A nursing-based infection risk assessment method and system

Abstract

The application discloses a kind of nursing-based infection risk assessment method and system, it is related to medical information technology field, including acquisition multidimensional nursing feature time sequence, nursing feature component of different sensitivity is separated out by multi-scale perception, and risk trajectory is independently quantified generation.According to the dynamic association mapping network between features constructed by infection risk knowledge graph, the risk trajectory of each component is interactively corrected and fused to form a comprehensive risk evolution path.By analyzing the time sequence pattern of the path, the confidence of the potential infection risk pattern fragment is identified and calculated, and after screening and sorting, it is reconstructed into a warning signal with a time sequence marker, and finally matches the clinical intervention strategy library to output a structured report and suggestion.This method realizes the accurate description and early warning of the infection risk evolution process through dynamic association correction and high-confidence pattern analysis.

Description

Technical Field

[0001] This invention belongs to the field of medical information technology, specifically a nursing-based infection risk assessment method and system. Background Technology

[0002] In current clinical practice, infection risk assessment for hospitalized patients largely relies on static risk scoring scales or monitoring systems based on fixed rules. These methods typically measure multiple nursing characteristics independently, such as body temperature, white blood cell count, and incision condition, and then obtain a comprehensive risk score through weighted summation or simple logical judgment. This type of approach simplifies the complex clinical physiology and nursing process into isolated combinations of indicators, ignoring the dynamic interactions and synergistic effects between different risk factors as the disease progresses. Its inherent flaw lies in the fact that the assessment model cannot reflect the medical logical relationships between nursing characteristics over time and in the patient's condition, leading to a distorted depiction of risk trajectories and making it difficult to predict infection events caused by complex coupling of multiple factors.

[0003] Existing early warning mechanisms primarily rely on setting fixed thresholds for triggering comprehensive risk scores. This method can only indicate the level of risk, failing to identify clinically significant risk evolution patterns from continuous time-series data. Even if the risk value briefly rises, the system cannot distinguish between occasional fluctuations caused by non-infectious factors and persistent abnormal patterns indicating pre-infection symptoms, thus generating numerous invalid alerts or missing genuine early signals. Clinical practice requires a technology capable of automatically analyzing, quantifying, and filtering high-confidence clinical risk patterns from multi-dimensional time-series data to achieve a shift from coarse risk level indications to precise risk event warnings. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art;

[0005] Therefore, this invention proposes a nursing-based infection risk assessment method, comprising:

[0006] Obtain time-series sequences of multidimensional nursing characteristics associated with infection risk;

[0007] Multi-scale data perception is performed on the time-series sequence of the multi-dimensional nursing features to separate nursing feature components that reflect different risk sensitivities;

[0008] The risk status of each of the multiple nursing feature components is independently quantified to generate a risk quantification trajectory corresponding to each nursing feature component.

[0009] Based on a pre-defined infection risk knowledge graph, a dynamic correlation mapping network between nursing feature components is constructed;

[0010] Based on the dynamic correlation mapping network, the risk quantification trajectory of each nursing feature component is interactively corrected and fused to form a comprehensive risk evolution path;

[0011] The comprehensive risk evolution path is analyzed by time-series pattern analysis to identify risk pattern fragments that characterize potential infection events;

[0012] Calculate the pattern confidence of each risk pattern fragment, and filter and sort the risk pattern fragments based on the pattern confidence.

[0013] By integrating and sorting the risk pattern fragments, a complete infection risk early warning signal with time-series markers is reconstructed.

[0014] The infection risk warning signals are matched with a pre-set clinical intervention strategy library to output a structured risk assessment report and nursing decision recommendations.

[0015] Preferably, the step of performing multi-scale data perception on the time-series sequence of the multi-dimensional nursing features to separate nursing feature components reflecting different risk sensitivities includes:

[0016] Define a set of perception scale parameters, with each perception scale parameter corresponding to a risk sensitivity resolution granularity;

[0017] For each nursing feature dimension in the multi-dimensional nursing feature time sequence, convolutional filtering is performed using the set of perception scale parameters to generate a set of subsequences for each nursing feature dimension at multiple perception scales.

[0018] Collaborative clustering analysis was performed on the subsequences of all nursing feature dimensions under each perception scale, and the subsequences representing similar risk sensitivities were grouped into the same nursing feature component.

[0019] Redundancy removal is performed on all nursing feature components generated at all perception scales to ensure that each nursing feature component is unique in terms of risk sensitivity.

[0020] Preferably, the step of independently quantifying the risk status of multiple nursing feature components to generate a risk quantification trajectory corresponding to each nursing feature component includes:

[0021] Define a state space containing multiple discrete risk levels for each nursing feature component;

[0022] Based on historical infection event data, the observed data of each nursing feature component are learned to form a transition probability model for each risk level in the state space.

[0023] The time-series observation data of the nursing feature components are input into the corresponding transition probability model, and the probability distribution of each risk level at each time moment is calculated by the forward algorithm.

[0024] The risk level with the highest probability value is extracted from the probability distribution, and the risk quantification trajectory of the nursing feature component is formed by connecting them in time sequence.

