Postoperative infection monitoring and early warning method and system

Abstract

The invention discloses a postoperative infection monitoring and early warning method and system, and relates to the technical field of clinical monitoring, and the method comprises the steps: collecting a postoperative infection multi-source alignment data set, building an individual reference model, dynamically extracting postoperative change features, building a behavioral physiological coupling model, recognizing abnormal interaction, and quantifying the infection risk. According to the method, multi-source data are fused, an individualized dynamic baseline model and a behavior-physiological coupling model are established, intelligent recognition and grading early warning of postoperative infection risks are achieved, and the method is high in practicability and easy to popularize. Model parameters are automatically corrected by utilizing a self-learning and feedback recharge mechanism, a closed-loop process of monitoring, prediction, intervention and optimization is formed, infection symptoms can be recognized in advance, manual dependence is reduced, the real-time performance and accuracy of postoperative infection prevention and control are improved, and intelligent decision support is provided for clinic.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of clinical monitoring, in particular to a postoperative infection monitoring and early warning method and system. BACKGROUND

[0002] Postoperative infection is one of the most common and highest-risk complications in clinical treatment, especially in large-scale surgical operations, elderly patients and immunocompromised populations. Infection prevention and control directly affects the recovery speed and medical quality. The existing postoperative infection monitoring mainly relies on laboratory test results and manual observation by medical staff. Therefore, there is an urgent need for a technical solution that can integrate multi-source heterogeneous data to achieve early intelligent identification, dynamic early warning and closed-loop intervention feedback of postoperative infection, in order to make up for the shortcomings of existing technologies in data correlation, intelligent decision-making and self-learning optimization.

[0003] At present, the Chinese invention patent with application number CN202510664838.9 discloses a wound flora monitoring method and system. The method comprises the following steps: acquiring wound physiological characteristic parameters, real-time flora monitoring data and multi-spectral imaging data, and constructing a three-dimensional flora distribution model and an initial flora dynamic model after preprocessing; applying metabolic and environmental conditions to obtain a comprehensive model and flora dynamic coupling data; training a flora classification model to obtain abnormal characteristic parameters; simulating flora abnormal evolution to obtain diffusion trend information; constructing an optimization model to determine the optimal configuration of the monitoring device; and adjusting the monitoring strategy based on a risk grading model. The system comprises data acquisition, model construction and analysis, flora classification, simulation analysis, configuration optimization and risk control modules.

[0004] The above-mentioned technology cannot realize multi-source dynamic perception and individualized real-time early warning of postoperative infection risk, and cannot form an intelligent closed-loop management from monitoring, identification to intervention feedback. SUMMARY

[0005] The technical problem solved by the present application is that the existing technology cannot realize multi-source dynamic perception and individualized real-time early warning of postoperative infection risk, and cannot form an intelligent closed-loop management from monitoring, identification to intervention feedback.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] A postoperative infection monitoring and early warning method, comprising the following steps:

[0008] Step S1: Multi-source data acquisition and alignment to form a postoperative infection multi-source aligned data set;

[0009] Step S2: Establishing an individual reference model based on preoperative health records using feature dimension reduction, clustering and regression modeling, and dynamically extracting postoperative change characteristics;

[0010] Step S3, construct a behavior-physiology coupling model to identify abnormal interactions and quantify infection risk;

[0011] Step S4, fuse environmental factors and risk scores to generate hierarchical warning labels through spatial clustering and rule mining;

[0012] Step S5, use task scheduling and intelligent recommendation mechanisms to allocate intervention actions and track execution effects;

[0013] Step S6, correct threshold values and model parameters based on actual infection outcomes.

[0014] Preferably, the step S1 includes the following sub-steps:

[0015] Step S101, collect patient's body temperature, heart rate, respiratory rate, wound infrared temperature, skin electric response and blood oxygen saturation data, output as patient's vital sign data set;

[0016] Step S102, collect dressing frequency, aseptic operation compliance record, touch event log and order execution time, output as nursing behavior data set;

[0017] Step S103, collect air temperature and humidity, microbial culture frequency, air particulate matter concentration and disinfection record, output as environmental monitoring data set;

[0018] Step S104, based on the ward information system control clock, synchronize the patient's vital sign data set, nursing behavior data set and environmental monitoring data set in time and unify the spatial identification, forming a postoperative infection multi-source alignment data set.

[0019] Preferably, the step S2 includes the following sub-steps:

[0020] Step S201, based on the postoperative infection multi-source alignment data set, analyze the temperature fluctuation amplitude, respiratory rate change rate and skin electric response trend in the patient's vital sign data set, extract the physiological dynamic feature set;

[0021] Step S202, based on the preoperative health record and the same period rehabilitation patient data, establish an individual baseline model, the establishment process includes using principal component analysis to reduce the dimensionality of preoperative data, extracting body temperature stability, heart rate fluctuation rate and immune index distribution, outputting patient feature data;

[0022] Use K-means clustering algorithm to group the same period rehabilitation patient data, determine the normal recovery mode distribution interval;

[0023] Map the patient's vital sign data set to the normal recovery mode distribution interval, generate an individual reference curve, and form an individual baseline model through support vector regression training, output the individual baseline feature set;

[0024] Step S203: Perform a difference analysis between the physiological dynamic feature set and the individual baseline feature set. The analysis process includes first standardizing the two feature sets to eliminate unit differences and dimensional differences.

[0025] Mahalanobis distance is used to calculate the degree of deviation between patient feature data, output the feature deviation magnitude, and identify the feature group with significant deviation using hierarchical clustering.

[0026] By analyzing the deviation trends in different time periods using time series similarity metrics, we can extract the characteristics of abnormal persistence.

[0027] A postoperative deviation feature dataset is generated based on the magnitude and duration of the deviation.

[0028] Step S204: During the difference analysis process, the sliding window mechanism is used to capture the periodicity of vital signs fluctuations and the duration of abnormalities, identify potential abnormal bands, and generate an abnormal trend label dataset.

[0029] Preferably, step S3 includes the following sub-steps:

[0030] Step S301, establish the input layer of the behavioral-physiological coupling model:

[0031] Step S302, construct the temporal inference layer of the behavioral-physiological coupling model:

[0032] Establish a coupling mapping matrix between the nursing behavior dataset and the patient vital signs dataset:

[0033] The nursing behavior dataset and the patient vital signs dataset were synchronized over time, and the correlation coefficient matrix was used to calculate the coupling strength between dressing change frequency, aseptic operation compliance rate, touch event density and wound temperature, and skin conductance response.

[0034] By identifying nonlinear dependencies through mutual information analysis, a multidimensional coupled feature vector is formed, and a set of behavioral and physiological coupled features is output.

