Postoperative medication management system for acute abdominal disease

Abstract

The invention relates to the technical field of medical care informatics, and discloses an acute abdominal postoperative medication management system, which comprises a multi-modal data flow credibility identification unit, a state vector generation unit, a state vector generation unit and a medication management unit, the system comprises a data flow purification unit which receives a purified data flow and calculates a multi-order derivative of the purified data flow to generate a real-time state vector representing an illness state evolution direction and rate, and a feed-forward regulation and online learning unit which carries out prospective decision-making based on the state vector and continuously carries out individualized calibration according to vector change after decision-making. According to the method, discrete and possibly polluted original data points are reconstructed into an information entity which is subjected to logic purification and dimensionality raising to represent a process trend, so that the problem of misjudgment caused by data artifacts or decision lag caused by data isolation is avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to a postoperative management system for acute abdominal pain, belonging to the technical field of medical informatics. BACKGROUND

[0002] Currently, a common technical method is that the system receives multi-source physiological parameter data streams such as heart rate and blood pressure, compares the instantaneous measurement values of each parameter with the preset threshold, and triggers an alarm when the measurement value exceeds the range; this method can provide direct and effective warning for acute events caused by sharp changes in a single indicator.

[0003] However, in the recovery period of physiological instability such as postoperative acute abdominal pain, the disease evolution often shows slow and interrelated dynamic changes of multiple physiological parameters within their respective threshold ranges; in such applications, the information processing logic is based on the instantaneous value judgment of isolated data points, and there is a lack of effective recognition mechanism for the overall trend composed of multiple parameter collaborative evolution, which can cause the potential risk process to have developed before the hard alarm indicator is triggered, resulting in delayed clinical intervention.

[0004] To solve this problem, if the alarm thresholds of each parameter are simply tightened, in practice, excessive invalid alarms will be triggered due to normal physiological fluctuations of patients, reducing the clinical application value of the system; while introducing a single-parameter trend analysis method, although it improves the predictability to some extent, it still cannot integrate the pathophysiological relevance between different parameter change trends due to its analysis dimension still being limited to isolated parameter sequences; as can be seen from the analysis, the above improvement methods do not solve the technical problem of mismatch between the state snapshot type of information processing and the nature of continuous process type of disease evolution when dealing with dynamic multi-dimensional physiological data in the prior art; specifically, the prior art mainly has the following deficiencies: 1. Its data processing method is reactive, lacking a way to judge the overall evolution direction and rate of the disease; 2. Its decision is based on a single dimension threshold, which cannot integrate multi-dimensional information to identify systemic risks; 3. The system lacks an internal mechanism to adaptively adjust its monitoring strategy according to changes in patient physiological stability. Therefore, how to establish a new information processing system that can generate a real-time state vector representing the direction and rate of pathophysiological evolution by calculating the change speed and acceleration of multi-source dynamic physiological parameter data streams, and make forward-looking management decisions based on this vector, while having the ability to identify the credibility of the input data stream, online learning of individual drug response, and safe degradation when the system internal logic fails, has become a technical problem to be solved by the present application. SUMMARY

[0005] The application provides a postoperative drug management system for acute abdominal pain, which mainly aims to solve the problem that the existing medical information system adopts a judgment mode based on discrete data points and static thresholds, thus causing decision lag in drug management, being easily interfered by data artifacts, and lacking individualized adaptive and systematic security capabilities.

[0006] To achieve the above-mentioned purpose, the application provides a postoperative drug management system for acute abdominal pain, which comprises: A multi-modal data stream credibility identification unit is configured to receive multi-source physiological parameter data streams obtained from monitoring devices, and to perform artifact identification on the data streams according to preset logical consistency rules between at least two physiological parameters, so as to generate data streams that have been subjected to credibility evaluation. A multi-dimensional physiological state vector generation unit is configured to receive only the data streams that have been subjected to credibility evaluation, and to generate a real-time state vector that uniformly represents the evolution direction and evolution rate of the patient's pathological physiology by calculating the first and second derivatives of the physiological parameters reflected by the data streams. A drug efficacy coupling feedforward regulation unit is internally provided with a regulation model, and is configured to receive the real-time state vector as a decision input, and to generate or adjust the drug management instruction based on the direction of the real-time state vector and the change rate of the modulus thereof. An individualized drug efficacy response feature online self-learning unit is configured to take the drug management instruction as an input, and to take the actual change of the real-time state vector within a preset time window after the execution of the instruction as an output, so as to continuously calibrate the regulation model of the drug efficacy coupling feedforward regulation unit online.