[0025] Preferably, the step of constructing a dynamic association mapping network between nursing feature components based on a preset infection risk knowledge graph includes:

[0026] The infection risk knowledge graph includes entities, attributes, and relationships between entities. Entity types include pathogens, susceptible sites, clinical symptoms, and nursing procedures.

[0027] From the infection risk knowledge graph, extract the entities and relational subgraphs associated with all nursing feature components;

[0028] Analyze the topological connection strength and path of entities corresponding to each nursing feature component in the relation subgraph, and quantify the connection strength and path information into association weights;

[0029] Using nursing feature components as nodes and quantized association weights as the weights of directed edges, an initial static association mapping network is constructed.

[0030] By introducing a time decay function, the weights of directed edges in the static association mapping network are dynamically adjusted according to the rate of change of the risk quantification trajectory of the nursing feature components, thus forming a dynamic association mapping network.

[0031] Preferably, the step of interactively correcting and fusing the risk quantification trajectories of each nursing feature component based on the dynamic correlation mapping network to form a comprehensive risk evolution path includes:

[0032] Select a nursing feature component as the current correction component, and locate other nursing feature components that are related to the current correction component according to the dynamic association mapping network as related components;

[0033] Align and compare the risk quantization trajectory of the current correction component with the risk quantization trajectory of each associated component to detect inconsistent risk state transition points in the trajectory;

[0034] Based on the weights of edges in the dynamic association mapping network, confidence-weighted voting is performed on the detected inconsistent risk state transition points to correct transition points with low confidence in the current correction component risk quantification trajectory.

[0035] Iterate through all nursing feature components, and for each nursing feature component, perform the following correction process in sequence: select the current correction component, locate the associated component, perform trajectory alignment comparison, detect inconsistency risk state transition points, and perform weighted voting correction based on the association weight, until the risk quantification trajectory of all nursing feature components tends to stabilize.

[0036] The risk quantification trajectories of all corrected nursing feature components are weighted and superimposed according to the correlation relationships shown in the dynamic correlation mapping network to generate a comprehensive risk evolution path.

[0037] Preferably, the step of performing time-series pattern analysis on the comprehensive risk evolution path to identify risk pattern fragments characterizing potential infection events includes:

[0038] Define a set of risk model templates. Each risk model template describes the constraints that the comprehensive risk evolution path should meet in terms of form, magnitude, and duration before a specific infection event occurs.

[0039] The comprehensive risk evolution path is traversed using a sliding time window, and the morphological features, statistical features, and trend features of the path segments within each window are extracted.

[0040] The feature set extracted from each window is matched with all risk pattern templates to calculate similarity.

[0041] When the similarity between the feature set of a certain window and any risk pattern template exceeds a preset threshold, the path segment corresponding to the window is determined to be a risk pattern segment, and the risk pattern template type, start time, and end time that it matches are recorded.

[0042] Preferably, the step of calculating the pattern confidence of each risk pattern fragment and filtering and sorting the risk pattern fragments based on the pattern confidence includes:

[0043] For each identified risk pattern fragment, its pattern confidence is determined by the following factors: the similarity of matching with the risk pattern template, the magnitude significance of the risk pattern fragment in the comprehensive risk evolution path, and the duration of the risk pattern fragment.

[0044] Weight coefficients were assigned to matching similarity, magnitude significance, and duration length, and the pattern confidence value of each risk pattern segment was calculated by weighted summation.

[0045] Set a pattern confidence threshold to filter out risky pattern segments with a pattern confidence value lower than the pattern confidence threshold;

[0046] The risk pattern fragments retained after screening are sorted from high to low according to their pattern confidence values, and for risk pattern fragments that overlap in time, the fragment with the highest pattern confidence value is retained.

[0047] Preferably, the integrated, filtered, and sorted risk pattern fragments are reconstructed into a complete, time-stamped infection risk early warning signal, including:

[0048] Based on the timeline, the filtered and sorted risk pattern segments are arranged according to their recorded start and end times;

[0049] For risk pattern segments that are adjacent on the timeline and have the same risk pattern type, the segments are merged to form a longer continuous warning interval;

[0050] Assign a globally unique warning identifier to each warning interval or independent risk pattern segment, and label its corresponding risk pattern template type, pattern confidence level, and time range;

[0051] In chronological order, all warning intervals and independent risk pattern segments and their annotation information are integrated to generate a structured infection risk warning signal, which contains a series of warning event records ordered by time.

[0052] Preferably, the step of matching the infection risk warning signal with a preset clinical intervention strategy library to output a structured risk assessment report and nursing decision recommendations includes:

[0053] The clinical intervention strategy library stores a set of nursing intervention measures corresponding to different risk patterns, different risk levels, and different patient baseline conditions;

[0054] Analyze the warning event records in the infection risk warning signal to extract its risk pattern type, time range, and pattern confidence level;

[0055] Based on the risk pattern type and pattern confidence level recorded in the early warning event records, a matching set of nursing intervention candidates is retrieved from the clinical intervention strategy database;

[0056] By combining the real-time basic information of the nursing subjects, the most applicable intervention measures are selected from the candidate set of nursing intervention measures, and specific nursing decision-making suggestions are generated.

