Basic nursing risk intelligent identification and early warning system and method

Abstract

The invention relates to the technical field of intelligent medical treatment, in particular to a basic nursing risk intelligent identification and early warning system and method. The system comprises a multi-source data perception and fusion module which is used for acquiring and fusing dominant risk data and implicit risk data of a patient in real time and outputting a structured patient state feature set; the risk intelligent identification core module simulates physiological state changes through a patient personalized digital twinborn model and is coupled with a risk conduction knowledge graph to deduce a risk evolution trajectory so as to realize intelligent identification of potential risks and intervention measure simulation; the intelligent early warning and intervention module is used for executing hierarchical early warning and active intervention linkage; and the man-machine collaborative feedback and self-evolution module continuously optimizes the system model and strategy through clinical feedback data. The problems that in the prior art, the risk perception dimension is single, early warning lags behind and intervention is disjointed are solved, and prospective and precise recognition, prevention and control of the nursing risk are achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical technology, and more specifically, to a basic nursing risk intelligent identification and early warning system and method. Background Technology

[0002] Basic nursing care is a core component in ensuring patient safety and treatment effectiveness. The prevention and control of nursing risk events such as pressure ulcers, falls from bed, and infusion reactions are long-standing key and challenging issues in clinical practice.

[0003] Current basic nursing risk identification relies primarily on manual assessment, using standardized scales to quantify high-risk factors such as falls and pressure ulcers. Existing basic nursing risk early warning systems are centered on tiered responses, combining static thresholds with dynamic monitoring to classify risks into high, medium, and low levels. Data linkage is achieved through hospital information systems and IoT devices, automatically triggering alarms when indicators are abnormal; in some scenarios, big data is used to capture risk deterioration trends. A "identification-early warning-treatment-feedback" mechanism is formed based on closed-loop reporting of adverse events and root cause analysis optimization strategies. However, the overall approach remains primarily passive, lacking forward-looking risk evolution simulation. Therefore, this paper proposes an intelligent basic nursing risk identification and early warning system and method.

[0004] The existing technology has the following technical defects, specifically:

[0005] 1. Limited data perception dimensions and lagging risk identification: Existing systems mainly rely on static assessment scales and macroscopic vital signs manually entered by nurses, which are "explicit data." They cannot capture "implicit risk data" such as patient behavior, posture, and facial expressions in real time and automatically. As a result, the system cannot detect weak, early physiological and behavioral abnormal signals before risks occur, leading to a serious lag in risk identification.

[0006] 2. Risk assessment is passive and static, lacking forward-looking projection: Existing technologies mostly adopt a passive "assessment-record-response" model, which is essentially a static pattern matching based on historical data. The system cannot proactively simulate and project the future evolution trajectory of risks based on the current state, nor can it predict "what might happen if the patient gets out of bed." Therefore, it cannot achieve true "early warning," but can only provide "alarms" after the fact or during the event.

[0007] 3. Disconnect between early warning and intervention decision-making, hindering system self-evolution: Existing systems typically only issue simple "risk" alerts, failing to provide personalized intervention recommendations based on specific risk causes, resulting in a severe disconnect between early warning and decision support. Furthermore, the systems are static and closed, unable to continuously learn and optimize their models and knowledge bases from feedback from healthcare professionals, leading to stagnant intelligence and difficulty adapting to personalized clinical scenarios. Summary of the Invention

[0008] The purpose of this invention is to provide a basic nursing risk intelligent identification and early warning system and method to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention aims to provide a basic nursing risk intelligent identification and early warning system, including: a multi-source data perception and fusion module, used to acquire and fuse multi-source heterogeneous data of patients in real time, and output a structured patient status feature set.

[0010] The core module for intelligent risk identification is used to identify risks based on a structured set of patient status features through simulation and deduction.

[0011] The intelligent early warning and intervention module is used to perform graded early warning and proactive intervention based on the output of the risk intelligent identification core module.

[0012] The human-machine collaborative feedback and self-evolution module is used to collect user feedback on the risk simulation results and the execution and effect of intervention strategies, and to use the feedback data to simultaneously optimize the risk intelligent identification core module and the intelligent early warning and intervention module.

[0013] As a further improvement to this technical solution, the patient's multi-source heterogeneous data includes explicit risk data and implicit risk data.

[0014] The explicit risk data includes at least the patient's vital signs, medical orders, and nursing records.

[0015] The hidden risk data includes at least patient behavior and posture data and facial expression data obtained based on computer vision analysis.