[0007] Preferably, the multi-modal data stream credibility identification unit is further configured to calculate a credibility weight for each data stream in real time, and when the credibility weight of a certain data stream is lower than a preset threshold, the multi-dimensional physiological state vector generation unit will reduce the weight of the data stream in the weighted fusion calculation or temporarily isolate the data stream when generating the real-time state vector.

[0008] Preferably, the individualized drug efficacy response feature online self-learning unit is further configured to estimate a response coefficient for quantifying the response sensitivity of the current patient to the preset drug in real time according to the drug management instruction and the actual change of the subsequent real-time state vector, and to take the response coefficient as the basis for calibrating the regulation model.

[0009] Preferably, the application further comprises a physiological stability monitoring module configured to generate a state parameter representing the current physiological stability of the patient by calculating the prediction error between the predicted state vector generated by the regulation model and the real-time state vector, and an adaptive threshold adjustment bus configured to receive the state parameter and dynamically and nonlinearly adjust the decision threshold used by the multi-modal data stream credibility identification unit when performing artifact identification according to the parameter.

[0010] Preferably, the adaptive threshold adjustment bus is configured to automatically lower the decision threshold when the value of the state parameter exceeds a first preset value, and to automatically raise the decision threshold when the value of the state parameter falls below a second preset value.

[0011] Preferably, the system further comprises a multi-level functional safety degradation monitor configured to continuously monitor whether the response coefficient output by the individualized pharmacodynamic response feature online self-learning unit is within a preset absolute safety range, and to automatically and orderly switch the operation mode of the entire system from a full-function adaptive regulation mode to a predefined, more basic functional safety mode when the response coefficient is diagnosed to be outside the absolute safety range and the unit is determined to be failed.

[0012] Preferably, the predefined, more basic functional safety mode at least includes one of the following two modes: a standard feedforward guardian mode, in which the monitor deprives the online self-learning unit of control and instructs the feedforward regulation unit to switch to a set of built-in, standardized conservative regulation strategy; or a core monitoring safety mode, in which the monitor deprives all automatic regulation functions, causing the system to switch to only visual monitoring and alarming of the real-time state vector.

[0013] Preferably, the real-time state vector is a vector defined by its direction and modulus in a multi-dimensional state space, wherein the direction represents the overall trend of improvement, deterioration or stability of the patient's condition, and the modulus represents the degree of severity of the trend.

[0014] Preferably, the multi-dimensional physiological state vector generation unit normalizes and weightedly fuses the first-order derivative and the second-order derivative when generating the real-time state vector.

[0015] Preferably, the vector-pharmacodynamic coupling feedforward regulation unit has a predefined decision mapping area built-in, and the feedforward regulation unit generates the medication management instruction by projecting the current value of the real-time state vector and its change trend into the decision mapping area.

[0016] Compared with the prior art, the present application has the following advantages: 1、Firstly, a multi-modal data stream credibility discriminator is used to identify artifacts and assess the credibility of multi-source physiological parameter data streams obtained from monitoring devices. This assessment is not based on the signal characteristics of a single data stream, but rather on the internal logical consistency between different physiological parameters. Then, a multi-dimensional physiological state vector generation engine processes only the data streams that pass the credibility assessment. By calculating the rate of change and acceleration of each parameter, a real-time state vector is generated that can uniformly represent the direction and rate of the patient's pathophysiological evolution. Finally, a pharmacodynamic coupling feedforward regulator receives the state vector as input for decision-making, generating medication management instructions. This information processing path ensures that the system's decision-making is based on a new information entity that has been logically purified and elevated to represent the process trend, rather than discrete raw data points that may be contaminated. This avoids the two problems that are difficult to reconcile in existing technologies: false judgments due to data artifacts and decision lag due to data isolation.

[0017] 2、A collaborative regulation mechanism is established between different functional modules within the system. During the continuous calibration of the medication regulation model, the individualized pharmacodynamic response feature online self-learning module generates a physiological fluctuation entropy index that quantifies the current patient's physiological stability and drug response predictability. A physiological fluctuation entropy guardian threshold adaptive bus takes this index as input to dynamically adjust the artifact judgment threshold in the aforementioned multi-modal data stream credibility discriminator. In this way, when the patient's state is stable and the physiological fluctuation entropy index is low, the discriminator's judgment threshold is automatically adjusted to filter out meaningless minor fluctuations. When the patient's state tends to be unstable and the index rises, the judgment threshold is automatically lowered to capture any weak signals that may be precursors of disease deterioration. This allows the system's data preprocessing link to adapt to the patient's macro-state, with its monitoring sensitivity optimized according to the actual diagnosis and treatment situation.