[0057] The system summarizes all warning event records, corresponding pattern confidence levels, and generated nursing decision recommendations from the current infection risk warning signals, and packages them according to a preset report format to generate a structured risk assessment report.

[0058] Preferably, the present invention also includes a nursing-based infection risk assessment system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the nursing-based infection risk assessment method described above.

[0059] Compared with the prior art, the beneficial effects of the present invention are:

[0060] This technology constructs a dynamic correlation mapping network between nursing feature components using a pre-defined infection risk knowledge graph, and interactively corrects and integrates the quantified trajectories of each independent risk based on this network. This approach enables the assessment process to simulate the complex reasoning involving multiple factors in clinical decision-making, with the risk contribution of different feature components dynamically adjusted in real time according to their medical relevance. The resulting comprehensive risk evolution path is no longer a simple superposition of static weights, but possesses dynamism and adaptability that reflects the interaction between real pathophysiology and nursing interventions, improving the realism of risk evolution representation and the accuracy of tracking in complex clinical scenarios.

[0061] This approach analyzes the temporal pattern of the integrated risk evolution path to identify and extract risk pattern fragments representing potential infection events. It then calculates the pattern confidence score for each fragment and filters and ranks them accordingly. This scheme transforms continuous risk curves into discrete clinically suspected events. The system proactively captures abnormal patterns consistent with clinical understanding, such as "persistent low-grade fever accompanied by specific fluctuations in inflammatory markers," and assigns a quantified confidence score to each identified pattern. The final reconstructed warning signal consists of a series of risk event fragments with clear temporal markers, ranked by confidence score. This ensures that the warning output directly corresponds to specific and credible clinical abnormal scenarios, enhancing the relevance and interpretability of risk alerts and effectively assisting clinicians in root cause investigation and precise intervention. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating the steps of the nursing-based infection risk assessment method described in this invention.

[0063] Figure 2 A flowchart for multi-scale data perception;

[0064] Figure 3 A flowchart for constructing a dynamic association mapping network;

[0065] Figure 4 A bar-line composite plot of confidence analysis for infection risk pattern fragments;

[0066] Figure 5 This is a comprehensive chart showing the evolution trend of infection risk and the warning range. Detailed Implementation

[0067] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0068] See Figure 1 The method acquires a time-series sequence of multi-dimensional nursing characteristics associated with infection risk. These nursing characteristics may include observational data changing over time, such as body temperature, white blood cell count, respiratory secretion characteristics, surgical incision condition, catheter care records, and hand hygiene practices. Multi-scale data perception is performed on this time-series sequence to separate nursing characteristic components reflecting different risk sensitivities from the raw data. For example, components reflecting long-term, slow risk trends and components reflecting short-term, acute risk fluctuations are identified. Each separated nursing characteristic component undergoes independent risk state quantification, transforming continuous observations of each component into a discrete risk level sequence, thereby generating a risk quantification trajectory corresponding to each nursing characteristic component. Based on a pre-defined infection risk knowledge graph, which encodes the medical logical relationships between entities such as pathogens, susceptible sites, and nursing procedures, a dynamic correlation mapping network is constructed between nursing characteristic components. This network characterizes the mutual influence strength of different components in risk assessment. Based on the dynamic correlation mapping network, the risk quantification trajectories of each nursing feature component are interactively corrected and fused. By comparing and correcting inconsistent risk transitions between correlated components, a comprehensive risk evolution path that reflects multi-dimensional information is formed. The comprehensive risk evolution path is then subjected to temporal pattern analysis. By matching predefined risk pattern templates, risk pattern segments representing potential infection events are identified. The pattern confidence of each risk pattern segment is calculated, which integrates factors such as matching degree and risk amplitude. Based on the pattern confidence, all identified risk pattern segments are screened and sorted, filtering out low-confidence segments and resolving overlapping issues between segments. The screened and sorted risk pattern segments are integrated, arranged on a timeline, and merged as necessary to reconstruct a complete infection risk warning signal with clear temporal markers for each segment. The infection risk warning signal is matched with a pre-set clinical intervention strategy library, which associates different risk patterns with specific nursing measures, outputting a structured risk assessment report and personalized nursing decision recommendations.

[0069] In one embodiment of the present invention, see [reference] Figure 2This study employs multi-scale data perception to separate nursing feature components from multi-dimensional nursing feature time-series sequences, and independently quantifies the risk status of each nursing feature component to generate risk quantification trajectories. It acquires multi-dimensional nursing feature time-series sequences associated with infection risk, including time-series data on vital signs, laboratory test results, invasive procedure records, and environmental monitoring indicators. A set of perception scale parameters is defined, including a short-term scale with a 4-hour period, a medium-term scale with a 24-hour period, and a long-term scale with a 7-day period. Each perception scale parameter corresponds to a risk sensitivity analysis granularity; the short-term scale analyzes acute and rapid changes, while the long-term scale analyzes chronic trend evolution. For each nursing feature dimension in the multi-dimensional nursing feature time series, convolutional filtering is performed using a set of perception scale parameters to generate a set of subsequences for each nursing feature dimension at multiple perception scales. For the body temperature dimension, a 4-hour convolutional kernel filter is used to obtain a subsequence reflecting intraday fluctuations, a 24-hour convolutional kernel filter is used to obtain a subsequence reflecting diurnal trends, and a 7-day convolutional kernel filter is used to obtain a subsequence reflecting weekly trends. Co-clustering analysis is performed on all subsequences of nursing feature dimensions at each perception scale, grouping subsequences representing similar risk sensitivities into the same nursing feature component. At the 4-hour scale, the intraday fluctuation subsequence of body temperature, the heart rate variability subsequence, and the instantaneous fluctuation subsequence of blood pressure are clustered into a single nursing feature component reflecting physiological acute stress. Redundancy removal is performed on all nursing feature components generated at all perception scales. By calculating the correlation coefficient between components and setting a threshold, highly redundant components with correlation coefficients exceeding the threshold are removed to ensure that each nursing feature component is unique in terms of risk sensitivity.