[0035] Step S302: The dynamic Bayesian inference method is used to perform time-series modeling of the behavioral and physiological coupling feature set. The hidden state nodes represent the patient's infection potential, and the visible state nodes correspond to nursing operations and physiological indicators. When a high touch event density and a wound temperature rise are detected to occur continuously within the same window, the model outputs the probability of potential infection abnormality.

[0036] When the probability of a potential infection exceeds a set threshold, it is marked as a potential infection event.

[0037] Step S303, construct the output layer of the behavioral-physiological coupling model:

[0038] In combination with the postoperative deviation feature dataset, the abnormal trend label dataset, and the potential infection abnormal event, a weighted fusion algorithm is used to calculate an infection risk impact factor, wherein the weight is dynamically adjusted by a feature importance learning model.

[0039] An infection risk score result is generated by normalized weighted summation, which is used to reflect the real-time coupling strength between physiological abnormalities and nursing behaviors.

[0040] Preferably, the step S4 includes the following sub-steps:

[0041] Step S401, time and space correlation between the infection risk score result and the environmental monitoring dataset is performed, a DBSCAN algorithm is used to identify high-risk areas, environmental sensor data is matched with patient bed coordinates, and a risk correlation matrix is constructed;

[0042] Step S402, an Apriori association rule mining algorithm is used to identify frequent co-occurrence patterns of air quality decline, disinfection delay, and infection events in the same ward in the risk correlation matrix, and significant relationships are screened through support and confidence thresholds, and an environmental association risk factor set is output;

[0043] Step S403, the infection risk score result and the environmental association risk factor set are integrated, a hierarchical weighted fusion method is used to perform multi-level risk judgment, and multi-level infection warning labels are divided according to the comprehensive score interval, including a first-level infection warning label, a second-level infection warning label, and a third-level infection warning label;

[0044] The multi-level infection warning labels are synchronized to the medical terminal through a data interface.

[0045] Preferably, the step S5 includes the following sub-steps:

[0046] Step S501, a high-risk patient is identified based on the multi-level infection warning label and an intervention task is generated, and the priority is sorted according to the label level, the infection risk score result, and the environmental association risk factor in the same ward;

[0047] The task is automatically assigned and the execution time limit and re-evaluation time point are generated in combination with the on-duty load and professional division of the nurse, wherein the on-duty load is estimated by a load index formed by the shift working hours, the current number of tasks in progress, the patient condition coefficient, and the walking distance, and the professional division is represented by the qualification label, the experience level, and the compliance score in the past three months;

[0048] Under the conditions of meeting the qualification matching, the maximum load threshold, and the cross-zone upper limit, a heuristic scheduling and linear programming hybrid solution is used to minimize the weighted overload and task delay, and an SLA task item is generated, including the person in charge, the time limit to bed, and the re-evaluation time point;

[0049] Step S502, combine the historical intervention effect database with the infection risk score result, and perform multi-objective scoring on the preset candidate intervention action, wherein the scoring dimensions of the multi-objective scoring include expected risk reduction amplitude, invasiveness to the patient, resource occupation, and execution accessibility;

[0050] The recommendation engine combining the rule template and the contextual multi-arm bandit selects the strategy combination with the best historical effect in the similar patient and similar risk situation adaptively, and outputs an intervention strategy suggestion set;

[0051] Step S503, record the response time delay of the medical staff to the intervention strategy suggestion set, the time to reach the bedside, the actual execution step, and the patient tolerance, collect the short-term changes of the signs and the test results after the execution;

[0052] The deviation reasons in the case where the execution is not according to the suggestion are recorded to form a response execution log data set, wherein the response execution log data set includes a time stamp, compliance degree, and immediate effect evaluation.

[0053] Preferably, the step S6 comprises the following sub-steps:

[0054] Step S601, collect the test results, sign changes, and infection outcome data of the patient after the intervention, and output an intervention feedback data set;

[0055] Step S602, compare the intervention feedback data set with the multi-level infection early warning label, and correct the infection risk score threshold and the weight parameter in combination with the response execution log data set;

[0056] Step S603, update the individual baseline model and the behavior-physiology coupled model according to the corrected weight parameter, and output a self-correcting parameter set.

[0057] Preferably, the step S303 further comprises an adaptive weight adjustment mechanism.

[0058] The adaptive weight adjustment mechanism adjusts the weight proportions of the body temperature, skin electric response, and touch event density in the infection risk score in real time according to the physiological fluctuation characteristic change trend of different postoperative periods, and increases the behavior class weight to a preset first percentage when the body temperature rise and the synchronous increase of the touch events are detected, wherein the first percentage is 60%-80%, and preferably 70%.

[0059] Preferably, the step S602 further performs a lag compensation process when correcting the infection risk score threshold.

[0060] The hysteresis compensation process corrects the risk threshold adjustment step by analyzing the time interval from policy issuance to sign improvement in the intervention execution log, automatically shortens the threshold update period to a preset second percentage when the intervention response of the last two times lags behind the risk increase event, wherein the second percentage is 50%-80% of the original threshold update period, preferably 67%.

[0061] A postoperative infection monitoring and early warning system comprises a data acquisition module, an individual analysis module, a model construction module, a fusion analysis module, a task scheduling module and a real-time correction module.

[0062] The data acquisition module is used for multi-source data acquisition and alignment to form a postoperative infection multi-source aligned data set.

[0063] The individual analysis module is used for establishing an individual reference model based on preoperative health records by using feature dimension reduction, clustering and regression modeling, and dynamically extracting postoperative change features.

[0064] The model construction module is used for constructing a behavior-physiology coupled model, identifying abnormal interactions and quantifying infection risks.

[0065] The fusion analysis module is used for fusing environmental factors and risk scores to generate hierarchical warning labels through spatial clustering and rule mining.

[0066] The task scheduling module is used for allocating intervention actions and tracking execution effects by using task scheduling and intelligent recommendation mechanisms.

[0067] The real-time correction module is used for correcting threshold values and model parameters based on actual infection outcomes.

[0068] The present application has the following advantages: the present application fuses multi-source data, establishes individualized dynamic baseline models and behavior-physiology coupled models, realizes intelligent identification and hierarchical early warning of postoperative infection risks, automatically corrects model parameters by using self-learning and feedback mechanisms, forms a closed-loop process of monitoring, prediction, intervention and optimization, can identify infection signs in advance, reduces artificial dependence, improves the real-time performance and accuracy of postoperative infection prevention and control, and provides intelligent decision support for clinics. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 A step flowchart of a postoperative infection monitoring and early warning method provided by an embodiment of the present application is provided.

[0070] Figure 2 A basic flowchart of a postoperative infection monitoring and early warning system provided by an embodiment of the present application is provided.