[0018] 3、To deal with the internal logic failure of complex software systems in extreme cases, the invention also sets up a multi-level functional safety degradation monitor. This monitor continuously monitors the health status of each functional module in the system, such as whether the adjustment coefficients output by the online self-learning module are within the preset safety range. Once a functional module is diagnosed as failed, the monitor will immediately deprive it of control and instruct the entire system to automatically and orderly switch to a predefined, more basic safety mode, such as from individualized adaptive regulation mode to standard conservative regulation mode, or further to a pure monitoring and alarm mode. This design provides a deterministic and automated failure protection and functional degradation path for the entire information processing system, ensuring that even if the core advanced modules fail, the system's behavior boundaries remain controllable and safe, thus solving the inherent fragility problem of advanced algorithms when applied in high-risk fields. BRIEF DESCRIPTION OF DRAWINGS

[0019] Fig. 1 Flow chart of adaptive adjustment and online learning process of the system of the present application; Fig. 2 Comparison chart of evolutionary trend of state vector for early risk identification of the present application; Fig. 3 Structure chart of functional modules and physical deployment of the present application. DETAILED DESCRIPTION

[0020] The present application will be further described below, and it is obvious that the described embodiments are part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0021] The acute abdominal postoperative drug management system claimed in the present application has an overall information processing architecture including a multi-modal data stream reliability identification unit, a multi-dimensional physiological state vector generation unit, a drug efficacy coupled feedforward adjustment unit, and an individualized drug efficacy response feature online self-learning unit. When the system is running, the multi-source physiological parameter data stream obtained from the monitoring device is first subjected to artifact identification and reliability weighting by the multi-modal data stream reliability identification unit, then the processed data stream is sent to the multi-dimensional physiological state vector generation unit, which generates a real-time state vector representing the disease evolution trend by calculating the first and second derivatives of the data stream. This vector is used as the core input and is transmitted to the drug efficacy coupled feedforward adjustment unit to generate prospective drug management instructions. Finally, the individualized drug efficacy response feature online self-learning unit continuously calibrates the internal model of the feedforward adjustment unit according to the execution effect of the drug administration instructions, thereby forming a closed loop of information processing and decision adjustment. In postoperative monitoring applications, the data stream from the monitoring device often contains non-physiological data artifacts due to patient movement or poor sensor contact, etc. Such artifacts can interfere with subsequent analysis and judgment. To address this challenge, the multi-modal data stream reliability identification unit configured in the present system is configured to receive multiple physiological parameter data streams obtained in parallel from the monitoring device, such as continuous data of heart rate , arterial blood pressure , and central venous pressure , and perform artifact identification based on pre-set logical consistency rules between physiological parameters. One rule is that rapid changes in heart rate are usually accompanied by corresponding fluctuations in blood pressure. According to this rule, when a blood pressure data point at time point exceeds a pre-set change rate threshold (such as ), but the heart rate data stream in the same time window ​ Volatility below the quiescent threshold (e.g.) When ), the unit will If identified as a data artifact, the unit calculates a confidence weight for each data point in real time after identification. Its value is in For an interval, the weight of the data points identified as artifacts is... Set to a lower value, such as For data points that pass the consistency check, their weights are... It is then set to a higher value, such as By executing this procedure, the unit outputs a data set that has been logically cleaned and accompanied by continuous confidence weight information, providing data input for subsequent processing modules.

[0022] After obtaining reliable data, the next technical challenge is how to convert discrete multidimensional parameter measurements into information entities that reflect the overall evolution trend of the disease. Therefore, the system's multidimensional physiological state vector generation unit only receives the aforementioned data stream that has undergone reliability assessment. To characterize the rate of disease evolution, this unit calculates the first derivative of each physiological parameter, such as heart rate. At the point of time first derivative To characterize the acceleration of disease progression, this unit further calculates the second derivatives of each parameter. After obtaining the derivative values ​​of each parameter, to eliminate the influence of differences in dimensions and numerical ranges among different physiological parameters, this unit normalizes all derivative values ​​and maps them to... Within a unified range; ultimately, this unit normalizes these values ​​and applies the confidence weights from the discrimination unit. The multidimensional derivative values ​​after weighted fusion are combined into a unified real-time state vector. The direction of this vector in the multidimensional state space represents the overall trend of improvement or deterioration of the patient's condition, and its magnitude... This characterizes the intensity of the trend; thus, the system transforms discrete data points into a unified, quantitative state vector that represents the evolution and direction of the disease.