[0016] As a further improvement to this technical solution, the core module for intelligent risk identification includes: a patient-personalized digital twin unit and a risk extrapolation and intervention simulation unit.

[0017] The patient-personalized digital twin unit is used to construct and update a parameterized patient physiological simulation model in real time based on the patient state feature set.

[0018] The risk extrapolation and intervention simulation unit, which is coupled with the digital twin unit and the risk transmission knowledge graph, is used to simulate the risk evolution trajectory, identify potential nursing risks, and simulate the effects of candidate intervention measures.

[0019] As a further improvement to this technical solution, the patient physiological simulation model is specifically implemented by: establishing a parameterized basic physiological mechanism model, which includes a series of adjustable physiological parameters.

[0020] Obtain the patient's static baseline characteristics, and based on the patient's static baseline characteristics, initialize and assign values ​​to the adjustable physiological parameters in the basic physiological mechanism model to form an initial personalized model.

[0021] Model calibration is performed using real-time data. The dynamic trend features of the patient state feature set are used as input. Through data assimilation algorithm, the adjustable physiological parameters in the initial personalized model are continuously optimized to minimize the error between the model's output value and the patient's actual real-time vital sign monitoring value, thus obtaining a calibrated high-fidelity physiological simulation model.

[0022] As a further improvement to this technical solution, the risk transmission knowledge graph is specifically implemented as follows: Constructing a basic medical ontology: Based on the standard medical terminology system, a three-layer ontology structure containing nursing risk factors, physiological intermediate states, and nursing risk events is constructed as the entity framework of the knowledge graph.

[0023] Establish multi-order risk transmission relationships: Based on clinical medical guidelines and expert knowledge, establish directed causal transmission edges between entities in the ontology framework. The transmission edges include at least first-order transmission from risk factors to physiological intermediate states and second-order transmission from physiological intermediate states to risk events.

[0024] Quantifying the strength of transmission relationships: Based on historical clinical data, statistical analysis is performed to assign initial probability weights and confidence levels to each causal transmission edge, transforming qualitative knowledge relationships into computable quantitative relationships, thus obtaining a risk transmission knowledge graph.

[0025] As a further improvement to this technical solution, the simulated risk evolution trajectory is specifically implemented as follows: based on the patient's planned nursing activities, real-time behavior recognition results, and historical behavior patterns, a sequence of multiple behavioral scenarios within a specific future time window is constructed. Each constructed behavioral scenario is used as an external stimulus and sequentially input into the high-fidelity physiological simulation model to drive its state evolution. Simultaneously, based on the risk transmission knowledge graph, the chain probability of downstream risk events triggered by changes in physiological state is queried and calculated, and a dynamic risk evolution trajectory curve with time as the horizontal axis and risk probability as the vertical axis is output.

[0026] As a further improvement to this technical solution, the specific method for identifying potential nursing risks is as follows: A preset risk probability threshold and risk trend slope threshold are obtained from the database. The risk probability value at any moment in the dynamic risk evolution trajectory curve is compared with the risk probability threshold to identify instantaneous high-risk points exceeding the threshold. Simultaneously, the instantaneous slope of the dynamic risk evolution trajectory curve is calculated and compared with the risk trend slope threshold to identify warning windows where the risk probability is increasing. Based on the risk transmission knowledge graph, the identified high-risk points and warning windows are analyzed to determine the main risk factors and key physiological intermediate states leading to the risks, thereby outputting specific potential nursing risk types, risk levels, and critical paths of risk evolution.

[0027] As a further improvement to this technical solution, the specific method for simulating the effects of candidate intervention measures is as follows: A set of candidate nursing intervention measures is obtained from the database, and this set is represented as a constraint on the parameter control logic of the high-fidelity physiological simulation model. Measures are applied in parallel during the risk trajectory deduction process to simulate their inhibitory and regulatory effects on the risk trajectory curve. A comprehensive evaluation function is constructed, whose variables include at least the risk reduction area, the cost of implementing the measures, and the patient's expected comfort. Based on the simulation results, a multi-objective optimization algorithm is used to perform Pareto front analysis on all candidate strategies, and the non-dominated solution set is output as the optimal intervention strategy recommendation.

[0028] The second aspect of the present invention provides a method for a basic nursing risk intelligent identification and early warning system, comprising: S1, multi-source data perception and fusion, for real-time acquisition and fusion of multi-source heterogeneous data of patients, and outputting a structured patient state feature set.

[0029] S2, the core processing for intelligent risk identification, is used to identify risks through simulation and deduction based on a structured set of patient status features.

