General surgery department patient postoperative complication risk prediction and nursing decision-making system

Abstract

The invention relates to the technical field of medical informatization and intelligent nursing, in particular to a general surgery department patient postoperative complication risk prediction and nursing decision-making system, which comprises a baseline manifold construction module for initializing an individualized steady state manifold in a preset multi-dimensional state space based on historical operation parameters and basic state data; the perturbation response topology evaluation module is used for calculating a recovery attenuation coefficient after the real-time phase trajectory deviates from the individualized steady state manifold so as to generate a steady state dissipation index and a growth rate; the offset course quantity decoding module is used for adjusting a dynamic threshold value based on a time delay parameter and an embedded dimension parameter on the basis of the basic state data; the risk prediction module is used for weighting, merging and normalizing the growth rate of the steady-state dissipation index and the modulus length of the course deviation vector to generate a risk prediction evaluation result; the decision closed-loop module is used as a secondary perturbation variable and sends the secondary perturbation variable to the perturbation response topology evaluation module for continuous monitoring; according to the invention, the baseline anchoring precision in different clinical scenes is improved.

Description

Technical Field

[0001] This invention relates to the fields of medical information technology and smart nursing technology, specifically to a system for predicting the risk of postoperative complications and making nursing decisions for general surgery patients. Background Technology

[0002] With the continuous development of clinical intensive care services, the risk prediction and nursing decisions for postoperative complications in general surgery patients have become increasingly complex. This complexity presents significant challenges in handling high-frequency physiological data streams and occasional nursing intervention events. Currently, clinical monitoring generally relies on traditional monitors to collect single-dimensional physiological signals and assess the patient's condition by setting fixed absolute physiological index thresholds. Medical staff make judgments about the patient's condition and nursing intervention decisions based on these absolute values.

[0003] However, fixed thresholds in traditional monitoring methods are prone to causing baseline drift when dealing with elderly or critically ill patients; single-dimensional signals often mask the potential worsening trend of disease cascade mechanisms; in addition, absolute physiological indicators have a significant time lag effect, resulting in a dynamic mismatch between high-frequency physiological data streams and occasional nursing intervention events in the spatiotemporal dimension, making it difficult to quantify the patient's physiological compensatory exhaustion process in advance. Therefore, how to effectively overcome baseline drift and solve dynamic mismatch, so as to accurately predict the risk of postoperative complications and provide early warning, has become an urgent problem to be solved in this field. Summary of the Invention

[0004] The objective of this invention can be achieved through the following technical solutions: A risk prediction and nursing decision-making system for postoperative complications in general surgery patients, including: The data acquisition and interface module collects the time-series operational data stream of the patients to be tested, intervention data including intervention events and timestamps, and historical operational parameters and basic status data obtained from the hospital information system; The baseline manifold construction module initializes an individualized steady-state manifold in a preset multidimensional state space based on the historical operating parameters and basic state data. The perturbation response topology evaluation module uses the delayed coordinate method to map the time-series running data stream into a real-time phase trajectory; it treats the intervention event as a perturbation variable and calculates the recovery decay coefficient after the real-time phase trajectory deviates from the individualized steady-state manifold to generate the steady-state dissipation index and growth rate. The yaw vector decoding module calculates and extracts the magnitude and direction of the yaw vector of the real-time phase trajectory if the steady-state dissipation index is greater than the dynamic threshold, and maps it to a preset abnormal dynamic subspace; otherwise, it sets the yaw vector and magnitude to zero; the dynamic threshold is adjusted based on the basic state data. The risk prediction module weights and normalizes the growth rate of the steady-state dissipation index and the magnitude of the yaw vector to calculate the probability of postoperative complications occurring within a preset time window and generates a risk prediction assessment result. The decision-making closed-loop module, based on the risk prediction and assessment results and the yaw vector direction and magnitude, outputs a decision instruction containing diagnostic intervention actions, and sends it as a secondary perturbation variable to the perturbation response topology assessment module for continuous monitoring.