[0023] After generating a real-time state vector representing the disease's evolution trend, how to make early intervention decisions based on this information is a problem that the system needs to solve. To this end, the system is equipped with a pharmacodynamic coupling feedforward adjustment unit that receives the real-time state vector. As its core decision input; it has a built-in adjustment model that divides the multi-dimensional state space into decision regions corresponding to different medication management instruction sets; when the unit receives The direction continuously points towards the preset region representing insufficient cycle capacity, and the rate of change of its modulus, i.e. When the value remains above a positive threshold, the unit projects this vector state into the decision region and matches it with a corresponding medication management instruction, such as increasing the current fluid infusion rate by 10% within the next 15 minutes. This adjustment method aims to intervene based on the evolution of the disease, rather than waiting for physiological parameters to exceed limits before responding. Considering the differences in physiological responses to drugs among different patients, a standardized adjustment model is difficult to adapt to all situations. To address this issue, the system also integrates an online self-learning unit for individualized drug efficacy response characteristics. This unit takes the medication management instruction issued by the drug efficacy coupling feedforward adjustment unit as input and transmits the real-time state within a preset time window after the instruction is executed (e.g., within 30 minutes after the instruction is issued) to the system. Actual change As output, by continuously analyzing command-response data pairs, this unit estimates online a response coefficient that quantifies the current patient's sensitivity to a preset drug response. The response coefficient is updated in real time. Feedback-feedback drug-efficacy coupling feedforward regulation unit is used to calibrate its built-in regulation model, so that subsequent medication management instructions are more closely matched with the patient's individualized drug-efficacy response characteristics.

[0024] To enhance system stability and adaptability, this system also includes other functional modules: a physiological stability monitoring module that generates state parameters characterizing the patient's current physiological stability by calculating the prediction error between the predicted state vector generated by the adjustment model and the real-time state vector; and an adaptive threshold adjustment bus that receives these state parameters and dynamically adjusts the judgment threshold used by the multimodal data stream reliability identification unit for artifact recognition based on these parameters. When the state parameter value exceeds a first preset value, it indicates that the patient's state is becoming unstable, and the bus automatically lowers the judgment threshold to improve monitoring sensitivity. Conversely, when the state parameter value is below a second preset value, the judgment threshold is raised. To filter out minor fluctuations, the system also incorporates a multi-level functional safety degradation monitor to ensure operational safety. This monitor continuously monitors whether the response coefficients output by the online self-learning unit are within the preset safety range. When the response coefficients are found to be outside the range, the monitor automatically switches the system's operating mode from the full-function adaptive adjustment mode to a predefined, more basic safety mode. For example, it can switch to the standard feedforward guardian mode, which employs a built-in standardized conservative adjustment strategy, or, in more severe cases, switch to the core monitoring safety mode, which only performs real-time state vector visualization monitoring and alarms. This design provides the system with a deterministic failure protection and functional degradation path.

[0025] In a monitoring scenario for a post-operative patient with acute abdomen, a patient's heart rate, arterial blood pressure, central venous pressure and urine output, which are collected by a monitoring device, are all within the normal threshold range set by the clinic, and no alarm based on single-parameter threshold setting is triggered. In this case, a potential deterioration process evolving from multiple parameters is invisible to the monitoring method relying on instantaneous values. The system receives the above parallel data streams, and the multi-modal data stream credibility identification unit identifies and reduces the credibility weight of the instantaneous false data points caused by the patient's body movement according to the logical consistency rules between physiological parameters , which provides purified data input for subsequent analysis, and this step is the premise of the accuracy of subsequent vector generation. Then, the multi-dimensional physiological state vector generation unit receives the purified data stream and calculates the first and second derivatives of each parameter. The calculation results show that the first derivative of heart rate is a small positive value and its second derivative is also positive, while the first and second derivatives of arterial blood pressure are small negative values. This indicates that the heart rate is slowly accelerating upward, while the blood pressure is slowly accelerating downward. Although the absolute values of each parameter do not reach the alarm limit, the real-time state vector composed of the multi-dimensional derivative values normalized and weighted by the credibility weight has a direction that has been continuously and stably pointing to a predefined area representing circulatory function deterioration in the multi-dimensional state space, and its modulus presents a monotonically increasing trend over time.

[0026] The pharmacodynamic coupling feedforward regulation unit receives the real-time state vector as input and, based on the identification of its direction and the rate of change of its modulus, generates a drug management instruction that suggests a small preventive dose adjustment of the vasoactive drug being used at the time, before any single parameter alarm is triggered. After the drug management instruction is adopted and executed by the clinical staff, the modulus growth trend of the real-time state vector is contained, and its direction gradually deviates from the deterioration area and turns to the area representing stable physiological state. At the same time, the individualized pharmacodynamic response characteristic online self-learning unit records the actual changes after the instruction and the instruction execution to calculate and update the patient's response coefficient This allows for the calibration of the feedforward regulation unit's regulation model. The system provides an early intervention window by identifying multi-parameter co-evolutionary trends, avoiding intervention delays caused by waiting for a single hard physiological indicator to break through a threshold. Furthermore, the synergistic effect of the online self-learning mechanism and the data credibility identification mechanism ensures that the decision-making basis for this prospective intervention is reliable and individualized.