[0030] S3, Intelligent Early Warning and Intervention, is used to execute graded early warning and proactive intervention based on the output of the aforementioned risk intelligent identification core processing.

[0031] S4, Human-Machine Collaborative Feedback and Self-Evolution, is used to collect user feedback on the risk projection results and the execution and effect of intervention strategies, and to use the feedback data to synchronously optimize the model and knowledge base during the process.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] 1. Multimodal data fusion for early risk perception: This invention constructs a comprehensive patient profile by simultaneously collecting explicit data such as vital signs and medical orders, as well as implicit data such as behavioral posture and facial expressions. This enables the system to capture subtle early risk signals that traditional methods cannot detect, achieving a fundamental shift from "post-event alarm" to "pre-event warning," greatly improving the timeliness of risk identification.

[0034] 2. Dynamic simulation and deduction for proactive risk warning: This invention innovatively introduces patient digital twins and risk transmission knowledge graphs, enabling proactive simulation of risk evolution trajectories under various future behavioral scenarios based on the current state. This changes the traditional static and passive assessment model, giving the system the "predictive ability" to anticipate risks, thereby achieving truly proactive intervention.

[0035] 3. Intelligent Decision-Making and Closed-Loop Evolution for Precise and Personalized Prevention and Control: This invention not only issues alerts but also recommends the optimal intervention strategy, based on quantitative evaluation, to medical staff through intervention effect simulation and multi-objective optimization. Simultaneously, through a human-machine collaborative feedback loop, the system continuously learns from clinical practice, optimizing its model and knowledge to become a self-evolving system that becomes increasingly accurate with use, thus achieving intelligent and personalized nursing risk management. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of the system structure connection of the present invention.

[0038] Figure 2 This is a schematic diagram of the implementation steps of the method of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Example: Please refer to Figure 1As shown, a basic nursing risk intelligent identification and early warning system is provided, including: a multi-source data perception and fusion module, which is used to acquire and fuse multi-source heterogeneous data of patients in real time and output a structured patient status feature set.

[0041] The realization of a structured patient status feature set is based on multi-source heterogeneous data. First, explicit risk data (vital signs, medical orders, nursing records, etc.) and implicit risk data (behavioral postures, facial expressions, etc. analyzed by computer vision) of patients are collected in real time through multi-source data sensing terminals. Then, the collected data is cleaned and denoised, outlier removal is performed, and the format is standardized. Among them, unstructured text data is extracted for entity features through natural language processing, and implicit behavioral data is transformed into quantitative indicators. Subsequently, static baseline features (such as medical history, underlying diseases, etc.) and dynamic trend features (such as fluctuations in vital signs, changes in behavioral patterns, etc.) are extracted in combination with nursing risk identification needs. Finally, various features are integrated through data assimilation or weighted fusion technology to form a standardized, structured set that can be directly adapted to subsequent risk identification, including feature names, quantitative values, timestamps, and data sources.

[0042] In one specific embodiment, the patient's multi-source heterogeneous data includes explicit risk data and implicit risk data.

[0043] The explicit risk data includes at least the patient's vital signs, medical orders, and nursing records.

[0044] The hidden risk data includes at least patient behavior and posture data and facial expression data obtained based on computer vision analysis.

[0045] The core module for intelligent risk identification is used to identify risks based on a structured set of patient status features through simulation and deduction.

[0046] In one specific embodiment, the core module for intelligent risk identification includes: a patient-personalized digital twin unit and a risk extrapolation and intervention simulation unit.

[0047] The patient-personalized digital twin unit is used to construct and update a parameterized patient physiological simulation model in real time based on the patient state feature set.

[0048] The risk extrapolation and intervention simulation unit, which is coupled with the digital twin unit and the risk transmission knowledge graph, is used to simulate the risk evolution trajectory, identify potential nursing risks, and simulate the effects of candidate intervention measures.

[0049] In one specific embodiment, the patient physiological simulation model is implemented by establishing a parameterized basic physiological mechanism model, which includes a series of adjustable physiological parameters.