[0071] Figure 3 A touch event log field and collection source example table provided by an embodiment of the present application is provided.

[0072] Figure 4 The sterile operation compliance record field and the collection source example table provided for an embodiment of the present application are shown in Table 1.

[0073] Figure 5 The model training and early warning inference key parameter setting example table provided for an embodiment of the present application is shown in Table 2. DETAILED DESCRIPTION

[0074] In order to make the above objectives, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments.

[0075] Embodiment 1, with reference to Figure 1 provides a postoperative infection monitoring and early warning method, comprising the following steps:

[0076] Step S1, multi-source data collection and alignment, forming a postoperative infection multi-source aligned data set.

[0077] Step S2, based on the preoperative health record, using feature dimension reduction, clustering and regression modeling to establish an individual reference model and dynamically extract postoperative change features.

[0078] Step S3, constructing a behavior-physiology coupling model, identifying abnormal interactions and quantifying infection risks.

[0079] Step S4, fusing environmental factors and risk scores to generate hierarchical warning labels through spatial clustering and rule mining.

[0080] Step S5, using task scheduling and intelligent recommendation mechanism to allocate intervention actions and track the execution effect.

[0081] Step S6, correcting threshold values and model parameters based on actual infection outcomes.

[0082] The present application fuses multi-source data, establishes individualized dynamic baseline models and behavior-physiology coupling models, realizes intelligent identification and hierarchical early warning of postoperative infection risks, automatically corrects model parameters using self-learning and feedback mechanisms, forms a closed-loop process of monitoring, prediction, intervention and optimization, can identify infection signs in advance, reduces artificial dependence, improves the real-time and accuracy of postoperative infection prevention and control, and provides intelligent decision support for clinical practice.

[0083] Step S1 comprises the following sub-steps:

[0084] Step S101, collecting patient's body temperature, heart rate, respiratory rate, wound infrared temperature, skin electric response and blood oxygen saturation data, and outputting as a patient's sign data set.

[0085] Step S101 comprehensively reflects the patient's postoperative physiological state and potential stress response by collecting multiple vital signs data such as body temperature, heart rate, and respiratory rate in real time, forming a quantifiable physiological characteristic basis.

[0086] Step S102: Collect data on dressing change frequency, aseptic operation compliance records, touch event logs, and medical order execution time, and output them as a nursing behavior dataset.

[0087] The automatic detection of touch events and the collection and detection methods for aseptic operation compliance records are as follows:

[0088] Automatic touch event detection: Pressure and capacitive contact sensor pads are placed on the outer layer of the patient's wound dressing, or infrared and visible light depth cameras and area fences are set up at the bedside;

[0089] Healthcare workers wear RFID badges, wristbands, or gloves with IMUs. The sensor signals are synchronized with the ward's unified clock. A touch event is generated when a healthcare worker's identification is detected entering the bedside enclosure and the pressure of the dressing contact sensor exceeds a contact threshold and remains above a preset touch duration threshold. For cases of approach without contact, a proximity touch event is generated based on the minimum distance threshold of the distance sensor, and the confidence level is noted in the log. To reduce false alarms, patient movement signals (mattress pressure, wearable IMU) can be used for exclusion: if the patient's movement exceeds the movement threshold within the touch window and no operator identification is detected, the touch event is not counted.

[0090] Touch event log fields: The touch event log should include at least the event timestamp, patient identifier, operator identifier, bed and area identifier, touch type, touch duration, cumulative number of touches within the window, event confidence level, and sensor source. See the corresponding table below. Figure 3 .

[0091] Aseptic Procedure Compliance Record Collection: Aseptic procedure compliance records are used to quantify the execution of aseptic procedures by medical staff during dressing changes / medical dressing changes / invasive procedures. Data sources include: IoT counts and work badge association records of hand sanitizer dispensers, barcode or RFID opening and scanning records of sterile packs and dressing packs, glove box issuance records and replacement times, and operation step confirmation records from bedside terminals. These records are aggregated by operation task ID to form a compliance record for a single nursing task.

[0092] Aseptic technique compliance record fields should include at least the task ID, operator ID, patient ID, procedure type, key step completion marker, step timestamp, step interval duration, compliance score, and anomaly reason code. See the corresponding table below. Figure 4 .

[0093] Step S102 quantifies the operation intensity and infection exposure frequency in the nursing process by recording the behavior information such as dressing frequency, aseptic operation and touch events, and provides a basis for coupling analysis of behavior and infection risk.

[0094] Step S103 collects air temperature and humidity, microbial culture times, air particulate matter concentration and disinfection records, and outputs environmental monitoring data sets.

[0095] Step S103 identifies the influence of external infection transmission environment and air quality changes on patient recovery by collecting environmental parameters such as air temperature and humidity and microbial concentration.

[0096] Step S104, taking the ward information system control clock as the reference, synchronizes the time and unifies the space identification of the patient sign data set, the nursing behavior data set and the environmental monitoring data set, and forms the postoperative infection multi-source alignment data set.

[0097] Step S104 synchronizes the time and space identification of multiple types of data with a unified clock, eliminates the time offset caused by the difference between the collection sources, and ensures the corresponding accuracy and real-time relevance of each data dimension in subsequent model analysis.

[0098] Step S1 establishes a complete and traceable postoperative infection multi-source alignment data set through the synchronous collection and unified alignment of multi-source data, realizes the spatio-temporal consistency of physiological, nursing and environmental information, and provides a reliable data foundation for subsequent individualized modeling and infection risk identification.

[0099] Step S2 includes the following sub-steps:

[0100] Step S201, based on the postoperative infection multi-source alignment data set, analyzes the temperature fluctuation amplitude, respiratory rate change rate and skin electric reaction trend in the patient sign data set, and extracts a set of physiological dynamic features.

[0101] Step S201 extracts key features that can reflect the physiological dynamic changes of the patient by analyzing the temperature fluctuation, respiratory rate change rate and skin electric reaction trend, and establishes a basic feature space for anomaly detection.

[0102] Step S202, based on the preoperative health record and the same period rehabilitation patient data, establishes an individual baseline model, and the process includes using principal component analysis to reduce the dimensionality of preoperative data, extracting body temperature stability, heart rate fluctuation rate and immune index distribution, and outputting patient feature data.

[0103] The K-means clustering algorithm is used to group the same period rehabilitation patient data, and the normal recovery mode distribution interval is determined.

[0104] Mapping the patient's sign data set to the normal recovery mode distribution interval, generating an individual reference curve, and forming an individual baseline model through support vector regression training, outputting an individual baseline feature set.