[0027] Example 2: This example conducts a comparative experiment based on retrospective data to verify the performance of the technical solution of the present invention in identifying the trend of physiological deterioration. The experimental data comes from a publicly available anonymous intensive care physiological database, from which 100 data records of patients who meet the criteria for postoperative acute abdominal pain and have circulatory instability are selected. Each record contains continuous physiological parameters such as heart rate and arterial blood pressure.

[0028] The experiment established a control group and an experimental group, both using the same raw data input. The control group used a conventional alarm method based on instantaneous value thresholds, with the alarm rule set to trigger when either the arterial systolic blood pressure was below 90 mmHg or the heart rate was above 120 beats / min. The experimental group used the complete system of this invention, where the intervention command generation threshold of the internal pharmacodynamic coupling feedforward regulation unit was set to the threshold when the real-time state vector... Length of the module Triggered when the data continues to exceed an offline calibrated baseline value for 30 minutes; the sampling period for data processing is set to 1 minute. This period is determined after balancing the real-time nature of data acquisition with the system data processing load, aiming to capture physiological changes at the minute level. In a typical sample data recording and processing process, the operating status of the control group and the experimental group showed differences, as shown in Table 1. Table 1: A comparison of the operational status of the experimental group and the control group on a key physiological data point.

[0029]

[0030] As shown in Table 1, after the start of the experiment, heart rate and systolic blood pressure showed a slow and continuous deterioration trend. At time point 50 minutes, the instantaneous values ​​of heart rate (110 bpm) and systolic blood pressure (96 mmHg) were both within the alarm threshold of the control group, and the control group did not generate an alarm. However, the magnitude of the real-time state vector calculated by the experimental group had increased to 51.5, which met the conditions for generating the intervention command. The alarm of the control group was not triggered until time point 80 minutes, when the heart rate reached 121 bpm.

[0031] The experimental results show that, when processing the sample data, compared with the judgment method based on a single parameter instantaneous value threshold, the experimental group using the technical solution of this invention can generate an intervention signal 30 minutes earlier; statistical analysis of all 100 data records shows that the experimental group generates an intervention signal on average 25.2 minutes earlier (standard deviation 8.1 minutes); this experimental data indicates that by processing multi-source physiological parameter data streams into real-time state vectors characterizing the direction and rate of pathophysiological evolution, the technical solution of this invention can identify systemic risks earlier than the method relying on instantaneous value threshold judgment.

[0032] Example 3: This example combines Figs. 1 to 3 Instructions for a postoperative medication management system for acute abdominal conditions, such as... Fig. 1 As shown, the data first enters the data credibility identification unit. This unit processes the data stream according to the logical consistency rules called from the A1 physiological parameter logical rule base, and outputs the purified and weighted data stream to the state vector generation unit. The state vector generation unit generates a real-time state vector based on this and transmits the vector to the feedforward regulation decision unit on one hand, and to the online learning and calibration unit as a feedback signal on the other hand. The feedforward regulation decision unit solves the received real-time state vector according to the current decision model obtained from the A2 regulation model and decision area, and generates a medication management instruction to output to the clinical staff. At the same time, the instruction is also transmitted to the online learning and calibration unit as input. The online learning and calibration unit integrates the medication management instruction and the subsequent real-time state vector feedback to update the regulation model, and stores the updated regulation model in the A2 regulation model and decision area, thus forming a complete closed loop of information processing and decision optimization.

[0033] like Fig. 2 As shown, the horizontal axis represents time (minutes), the left vertical axis represents heart rate and blood pressure, and the right vertical axis represents the magnitude of the state vector. The curve marked with a solid line and a circle represents heart rate (bpm), which increases from 95 bpm to approximately 123 bpm over time. The curve marked with a dashed line and a diamond represents systolic blood pressure (mmHg), which decreases from 110 mmHg to below 90 mmHg. The curve marked with a dotted line and a triangle represents the real-time magnitude of the state vector generated by the system of this invention. The value shows an accelerating growth trend, increasing from an initial 12.5 dimensionless units to over 75 dimensionless units. The figure clearly shows that although neither the heart rate nor blood pressure parameter touched the traditional alarm threshold at the 50th minute, the magnitude of the state vector representing the co-evolution trend of multiple parameters has increased significantly, thus enabling early identification of risks.