[0070] In some embodiments, a state space containing multiple discrete risk levels is defined for each nursing feature component. The state space consists of three ordered levels: "low risk," "medium risk," and "high risk." Based on historical infection event data, a transition probability model from the observed data of each nursing feature component to each risk level in the state space is learned. The transition probability model is trained using a Hidden Markov Model, and its state transition matrix A and observation probability matrix B are estimated from labeled historical time-series data using the Baum-Welch algorithm. The time-series observed data of the nursing feature component are input into the corresponding transition probability model, and the probability distribution of being at each risk level at each time step is calculated using a forward algorithm. The recursive formula for the forward algorithm is:

[0071]

[0072] in: Indicates at time Observed sequence And in a state The probability, It is the total number of states. From state Transition to state The probability, It is in state The following observations The probability, It is a moment Nursing characteristic component observation values.

[0073] In one embodiment of the present invention, see [reference] Figure 3 This study constructs a dynamic association mapping network based on an infection risk knowledge graph, and interactively corrects and fuses the risk quantification trajectories of each nursing feature component based on this network. The infection risk knowledge graph contains entities, attributes, and relationships between entities. Entity types include pathogens, susceptible sites, clinical symptoms, and nursing procedures. From the infection risk knowledge graph, entities and relationship subgraphs associated with all nursing feature components are extracted. If there are nursing feature components such as "central venous catheter maintenance compliance" and "body temperature fluctuation," then subgraphs containing entities such as "central venous catheter," "bloodstream infection," "bacteremia," and "fever," along with relationships such as "may lead to" and "clinical manifestations," are extracted. The topological connection strength and path of the entities corresponding to each nursing feature component in the relationship subgraphs are analyzed. Connection strength is obtained by calculating the reciprocal of the shortest path between entities and combining it with the preset weights of the relationship types, quantifying the connection strength and path information into an association weight value between 0 and 1. Using nursing feature components as nodes and quantified association weights as the weights of directed edges, an initial static association mapping network is constructed. In this network, the weight of the directed edge pointing from the "Central Venous Catheter Maintenance Compliance" node to the "Body Temperature Fluctuation" node may be 0.7. A time decay function is introduced, which dynamically adjusts the weights based on the rate of change of the risk quantification trajectory of the nursing feature components. The rate of change is obtained by calculating the frequency and magnitude of risk level changes in the trajectory within the most recent time window, forming a dynamic association mapping network. A specific form of the time decay function is:

[0074]

[0075] in: These are the adjusted directed edge weights in a dynamic association mapping network. These are the initial directed edge weights in a static association mapping network. It is a time decay factor parameter greater than 0. This represents the difference in risk level of the risk quantification trajectory of the nursing characteristic component pointed to by the associated component within a recent fixed time interval. Take its absolute value.

[0076] In some embodiments, a nursing feature component is selected as the current correction component, such as "white blood cell count trend". Other nursing feature components associated with the current correction component are located using a dynamic association mapping network. The network shows that the "body temperature" and "procalcitonin level" nursing feature components have directed edge connections with the "white blood cell count trend" nursing feature component, with weights of 0.6 and 0.8 respectively. Therefore, "body temperature" and "procalcitonin level" are identified as associated components. The risk quantization trajectory of the current correction component is compared with the risk quantization trajectories of each associated component in time alignment to detect inconsistent risk state transition points. In the risk quantization trajectory of the current correction component "white blood cell count trend", time t is marked as a transition from "medium risk" to "low risk", but in the risk quantization trajectory of the associated component "procalcitonin level", the state remains "high risk" near time t. This time t is detected as an inconsistent risk state transition point. Based on the edge weights in the dynamic association mapping network, a confidence-weighted voting process is used to evaluate the detected inconsistent risk state transitions. Higher edge weights result in greater voting weights for their corresponding associated component trajectories, thus correcting transitions with lower confidence in the current correction component's risk quantification trajectory. For example, if the association weight of "procalcitonin level" (0.8) is higher than that of "body temperature" (0.6), the weighted voting result may support correcting the "white blood cell count trend" state at time t to "medium risk" instead of "low risk." All nursing feature components are traversed, and for each component, a correction process is sequentially performed, including selecting the current correction component, locating associated components, performing trajectory alignment comparison, detecting inconsistent risk state transitions, and weighted voting correction based on association weights. This process continues until the risk quantification trajectories of all nursing feature components no longer change in two consecutive correction rounds, indicating stability. The corrected risk quantification trajectories of all nursing feature components are then weighted and superimposed according to the association relationships shown in the dynamic association mapping network. The weighting coefficient of each nursing feature component node is the normalized value of all its in-degree weights in the dynamic association mapping network, generating a comprehensive risk evolution path.