[0005] Preferably, the baseline manifold building block includes: The parameter parsing unit is used to parse the historical operating parameters of the patient under test in order to extract system stress characteristics; The manifold initialization unit is used to construct a safe steady-state manifold with attraction basin characteristics in the multidimensional state space based on the system stress characteristics and the basic state data, according to a preset medical feature mapping rule, and to use the safe steady-state manifold as the individualized steady-state manifold.

[0006] Preferably, the perturbation response topology evaluation module uses the delayed coordinate method to map the time-series running data stream into a real-time phase trajectory in the multi-dimensional state space, including: Obtain the time delay parameter and embedding dimension parameter of the time-series running data stream; Based on the time delay parameter and the embedding dimension parameter, the phase space of the time-series running data stream is reconstructed to generate a high-dimensional state vector; The high-dimensional state vectors from the continuous time series are concatenated to generate the real-time phase trajectory.

[0007] Preferably, the perturbation response topology evaluation module calculates the recovery decay coefficient after the real-time phase trajectory deviates from the individualized steady-state manifold to generate a steady-state dissipation index, including: Extract the current state point corresponding to the real-time phase trajectory at the current moment, and calculate the curvature tensor of the current state point on the individualized steady-state manifold; By combining the curvature tensor and the perturbation variable, the Lyapunov exponent for the regression of the real-time phase trajectory to the equilibrium point of the individualized steady-state manifold is calculated; The Lyapunov exponent is used as the recovery decay coefficient, and the reciprocal of the recovery decay coefficient is used as the steady-state dissipation exponent.

[0008] Preferably, the yaw vector decoding module calculates the yaw vector of the real-time phase trajectory in the multi-dimensional state space, and extracts the magnitude and direction of the yaw vector, including: Extract the directional drift direction and drift distance of the real-time phase trajectory deviating from the individualized steady-state manifold; The directional drift direction is taken as the direction of the yaw vector; The drift distance is used as the magnitude of the yaw vector.

[0009] Preferably, the predefined abnormal dynamics subspace includes a low-capacity subspace, an infected subspace, and a blocking subspace; The yaw vector decoding module maps the yaw vector to a preset abnormal dynamics subspace, including: Calculate the spatial angle between the yaw vector and the low-capacity subspace, the infected subspace, and the blocked subspace; The subspace corresponding to the smallest spatial angle is selected as the target anomaly dynamics subspace, and the yaw vector is mapped to the target anomaly dynamics subspace.

[0010] Preferably, the decision-making closed-loop module, based on the risk prediction and assessment results and the direction and magnitude of the yaw vector, outputs decision instructions containing diagnostic intervention actions, including: Based on the direction of the yaw vector, the target anomaly type is determined; Based on the magnitude of the yaw vector, the degree of compensatory exhaustion of the target anomaly type is determined; Based on the target anomaly type, the degree of compensatory exhaustion, and the risk prediction and assessment results, the corresponding diagnostic intervention action is matched from the preset intervention strategy mapping table, and the decision instruction is generated.

[0011] Preferably, it also includes edge computing nodes and a central server; The perturbation response topology evaluation module is deployed on the edge computing node, which is used to perform the mapping of the real-time phase trajectory and the calculation of the steady-state dissipation index locally. The yaw vector decoding module and the decision closed-loop module are deployed on the central server. The central server is used to receive the steady-state dissipation index, the growth rate of the steady-state dissipation index and the yaw vector transmitted by the edge computing node, and to issue the decision command.

[0012] Preferably, the historical operating parameters of the patient to be tested include surgical record data, the basic status data includes preoperative physiological index data, the time-series operating data stream includes continuous vital sign waveform data, and the intervention events include nursing operation data.