[0034] like Fig. 3As shown, the physiological data stream generated by the bedside monitoring device cluster is aggregated and preliminarily processed by the data acquisition and gateway server, and then sent to the core application server through a real-time data transmission link. The core application server integrates core functional modules such as credibility identification service, state vector generation service, feedforward regulation and decision-making service, and online self-learning service. The server interacts with the database server through model read / write and data storage interfaces. The database server contains a physiological parameter database, regulation model, and patient feature database. The medication instructions and visualization data generated by the core application server are pushed to the clinical terminal, which can take various forms such as a doctor's workstation, a nurse's station large screen, or a mobile monitoring tablet to support clinical decision-making and monitoring.

[0035] Example 4: This example discloses an engineering process for offline calibration of artifact recognition rules in a multimodal data stream credibility identification unit and an initial regulation model in a pharmacodynamic coupling feedforward regulation unit. In the configuration scenario before system deployment, a technical problem is how to provide a set of engineering-based initial parameters for the system's built-in algorithm model to ensure its accuracy during initial system startup. For this purpose, a reproducible offline calibration process is needed. The input to this process is a physiological database containing at least 1000 anonymized historical patient data points, where each data point contains at least 72 hours of time-synchronized multi-source physiological parameter data streams, such as heart rate. With arterial blood pressure Furthermore, the database has been labeled by clinical experts with two types of events: one is signal artifact events caused by non-physiological factors, and the other is events where a confirmed physiological state has significantly deteriorated clinically. The first step in the labeling process is to determine the specific threshold of the logical consistency rule on which the multimodal data stream credibility identification unit bases its artifact recognition. The system traverses all normal physiological data segments labeled as non-artifacts in the database and performs statistical analysis on the rate of change of the two parameters, heart rate and arterial blood pressure, to determine that within a 99.5% confidence interval, when rate of change Greater than At that time, heart rate rate of change Also greater than Based on these statistical results, the system sets these two values ​​as the judgment thresholds in the logical consistency rule. Any data point that violates this association rule will have its credibility weight reduced. It will be set to a low value below 0.2.

[0036] The second step of the calibration procedure is to build the initial regulation model for the pharmacodynamic coupling feedforward regulation unit, i.e. a basic decision-making zone; the system extracts all the events labeled as physiological state deterioration from the database, and intercepts the data segments 60 minutes before these events; then, the system processes these data segments using the multidimensional physiological state vector generation unit to calculate the corresponding real-time state vectors The system determines a high-density distribution area in the multidimensional state space by statistical clustering analysis of the vector set pointing to deterioration events, and defines the direction and range of this area as the initial decision-making zone representing circulatory function deterioration; when the vector generated in real-time monitoring enters this area, the regulation unit will trigger an initial, standardized medication management instruction; by executing this offline calibration procedure, the system's internal key recognition thresholds and decision-making models are determined through statistical analysis of large-scale historical data before it is formally put into real-time operation, and this way changes the parameter setting process from relying on the operator's personal experience to a data-driven repeatable engineering step.

[0037] Embodiment 5: This embodiment aims to supplement the description of the internal operation mechanism of the system's adaptive regulation and safety guarantee involved in the foregoing technical solutions; when the system is applied to a patient with severe physiological state fluctuations, the individualized pharmacodynamic response characteristic online self-learning unit cannot establish a stable response model, resulting in a persistent high prediction error between the predicted state vector generated by the unit and the real-time state vector; at this time, the system's built-in physiological stability monitoring module quantifies this error as a high-value state parameter, and the adaptive threshold adjustment bus automatically reduces the decision threshold used by the multi-modal data stream credibility identification unit when identifying artifacts after receiving this high-value parameter, which makes the system more sensitive to the input data stream during the patient's unstable state to capture any weak signals that may be precursors of disease deterioration.

[0038] If the patient's physiological state further changes into a rare data mode, causing the regulation model inside the individualized pharmacodynamic response characteristic online self-learning unit to have calculation abnormalities and output a response coefficient ; the multi-level functional safety degradation monitor in the system identifies this abnormal output in continuous monitoring, and determines that the self-learning unit is malfunctioning; the monitor automatically deprives the unit of control, and instructs the pharmacodynamic coupling feedforward regulation unit to abandon the use of individualized parameters, and instead switches to a set of built-in, standardized conservative regulation strategies based on large-scale data statistics, while clearly prompting the user interface that the system is currently running in the standard feedforward guardian mode; this orderly functional degradation mechanism aims to constrain the running state of the system when facing internal failures of the core advanced module within a pre-set controllable range.