[0105] The sample construction, feature vector, and parameter selection method for support vector regression training are as follows:

[0106] Training sample construction: Based on the postoperative infection multi-source alignment data of each patient, time window samples are generated at a fixed time granularity Δt, which can be 1 minute, 5 minutes, or 10 minutes, preferably 10 minutes. For each time window k, an input feature vector Xk is constructed and a prediction target Yk is constructed. The Yk is used to fit the expected value or trend item of the key physiological indicators of the individual reference curve, preferably the expected value of the wound infrared temperature, the expected value of the body temperature, or the trend slope of the wound temperature within the future H time span (H can be 3-6 hours).

[0107] Input feature vector Xk: at least includes:

[0108] Physiological dynamic features: mean body temperature, body temperature slope, mean heart rate, heart rate standard deviation, respiratory rate change rate, mean blood oxygen saturation, mean and slope of wound infrared temperature, mean and rising rate of skin electrical response;

[0109] Nursing behavior features: dressing change flag, touch event density (number of touches within the window), sterile operation compliance score, order execution delay;

[0110] Static profile features: age, BMI, underlying disease label, surgery type and surgery duration, etc.

[0111] Each feature is subjected to missing value filling and standardization before entering the model, and is indexed with patient identification and time stamp.

[0112] Each sample is represented as a structured record of patient identification, time window start and end, feature vector, and target value. Example: for 300 patients' postoperative 72-hour monitoring data, windowing at Δt=10 minutes can obtain about 300x(72x6)=129600 time window samples, of which infection-positive patients are used to construct abnormal situation samples, and recovered patients are used to construct normal baseline samples. The above sample size is an example, and the actual hospital data size can be adjusted.

[0113] The kernel function of SVR can be radial basis function (RBF) or polynomial kernel function, and RBF kernel is preferred to process the nonlinear recovery curve. The key parameters include penalty coefficient C, kernel width parameter γ, and insensitive interval ε. The parameter selection method is: using grid search or Bayesian optimization on the training set, and using K-fold cross-validation (K can be 5 or 10) to minimize the mean absolute error MAE or mean square error MSE of the validation set; the example value range is: C∈[1,100], γ∈[10^-4,10^-1], ε∈[0.01,0.2]. When the verification error converges, the optimal parameter group is taken as the individual baseline model parameter.

[0114] After obtaining the new intervention feedback data set in step S6, the new samples can be incorporated into the training set in a "daily" cycle, and the SVR model parameters are retrained or incrementally updated to adapt to the seasonal environmental changes and patient composition changes in the ward.

[0115] Step S202 establishes an individual baseline model through principal component analysis, clustering, and regression modeling, realizes individualized comparison of postoperative signs and normal recovery mode, and ensures that the model can accurately reflect individual differences and postoperative recovery state.

[0116] Step S203, difference analysis is performed on the physiological dynamic feature set and the individual baseline feature set. The analysis process includes first standardizing the two feature sets to eliminate unit differences and dimensional differences.

[0117] The Mahalanobis distance is used to calculate the deviation degree between the feature data of the patients, and the feature deviation amplitude is output, and the hierarchical clustering method is used to identify the feature groups that deviate significantly.

[0118] The deviation trend of different time periods is analyzed by time series similarity measurement, and the abnormal persistence feature is extracted.

[0119] The postoperative deviation feature data set is generated according to the feature deviation amplitude and the duration.

[0120] Step S203 calculates the sign deviation degree and feature group difference through Mahalanobis distance and hierarchical clustering, identifies the persistent abnormal change by combining time series similarity analysis, forms a quantitative postoperative deviation feature data set, and is used to identify potential infection risk signals.

[0121] In the difference analysis process, the sign fluctuation period and the abnormal duration are captured through the sliding window mechanism, the potential abnormal waveband is identified, and the abnormal trend label data set is generated.

[0122] Step S204 tracks the sign fluctuation period through the sliding window mechanism, identifies the abnormal duration interval, and generates the abnormal trend label data set, realizes continuous monitoring and early abnormal marking of the patient state change.

[0123] Step S2 realizes dynamic quantification and abnormal identification of the physiological state of the patient by constructing an individualized baseline model and performing difference analysis on postoperative sign changes, forming a feature data set that can reflect the deviation degree of individual recovery, and providing accurate input for subsequent infection risk prediction.

[0124] Step S3 includes the following sub-steps:

[0125] Step S301, establish the input layer of the behavior-physiology coupling model:

[0126] Establish the coupling mapping matrix of the nursing behavior data set and the patient sign data set:

[0127] Synchronize the time series of the nursing behavior data set and the patient sign data set, and use the correlation coefficient matrix to calculate the coupling strength between the dressing change frequency, the sterile operation compliance rate, the touch event density, and the wound temperature, and the skin electric reaction.

[0128] Identify the nonlinear dependence relationship through mutual information analysis, form a multi-dimensional coupling feature vector, and output the behavior-physiology coupling feature set.

[0129] Step S301 calculates the correlation and nonlinear relationship between the dressing change frequency, the sterile operation compliance rate, the touch event density, and the wound temperature, and the skin electric reaction by constructing the coupling mapping matrix of the nursing behavior and sign data, extracts the coupling feature set reflecting the interaction between behavior and physiology, and provides multi-dimensional feature input for abnormal identification.

[0130] Step S302, construct the time series inference layer of the behavior-physiology coupling model:

[0131] Use dynamic Bayesian inference method to model the behavior-physiology coupling feature set in time series, represent the patient's infection potential through hidden state nodes, and correspond nursing operation and physiological indicators through visible state nodes. When high touch event density and rising wound temperature are detected to appear continuously in the same window, the model outputs the abnormal probability of potential infection.

[0132] The definition of hidden state and visible state of dynamic Bayesian inference, state transition matrix initialization and learning method are as follows:

[0133] Visible state (observation variable) Ot: The behavior-physiology coupling feature set output by step S301, at least including touch event density, sterile operation compliance score, dressing change frequency, wound infrared temperature slope, body temperature slope, skin electric reaction rising rate, and postoperative deviation feature amplitude, etc.; for continuous variables, they can be discretized into low, medium and high levels according to quantile points or clinical thresholds to meet the needs of discrete dynamic Bayesian network modeling, or they can be kept as continuous variables and use Gaussian emission model.

[0134] Hidden state (latent variable) St: used to represent the patient's infection potential, preferably set to three states: S0 = normal recovery, S1 = suspicious infection, S2 = high-risk infection; when higher resolution is needed, it can be extended to continuous potential values and implemented with particle filtering to infer. The model output of the latent infection anomaly probability is defined as P(St=S2|O1:t).

[0135] State transition probability matrix A initialization: uniform initialization or clinical prior initialization can be used. Example clinical prior: the probability of a patient remaining in the original state within adjacent time windows is high, and the probability of jumping to adjacent risk levels is low, set to A=[[0.90,0.10,0.00],[0.05,0.90,0.05],[0.00,0.10,0.90]]; where the matrix elements can be initially estimated according to the historical data frequency.