[0013] The beneficial effects of this invention are: 1) This invention solves the problem of spatiotemporal dynamic mismatch between high-frequency physiological data and occasional nursing interventions by replacing the traditional absolute numerical comparison with the calculation of manifold topological dimensions; by using the Lyapunov index to establish a steady-state dissipation index, it can quantify the physiological compensation exhaustion process in advance before the patient's vital signs deviate substantially; this mechanism overcomes the time lag effect of traditional indicators and provides medical staff with a critical intervention window. 2) This invention uses a baseline manifold construction module to initialize an individualized steady-state manifold in situ in a multidimensional space, taking into account the patient's preoperative baseline condition and surgical stress characteristics. This method solves the baseline drift problem caused by fixed thresholds in traditional monitors when dealing with elderly or critically ill patients. By establishing an individualized physiological reference coordinate system, the system can accurately eliminate acquisition artifacts and routine physiological fluctuations, improving the baseline anchoring accuracy in different clinical scenarios. 3) This invention uses the delayed coordinate method to perform phase space reconstruction and extract the autonomic nervous system imbalance features that are masked by conventional fluctuations; the real-time state is mapped to a specific abnormal dynamic subspace through the yaw vector decoding module, which effectively avoids the diagnostic ambiguity caused by single modulus evaluation; this enables the system to output pathologically oriented classification results and significantly optimizes the triage efficiency during multidisciplinary consultations. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a schematic diagram of the module of the postoperative complication risk prediction and nursing decision-making system for general surgery patients provided in the embodiments of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 The logic of the data acquisition and interface module of the general surgery patient postoperative complication risk prediction and nursing decision system aims to solve the problem of dynamic mismatch between high-frequency physiological data flow and occasional nursing intervention events in the spatiotemporal dimension in the clinical intensive care business scenario; The system acquires high-frequency time-series operational data streams and intervention data of the patients under test, and after acquiring historical operational parameters and baseline state data, it starts the baseline manifold construction module; This module initializes an individualized steady-state manifold in a multidimensional state space to establish a physiological baseline coordinate system for a specific patient under postoperative stress. The perturbation response topology assessment module receives the above data stream, uses the delayed coordinate method to map the one-dimensional sequence to a real-time phase trajectory in a multidimensional state space, and takes the intervention event as a perturbation variable input to calculate the recovery decay coefficient after the real-time phase trajectory deviates from the individualized steady-state manifold, generating a steady-state dissipation index and its growth rate. Specifically, the calculation logic for the growth rate of the steady-state dissipation index is as follows: obtain the current steady-state dissipation index at the end of the current observation time window. Subtract the base period steady-state dissipation index at the beginning of the window. The difference is then divided by the time span of the observation window. This allows us to quantify the rate of deterioration of the dissipative state and thus obtain the growth rate of the steady-state dissipation index. The corresponding calculation formula is: When dealing with concurrent alarms from multiple beds in a ward, the yaw vector decoding module dynamically adjusts the threshold based on the basic status data; The adjustment logic of the dynamic threshold is as follows: extract the age characteristics and preoperative baseline heart rate characteristics from the baseline data, and use the preset population standard dissipation index as a benchmark. When the age is greater than 65 years or the baseline heart rate deviates from the normal range, the benchmark is adjusted downward according to the preset proportional coefficient to generate the dynamic threshold. The specific rules for setting the preset proportional coefficient are as follows: with the patient's age of 65 years as the baseline, the adjustment coefficient increases by 1% for each year of age increase; with a baseline heart rate of 60-100 beats / minute as the normal range, the adjustment coefficient increases by 2% for every 10 beats / minute that the heart rate exceeds the upper limit of the normal range or falls below the lower limit; the overall downward adjustment correction shall not exceed 30% of the baseline. When the steady-state dissipation index exceeds the dynamic threshold, the magnitude and direction of the yaw vector of the real-time phase trajectory are extracted and mapped to the preset abnormal dynamic subspace for classification. If the threshold is not exceeded, the yaw vector is forced to be a zero vector, thereby filtering out invalid alarm triggers at the underlying computing power allocation. The risk prediction module integrates the growth rate of the steady-state dissipation index with the magnitude of the yaw vector to output the probability assessment result of postoperative complications in the patient under test within a preset time window. Specifically, it calculates the comprehensive risk scalar after weighted summation by weighting the growth rate of the steady-state dissipation index with the magnitude of the yaw vector and mapping it through a normalization function. The weighted summation is then converted into a probability value of 0 to 100%. Specifically, let the growth rate of the steady-state dissipation exponent be... yaw vector The modulus is The comprehensive risk scalar is The system is configured with a preset reference module constant. In this embodiment, a dimensionless division operation is performed on the magnitude of the yaw vector. The value is taken as the 95th quantile of the magnitude of the phase trajectory yaw vector of the healthy control group samples; prior weights are configured. and satisfy Calculate the comprehensive risk scalar The formula is: Input it as a natural constant The base-1 Sigmoid function is used to convert the probability of postoperative complications occurring in the patient within a pre-defined time window. The calculation incorporates a risk decision midpoint constant derived from historical complication case statistics. Preferably, The value ranges from 0.4 to 0.6, representing the population average risk scalar threshold when the complication rate reaches 50%. Its calculation formula is as follows: Based on the evaluation results and yaw vector characteristics, the final decision-making closed-loop module outputs a decision instruction containing diagnostic intervention actions, and feeds this intervention action as a secondary perturbation variable back to the perturbation response topology evaluation module to form a closed-loop monitoring. This embodiment demonstrates the system's responsiveness in complex nursing scenarios involving multiple tasks. By replacing absolute numerical comparisons with calculations of manifold topological dimensions, it effectively suppresses interference from acquisition artifacts and routine physiological fluctuations, verifying the robustness of this technical solution under different complication evolution paths.