[0039] Embodiment 6: This embodiment describes a specific calculation procedure for the individualized pharmacodynamic response characteristic online self-learning unit to update the response coefficient online; in the real-time running of the system, in order to enable the decision-making of the pharmacodynamic coupling feedforward regulation unit to continuously adapt to the individualized differences of the patient, an operable procedure is needed to quantitatively update the response coefficient online; the procedure aims to smooth the estimated value of according to the latest medication and response events; after the pharmacodynamic coupling feedforward regulation unit generates and executes a medication management instruction, the individualized pharmacodynamic response characteristic online self-learning unit starts an update calculation; the calculation uses an iterative weighted average method, and its update logic is shown in the following formula: In the formula, is the response coefficient value after this update; is the response coefficient value before the update; is the learning rate factor, which has a value range of (0, 1), and is used to balance the weight of historical data and the latest observation data in the update calculation; a lower value (set to 0.1 in this embodiment) can increase the stability of the coefficient and avoid drastic fluctuations due to a single abnormal response; is the change in the drug dose corresponding to the medication management instruction; is the projection change in the preset time window after the execution of the instruction, which is observed in the real-time state vector in the preset utility direction.

[0040] In a specific update calculation example, the response coefficient before the system update is set to 1.2, and the learning rate factor is 0.1; at this time, the regulation unit issues an instruction to increase the drug dose by one unit; in the observation time window after the execution of the instruction, the system measures the projection change in the preset utility direction of the real-time state vector to be 0.08 units; then the response coefficient after this update is Calculated as ; This updated response coefficient This was then used to calibrate the adjustment model of the feedforward adjustment unit; through this iterative procedure, the system can continuously fine-tune its quantitative understanding of the individual patient's drug efficacy response characteristics based on continuous clinical feedback.

[0041] To further verify the necessity and non-obviousness of the core technical feature of real-time state vector in the technical solution of this invention from the reverse perspective, the following comparative example is set up.

[0042] Comparative Example 1: This comparative example uses 100 sample data from a publicly available intensive care database, the same as in Example 2. It aims to simulate a conventional technical path that is closest in technical logic to what a person skilled in the art can achieve by improving upon existing technology. The only essential difference between this path and the system used in the experimental group of Example 2 of this invention is that it does not synthesize a unified real-time state vector by calculating multidimensional derivatives, but instead adopts a conventional monitoring and early warning mechanism based on multi-parameter independent trend analysis and threshold judgment.

[0043] The system used in Comparative Example 1 was configured to receive the same raw data stream as in Example 2, including continuous heart rate (HR) and systolic blood pressure (BPart). The warning rules consisted of two parts: instantaneous value warning, triggered when the instantaneous BPart value was below 90 mmHg or the instantaneous HR value was above 120 bpm; and trend warning, triggered when the average rate of increase of HR over the past 15 minutes consistently exceeded 0.5 bpm / min, or the average rate of decrease of BPart over the past 15 minutes consistently exceeded 0.8 mmHg / min. This trend threshold was set as an optimal value with engineering rationality, aiming to balance sensitivity and specificity, after statistical analysis of normal physiological fluctuation data from another 500 non-deteriorating events in the database. The system was used to process the typical sample data shown in Table 1 of Example 2. The comparison results of its operating status with the technical solution of the present invention (experimental group) during a typical sample data recording and processing process are shown in Table 2.

[0044] Table 2: Comparison of the operating status of the system in Comparative Example 1 and the technical solution of the present invention in terms of key physiological data.

[0045]

[0046] From the data in Table 2, it can be seen that throughout the entire process from 0 minutes to 70 minutes, although the deteriorating trend of heart rate and arterial systolic pressure objectively exists, since the respective change rates are relatively flat, the 15-minute average change rate calculated by the independent trend analysis module of the comparative example 1 system has never broken through its preset threshold of 0.5 bpm / min or 0.8 mmHg / min. During this period, the independent judgments of heart rate trend acceptable and blood pressure trend acceptable are continuously recorded in the operation log of the system, and the strong correlation risk between the two seemingly acceptable trends indicating the deterioration of circulatory function is not identified until the 80th minute, when the instantaneous measurement value of heart rate (121 bpm) finally breaks through the traditional absolute threshold (120 bpm), the system triggers an alarm for the first time, at which time the best early intervention window has been missed; statistical analysis of all 100 data records shows that the average early warning time of the system of comparative example 1 is only 2.1 minutes (standard deviation 6.5 minutes) earlier than that of the conventional alarm method in example 2 (control group), but it is 23.1 minutes later than that of the technical solution of the present application (experimental group); the results of comparative example 1 show that even if the single-parameter trend analysis is introduced as an improvement means on the basis of the conventional technology, this technical path has essential limitations in identifying the systemic and slow-onset pathophysiological risks driven by multiple factors due to the lack of a mechanism for integrated and collaborative quantitative evaluation of the evolution trends of multiple physiological parameters, and compared with the technical solution of the present application based on real-time state vector for judgment, there is still a delay in decision-making.