[0077] In one embodiment of the present invention, a time-series pattern analysis is performed on the comprehensive risk evolution path to identify risk pattern segments. A set of risk pattern templates is defined, which describe the constraints that the comprehensive risk evolution path should meet in terms of morphology, amplitude, and duration before a specific infection event occurs. For example, the "ventilator-associated pneumonia risk pattern" template may stipulate that the path morphology must show a continuous upward trend, the amplitude must exceed the risk threshold of 0.6, and the duration must be maintained for at least 12 hours. A sliding time window is used to traverse the comprehensive risk evolution path. The length of the sliding time window is set to 6 hours and the step size is set to 1 hour. The morphological features, statistical features, and trend features of the path segments within each window are extracted. The morphological features include the slope, curvature, and number of extreme points of the path segment. The statistical features include the mean, variance, and maximum risk value of the path segment. The trend features include the mean of the first difference and the frequency of the sign change of the second difference of the path segment. The feature set extracted from each window is compared with all risk pattern templates for similarity matching calculation. The similarity matching calculation is achieved by comparing the spatial distance or angle between the window feature vector and the template feature vector.

[0078] In some embodiments, the similarity matching calculation uses the cosine similarity formula, which is expressed as:

[0079]

[0080] in: Represents the feature set of the ω-th sliding time window The feature vector of the κ-th risk pattern template The similarity value between them It is a feature set The normalized value of the ι-th eigencomponent, It is a risk model template The normalized value of the ι-th eigencomponent, It is the total number of dimensions of the feature set.

[0081] It is understood that the constraints of the risk pattern template can be derived from historical infection event data through expert experience or machine learning methods. Optionally, the length and step size of the sliding time window can be adjusted according to the expected development speed of different infection risk patterns; for rapidly progressing infection risks, the window length can be shortened to 2 hours. In some embodiments, the feature extraction process also includes standardization preprocessing of path segments to eliminate the influence of different risk value dimensions and ranges. It is understood that similarity matching calculation can be implemented using the reciprocal of Euclidean distance or Pearson correlation coefficient, in addition to cosine similarity. Optionally, the preset threshold can be set with different values ​​for different risk pattern template types to improve the specificity of identification. When multiple sliding time windows match the same risk pattern template and are continuous or overlapping in time, these windows can be merged into a longer risk pattern segment for processing.

[0082] In one embodiment of the present invention, the pattern confidence of risk pattern fragments is calculated, and the risk pattern fragments are screened and sorted based on the pattern confidence. For each identified risk pattern fragment, its pattern confidence is determined by three factors: the matching similarity with the risk pattern template, the magnitude significance of the risk pattern fragment in the comprehensive risk evolution path, and the duration of the risk pattern fragment. Matching similarity refers to the feature similarity value between the fragment and the matched risk pattern template; magnitude significance refers to the degree to which the peak risk value of the fragment exceeds the global baseline risk level in the comprehensive risk evolution path; and duration refers to the time span from the start time to the end time of the fragment. Weighting coefficients are assigned to matching similarity, magnitude significance, and duration, respectively. The weighting coefficients are determined by expert scoring or regression analysis based on the incidence rate of historical infection events. The pattern confidence value of each risk pattern fragment is calculated by weighted summation. The formula for calculating the pattern confidence value is expressed as:

[0083]

[0084] in: This represents the final calculated pattern confidence score. This represents the normalized matching similarity factor. This represents the magnitude significance factor after normalization. This represents the normalized duration factor. These are the weighting coefficients of the matching similarity factor. It is the weighting coefficient of the magnitude significance factor. It is the weighting coefficient of the duration factor, and satisfies .

[0085] A pattern confidence threshold of 0.75 is set to filter out risky pattern segments with a pattern confidence value lower than the threshold. The remaining risky pattern segments are then sorted in descending order of their pattern confidence values. For risky pattern segments with temporal overlap, the segment with the highest pattern confidence value is retained, and other overlapping segments are removed. Refer to Table 1. The following example illustrates the specific filtering and sorting process, assuming five identified risky pattern segments, their relevant information, and calculated pattern confidence values.

[0086] Table 1: Risk Pattern Fragment Screening Table

[0087]

[0088] In some embodiments, filtering is performed based on a pattern confidence threshold of 0.75. The pattern confidence value of risk pattern fragment P003 is 0.70, which is lower than the threshold of 0.75, so it is filtered out. The remaining fragments P001, P002, P004, and P005 are sorted from highest to lowest pattern confidence value, resulting in P001 (0.92), P002 (0.88), P004 (0.81), and P005 (0.79). The time overlap between the sorted fragments is checked. Fragment P001 (time range 08:00-16:00) overlaps with fragments P002 (time range 10:00-18:00) and P004 (time range 15:00-19:00). P001 has the highest pattern confidence value of 0.92 among the three, so P001 is retained, and P002 and P004 are removed. The time range of fragment P005 (22:00 to 06:00 the next day) does not overlap with the retained fragment P001, so it is retained. Finally, after screening and sorting, the risk pattern fragments retained are P001 and P005.