[0018] In a preferred embodiment of the present invention, this embodiment is a further specification of the internal structure of the baseline manifold construction module in the above-mentioned general surgery patient postoperative complication risk prediction and nursing decision-making system; In the scenario of electronic medical record information flow when patients are transferred to the ward after surgery, the parameter parsing unit is configured to perform dimensionality reduction extraction on unstructured historical records and parse the historical operating parameters of the patient to be tested in order to quantitatively extract the system stress characteristics. The manifold initialization unit then intervenes, and based on the system stress characteristics and basic state data, constructs a safe steady-state manifold with local convergence characteristics in situ in the multidimensional state space according to the preset medical feature mapping rules, and establishes this safe steady-state manifold as an individualized steady-state manifold. Specifically, the preset medical feature mapping rule is as follows: extract the mean of stable physiological indicators from the baseline state data as the origin of the multidimensional state space; analyze the trauma score in the system stress features as... The indicators of blood loss are The radius of the attraction basin boundary of the manifold is calculated using a preset negative exponential decay function. The specific calculation formula is as follows: in, In one specific embodiment, the maximum boundary radius of the preset baseline state is... The value is 10.0; and These are the trauma attenuation coefficient and the blood loss attenuation coefficient, respectively, defined based on historical statistical data. For example... 0.05 is acceptable. 0.002 is acceptable; The radius of the attraction basin boundary was calculated quantitatively based on this. To constrain the space, a hypersphere with a quadratic potential energy function is constructed with the origin of the coordinate system as the center, serving as a safe and steady-state manifold with local convergence characteristics; The potential energy function of this safe steady-state manifold reaches a minimum at the origin, representing the most stable physiological state; an appropriate manifold curvature setting helps the system establish reliable reference coordinates when there are large individual differences among patients. This embodiment constructs a safe steady-state manifold by extracting system stress features, which overcomes the baseline drift problem caused by fixed thresholds in traditional monitors when dealing with elderly or critically ill patients, and improves the baseline anchoring accuracy of the system in cross-departmental transfer scenarios.

[0019] In a preferred embodiment of the present invention, this embodiment discloses in detail the specific mathematical implementation steps of the phase trajectory mapping performed by the perturbation response topology evaluation module in the general surgery patient postoperative complication risk prediction and nursing decision system; in the scenario of buffering high-frequency sampled waveform data of bedside monitors, single-dimensional physiological signals often mask the potential cascading mechanism deterioration trend; To this end, the system obtains the time delay parameters and embedding dimension parameters of the time-series running data stream, and performs phase space reconstruction on the time-series running data stream; specifically, the system uses historical observation data and the current time... Specific single-dimensional sampling scalar value ,in Relative to time Backward delay delay parameters The sampled scalar value, and Delay parameters and embedding dimension A high-dimensional state vector is generated using a backward delay reconstruction formula. : The system performs temporal concatenation of high-dimensional state vectors in continuous time series to generate real-time phase trajectories in multi-dimensional state space. This reconstruction process can extract the autonomic nervous system imbalance features that are masked by conventional heart rate fluctuations, reduce the interference of missing single-point data on the overall assessment, and demonstrate the robustness of the topology mapping algorithm in dealing with clinical data loss scenarios.

