.
[0079] It should be further noted that, due to differences in basic physiological conditions such as age, gender, and weight among different patients, their physiological response levels can vary significantly even under the same medication conditions. Therefore, when selecting historical patients as reference subjects, it is necessary to prioritize screening patient groups with similar basic characteristics to the current patients in order to improve the reliability and suitability of the reference data.
[0080] For example, the specific method for calculating the similarity between the current patient and each historical patient can be: ,in, Indicates the current patient and the first Similarity between historical patients This represents an exponential function with the natural constant as its base. Indicates the current patient and the first The Euclidean distance between the basic feature vectors corresponding to each historical patient.
[0081] For example, the specific calculation method for the reference influence coefficient can be:
[0082]
[0083] in, This indicates the current patient's medication-related events. Physiological parameters Reference influence coefficient; Indicates the current patient and the first Similarity between historical patients This represents the set of all target historical patients of the current patient; This represents the number of target historical patients in the set; Indicates the current patient's number A target patient's history of medication events Physiological parameters The changes in physiological indicators; Indicates the current patient's number A target patient's history of medication events The corresponding medication dosage is as follows.
[0084] Similarity is used as a reference weight for historical samples to the current patient, and is used to calculate the influence coefficient of changes in physiological indicators of the current patient under medication conditions. The specific calculation method for the reference influence coefficient uses a similarity-weighted approach, ensuring that historical samples more closely related to the current patient (i.e., the target historical patient) contribute more to the influence coefficient, thereby improving the model's individualization.
[0085] It is necessary to further explain that the reason for adopting the above-mentioned similarity-weighted reference influence coefficient calculation method is that in clinical pharmacology, the effects of the same dose of drug vary greatly among different cardiovascular postoperative patients. For example, elderly patients or those with renal insufficiency have significantly reduced clearance rates of vasoactive drugs (such as norepinephrine), resulting in a much stronger pressor effect per unit dose compared to younger patients. If empirical values from fixed pharmacology manuals are directly used for baseline correction, it will inevitably lead to over- or under-correction. The embodiments of this invention, through basic feature similarity screening, actually match historical patient groups with similar drug metabolism capabilities and target organ sensitivities, thereby making the extracted unit influence ratio (unit influence coefficient) more consistent with the current patient's true pathophysiological compensation state, ensuring the accuracy of subsequent baseline update results.
[0086] Thus, the reference influence coefficients of each physiological parameter of the current patient after each medication administration are obtained through the above methods.
[0087] The dynamic baseline module 103 is used to predict the trend of physiological parameter changes based on the reference influence coefficient, mask window and drug concentration changes, and obtain the baseline of the previous moment. Then, the baseline of the previous moment is iteratively updated by the predicted value of the physiological parameters to obtain the updated baseline.
[0088] It should be noted that the above influence coefficients reflect the static impact of a unit dose on physiological indicators. However, in actual clinical practice, drug effects exhibit dynamic changes over time. For example, bolus-administered drugs have rapid onset and rapid attenuation, while continuous infusion drugs exhibit a slow process of reaching steady state and slow decline. Therefore, when performing baseline correction, it is necessary to dynamically adjust the influence coefficients in conjunction with the time dimension after medication administration.
[0089] As a preferred embodiment, the method for predicting the trend of physiological parameter changes includes: constructing a piecewise time response function characterizing the drug's onset and decay processes for the currently acting medication event; multiplying the reference influence coefficient corresponding to the medication event, the current medication dose, and the time response function to obtain the predicted influence of a single medication administration; obtaining the statistical mean of physiological state data that were not within the mask window before the first baseline update as the initial baseline; and summing the predicted influence of all medication events currently within the mask window with the initial baseline to obtain the predicted value of the physiological parameter.
[0090] For example, the specific calculation formula for the piecewise time response function can be:
[0091]
[0092] in, This indicates the current patient's medication-related events. The next moment Time response weights; For current patients' medication events The onset rate parameter is below. For current patients' medication events The decay rate parameter is as follows; For current patients' medication events The following is the start time for drug administration; For current patients' medication events The following is the end time of drug administration. This represents an exponential function with the natural constant as its base.