[0089] Understandably, normalization maps the original values ​​of the three factors—match similarity, magnitude significance, and duration—to the interval [0,1]. Optionally, weighting coefficients... , , Different configurations can be made based on the risk assessment focus of different clinical departments. In some embodiments, for time-overlapping segments with different risk pattern types, if their pattern confidence values ​​are all above a threshold and differ within a very small range, a composite risk warning can be issued simultaneously. It is understood that the pattern confidence threshold can be dynamically adjusted according to the system's need for balancing warning sensitivity and specificity. Optionally, an amplitude significance factor... The calculation can be determined based on the ratio of the peak value of the comprehensive risk evolution path within the risk pattern segment to the preset departmental or patient's historical risk baseline value.

[0090] See Figure 4 This is a bar-line composite chart analyzing the confidence levels of five risk pattern segments. It clearly shows the confidence composition and selection criteria for each of the five risk pattern segments, representing a crucial analytical step in nursing-based infection risk assessment. P003 was excluded because all three normalization factors were at low levels, resulting in a composite confidence level below the warning threshold of 0.75. P001 had the highest confidence level because its matching similarity and amplitude significance were the highest among all segments, and its duration was also the longest. The synergistic effect of these three factors made it the most reliable risk warning signal. This chart provides clinicians with quantitative risk assessment data, enabling them not only to identify which risk patterns require attention but also to understand the composition of confidence levels, thus allowing for more precise development of nursing intervention strategies.

[0091] In one embodiment of the present invention, the risk pattern fragments after screening and sorting are integrated and reconstructed into an infection risk warning signal. The infection risk warning signal is then matched with a preset clinical intervention strategy library to output a structured risk assessment report and nursing decision recommendations. Based on the time axis, the screened and sorted risk pattern fragments are arranged according to their recorded start and end times. Assuming that the risk pattern fragments retained after processing in the embodiment are P001 and P005, fragment P001 has a start time of 08:00 on 2023-10-05 and an end time of 16:00 on 2023-10-05, and fragment P005 has a start time of 22:00 on 2023-10-05 and an end time of 06:00 on 2023-10-06. They are arranged on the time axis in chronological order, with P001 first and P005 second. For risk pattern segments that are adjacent on the timeline and have the same risk pattern type, the segments are merged to form a longer continuous warning interval. In this example, segments P001 and P005 have different risk pattern types and the time interval is more than 6 hours, so they are not merged and each is an independent warning interval. Each warning interval or independent risk pattern segment is assigned a globally unique warning identifier, and its corresponding risk pattern template type, pattern confidence level, and time range are labeled. For example, the warning identifier "Alert-20231005-001" is assigned to segment P001, with the risk pattern template type labeled as "catheter-related bloodstream infection risk pattern", the pattern confidence level as 0.92, and the time range as "2023-10-05 08:00 to 2023-10-05 16:00"; and the warning identifier "Alert-20231005-002" is assigned to segment P005, with the risk pattern template type labeled as "catheter-related urinary tract infection risk pattern", the pattern confidence level as 0.79, and the time range as "2023-10-05 22:00 to 2023-10-06 06:00". In chronological order, all warning intervals and independent risk pattern segments and their annotation information are integrated to generate a structured infection risk warning signal. The infection risk warning signal contains a series of warning event records ordered by time. Its data structure can be represented as an ordered list. Each element in the list contains fields such as warning identifier, risk pattern type, start time, end time, and pattern confidence.

[0092] The clinical intervention strategy database stores sets of nursing interventions corresponding to different risk pattern types, risk levels, and patient baseline conditions. For example, under the risk pattern type "catheter-related bloodstream infection risk pattern," the set of nursing interventions corresponding to patients with "high risk level" and "neutropenia" baseline conditions might include specific measures such as "immediately collect bilateral peripheral blood cultures," "assess the necessity of central venous catheter placement and prepare for removal," and "empirical administration of vancomycin." The database analyzes the warning event records in the infection risk warning signals, extracting their risk pattern type, time range, and pattern confidence. From the warning event record "Alert-20231005-001," the database extracts the risk pattern type as "catheter-related bloodstream infection risk pattern," the time range as "2023-10-05 08:00 to 2023-10-05 16:00," and the pattern confidence as 0.92. Based on the risk pattern type and pattern confidence level recorded in the early warning event records, a matching set of nursing intervention candidates is retrieved from the clinical intervention strategy database. A pattern confidence level of 0.92 is mapped to "high risk level". Using "catheter-related bloodstream infection risk pattern" and "high risk level" as joint query conditions, the corresponding set of nursing intervention candidates is retrieved from the clinical intervention strategy database.