[0020] In a preferred embodiment of the present invention, this embodiment reveals the core calculation logic of the steady-state dissipation index in the risk prediction and nursing decision-making system for postoperative complications in general surgery patients; in the emergency dispatching of medical resources to deal with sudden stress reactions in critically ill patients, traditional absolute physiological indicators often have obvious time lag effects; The perturbation response topology evaluation module extracts the current state point corresponding to the real-time phase trajectory at the current moment, uses the central difference method to extract the multidimensional coordinate data in the neighborhood of the current state point, constructs a local Hessian matrix and solves its eigenvalues, and uses the diagonal matrix formed by the eigenvalues ​​as the curvature tensor of the current state point on the individualized steady-state manifold to evaluate the local topological steepness of the current physiological state. By combining the curvature tensor with the input perturbation variables, the system calculates the Lyapunov exponent for the regression of real-time phase trajectories to individualized steady-state manifold equilibrium points; Before calculation, the system converts the input perturbation variable, i.e. the intervention event, into a scalar amplitude in a multidimensional state space according to a preset intervention intensity equivalent table. The intervention intensity equivalent scale contains mapping rules between discrete clinical intervention actions and scalar amplitudes. It assigns numerical values ​​to the volume or concentration of the physical impact of the clinical intervention action on the patient's hemodynamics or metabolic load, and extracts them as scalar amplitudes. For example, intravenous injection of vasoactive drugs is mapped to a scalar amplitude of 5.0, rapid rehydration of 250ml of crystalloid solution is mapped to a scalar amplitude of 2.5, and adjusting the positive end-expiratory pressure of the ventilator to increase 2cmH2O is mapped to a scalar amplitude of 1.5. The system combines the direction of the normal vector of the current state point on the individualized steady-state manifold and maps the scalar magnitude to an initial perturbation vector in the multidimensional state space. ; To accommodate the limitations of the limited observation time window in clinical real-time monitoring, the system adopts a preset limited evolution time window. Approximate calculations are performed by replacing the infinity limit condition with the natural constant. logarithmic function with base Calculate the evolution perturbation vector at the end of the evolution time window. With the initial disturbance vector Through curvature tensor The Lyapunov exponent, with dimensions of the reciprocal of time, is obtained from the ratio of the modulus after linear mapping. The specific calculation model is as follows: In this formula, the operators Represents the curvature tensor The linear mapping effect of matrix multiplication on the perturbation vector in each direction of the multidimensional state space; the evolution of the perturbation vector. The logic for obtaining the evolution time window is as follows: Since the system state is represented by the discrete reconstructed phase trajectory, the system adopts the local linear fitting method, selects several historical state points in the neighborhood of the current state point in the multidimensional state space, and fits the local linear evolution mapping matrix by the least squares method, which is used as an approximation of the local Jacobian matrix of the system dynamic evolution equation at the current state point. Based on this approximate Jacobian matrix, a discrete linearized variational equation is constructed, and the initial perturbation vector is... Substituting into the variational equation, the iterative evolution with the discrete time step is obtained. The window length is used to obtain the evolution perturbation vector at the end of the evolution time window. ; The system establishes the calculated Lyapunov exponent as the recovery decay coefficient and defines its reciprocal as the steady-state dissipation exponent. The appropriate perturbation vector capture mechanism helps the system quantify the physiological compensation exhaustion process in advance before the patient's vital signs deviate substantially, thus gaining a critical intervention window for medical staff and demonstrating the high sensitivity of the calculation model in early warning applications.