[0093] For example, the specific method for obtaining the onset rate parameter is as follows: during the medication event... The time when the drug concentration reaches its maximum is recorded as the peak concentration time. The rate of change of drug concentration between the start of drug administration and the peak concentration time is used as the onset rate parameter.
[0094] For a piecewise time response function, the onset rate parameter describes the growth process of the drug from administration to peak concentration, while the decay rate parameter reflects the drug's metabolic process after administration. Therefore, by constructing a piecewise function based on changes in drug concentration during the administration and metabolism phases and incorporating an exponential function, the process of continuously increasing drug concentration after administration until administration ceases, followed by a gradual decrease in concentration due to metabolism, is realized. Thus, the piecewise time response function exhibits exponential growth in the initial segment, simulating the drug's onset process, and exponential decay in the later segment, simulating the drug's metabolic process, which can better fit the actual trend of drug efficacy changes. As time extends, in... As the function value gradually approaches 1, it indicates that the drug's effect is gradually increasing. Subsequently, the function value gradually decreased, reflecting the weakening of the drug's effect.
[0095] For example, the specific calculation method for the predicted values of physiological parameters can be as follows:
[0096]
[0097] in, This indicates the effect of the current medication status on the first... A physiological parameter at time The predicted value below; This indicates the current patient's medication-related events. Physiological parameters Reference influence coefficient, This indicates the current patient's medication-related events. The next moment Time response weights This serves as the initial baseline; For current patients' medication events The following dosage; This represents the set of medication events for the current patient.
[0098] It should be noted that the predicted values calculated based on the reference influence coefficient and time response function reflect the physiological change trend under the ideal constraint of only considering known drug interventions. However, in real postoperative monitoring, patients also experience non-drug-driven pathological evolution, such as progressive anemia caused by occult blood loss and natural decline in ventricular diastolic function. These factors cannot be predicted through medication records. If the prediction results are solely based on pure mathematical superposition, the predicted baseline will become severely deviated from the actual state when the patient experiences a sudden deterioration due to non-drug factors. Therefore, this embodiment further introduces the baseline updated at the previous moment for smoothing and iterative correction. Essentially, this endows the baseline system with memory capabilities, allowing the baseline to follow the trend of drug efficacy prediction while always being constrained by the patient's recent actual physiological state. When non-drug factors worsen, the real-time observation value will continuously deviate from the predicted value. Through the iteration of the smoothing coefficient, the baseline will gradually move closer to the deteriorating actual state, thereby improving the robustness of the system in complex clinical scenarios.
[0099] As a preferred embodiment, the specific method for obtaining the updated baseline is as follows: obtaining the updated baseline of the previous moment; performing weighted fusion calculation on the updated baseline of the previous moment and the predicted value of the physiological parameters at the current moment according to a preset smoothing coefficient to obtain the updated baseline at the current moment; wherein, the smoothing coefficient is used to control the weight ratio of the updated baseline of the previous moment and the predicted value in the update process.
[0100] For example, the specific calculation method for the updated baseline can be:
[0101]
[0102] in, For the first A physiological parameter at time The updated baseline below; For the first A physiological parameter at time The updated baseline below; The preset smoothing coefficient, This indicates the effect of the current medication status on the first... A physiological parameter at time The predicted value is below.
[0103] The smoothing coefficient, used to control the fusion ratio of historical data and predicted data, is set to 0.7 in one specific embodiment of the invention. In other embodiments, it can be adjusted according to actual conditions; the embodiments of the invention do not impose specific limitations. Regarding the calculation method of the updated baseline, this formula incorporates historical baselines. This allows the system to maintain its predictive ability while avoiding drastic fluctuations, thereby improving the stability and robustness of baseline estimation. After updating the baseline using the above methods, the resulting dynamic baseline is no longer a simple statistical average, but a conditional expectation that comprehensively considers individual patient characteristics, current medication status, and time response patterns, and can more realistically reflect the normal physiological level under the current intervention environment.
[0104] Thus, the updated baseline is obtained through the above method.
[0105] The risk assessment module 104 is used to determine the current patient's risk level based on the deviation of physiological parameters from the updated baseline, and to identify the corresponding risk type using a pre-trained classification model.