[0050] The establishment of a parameterized basic physiological mechanism model is based on "comprehensive coverage of physiological systems highly correlated with nursing risks + support from medical mechanism equations + explicitness of adjustable parameters," specifically achieved through the following process: First, core physiological systems directly associated with basic nursing risks (such as pressure ulcers, falls, metabolic disorders, etc.) are screened, including the circulatory system, respiratory system, metabolic system, skin pressure transmission system, and motor sensory system; then, for each system, quantitative mechanism equations are constructed based on classical medical mechanisms and clinical consensus (e.g., the circulatory system uses hemodynamic equations to describe the relationship between cardiac output and blood pressure, and the skin system uses pressure transmission equations to characterize the relationship between body pressure and skin tissue stress), and from... Key physiological variables are extracted from the equation as adjustable parameters (such as heart rate and peripheral resistance in the circulatory system, and tissue elasticity coefficient in the skin system). Subsequently, a parameter system is constructed, clarifying the medical definition, value range, and unit of each parameter (such as the age-related basal metabolic rate parameter with a value range of 10-30 kcal / (kg・d)). Parameter interfaces are designed to support subsequent personalized initialization and dynamic calibration. Finally, the model is validated using clinical data from healthy individuals and standard values ​​from medical guidelines to ensure that the physiological indicators output by the model (such as resting heart rate and baseline blood pressure) conform to the normal medical range when no personalized parameters are input, thus forming a basic physiological mechanism model.

[0051] Obtain the patient's static baseline characteristics, and based on the patient's static baseline characteristics, initialize and assign values ​​to the adjustable physiological parameters in the basic physiological mechanism model to form an initial personalized model.

[0052] The patient's static baseline characteristics include at least the following: demographic characteristics (age, sex, height, weight, BMI), past medical history and underlying disease characteristics (type and course of chronic diseases, surgical history, trauma history, infectious disease history), physiological baseline characteristics (blood type, allergy history, organ function baseline classification, immune function baseline level), lifestyle and individual differences (long-term smoking / drinking history, dietary habits, exercise frequency, sleep patterns), and genetic and family history characteristics (family history of genetic diseases, history of high-incidence underlying diseases among relatives).

[0053] Model calibration is performed using real-time data. The dynamic trend features of the patient state feature set are used as input. Through data assimilation algorithm, the adjustable physiological parameters in the initial personalized model are continuously optimized to minimize the error between the model's output value and the patient's actual real-time vital sign monitoring value, thus obtaining a calibrated high-fidelity physiological simulation model.

[0054] The data assimilation algorithm is existing technology and will not be described in detail here.

[0055] In one specific embodiment, the risk transmission knowledge graph is implemented by: constructing a basic medical ontology: based on the standard medical terminology system, constructing a three-layer ontology structure that includes nursing risk factors, physiological intermediate states, and nursing risk events, as the entity framework of the knowledge graph.

[0056] First layer: Nursing risk factors (entities that are the sources of risk)

[0057] Definition: A collection of factors that directly or indirectly induce nursing risks, subdivided according to their source:

[0058] Individual factors: Risk factors based on the patient's own characteristics (such as age ≥65 years, BMI ≥28, history of diabetes, cognitive impairment, decreased skin sensation, etc., corresponding to the standard terms "elderly state", "obesity", "type 2 diabetes", and "cognitive impairment").

[0059] Environmental factors: External factors in the nursing setting that affect patient safety (such as slippery ward floors, bed rails not raised, insufficient lighting, equipment malfunction, etc., corresponding to the standard terms "environmental safety hazards" and "medical equipment malfunctions").

[0060] Nursing operation factors: operational deviations or improper procedures that may cause risks during the nursing process (such as delayed turning over, excessively rapid infusion rate, incorrect medication administration, improper catheter fixation, etc., corresponding to the standard terms "non-standard nursing operation" and "medication error").

[0061] Second layer: Physiological intermediate state (entity that acts as a medium for risk transmission)

[0062] Definition: Quantifiable and monitorable abnormal physiological states triggered by nursing risk factors in patients, categorized by core physiological system:

[0063] Circulatory system abnormalities: such as sudden drop in blood pressure, arrhythmia, and insufficient cerebral blood supply (corresponding to the standard terms "hypotension", "arrhythmia", and "insufficient cerebral perfusion").

[0064] Abnormalities in the skin system: such as continuous pressure on the skin, skin damage, and moist skin (corresponding to the standard terms "pre-injury of skin pressure" and "impaired skin integrity").

[0065] Metabolic system abnormalities: such as blood glucose fluctuations, electrolyte imbalances, nutritional deficiencies, etc. (corresponding to standard terms "unstable blood glucose" and "hypokalemia");

[0066] Abnormalities in the nervous / consciousness system: such as confusion, limb movement disorders, and decreased balance (corresponding to the standard terms "disorder of consciousness" and "limb dysfunction").