[0021] In a preferred embodiment of the present invention, this embodiment illustrates the cascade mechanism by which the yaw vector decoding module extracts the yaw vector and maps it in the abnormal dynamics subspace in the yaw vector risk prediction and nursing decision-making system for general surgery patients; in the scenario of electronic triage of suspected cases before multidisciplinary joint consultation, accurate initial classification can significantly optimize the intervention efficiency of specialist doctors. The yaw vector decoding module extracts the directional drift direction and drift distance of the real-time phase trajectory deviating from the individualized steady-state manifold, and establishes them as the direction and magnitude of the yaw vector, respectively. In terms of specific calculation logic, the system obtains the current state point on the real-time phase trajectory and calculates the orthogonal projection point of the current state point on the individualized steady-state manifold. Construct a difference vector pointing from the orthogonal projection point to the current state point; the system extracts the unit direction vector of the difference vector as the directional drift direction, and extracts the Euclidean norm of the difference vector as the drift distance; The system internally pre-defines anomaly dynamics subspace matrix comprising a low-capacity subspace, an infection subspace, and an obstruction subspace. These subspaces are constructed from orthogonal basis matrices extracted by principal component analysis of pre-collected characteristic vectors of typical phase trajectories of various complications. Let the first... The orthogonal basis matrix corresponding to the anomalous dynamic subspace is: ,in These correspond to the low-capacity subspace, the infected subspace, and the blocked subspace, respectively. Let be the transpose of this orthogonal basis matrix; The module calculates the spatial angle between the yaw vector and the three subspaces mentioned above, i.e., using the inverse cosine function. Calculate the yaw vector In the Orthogonal projection vectors on anomalous dynamic subspaces The modulus and the original yaw vector The ratio of the modulus length is used to obtain the yaw vector and the first... The spatial angle between the anomalous dynamic subspaces The specific calculation formula is as follows: The subspace corresponding to the smallest spatial angle is selected as the target abnormal dynamic subspace. The yaw vector is mapped to this target abnormal dynamic subspace, and the projection component is extracted as a specific pathological feature parameter. This mapping mechanism avoids the diagnostic ambiguity caused by single modulus evaluation, enabling the system to directly output pathologically oriented classification results, and verifying the robustness of this technical solution in the differential diagnosis process of complex complications.

[0022] In a preferred embodiment of the present invention, this embodiment discloses the logic of the decision-making closed-loop module generating specific executable instructions in the risk prediction and nursing decision-making system for postoperative complications of general surgery patients; in the scenario of automatic distribution of standardized nursing pathways in ward nursing workstations, the system needs to transform abstract risk probabilities into nursing actions that nurses can directly perform; The decision-making closed-loop module determines the target abnormality type based on the direction of the yaw vector and determines the degree of compensatory exhaustion of the target abnormality type based on the magnitude of the yaw vector. Specifically, the system pre-divides the magnitude of the yaw vector into three intervals of mild, moderate and severe based on the physiological deterioration trajectory of historical patients under each target abnormality type using the quantile method and sets corresponding distance thresholds. When the magnitude of the yaw vector calculated in real time falls into a specific range, it is determined to be the corresponding degree of compensatory exhaustion. The module integrates the target anomaly type, the degree of compensatory exhaustion and the risk prediction and assessment results, matches the corresponding diagnostic intervention action from the preset intervention strategy mapping table, and generates decision instructions. The intervention strategy mapping table adopts a multidimensional lookup table structure. For example, when the target abnormality type is determined to be low volume state, the degree of compensation exhaustion reaches the moderate threshold, and the risk prediction assessment result shows high risk, the system directly queries the mapping table and locks the corresponding action of rapid fluid resuscitation with 250ml crystalloid solution and monitoring central venous pressure as a diagnostic intervention action. During this period, by introducing diagnostic intervention actions as secondary perturbation variables, the system constructed a dynamic verification loop in situ with the patient's physiological state. Although this closed-loop design increased the computational load on the edge side, it effectively reduced the uncertainty risk brought about by empirical treatment through dynamic verification and improved the scientific nature of overall nursing decisions.