[0106] It should be noted that although the interference of drug intervention has been eliminated after the reconstruction of the dynamic baseline, the physical characteristics of the intensive care environment itself still introduce a large amount of high-frequency noise. For example, a nurse turning a patient over and patting their back, a patient holding their breath briefly due to pain, or a brief period of poor contact of the ECG electrodes can all cause transient spikes in heart rate or blood oxygen saturation lasting from a few seconds to tens of seconds. If alarms are triggered based solely on the degree of abnormality at a single point in time, a large number of invalid noise alarms will be generated, leading to clinical alarm fatigue and a decrease in the sensitivity of medical staff to truly dangerous signals. Therefore, to solve the above problems, this embodiment of the invention compares the real-time collected physiological parameters with the updated baseline at the corresponding time to determine the degree of deviation of the current physiological state from the normal state that should occur under individual conditions, so as to facilitate further risk level determination and risk type identification.
[0107] Specifically, firstly, any physiological parameter in the physiological state data is compared with the corresponding updated baseline, and the absolute value of the difference between the physiological parameter and the updated baseline at each time point is obtained as the deviation of the physiological parameter at the corresponding time point. The coefficient of variation of the physiological parameter at all times is used as the degree of abnormality. The average value of the degree of abnormality of all physiological parameters is taken as the comprehensive abnormality index of the current patient.
[0108] The degree of abnormality essentially characterizes the extent to which current physiological parameters deviate from the normal state that an individual should be in. A comprehensive abnormality index is obtained by merging and averaging the abnormality degrees of multiple physiological parameters, reflecting the degree of overall physiological imbalance in the current patient. Furthermore, to avoid false alarms caused by transient noise or brief fluctuations, the system performs continuous time series analysis. A cumulative analysis was performed over time, and the median within the window was calculated using a sliding window of 60 seconds. And use it as a measure of steady-state risk. When If a risk level consistently exceeds a preset risk threshold (0.3 for low risk, 0.6 for medium risk, and 0.85 for high risk) for more than 10 seconds, the user is considered to have entered the corresponding risk level. The start time and duration of each risk level are recorded to form a continuous sequence of risk events, providing structured evidence for subsequent clinical analysis.
[0109] Then, all physiological parameters are traversed through a sliding window of a preset length, and the comprehensive abnormal indicators of the current patient are obtained through a comprehensive abnormal indicator acquisition method. The median of the comprehensive abnormal indicators at all times within the time range corresponding to the sliding window is used as the steady-state risk measure. Different risk thresholds are preset. When the steady-state risk measure exceeds the highest risk threshold that the steady-state risk measure can reach at several consecutive times, the risk level corresponding to the highest risk threshold is used as the final risk level.
[0110] Secondly, the degree of abnormality corresponding to each of the physiological parameters is obtained and formed into a vector, denoted as the degree of abnormality vector. The degree of abnormality vector is then input into the trained neural network model to identify the corresponding risk type.
[0111] The model is trained based on real clinical events labeled in historical patients. It can learn the mapping relationship between different combinations of physiological abnormalities and specific risk types, thereby outputting the probability distribution of various risks (such as hypotension-related risks, arrhythmia risks, respiratory failure, etc.) and selecting the category with the highest probability as the current risk type.
[0112] As an optional embodiment, the specific network structure of the neural network model can be: including an input layer, several fully connected hidden layers, and an output layer; wherein, the number of nodes in the input layer matches the dimension of the anomaly degree vector; each fully connected hidden layer is used to perform high-dimensional feature extraction and nonlinear mapping on the input feature vector, and a ReLU activation function is configured after each hidden layer to increase the nonlinear expressive power of the model; the number of nodes in the output layer matches the preset total number of risk types, and a Softmax activation function is configured in the output layer to transform the output result into a normalized vector representing the probability distribution of various risks.
[0113] For example, the specific training method of the neural network model can be as follows: obtain a training sample set of labeled real clinical events from historical patients; use the anomaly degree vector as a sample and the real risk type as a sample label; use the cross-entropy loss function as the objective function of the neural network model to measure the difference between the probability distribution predicted by the model and the real label; use the Adam gradient descent optimizer to calculate the loss value through forward propagation, and then update the weights and bias parameters of each fully connected layer in the network through the backpropagation algorithm, iteratively training until the loss function converges, thereby obtaining a pre-trained classification model with classification reasoning ability.