[0067] Third level: Nursing risk events (the final outcome entity of the risk)

[0068] Definition: Adverse events that occur during the nursing process and may cause harm to the patient, categorized by type of harm:

[0069] Pressure injuries (pressure ulcers), falls, infusion reactions, catheter slippage, hypoglycemic coma, infections (such as lung infections, surgical site infections), etc. (strictly correspond to the "medical-related injuries" classification terms, such as "pressure ulcers", "accidental falls", "catheter-related infections").

[0070] Establish multi-order risk transmission relationships: Based on clinical medical guidelines and expert knowledge, establish directed causal transmission edges between entities in the ontology framework. The transmission edges include at least first-order transmission from risk factors to physiological intermediate states and second-order transmission from physiological intermediate states to risk events.

[0071] First-order conduction edge (risk factors → physiological intermediate state):

[0072] Definition: The transmission relationship of the first abnormal physiological state caused by the direct action of risk factors on the patient, with the transmission direction being "risk factor → intermediate physiological state";

[0073] Specific examples:

[0074] Individual factors → Physiological intermediate state: <Age ≥ 65 years → Decreased balance>, <History of diabetes → Blood glucose fluctuations>, <Cognitive impairment → Confusion>;

[0075] Environmental factors → Physiological intermediate state: <Slippery ward floor → unstable limb support>, <Insufficient lighting → visual judgment deviation>;

[0076] Nursing procedural factors → Physiological intermediate state: <Delayed turning over → continuous pressure on the skin>, <Excessive infusion rate → sudden increase in blood volume>.

[0077] Second-order conduction boundary (physiological intermediate state → nursing risk event):

[0078] Definition: The transmission relationship of a physiological intermediate state that continues to develop or is not intervened in, ultimately leading to an adverse nursing event. The direction of the transmission is "physiological intermediate state → nursing risk event".

[0079] Specific examples:

[0080] Circulatory system abnormalities → risk events: <sudden drop in blood pressure → syncope>, <sudden increase in blood volume → heart failure>;

[0081] Skin system abnormalities → risk events: <Prolonged pressure on the skin → pressure ulcers>, <Skin damage → infection>;

[0082] Nervous / consciousness system abnormalities → risk events: <decreased balance → fall from bed>, <confusion → aspiration>;

[0083] Metabolic system abnormalities → risk events: <blood glucose fluctuations → hypoglycemic coma>, <electrolyte imbalances → cardiac arrhythmias>.

[0084] Conduction edge attribute annotation: Add core attributes to each directed conduction edge, including "conduction direction", "medical basis (guideline clause / expert consensus number)" and "applicable population" to ensure that the relationship is traceable (e.g., the attribute annotation for conduction edge <delayed turning over → continuous skin pressure> is: direction = risk factor → physiological intermediate state, applicable population = bedridden patients).

[0085] Hierarchical constraint rules:

[0086] Forced transmission pathway: Only second-order transmission pathways of "risk factor → physiological intermediate state → risk event" are allowed. Risk factors are prohibited from directly pointing to risk events (e.g., <prolonged bed rest → pressure ulcer> is an invalid association and the complete transmission chain of <prolonged bed rest → continuous pressure on the skin → pressure ulcer> needs to be supplemented).

[0087] Many-to-many association restrictions: The same risk factor can correspond to multiple physiological intermediate states (e.g., <history of diabetes → blood glucose fluctuation>, <history of diabetes → peripheral neuropathy>), and the same physiological intermediate state can correspond to multiple risk events (e.g., <confusion → fall from bed>, <confusion → aspiration>), but must conform to clinical logic;

[0088] Exclusion of mutual exclusion: If there is a medically exclusive relationship between entities (such as <dry skin → moist skin>), the establishment of a transmission association is prohibited.

[0089] Quantifying the strength of transmission relationships: Based on historical clinical data, statistical analysis is performed to assign initial probability weights and confidence levels to each causal transmission edge, transforming qualitative knowledge relationships into computable quantitative relationships, thus obtaining a risk transmission knowledge graph.

[0090] In one specific embodiment, the simulated risk evolution trajectory is implemented as follows: based on the patient's planned nursing activities, real-time behavior recognition results, and historical behavior patterns, a sequence of multiple behavioral scenarios within a specific future time window is constructed. Each constructed behavioral scenario is used as an external stimulus and sequentially input into the high-fidelity physiological simulation model to drive its state evolution. Simultaneously, based on the risk transmission knowledge graph, the chain probability of downstream risk events triggered by changes in physiological state is queried and calculated, and a dynamic risk evolution trajectory curve with time as the horizontal axis and risk probability as the vertical axis is output.