[0023] In a preferred embodiment of the present invention, this embodiment clarifies the physical deployment architecture and data processing scope of the postoperative complication risk prediction and nursing decision-making system for general surgery patients; in the scenario of bandwidth-constrained medical IoT with hundreds of beds in the entire ward of a large tertiary hospital accessing the network concurrently, centralized data uploading often causes severe network congestion and dynamic mismatch. To this end, the system is configured with edge computing nodes and a central server. The perturbation response topology evaluation module is deployed on the edge computing nodes to perform real-time phase trajectory mapping and steady-state dissipation index calculation locally. The yaw vector decoding module and the decision closed-loop module are deployed on the central server to receive the steady-state dissipation index, the growth rate of the steady-state dissipation index and the yaw vector transmitted by the edge computing nodes, and issue decision instructions. It should be noted that, in order to ensure the integrity and logical consistency of data flow under the distributed architecture, the data acquisition and interface module and the baseline manifold construction module are also deployed on edge computing nodes so as to directly acquire and process high-frequency data from bedside devices locally; while the risk prediction module is deployed on the central server, working in conjunction with the decision-making closed-loop module to complete the probability calculation for high computing power requirements. Regarding the logic of edge computing nodes transmitting yaw vectors to the central server, in actual deployment, after the edge computing node calculates the original coordinate offset of the real-time phase trajectory deviating from the steady-state manifold locally, it transmits it as the initial yaw vector data packet to the central server. The yaw vector decoding module deployed on the central server receives the yaw vector data packet and performs a complete decoding analysis operation to extract the modulus and direction and map the abnormal dynamic subspace. In terms of data processing, the historical operating parameters of the patients under test specifically include surgical record data, basic status data includes preoperative physiological index data, time-series operating data stream includes continuous vital sign waveform data, and intervention events include nursing operation data. This asymmetric computing power distribution architecture enables high-frequency waveform data to be reduced in dimensionality at the edge, and only low-frequency scalars and coordinate features are transmitted to the center, effectively avoiding the input-output blocking problem when monitoring concurrently on a local area network, and verifying the robustness of the system in a large-scale deployment environment.

[0024] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A risk prediction and nursing decision-making system for postoperative complications in general surgery patients, characterized in that, include: The data acquisition and interface module collects the time-series operational data stream of the patients to be tested, intervention data including intervention events and timestamps, and historical operational parameters and basic status data obtained from the hospital information system; The baseline manifold construction module initializes an individualized steady-state manifold in a preset multidimensional state space based on the historical operating parameters and basic state data. The perturbation response topology evaluation module uses the delayed coordinate method to map the time-series running data stream into a real-time phase trajectory; it treats the intervention event as a perturbation variable and calculates the recovery decay coefficient after the real-time phase trajectory deviates from the individualized steady-state manifold to generate the steady-state dissipation index and growth rate. The yaw vector decoding module calculates and extracts the magnitude and direction of the yaw vector of the real-time phase trajectory if the steady-state dissipation index is greater than the dynamic threshold, and maps it to a preset abnormal dynamic subspace. If not, set the yaw vector and magnitude to zero; the dynamic threshold is adjusted based on the basic state data; The risk prediction module weights and normalizes the growth rate of the steady-state dissipation index and the magnitude of the yaw vector to calculate the probability of postoperative complications occurring within a preset time window and generates a risk prediction assessment result. The decision-making closed-loop module, based on the risk prediction and assessment results and the yaw vector direction and magnitude, outputs a decision instruction containing diagnostic intervention actions, and sends it as a secondary perturbation variable to the perturbation response topology assessment module for continuous monitoring.

2. The postoperative complication risk prediction and nursing decision-making system for general surgery patients according to claim 1, characterized in that, The baseline manifold construction module includes: The parameter parsing unit is used to parse the historical operating parameters of the patient under test in order to extract system stress characteristics; The manifold initialization unit is used to construct a safe steady-state manifold with attraction basin characteristics in the multidimensional state space based on the system stress characteristics and the basic state data, according to a preset medical feature mapping rule, and to use the safe steady-state manifold as the individualized steady-state manifold.