[0047] It is apparent for those skilled in the art that the present application is not limited to the details of the foregoing exemplary embodiments, and the present application can be implemented in other concrete forms without departing from the spirit or essential characteristics of the present application.

[0048] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A postoperative medication management system for acute abdominal pain, characterized in that, include: The multimodal data stream credibility identification unit is configured to: receive multi-source physiological parameter data streams obtained from monitoring devices, and perform artifact identification on the data streams according to preset logical consistency rules between at least two physiological parameters, so as to generate a data stream with credibility assessment. The multidimensional physiological state vector generation unit is configured to receive only the data stream with confidence assessment and generate a real-time state vector that uniformly represents the direction and rate of the patient's pathophysiological evolution by calculating the first and second derivatives of the physiological parameters reflected therein. The drug-efficacy coupling feedforward regulation unit has a built-in regulation model. The regulation unit is configured to receive a real-time state vector as a decision input and generate or adjust medication management instructions based on the direction of the real-time state vector and the rate of change of its magnitude. The individualized drug efficacy response feature online self-learning unit is configured to take medication management instructions as input and the actual changes in the real-time state vector within a preset time window after the instructions are executed as output, and continuously calibrate the regulation model of the drug efficacy coupling feedforward regulation unit online.

2. The postoperative medication management system for acute abdomen according to claim 1, characterized in that, The multimodal data stream credibility identification unit is further configured to: calculate a credibility weight for each data stream in real time, and when the credibility weight of a data stream is lower than a preset threshold, the multidimensional physiological state vector generation unit will reduce the weight of that data stream in the weighted fusion calculation or temporarily isolate it when generating the real-time state vector.

3. The postoperative medication management system for acute abdomen according to claim 1, characterized in that, The online self-learning unit for individualized drug efficacy response characteristics is further configured to: estimate a response coefficient in real time based on the actual changes in medication management instructions and subsequent real-time state vectors, and use this response coefficient as the basis for calibrating and adjusting the model.

4. The postoperative medication management system for acute abdomen according to claim 1, characterized in that, Also includes: A physiological stability monitoring module is configured to generate a state parameter characterizing the patient's current physiological stability by calculating the prediction error between the predicted state vector generated by the conditioning model and the real-time state vector; and an adaptive threshold adjustment bus is configured to receive the state parameter and dynamically and nonlinearly adjust the judgment threshold used by the multimodal data stream credibility discrimination unit for artifact recognition based on the parameter.

5. The postoperative medication management system for acute abdomen according to claim 4, characterized in that, The adaptive threshold adjustment bus is configured to automatically lower the judgment threshold when the value of the status parameter exceeds the first preset value, and automatically raise the judgment threshold when the value of the status parameter is lower than the second preset value.

6. The postoperative medication management system for acute abdomen according to claim 1, characterized in that, Also includes: A multi-level functional safety degradation monitor is configured to continuously monitor whether the response coefficient output by the online self-learning unit of individualized drug efficacy response characteristics is within a preset absolute safety range, and when the response coefficient is diagnosed as exceeding the absolute safety range and the unit is determined to have failed, automatically and orderly switch the operating mode of the entire system from a full-function adaptive adjustment mode to a predefined safety mode.

7. The postoperative medication management system for acute abdomen according to claim 6, characterized in that, The predefined, more basic security modes include at least one of the following two modes: Standard Feedforward Guardian Mode, in which the monitor deprives the online self-learning unit of control and instructs the feedforward adjustment unit to switch to a set of built-in conservative adjustment strategies; or Core Monitor Security Mode, in which the monitor deprives the automatic adjustment function, causing the system to switch to only perform real-time state vector visualization monitoring and alerts.

8. The postoperative medication management system for acute abdomen according to claim 1, characterized in that, The real-time state vector is a vector defined by its direction and magnitude in a multidimensional state space, where the direction represents the overall trend of the patient's condition improving, deteriorating, or stabilizing, and the magnitude represents the intensity of the trend.

9. A postoperative medication management system for acute abdomen according to claim 1, characterized in that, When generating real-time state vectors, the multidimensional physiological state vector generation unit normalizes and weights the first and second derivatives.

10. A postoperative medication management system for acute abdomen according to claim 1, characterized in that, The vector-pharmacokinetics coupling feedforward adjustment unit has a predefined decision mapping area built in. The feedforward adjustment unit generates medication management instructions by projecting the current value of the real-time state vector and its changing trend onto the decision mapping area.

Images