[0091] The calculation of the chain probability of downstream risk events triggered by changes in physiological state is as follows:

[0092] Step 1: State Mapping and Node Activation

[0093] When a high-fidelity physiological simulation model evolves under the drive of behavioral scenarios, its output physiological state variables (such as blood pressure and heart rate) need to be mapped to entities in the risk transmission knowledge graph.

[0094] Specific operation: The system continuously monitors the physiological parameters output by the digital twin. For example, at a certain time point... The model calculated the patient's systolic blood pressure to be 90 mmHg. Based on a preset threshold rule, this state was mapped and activated as the "physiological intermediate state" node for hypotension in the knowledge graph. We define the node... At the point of time It is activated, and its state is (Activated) or 0 (Not activated).

[0095] Step 2: Parallel Probabilistic Inference

[0096] For each activated physiological intermediate state node, the system performs forward probabilistic reasoning in parallel along directed edges in the knowledge graph to calculate the conditional probability of it leading to downstream risk events.

[0097] Core formula (based on the chain rule of conditional probability):

[0098] Each edge in a knowledge graph Each is accompanied by a conditional probability table or a weighting function, expressed as follows: Its meaning is "when the cause When it happens, the result The probability of occurrence.

[0099] When a risk event have An independent leader state When this occurs, the total probability of it happening can be calculated in the following way:

[0100] 1. Calculate the contribution of a single path:

[0101] For each leading state Its risk The contribution is the conditional probability. .

[0102] 2. Aggregate the effects of multiple paths (using the "Noisy-OR" model, which is a commonly used and reasonable model for handling this type of problem):

[0103] When multiple causes independently lead to the same result, the probability that the result does not occur is the product of the probabilities that each cause does not lead to it.

[0104] ;

[0105] This formula calculates the probability that "at least one leading state successfully triggers a risk event." It cleverly handles multiple influencing factors and does not require all causes to exist simultaneously.

[0106] Step 3: Time Series Integration and Trajectory Generation

[0107] Within the entire time window of the scenario simulation, steps one and two are repeated with fixed time intervals as steps, for each time point. Calculate a current risk probability value All time points When connected, this forms the final "dynamic risk evolution trajectory curve with time as the horizontal axis and risk probability as the vertical axis".

[0108] ;

[0109] in, Indicates at a point in time The activated first One leading state node This indicates the number of the leader state node.

[0110] In one specific embodiment, the identification of potential nursing risks is implemented as follows: A preset risk probability threshold and risk trend slope threshold are obtained from a database; the risk probability value at any moment in the dynamic risk evolution trajectory curve is compared with the risk probability threshold to identify instantaneous high-risk points exceeding the threshold; simultaneously, the instantaneous slope of the dynamic risk evolution trajectory curve is calculated and compared with the risk trend slope threshold to identify warning windows where the risk probability is increasing; based on the risk transmission knowledge graph, source analysis is performed on the identified high-risk points and warning windows to determine the main risk factors and key physiological intermediate states leading to the risk; and then, the specific potential nursing risk type, risk level, and critical path of risk evolution are output.

[0111] Risk probability threshold: Based on historical adverse nursing event datasets, the optimal critical point for different risk types (such as falls and pressure sores) is determined through statistical models (such as ROC curve analysis). This threshold is a static benchmark used to determine the absolute severity of the risk.

[0112] Risk trend slope threshold: This threshold is set based on clinical expert experience and the urgency of risk intervention. It is a dynamic benchmark used to judge the rate of risk deterioration, aiming to detect signals of accelerated deterioration before the risk probability reaches an absolute high level.

[0113] The instantaneous slope (i.e., derivative) is calculated using numerical differentiation.

[0114] The warning window is identified as a period of time during which specific conditions are continuously met, rather than a single point in time. The identification logic is as follows:

[0115] Step 1: Slope Determination. The system iterates through the calculated instantaneous slope sequence and marks all time points where the instantaneous slope is higher than the risk trend slope threshold.

[0116] Step 2: Continuity Assessment. The system checks whether these marked time points constitute a continuous time period. For example, it is stipulated that a meaningful trend is considered to exist only when the slope of three or more consecutive calculation points exceeds a threshold.

[0117] Step 3: Window Delineation. The start and end points of this continuous time period are defined as an "early warning window." Within this window, if the system believes the risk is accumulating continuously and rapidly, an early warning must be issued even if the absolute probability has not yet reached the high-risk threshold.