[0093] In some embodiments, by combining the real-time baseline information of the nursing patient, the most applicable intervention is selected from a candidate set of nursing interventions to generate specific nursing decision recommendations. The real-time baseline information of the nursing patient includes age, immune status, renal function, history of drug allergies, etc. The selection process involves calculating a matching score between each candidate intervention and the current baseline information of the nursing patient. The calculation formula is:

[0094]

[0095] in: This represents a score indicating the overall match between a candidate intervention and the current patient. It is the total number of dimensions for basic situation assessment. It is the first The matching coefficient for each evaluation dimension (1 for a match, 0.5 for a partial match, and 0 for no match). It is the first Preset weighting coefficients for each evaluation dimension in the overall evaluation. Select matching score. The highest-ranking candidate intervention is selected as the most applicable intervention, and specific nursing decision recommendations are generated accordingly. Considering the patient's current baseline condition of "moderate renal insufficiency," "assess the necessity of central venous catheter placement and prepare for removal" and "adjust vancomycin dosage and administer medication based on creatinine clearance" are selected from the candidate set as specific nursing decision recommendations. All warning event records, corresponding pattern confidence levels, and generated nursing decision recommendations in the current infection risk warning signal are summarized and packaged according to a preset report format to generate a structured risk assessment report. The preset report format includes structured fields such as patient identifier, report generation time, warning event list, overall risk assessment level, and sub-item nursing decision recommendations.

[0096] It is understandable that the generation rules for warning identifiers can combine timestamps, serial numbers, and patient identifiers to ensure global uniqueness. Optionally, for time-closely adjacent segments of the same type of risk pattern separated by brief low-risk intervals, a time interval tolerance threshold can be set; if the interval is less than the threshold, they are merged. In some embodiments, the retrieval of the candidate set of nursing interventions in the clinical intervention strategy library can be further filtered based on the nursing shift or ward workflow within which the warning time range falls.

[0097] See Figure 5 This is a comprehensive infection risk evolution trend and warning interval chart, fully presenting the dynamic changes in a patient's infection risk over approximately 36 hours, and accurately marking two types of high-risk warning periods. It clearly demonstrates the occurrence, development, and decline of risk, helping healthcare professionals understand the temporal patterns of infection risk. Different colored warning intervals clearly distinguish infection types, providing precise direction for clinical intervention. Around 10:00-05:12:00 in the pink warning interval, the risk value reaches its peak, representing the optimal time to implement preventative interventions such as catheter care and assess extubation indications. It can be used to evaluate the effectiveness of infection control measures. For example, if the peak value of the risk curve decreases or its duration shortens after an intervention, the clinical value of that measure can be quantitatively demonstrated.

[0098] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A nursing-based infection risk assessment method, characterized in that, include: Obtain time-series sequences of multidimensional nursing characteristics associated with infection risk; Multi-scale data perception is performed on the time-series sequence of the multi-dimensional nursing features to separate nursing feature components that reflect different risk sensitivities; The risk status of each of the multiple nursing feature components is independently quantified to generate a risk quantification trajectory corresponding to each nursing feature component. Based on a pre-defined infection risk knowledge graph, a dynamic correlation mapping network between nursing feature components is constructed; Based on the dynamic correlation mapping network, the risk quantification trajectories of each nursing characteristic component are interactively corrected and fused to form a comprehensive risk evolution path, including: Select a nursing feature component as the current correction component, and locate other nursing feature components that are related to the current correction component according to the dynamic association mapping network as related components; Align and compare the risk quantization trajectory of the current correction component with the risk quantization trajectory of each associated component to detect inconsistent risk state transition points in the trajectory; Based on the weights of edges in the dynamic association mapping network, confidence-weighted voting is performed on the detected inconsistent risk state transition points to correct transition points with low confidence in the current correction component risk quantification trajectory. Iterate through all nursing feature components, and for each nursing feature component, perform the following correction process in sequence: select the current correction component, locate the associated component, perform trajectory alignment comparison, detect inconsistency risk state transition points, and perform weighted voting correction based on the association weight, until the risk quantification trajectory of all nursing feature components tends to stabilize. The risk quantification trajectories of all corrected nursing feature components are weighted and superimposed according to the correlation relationships shown in the dynamic correlation mapping network to generate a comprehensive risk evolution path. The comprehensive risk evolution path is analyzed using time-series pattern analysis to identify risk pattern fragments characterizing potential infection events, including: Define a set of risk model templates. Each risk model template describes the constraints that the comprehensive risk evolution path should meet in terms of form, magnitude, and duration before a specific infection event occurs. The comprehensive risk evolution path is traversed using a sliding time window, and the morphological features, statistical features, and trend features of the path segments within each window are extracted. The feature set extracted from each window is matched with all risk pattern templates to calculate similarity. When the similarity between the feature set of a certain window and any risk pattern template exceeds a preset threshold, the path segment corresponding to the window is determined to be a risk pattern segment, and its matched risk pattern template type, start time, and end time are recorded. Calculate the pattern confidence of each risk pattern fragment, and filter and sort the risk pattern fragments based on the pattern confidence. By integrating and sorting the risk pattern fragments, a complete infection risk early warning signal with time-series markers is reconstructed. The infection risk warning signals are matched with a pre-set clinical intervention strategy library to output a structured risk assessment report and nursing decision recommendations.