[0114] Finally, the risk level and risk type are output to the central monitoring interface, triggering the corresponding alarm mechanism. At the same time, the relevant time series data and identification results are stored for subsequent model optimization and clinical quality review analysis.
[0115] Through the above steps, this invention achieves precise time alignment between medication behavior and changes in physiological parameters by constructing a real-time window based on pharmacokinetics. Furthermore, by combining this with a patient baseline characteristic similarity screening mechanism, it performs individualized modeling of the impact of different medication conditions on physiological indicators. Simultaneously, it introduces a time response function to dynamically correct the drug action process, thereby completing the adaptive reconstruction of the physiological parameter baseline. Further, by comparing the corrected baseline with real-time monitoring data and combining a multi-parameter abnormality fusion and time accumulation judgment mechanism, it effectively improves the stability and accuracy of risk identification, reduces the interference of short-term fluctuations and intervention behaviors on the judgment results, and realizes continuous, reliable, and individually adaptable risk monitoring and early warning for post-cardiovascular surgery patients.
[0116] This concludes the embodiment.
[0117] This invention also provides an electronic device. Please refer to [link to relevant documentation]. Figure 2 The electronic device may include a processor 201, a memory 202, and a program 2021 stored in the memory 202 and executable on the processor 201.
[0118] When program 2021 is executed by processor 201, it can achieve the following: Figure 1 The corresponding system embodiments and any modules that achieve the same beneficial effects will not be described in detail here.
[0119] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by hardware related to program instructions, and the program can be stored in a readable medium.
[0120] This invention also provides a readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described functions. Figure 1 Any module in the corresponding system embodiment that can achieve the same technical effect will not be described again here to avoid repetition.
[0121] The computer-readable storage medium of this invention can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
[0122] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.
[0123] The program code contained on the storage medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.
[0124] Computer program code for performing the operations of this invention can be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, and C++, as well as conventional procedural programming languages such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or terminal. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0125] This invention also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned steps to realize the cardiovascular postoperative risk identification system based on multi-physiological parameter feedback provided in the above embodiments.
[0126] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0127] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0128] It should also be noted that the method used in this embodiment The model is only used to represent negative correlations and the results of the constraint model output are in Within this range, in specific implementations, other models with the same purpose can be substituted; this embodiment is merely an example. The description will be based on a model, without making specific limitations on it. This refers to the input of the model.
[0129] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A cardiovascular postoperative risk identification system based on multi-physiological parameter feedback, characterized in that, The system includes the following modules: The medical data module is used to acquire physiological status data, medication record data, and basic characteristic data for the current patient and several historical patients. The influence analysis module is used to construct a mask window and perform pharmacokinetic characteristic analysis based on drug record data to determine the window size of the mask window. It also calculates the similarity by combining the basic characteristic data of the current patient and historical patients to obtain the reference influence coefficient of each physiological parameter of the current patient at any time of drug administration. The specific method for obtaining the reference influence coefficient is as follows: Based on the similarity between the current patient and each historical patient, several target historical patients are selected from all historical patients. Then, through medication record data and physiological state data, the changes in physiological indicators and corresponding medication dosages of each target historical patient under the medication event in the current patient are obtained. Calculate the ratio of changes in physiological indicators to drug dosage for each target patient's history to obtain the unit effect ratio; Using the similarity as a weight, the unit influence ratios of several target historical patients are weighted and summed to obtain the reference influence coefficient of the current patient on each physiological parameter under the medication event. The dynamic baseline module is used to predict the trend of physiological parameters based on the reference influence coefficient, mask window and drug concentration changes, and obtain the baseline of the previous moment. Then, the baseline of the previous moment is iteratively updated by using the predicted values of physiological parameters to obtain the updated baseline. The specific methods for predicting the changing trends of physiological parameters are as follows: For current drug use events, a piecewise time response function is constructed to characterize the drug's onset and decay processes; The predicted impact of a single medication administration is obtained by multiplying the reference impact coefficient corresponding to the medication event, the current medication dosage, and the time response function. The statistical mean of physiological state data that were not within the mask window before the first baseline update was obtained as the initial baseline. The predicted impact of all medication events within the mask window at the current moment was summed with the initial baseline to obtain the predicted value of the physiological parameter. The risk assessment module is used to determine the current patient's risk level based on the deviation of physiological parameters from the updated baseline, and to identify the corresponding risk type using a pre-trained classification model.