[0118] Method for identifying potential nursing risk types: Directly query within the risk transmission knowledge graph. The system performs source tracing analysis on identified high-risk points or warning windows to find the ultimate "nursing risk event" entity (such as "fall" or "pressure sore"). The name of this entity is the risk type.

[0119] Risk level determination method: Multidimensional rule mapping is used. The system takes "risk probability value" and "risk trend slope" as inputs and determines the final level through a preset decision matrix or rule set.

[0120] Example rules:

[0121] High risk: Probability > probability threshold OR (Probability > medium threshold AND slope > slope threshold)

[0122] Medium risk: Probability > Medium threshold OR Slope > Slope threshold

[0123] Low risk: Probability > Low threshold

[0124] Method for determining the critical path of risk evolution: Tracing the critical path on the risk transmission knowledge graph. Starting from the identified risk event node (such as "falling"), the system traverses all its incoming edges (i.e., the causes that led to it) in reverse, and finds one or more causal chains with the greatest contribution based on the weight of the edges (conditional probability) and the activation state of the nodes.

[0125] In one specific embodiment, the method for simulating the effects of candidate intervention measures is as follows: A set of candidate nursing intervention measures is obtained from a database; this set is then represented as a constraint on the parameter control logic of the high-fidelity physiological simulation model; measures are applied in parallel during the risk trajectory deduction process to simulate their inhibitory and regulatory effects on the risk trajectory curve; a comprehensive evaluation function is constructed, whose variables include at least the risk reduction area, the cost of implementing the measures, and the patient's expected comfort; based on the simulation results, a multi-objective optimization algorithm is used to perform Pareto front analysis on all candidate strategies, and the non-dominated solution set is output as the optimal intervention strategy recommendation.

[0126] The process of constructing the candidate nursing intervention set:

[0127] Knowledge base source: The set of measures is derived from a structured nursing measures knowledge base, which is built based on clinical guidelines, expert consensus and best practices, and is continuously updated.

[0128] Personalized search: Once the system identifies a specific risk type (such as "fall") and a key risk path (such as "orthostatic hypotension -> fall"), it will automatically retrieve from the knowledge base all interventions that are associated with these risk factors and intermediate states and are applicable to the current patient's condition.

[0129] Feasibility filtering: The retrieved measures will be further filtered based on the hospital's available resources (such as whether there are corresponding smart devices) and the patient's specific contraindications (such as allergies to a certain drug), ultimately forming a personalized set of candidate intervention measures for the current situation.

[0130] The construction of the comprehensive evaluation function is specifically as follows:

[0131] Risk reduction area: measures the effectiveness of the measure. It is calculated as the area between the original risk curve and the risk curve after the measure is implemented.

[0132] ;

[0133] in, This indicates the area where the risk has been reduced. This represents the original risk probability function. This represents the risk probability function after intervention. Indicates the starting point of the time frame for risk projection analysis. This indicates the end point of the time frame for risk projection analysis.

[0134] Implementation cost: Measures the resource consumption of an action. It can be quantified into a comprehensive score based on factors such as time required, manpower level, and material consumption.

[0135] ;

[0136] in, Indicates the cost of implementing the measures. Indicates the normalized time cost. This represents normalized labor costs. This represents the normalized material cost. , , These represent the weights for time cost, labor cost, and material cost, respectively.

[0137] Patient expected comfort level: This measures the degree of humanistic care provided by the measures. Based on historical feedback data and expert ratings, a comfort score is preset for each measure. .

[0138] The risk reduction area, the cost of implementing the measures, and the patient's expected comfort are normalized and weighted to construct a comprehensive evaluation function.

[0139] The intelligent early warning and intervention module is used to perform graded early warning and proactive intervention based on the output of the risk intelligent identification core module.

[0140] The human-machine collaborative feedback and self-evolution module is used to collect user feedback on the risk simulation results and the execution and effect of intervention strategies, and to use the feedback data to simultaneously optimize the risk intelligent identification core module and the intelligent early warning and intervention module.

[0141] Optimization of the digital twin model: Compare the actual clinical results with the predictions of the digital twin, and adjust the model parameters through a differential algorithm;

[0142] Optimization of the risk transmission knowledge graph: Transform user feedback on the effectiveness of intervention measures into signals that strengthen or weaken the weights of causal relationships in the knowledge graph;

[0143] The optimization of the risk transmission knowledge graph specifically involves adopting a reinforcement learning-based strategy, using successful intervention feedback as a positive reward to enhance the weight of the corresponding causal path, and using ineffective or negative intervention feedback as a negative reward to weaken the weight of the corresponding causal path.