[0136] Parameter learning: when there is an infection outcome label (such as blood culture positive, imaging evidence or clinical diagnosis conclusion), the Expectation Maximization (EM) or Forward-Backward algorithm (Baum-Welch) is used to learn the maximum likelihood of the A matrix and the emission distribution parameters; when there is no explicit label, a small amount of labeled samples and a large amount of unlabeled samples are jointly trained in a semi-supervised manner, and the parameters are updated through the feedback backfilling of step S6.

[0137] Inference and alarm: after performing forward recursion to calculate the posterior probability for each time window, when P(St=S2|O1:t) continuously exceeds the infection anomaly threshold (the threshold can be 0.6-0.8, preferably 0.7) and lasts for more than a duration threshold (such as 30-60 minutes), it is marked as a potential infection anomaly event and enters step S303 for fusion scoring.

[0138] When the latent infection anomaly probability exceeds the set threshold, it is marked as a potential infection anomaly event.

[0139] Step S302 uses dynamic Bayesian inference to model the time series of the coupled feature set, quantifies the infection potential in the form of hidden states, and outputs the latent infection anomaly probability when a high-frequency contact and temperature rise linkage pattern occurs, achieving intelligent identification and labeling of early infection trends.

[0140] Step S303, construct the output layer of the behavior-physiology coupling model:

[0141] Combine the postoperative deviation feature dataset, anomaly trend label dataset, and potential infection anomaly event to calculate the infection risk impact factor using a weighted fusion algorithm, where the weights are dynamically adjusted by the feature importance learning model.

[0142] Generate the infection risk score result by normalized weighted summation, which is used to reflect the real-time coupling strength between physiological abnormalities and nursing behaviors.

[0143] Step S303 combines the deviation feature data, abnormal trend and potential infection event by weighting, dynamically adjusts the weight through the feature importance learning model, generates an infection risk score result reflecting the real-time correlation strength of behavior and physiology, and realizes quantitative evaluation and trend prediction of infection risk.

[0144] Step S303 further includes an adaptive weight adjustment mechanism.

[0145] The adaptive weight adjustment mechanism adjusts the weight proportion of body temperature, skin electric response and touch event density in the infection risk score in real time according to the physiological fluctuation feature change trend of different postoperative periods, and increases the behavior class weight to a preset first percentage when detecting that the body temperature rises and the touch event increases synchronously, wherein the first percentage is 60%-80%, preferably 70%.

[0146] The determination method of the first percentage is as follows: before going online, a training set and a validation set are constructed by historical samples or simulation samples, candidate values: 60%, 65%, 70%, 75%, 80% are traversed, early warning lead time, false positive rate and intervention execution load are calculated respectively, and the candidate value with the optimal comprehensive score is taken as the first percentage; when lacking historical samples, the first percentage is defaulted to 70%, and is updated monthly after feedback in step S6.

[0147] The adaptive weight adjustment mechanism dynamically adjusts the weight proportion of body temperature, skin electric response and touch event density in the risk score according to the physiological fluctuation feature change trend of different postoperative periods, so that the model is more sensitive to key risk factors at different stages, and the accuracy and sensitivity of early infection identification are improved.

[0148] Step S3 realizes quantitative analysis of the causal correlation between nursing operation and physiological abnormality by establishing a coupling model of nursing behavior and patient physiological state, can identify potential infection signals in real time and generate an infection risk score, and provides accurate basis for subsequent risk grading and intervention decision.

[0149] Step S4 includes the following substeps:

[0150] Step S401 time and space correlates the infection risk score result with the environmental monitoring data set, identifies high-risk areas using the DBSCAN algorithm, matches the environmental sensor data with the patient bed coordinates, and constructs a risk correlation matrix.

[0151] Step S401 time and space correlates the infection risk score result with the environmental monitoring data set, identifies high-risk areas using the DBSCAN algorithm, matches the environmental sensor data with the patient bed coordinates, and constructs a risk correlation matrix.

[0152] Step S402, using Apriori association rule mining algorithm in the risk association matrix to identify frequent co-occurrence patterns of air quality decline, disinfection delay and infection events in the same ward, and to screen significant relationships through support and confidence thresholds, and output the environmental associated risk factor set.

[0153] Step S402, using Apriori association rule mining algorithm to find frequent co-occurrence patterns between air quality decline, disinfection delay and infection events from the risk association matrix, and to screen significant associated factors according to support and confidence, and generate the environmental associated risk factor set, which is used to identify the potential driving effect of external environment on infection transmission.

[0154] Step S403, integrating the infection risk score results and the environmental associated risk factor set, using hierarchical weighted fusion method to perform multi-level risk judgment, and dividing multi-level infection warning labels according to the comprehensive score interval, including first-level, second-level and third-level infection warning labels.

[0155] The multi-level infection warning labels are synchronized to the medical terminal through the data interface.

[0156] Step S403 integrates the infection risk score and the environmental associated risk factor, uses hierarchical weighted fusion method for multi-level risk judgment, divides the comprehensive score into first, second and third warning labels, realizes the hierarchical expression of infection risk intensity, and synchronizes to the medical terminal in real time through the data interface, which is used to guide the hierarchical response and prevention and control decision.

[0157] Step S4, by spatial and temporal correlation analysis of infection risk score and environmental monitoring data, realizing the comprehensive evaluation of patient individual infection risk and ward environmental factors, establishing a multi-level infection warning mechanism, so as to find the environmental driven infection aggregation trend in the early stage and send hierarchical warning signals to the medical terminal.

[0158] Step S5 includes the following sub-steps:

[0159] Step S501, identifying high-risk patients based on multi-level infection warning labels and generating intervention tasks, prioritizing by label level, infection risk score results and environmental associated risk factors in the same ward.

[0160] The tasks are automatically assigned according to the on-duty load and professional division, and the execution time limit and re-evaluation time point are generated, wherein the on-duty load is estimated by the load index formed by the shift working hours, the number of current tasks, the patient condition coefficient and the walking distance, and the professional division is represented by the qualification label, the experience level and the compliance score in the past three months.

[0161] In the case of meeting the qualification matching, the maximum load threshold and the cross-zone upper limit, a heuristic scheduling and linear programming hybrid solution is adopted to minimize the weighted overload and task delay, and SLA task entries are generated, which include the responsible person, the to-bed time limit and the re-evaluation time point.