2. The cardiovascular postoperative risk identification system based on multi-physiological parameter feedback according to claim 1, characterized in that, The specific method for obtaining the mask window is as follows: Based on the medication record data, the start and end times of drug administration are determined, and the pharmacokinetic parameters of the corresponding drugs are obtained. Based on the elimination rate constant, steady-state concentration during continuous infusion, and preset minimum effective concentration threshold in the pharmacokinetic parameters, the decay time when the drug concentration drops below the minimum effective concentration threshold is calculated. The time interval between the drug administration start time and the decay time is used as the mask window.
3. The cardiovascular postoperative risk identification system based on multi-physiological parameter feedback according to claim 1, characterized in that, The specific methods for calculating similarity by combining the basic characteristic data of current patients and historical patients are as follows: Standardize the basic feature data of the current patient and each historical patient to obtain the basic feature vector of the current patient and the basic feature vector of each historical patient. Calculate the distance metric between the current patient's baseline feature vector and the baseline feature vectors of each historical patient; The distance metric is transformed by a preset decreasing mapping function to obtain the similarity between the current patient and each historical patient.
4. The cardiovascular postoperative risk identification system based on multi-physiological parameter feedback according to claim 1, characterized in that, The specific method for obtaining the updated baseline is as follows: Get the baseline updated from the previous moment; According to a preset smoothing coefficient, the baseline updated at the previous time moment and the predicted values of physiological parameters at the current time moment are weighted and fused to obtain the baseline updated at the current time moment; wherein, the smoothing coefficient is used to control the weight ratio of the baseline updated at the previous time moment moment and the predicted values during the update process.
5. The cardiovascular postoperative risk identification system based on multi-physiological parameter feedback according to claim 1, characterized in that, The method for determining the current patient's risk level based on the deviation of physiological parameters from the updated baseline includes: Calculate the difference between each physiological parameter at the current moment and the updated baseline, and normalize the difference by combining the historical fluctuation range to obtain the degree of abnormality of each physiological parameter; The degree of abnormality of each physiological parameter is integrated to obtain a comprehensive abnormality index; A sliding calculation is performed on the time series of the comprehensive anomaly index using a sliding window of preset duration to obtain a steady-state risk measurement value. When the steady-state risk metric is greater than a preset risk threshold and the duration reaches a preset condition, the corresponding risk level is determined.
6. The cardiovascular postoperative risk identification system based on multi-physiological parameter feedback according to claim 1, characterized in that, The medical data module also includes data preprocessing before acquiring data, including the following specific methods: Heterogeneous data is received and timestamped via a medical information bus. Physiological state data with different sampling frequencies are resampled to a uniform time scale; For any missing physiological state data segments, interpolation methods are used to complete them.
7. The cardiovascular postoperative risk identification system based on multi-physiological parameter feedback according to claim 1, characterized in that, The method for obtaining the pre-trained classification model is as follows: Obtain training samples from historical patients whose real clinical events have been labeled. The characteristics of the training samples include combinations of different physiological parameter abnormalities and are labeled with specific risk types. The initial multilayer neural network model is trained under supervision using the training samples, enabling the multilayer neural network model to learn the mapping relationship between physiological abnormality combinations and risk types, thus obtaining the pre-trained classification model.
8. The cardiovascular postoperative risk identification system based on multi-physiological parameter feedback according to claim 5, characterized in that, The specific methods for identifying the corresponding risk type using a pre-trained classification model are as follows: Extract the degree of abnormality of each physiological parameter at the current moment and construct a multidimensional abnormal feature vector; The multidimensional anomaly feature vector is input into the pre-trained classification model, and the probability distribution of each candidate risk type is output through the inference of the classification model. The candidate risk type corresponding to the maximum probability value in the probability distribution is determined as the risk type of the current patient.