3. The postoperative complication risk prediction and nursing decision-making system for general surgery patients according to claim 1, characterized in that, The perturbation response topology evaluation module uses the delayed coordinate method to map the time-series running data stream into a real-time phase trajectory in the multi-dimensional state space, including: Obtain the time delay parameter and embedding dimension parameter of the time-series running data stream; Based on the time delay parameter and the embedding dimension parameter, the phase space of the time-series running data stream is reconstructed to generate a high-dimensional state vector; The high-dimensional state vectors from the continuous time series are concatenated to generate the real-time phase trajectory.

4. The postoperative complication risk prediction and nursing decision-making system for general surgery patients according to claim 1, characterized in that, The perturbation response topology evaluation module calculates the recovery decay coefficient after the real-time phase trajectory deviates from the individualized steady-state manifold, in order to generate a steady-state dissipation index, including: Extract the current state point corresponding to the real-time phase trajectory at the current moment, and calculate the curvature tensor of the current state point on the individualized steady-state manifold; By combining the curvature tensor and the perturbation variable, the Lyapunov exponent for the regression of the real-time phase trajectory to the equilibrium point of the individualized steady-state manifold is calculated; The Lyapunov exponent is used as the recovery decay coefficient, and the reciprocal of the recovery decay coefficient is used as the steady-state dissipation exponent.

5. The postoperative complication risk prediction and nursing decision-making system for general surgery patients according to claim 1, characterized in that, The yaw vector decoding module calculates the yaw vector of the real-time phase trajectory in the multi-dimensional state space, and extracts the magnitude and direction of the yaw vector, including: Extract the directional drift direction and drift distance of the real-time phase trajectory deviating from the individualized steady-state manifold; The directional drift direction is taken as the direction of the yaw vector; The drift distance is used as the magnitude of the yaw vector.

6. The postoperative complication risk prediction and nursing decision-making system for general surgery patients according to claim 1, characterized in that, The preset abnormal dynamics subspace includes a low-capacity subspace, an infection subspace, and a blocking subspace; The yaw vector decoding module maps the yaw vector to a preset abnormal dynamics subspace, including: Calculate the spatial angle between the yaw vector and the low-capacity subspace, the infected subspace, and the blocked subspace; The subspace corresponding to the smallest spatial angle is selected as the target anomaly dynamics subspace, and the yaw vector is mapped to the target anomaly dynamics subspace.

7. The postoperative complication risk prediction and nursing decision-making system for general surgery patients according to claim 1, characterized in that, The decision-making closed-loop module, based on the risk prediction and assessment results and the direction and magnitude of the yaw vector, outputs decision instructions containing diagnostic intervention actions, including: Based on the direction of the yaw vector, the target anomaly type is determined; Based on the magnitude of the yaw vector, the degree of compensatory exhaustion of the target anomaly type is determined; Based on the target anomaly type, the degree of compensatory exhaustion, and the risk prediction and assessment results, the corresponding diagnostic intervention action is matched from the preset intervention strategy mapping table, and the decision instruction is generated.

8. The postoperative complication risk prediction and nursing decision-making system for general surgery patients according to claim 1, characterized in that, It also includes edge computing nodes and a central server; The perturbation response topology evaluation module is deployed on the edge computing node, which is used to perform the mapping of the real-time phase trajectory and the calculation of the steady-state dissipation index locally. The yaw vector decoding module and the decision closed-loop module are deployed on the central server. The central server is used to receive the steady-state dissipation index, the growth rate of the steady-state dissipation index and the yaw vector transmitted by the edge computing node, and to issue the decision command.

9. The postoperative complication risk prediction and nursing decision-making system for general surgery patients according to claim 1, characterized in that, The historical operating parameters of the patient to be tested include surgical record data, the basic status data includes preoperative physiological index data, the time-series operating data stream includes continuous vital sign waveform data, and the intervention events include nursing operation data.

Images