[0162] Step S501 intelligently identifies high-risk patients and assigns intervention tasks by analyzing multi-level infection early warning labels, risk scores and environmental factors. Combined with the nurse load index and professional division, the heuristic scheduling and linear programming algorithm is used to optimize task allocation, realize automatic scheduling and rational division of tasks, and generate SLA task entries containing the responsible person, the to-bed time limit and the re-evaluation time point, to ensure the timeliness and balance of intervention execution.

[0163] Step S502, combined with the historical intervention effect database and the infection risk score result, multi-objective scoring is performed on the pre-set candidate intervention actions, and the scoring dimensions of multi-objective scoring include the expected risk reduction amplitude, the invasiveness to patients, the resource occupation and the execution accessibility.

[0164] A recommendation engine combining rule templates and contextual multi-armed bandit is used to adaptively select the strategy combination with the best historical effect in similar patients and similar risk situations, and output the intervention strategy suggestion set.

[0165] Step S502 integrates historical intervention effects and current risk scores to perform multi-objective scoring and dynamic recommendation on different candidate intervention actions, and intelligently selects the optimal intervention strategy combination using rule templates and contextual multi-armed bandit algorithm, to realize individualized, data-driven intervention decisions for different patient states.

[0166] Step S503 records the response time delay of medical staff to the intervention strategy suggestion set, the arrival time at the bedside, the actual execution steps and the patient's tolerance, and collects the short-term changes of signs and test results after execution.

[0167] The deviation reasons for not executing according to the suggestions are recorded, and a response execution log data set is formed, which includes timestamp, compliance degree and immediate effect evaluation.

[0168] Step S503 real-time tracks the execution process of medical staff, records the response time delay, execution steps and patient physiological feedback, and generates a response execution log data set containing timestamp, compliance degree and effect evaluation, to provide high-quality feedback data support for subsequent model optimization and clinical intervention effect verification.

[0169] Step S5 realizes rapid response and optimal resource scheduling after postoperative infection warning through automatic allocation of intervention tasks for high-risk patients and intelligent strategy recommendation, forms a traceable and evaluable intervention execution closed loop, and improves the response efficiency and intervention accuracy of medical staff.

[0170] Step S6 includes the following sub-steps:

[0171] Step S601, collect the test results, sign changes and infection outcome data of the patient after the intervention, and output the intervention feedback data set.

[0172] Step S601 forms an intervention feedback data set by collecting the test results, sign changes and final infection outcome data of the patient after the intervention, realizes the quantitative closed-loop tracking from the intervention behavior to the clinical results, and provides a true label basis for model correction.

[0173] Step S602 compares the intervention feedback data set with the multi-level infection warning label, and corrects the infection risk score threshold and weight parameter in combination with the response execution log data set.

[0174] Step S602 further performs a lag compensation process when correcting the infection risk score threshold.

[0175] The lag compensation process corrects the risk threshold adjustment step by analyzing the time interval from policy issuance to sign improvement in the intervention execution log. When the response of the intervention lags behind the risk rising event for two consecutive times, the threshold update period is automatically shortened to a preset second percentage, wherein the second percentage is 50%-80% of the original threshold update period, preferably 67%.

[0176] The determination method of the second percentage is as follows: the time interval distribution from policy issuance to sign improvement is counted, and the upper quartile or the mean plus one standard deviation is taken as the response lag threshold; when the threshold is exceeded for two consecutive times, the threshold update period is shortened by the second percentage. The second percentage can be discretely searched in the range of 50%-80%, preferably 67%, to reduce false negatives while controlling the fluctuations caused by frequent threshold updates.

[0177] Step S602 compares the intervention feedback data with the multi-level warning label and the response log, analyzes the deviation between the model prediction and the actual result, dynamically corrects the infection risk score threshold and the weight parameter, and adjusts the threshold step according to the time difference between policy execution and sign improvement through the lag compensation process, realizing the adaptive correction of the model to the delayed clinical response.

[0178] Step S603 updates the individual baseline model and the behavior-physiology coupled model according to the corrected weight parameter, and outputs the self-corrected parameter set.

[0179] Step S603 re-trains and corrects the parameters of the individual baseline model and the behavior-physiology coupled model according to the updated weight parameter, and outputs the self-corrected parameter set, so that the system realizes self-optimization of model performance and improvement of cross-batch generalization ability in continuous operation.

[0180] Step S6 corrects by comparing the post-intervention feedback data with the model output, realizes self-learning and continuous optimization of the system, enables the infection risk prediction model to dynamically adapt to clinical changes, continuously improves the early warning accuracy and response timeliness, and constructs a data-driven intelligent closed-loop optimization mechanism.

[0181] Embodiment 2, with reference to Figure 2 provides a postoperative infection monitoring and early warning system, comprising a data acquisition module, an individual analysis module, a model construction module, a fusion analysis module, a task scheduling module and a real-time correction module.

[0182] The data acquisition module is used for multi-source data acquisition and alignment to form a postoperative infection multi-source aligned data set.

[0183] The individual analysis module is used for establishing an individual reference model based on the preoperative health record by using feature dimension reduction, clustering and regression modeling, and dynamically extracting postoperative change characteristics.

[0184] The model construction module is used for constructing a behavior physiology coupled model, identifying abnormal interactions and quantifying infection risks.

[0185] The fusion analysis module is used for fusing environmental factors and risk scores to generate a hierarchical early warning label through spatial clustering and rule mining.

[0186] The task scheduling module is used for allocating intervention actions and tracking execution effects by using task scheduling and intelligent recommendation mechanisms.

[0187] The real-time correction module is used for correcting threshold values and model parameters based on actual infection outcomes.

[0188] The application collects three types of data of patient signs, nursing behaviors and environmental monitoring, and performs unified space-time alignment to establish a high-precision postoperative infection multi-source alignment data set, realizes the basis of collaborative analysis of physiological, behavioral and environmental signals, introduces principal component analysis (PCA), K-means clustering and support vector regression (SVR) methods to model the characteristics of preoperative and rehabilitation samples, form individual baseline models, realize individualized quantitative evaluation of postoperative sign changes, significantly improve the sensitivity and specificity of abnormal detection, and the behavior-physiology coupling modeling mechanism proposed correlates the nursing operation log and the physiological signal in time sequence, identifies potential infection abnormal events by using dynamic Bayesian inference, and realizes real-time risk scoring through weighted fusion and adaptive weight adjustment, so that the identification of early infection signals is changed from passive observation to active inference, environmental pollution, disinfection delay and local infection events are included in the comprehensive analysis by DBSCAN spatial clustering and Apriori rule mining, multi-level warning labels are generated, and an infection transmission risk identification system of three-dimensional interaction of environment, patient and nursing is constructed, combined with heuristic scheduling and linear programming algorithm, the intervention tasks are intelligently distributed according to the nurse load index and professional division; and the intervention strategy is adaptively recommended by using the Thompson Sampling algorithm, realizing the automation of the closed loop from risk identification to clinical execution, using the intervention feedback data and the execution log, combining with the lag compensation mechanism to automatically correct the infection risk threshold and weight parameter, and back-feeding to the individual baseline model and the coupling model, realizing the self-evolution and cross-patient migration optimization of the model, and ensuring the accuracy and robustness of the system in long-term operation.