2. The infection risk assessment method based on nursing care according to claim 1, characterized in that, The process of performing multi-scale data perception on the time-series sequence of the multi-dimensional nursing characteristics to separate nursing characteristic components reflecting different risk sensitivities includes: Define a set of perception scale parameters, with each perception scale parameter corresponding to a risk sensitivity resolution granularity; For each nursing feature dimension in the multi-dimensional nursing feature time sequence, convolutional filtering is performed using the set of perception scale parameters to generate a set of subsequences for each nursing feature dimension at multiple perception scales. Collaborative clustering analysis was performed on the subsequences of all nursing feature dimensions under each perception scale, and the subsequences representing similar risk sensitivities were grouped into the same nursing feature component. Redundancy removal is performed on all nursing feature components generated at all perception scales to ensure that each nursing feature component is unique in terms of risk sensitivity.

3. The infection risk assessment method based on nursing care according to claim 2, characterized in that, The step of independently quantifying the risk status of multiple nursing feature components to generate a risk quantification trajectory corresponding to each nursing feature component includes: Define a state space containing multiple discrete risk levels for each nursing feature component; Based on historical infection event data, the observed data of each nursing feature component are learned to form a transition probability model for each risk level in the state space. The time-series observation data of the nursing feature components are input into the corresponding transition probability model, and the probability distribution of each risk level at each time moment is calculated by the forward algorithm. The risk level with the highest probability value is extracted from the probability distribution, and the risk quantification trajectory of the nursing feature component is formed by connecting them in time sequence.

4. The infection risk assessment method based on nursing care according to claim 3, characterized in that, The process of constructing a dynamic correlation mapping network between nursing feature components based on a preset infection risk knowledge graph includes: The infection risk knowledge graph includes entities, attributes, and relationships between entities. Entity types include pathogens, susceptible sites, clinical symptoms, and nursing procedures. From the infection risk knowledge graph, extract the entities and relational subgraphs associated with all nursing feature components; Analyze the topological connection strength and path of entities corresponding to each nursing feature component in the relation subgraph, and quantify the connection strength and path information into association weights; Using nursing feature components as nodes and quantized association weights as the weights of directed edges, an initial static association mapping network is constructed. By introducing a time decay function, the weights of directed edges in the static association mapping network are dynamically adjusted according to the rate of change of the risk quantification trajectory of the nursing feature components, thus forming a dynamic association mapping network.

5. The infection risk assessment method based on nursing care according to claim 4, characterized in that, The calculation of the pattern confidence score for each risk pattern fragment, and the filtering and sorting of risk pattern fragments based on the pattern confidence scores, includes: For each identified risk pattern fragment, its pattern confidence is determined by the following factors: the similarity of matching with the risk pattern template, the magnitude significance of the risk pattern fragment in the comprehensive risk evolution path, and the duration of the risk pattern fragment. Weight coefficients were assigned to matching similarity, magnitude significance, and duration length, and the pattern confidence value of each risk pattern segment was calculated by weighted summation. Set a pattern confidence threshold to filter out risky pattern segments with a pattern confidence value lower than the pattern confidence threshold; The risk pattern fragments retained after screening are sorted from high to low according to their pattern confidence values, and for risk pattern fragments that overlap in time, the fragment with the highest pattern confidence value is retained.

6. The infection risk assessment method based on nursing care according to claim 5, characterized in that, The integrated, filtered, and sorted risk pattern fragments are reconstructed into a complete, time-stamped infection risk early warning signal, including: Based on the timeline, the filtered and sorted risk pattern segments are arranged according to their recorded start and end times; For risk pattern segments that are adjacent on the timeline and have the same risk pattern type, the segments are merged to form a longer continuous warning interval; Assign a globally unique warning identifier to each warning interval or independent risk pattern segment, and label its corresponding risk pattern template type, pattern confidence level, and time range; In chronological order, all warning intervals and independent risk pattern segments and their annotation information are integrated to generate a structured infection risk warning signal, which contains a series of warning event records ordered by time.

7. The infection risk assessment method based on nursing care according to claim 6, characterized in that, The process of matching the infection risk warning signal with a pre-set clinical intervention strategy library to output a structured risk assessment report and nursing decision recommendations includes: The clinical intervention strategy library stores a set of nursing intervention measures corresponding to different risk patterns, different risk levels, and different patient baseline conditions; Analyze the warning event records in the infection risk warning signal to extract its risk pattern type, time range, and pattern confidence level; Based on the risk pattern type and pattern confidence level recorded in the early warning event records, a matching set of nursing intervention candidates is retrieved from the clinical intervention strategy database; By combining the real-time basic information of the nursing subjects, the most applicable intervention measures are selected from the candidate set of nursing intervention measures, and specific nursing decision-making suggestions are generated. The system summarizes all warning event records, corresponding pattern confidence levels, and generated nursing decision recommendations from the current infection risk warning signals, and packages them according to a preset report format to generate a structured risk assessment report.

8. A nursing-based infection risk assessment system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of a nursing-based infection risk assessment method as described in any one of claims 1 to 7.

Images