[0189] The method realizes intelligent management of the whole process from infection monitoring to risk prediction to strategy execution, can issue an early warning in the early stage of infection, reduces the dependence on manual operation, improves the real-time performance and scientific nature of infection prevention and control, and provides intelligent decision support for hospital infection management.

[0190] Embodiment 3

[0191] In order to illustrate the practicability of the application, the following gives a set of running examples based on historical record desensitization summary data and simulation completion data, which are not used to limit the application.

[0192] Continuous monitoring data of 120 postoperative patients in a certain surgical ward are selected, covering 0-72 hours after operation. The sign data is collected with a sampling granularity of 1 minute and is subjected to outlier elimination before entering the system; the nursing behavior data comes from the bedside terminal and the sensing log; the environmental monitoring data comes from the temperature and humidity and particulate matter sensors and the disinfection records. Time window samples are generated by alignment with Δt=10 minutes, a total of about 120x(72x6)=51840 time window samples, of which 18 cases are positive for infection outcome (clinical diagnosis combined with test results as outcome label).

[0193] Key parameter settings: sliding window length is 30 minutes; touch event judgment threshold is pressure exceeding threshold and lasting no less than 2 seconds; SVR adopts RBF kernel, C=20, gamma=0.01, epsilon=0.05; dynamic Bayesian model adopts three hidden states, state transition matrix is initialized according to clinical priori A=[[0.90,0.10,0.00],[0.05,0.90,0.05],[0.00,0.10,0.90]] and is updated by EM in training; latent infection abnormality probability alarm threshold is 0.7 and needs to be continuously exceeded for 3 time windows; the first percentage is 70%; the second percentage is 67%, and the corresponding table is shown in Figure 5 .

[0194] Running result example: under the above data conditions, in the output three-level warning label, the coverage rate of secondary and above warning to the positive patients of infection outcome is 83.3%, and the false positive rate to the negative patients is 21.0%; taking the clinical diagnosis time as a reference, the average advance of the warning is 18 hours. After step S6 backfill intervention and outcome data, the threshold and weight are recalibrated, and in the next month on the incremental data in the same ward, the coverage rate of secondary and above warning is increased to 86.0%, and the false positive rate is decreased to 19.5%.

[0195] The above running result shows that the present application can discover infection signs in advance by coupling modeling of touch events, sterility compliance and physiological abnormalities, and realize continuous optimization through feedback backfill, thereby meeting the actual deployment needs of postoperative infection monitoring and warning.

[0196] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (or computer- readable storage media) having computer-usable program code embodied in the medium. The medium can be any available storage media that can be accessed by a computer. By way of example, and not limitation, such computer-usable storage media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other storage medium(s) that can be used to carry or store desired computer program code in the form of instructions or data structures and that can be accessed by a computer. Also, the present application can be embodied in a computer program product that can be traded as goods or merchandise, through the storage medium described above or any other suitable medium. Accordingly, the computer medium can be any entity or device containing, or Figure 1 the functions specified in the flow or flows and / or blocks Figure 1 the functions specified in the flow or flows and / or blocks

[0197] It should be noted that the above-mentioned embodiments are only used to illustrate but not to limit the technical solutions of the present application. Although the present application is 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 equivalently without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for monitoring and early warning of postoperative infection, characterized in that, Includes the following steps: Step S1: Multi-source data acquisition and alignment to form a postoperative infection multi-source aligned dataset; Step S2: Based on the preoperative health record, establish an individual reference model using feature dimensionality reduction, clustering, and regression modeling, and dynamically extract postoperative change features; Step S3: Construct a behavioral-physiological coupling model to identify abnormal interactions and quantify the risk of infection; Step S4: Integrate environmental factors and risk scores to generate hierarchical early warning labels through spatial clustering and rule mining; Step S5: Use task scheduling and intelligent recommendation mechanisms to allocate intervention actions and track the execution effect; Step S6: Adjust the threshold and model parameters based on the actual infection outcome.

2. The postoperative infection monitoring and early warning method as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S101: Collect the patient's body temperature, heart rate, respiratory rate, wound infrared temperature, skin conductance and blood oxygen saturation data, and output them as a patient vital signs dataset. Step S102: Collect dressing change frequency, aseptic operation compliance records, touch event logs, and medical order execution time, and output them as a nursing behavior dataset; Step S103: Collect air temperature and humidity, number of microbial culture times, air particulate matter concentration and disinfection records, and output them as an environmental monitoring dataset; Step S104: Using the control clock of the ward information system as a reference, the patient vital signs dataset, nursing behavior dataset, and environmental monitoring dataset are synchronized in time and unified in spatial identification to form a postoperative infection multi-source aligned dataset.

3. The postoperative infection monitoring and early warning method as described in claim 2, characterized in that, Step S2 includes the following sub-steps: Step S201: Based on the postoperative infection multi-source aligned dataset, analyze the temperature fluctuation amplitude, respiratory rate change rate and skin conductance response trend in the patient's vital signs dataset, and extract a set of physiological dynamic features. Step S202: Based on preoperative health records and data of patients undergoing rehabilitation at the same time, establish an individual baseline model. The establishment process includes using principal component analysis to perform feature dimensionality reduction on preoperative data, extracting body temperature stability, heart rate variability and immune index distribution, and outputting patient characteristic data. K-means clustering algorithm was used to stratify the data of patients recovering in the same period to determine the distribution range of normal recovery patterns; The patient's vital signs dataset is mapped to the normal recovery pattern distribution range to generate individual reference curves. Then, an individual baseline model is formed by training through support vector regression, and the individual baseline feature set is output. Step S203: Perform a difference analysis between the physiological dynamic feature set and the individual baseline feature set. The analysis process includes first standardizing the two feature sets to eliminate unit differences and dimensional differences. Mahalanobis distance is used to calculate the degree of deviation between patient feature data, output the feature deviation magnitude, and identify the feature group with significant deviation using hierarchical clustering. By analyzing the deviation trends in different time periods using time series similarity metrics, we can extract the characteristics of abnormal persistence. A postoperative deviation feature dataset is generated based on the magnitude and duration of the deviation. Step S204: During the difference analysis process, the sliding window mechanism is used to capture the periodicity of vital signs fluctuations and the duration of abnormalities, identify potential abnormal bands, and generate an abnormal trend label dataset.

4. The postoperative infection monitoring and early warning method as described in claim 3, characterized in that, Step S3 includes the following sub-steps: Step S301, establish the input layer of the behavioral-physiological coupling model: Establish a coupling mapping matrix between the nursing behavior dataset and the patient vital signs dataset: The nursing behavior dataset and the patient vital signs dataset were synchronized over time, and the correlation coefficient matrix was used to calculate the coupling strength between dressing change frequency, aseptic operation compliance rate, touch event density and wound temperature, and skin conductance response. By identifying nonlinear dependencies through mutual information analysis, a multidimensional coupled feature vector is formed, and a set of behavioral and physiological coupled features is output. Step S302, construct the temporal inference layer of the behavioral-physiological coupling model: The dynamic Bayesian inference method is used to perform temporal modeling of the behavioral and physiological coupled feature set. The hidden state nodes represent the patient's infection potential, and the visible state nodes correspond to nursing operations and physiological indicators. When high touch event density and wound temperature rise are detected to occur continuously in the same window, the model outputs the probability of potential infection abnormality. When the probability of a potential infection exceeds a set threshold, it is marked as a potential infection event. Step S303, construct the output layer of the behavioral-physiological coupling model: By combining the postoperative deviation feature dataset, the abnormal trend label dataset, and potential infection abnormal events, a weighted fusion algorithm is used to calculate the infection risk impact factor, where the weights are dynamically adjusted by the feature importance learning model; Infection risk scores are generated by normalized weighted summation, and these scores are used to reflect the real-time coupling strength between physiological abnormalities and nursing behaviors.

5. The postoperative infection monitoring and early warning method as described in claim 4, characterized in that, Step S4 includes the following sub-steps: Step S401: The infection risk score results are correlated with the environmental monitoring dataset in time and space. The DBSCAN algorithm is used to identify high-risk areas. The environmental sensor data is matched with the patient bed coordinates to construct a risk correlation matrix. Step S402: In the risk association matrix, the Apriori association rule mining algorithm is used to identify the frequent co-occurrence patterns of air quality decline, disinfection delay and infection events in the same ward, and significant relationships are screened by support and confidence thresholds to output a set of environmental association risk factors. Step S403: Based on the comprehensive infection risk score results and the set of environmental associated risk factors, a hierarchical weighted fusion method is used to perform multi-level risk determination. Multi-level infection warning labels are divided according to the comprehensive score range. The multi-level infection warning labels include a first-level infection warning label, a second-level infection warning label, and a third-level infection warning label. The multi-level infection early warning tags are synchronized to medical staff terminals via a data interface.

6. The postoperative infection monitoring and early warning method as described in claim 5, characterized in that, Step S5 includes the following sub-steps: Step S501: Identify high-risk patients based on multi-level infection early warning labels and generate intervention tasks, prioritizing them according to label level, infection risk score results, and risk factors associated with the same ward environment; The system automatically assigns tasks and generates execution deadlines and review time points by combining nurses' on-duty workload and professional division of labor. The on-duty workload is estimated by combining shift hours, current number of tasks, patient condition coefficient, and walking distance to form a workload index. Professional division of labor is characterized by qualification labels, experience level, and compliance score of the past three months. Under the conditions of meeting qualification matching, maximum load threshold and cross-regional upper limit, a hybrid solution of heuristic scheduling and linear programming is adopted to minimize weighted overload and task delay, and SLA task entries are generated. The SLA task entries include the responsible person, bed arrival time limit and re-evaluation time point. Step S502: Combining the historical intervention effect database and infection risk score results, perform multi-objective scoring on the preset candidate intervention actions. The scoring dimensions of the multi-objective scoring include the expected risk reduction, invasiveness to patients, resource consumption, and execution accessibility. A recommendation engine that combines rule templates with contextual multi-armed gambling machines adaptively selects the strategy combination with the best historical performance in similar patients and similar risk situations, and outputs a set of intervention strategy suggestions; Step S503: Record the response delay of medical staff to the set of intervention strategy recommendations, the time to reach the bedside, the actual implementation steps and the patient's tolerance, and collect short-term changes in vital signs and test results after implementation; For cases where the recommendations are not followed, the reasons for the deviation are recorded to form a response execution log dataset, which includes timestamps, compliance information, and immediate effect assessments.

7. The postoperative infection monitoring and early warning method as described in claim 6, characterized in that, Step S6 includes the following sub-steps: Step S601: Collect patient test results, changes in vital signs and infection outcomes after intervention, and output intervention feedback dataset; Step S602: Compare the intervention feedback dataset with the multi-level infection warning labels, and adjust the infection risk score threshold and weight parameters in conjunction with the response execution log dataset; Step S603: Update the individual baseline model and the behavioral-physiological coupling model according to the corrected weight parameters, and output the self-calibration parameter set.

8. The postoperative infection monitoring and early warning method as described in claim 7, characterized in that, Step S303 further includes an adaptive weight adjustment mechanism; The adaptive weight adjustment mechanism adjusts the weight ratio of body temperature, skin conductance response and touch event density in the infection risk score in real time according to the physiological fluctuation characteristics of different postoperative periods. When a rise in body temperature and a simultaneous increase in touch events are detected, the weight of behavioral categories is increased to a preset first percentage, wherein the first percentage is 60%-80%, preferably 70%.

9. The postoperative infection monitoring and early warning method as described in claim 8, characterized in that, Step S602 further performs a lag compensation process when correcting the infection risk score threshold; The lag compensation process analyzes the time interval from the issuance of the strategy to the improvement of vital signs in the intervention execution log, and corrects the risk threshold adjustment step size. When two consecutive intervention responses lag behind the risk increase event, the threshold update cycle is automatically shortened to a preset second percentage, wherein the second percentage is 50%-80% of the original threshold update cycle, preferably 67%.

10. A postoperative infection monitoring and early warning system, applied in a postoperative infection monitoring and early warning method as described in any one of claims 1-9, characterized in that, It includes a data acquisition module, an individual analysis module, a model building module, a fusion analysis module, a task scheduling module, and a real-time correction module; The data acquisition module is used for multi-source data acquisition and alignment to form a postoperative infection multi-source aligned dataset. The individual analysis module is used to establish an individual reference model based on preoperative health records using feature reduction, clustering, and regression modeling, and to dynamically extract postoperative change features. The model building module is used to construct a behavioral-physiological coupling model to identify abnormal interactions and quantify infection risk. The fusion analysis module is used to integrate environmental factors and risk scores to generate hierarchical early warning labels through spatial clustering and rule mining. The task scheduling module is used to allocate intervention actions and track the execution effect using task scheduling and intelligent recommendation mechanisms; The real-time correction module is used to correct thresholds and model parameters based on actual infection outcomes.

Images