[0144] See Figure 2 As shown, a method for intelligent identification and early warning system of basic nursing risk is provided, including: S1, multi-source data perception and fusion, which is used to acquire and fuse multi-source heterogeneous data of patients in real time and output a structured patient state feature set.

[0145] S2, the core processing for intelligent risk identification, is used to identify risks through simulation and deduction based on a structured set of patient status features.

[0146] S3, Intelligent Early Warning and Intervention, is used to execute graded early warning and proactive intervention based on the output of the aforementioned risk intelligent identification core processing.

[0147] S4, Human-Machine Collaborative Feedback and Self-Evolution, is used to collect user feedback on the risk projection results and the execution and effect of intervention strategies, and to use the feedback data to synchronously optimize the model and knowledge base during the process.

[0148] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A basic nursing risk intelligent identification and early warning system, characterized in that, include: The multi-source data perception and fusion module is used to acquire and fuse multi-source heterogeneous patient data in real time and output a structured patient status feature set. The core module for intelligent risk identification is used to identify risks based on a structured set of patient status features through simulation and deduction. The intelligent early warning and intervention module is used to perform graded early warning and proactive intervention based on the output of the risk intelligent identification core module; The human-machine collaborative feedback and self-evolution module is used to collect user feedback on the risk simulation results and the execution and effect of intervention strategies, and to use the feedback data to simultaneously optimize the risk intelligent identification core module and the intelligent early warning and intervention module.

2. The basic nursing risk intelligent identification and early warning system according to claim 1, characterized in that, The patient's multi-source heterogeneous data includes explicit risk data and implicit risk data; The explicit risk data includes at least the patient's vital signs, medical orders, and nursing records; The hidden risk data includes at least patient behavior and posture data and facial expression data obtained based on computer vision analysis.

3. The basic nursing risk intelligent identification and early warning system according to claim 1, characterized in that, The core module for intelligent risk identification includes: a patient-personalized digital twin unit and a risk projection and intervention simulation unit; The patient-personalized digital twin unit is used to construct and update a parameterized patient physiological simulation model in real time based on the patient state feature set. The risk extrapolation and intervention simulation unit, which is coupled with the digital twin unit and the risk transmission knowledge graph, is used to simulate the risk evolution trajectory, identify potential nursing risks, and simulate the effects of candidate intervention measures.

4. The basic nursing risk intelligent identification and early warning system according to claim 3, characterized in that, The specific implementation method of the patient physiological simulation model is as follows: Establish a parameterized basic physiological mechanism model, which includes a series of adjustable physiological parameters; Obtain the patient's static baseline characteristics, and based on the patient's static baseline characteristics, initialize and assign values ​​to the adjustable physiological parameters in the basic physiological mechanism model to form an initial personalized model; Model calibration is performed using real-time data. The dynamic trend features of the patient state feature set are used as input. Through data assimilation algorithm, the adjustable physiological parameters in the initial personalized model are continuously optimized to minimize the error between the model's output value and the patient's actual real-time vital sign monitoring value, thus obtaining a calibrated high-fidelity physiological simulation model.

5. The basic nursing risk intelligent identification and early warning system according to claim 1, characterized in that, The specific implementation method of the risk transmission knowledge graph is as follows: Constructing a basic medical ontology: Based on the standard medical terminology system, a three-layer ontology structure is constructed, which includes nursing risk factors, physiological intermediate states, and nursing risk events, as the entity framework of the knowledge graph; Establish multi-order risk transmission relationships: Based on clinical medical guidelines and expert knowledge, establish directed causal transmission edges between entities in the ontology framework. The transmission edges include at least first-order transmission from risk factors to physiological intermediate states and second-order transmission from physiological intermediate states to risk events. Quantifying the strength of transmission relationships: Based on historical clinical data, statistical analysis is performed to assign initial probability weights and confidence levels to each causal transmission edge, transforming qualitative knowledge relationships into computable quantitative relationships, thus obtaining a risk transmission knowledge graph.

6. The basic nursing risk intelligent identification and early warning system according to claim 5, characterized in that, The specific implementation method for the simulated risk evolution trajectory is as follows: Based on the patient's planned nursing activities, real-time behavior recognition results, and historical behavior patterns, multiple behavioral scenario sequences within a specific future time window are constructed. Each constructed behavioral scenario is used as an external stimulus and is sequentially input into the high-fidelity physiological simulation model to drive its state evolution. Simultaneously, based on the risk transmission knowledge graph, the chain probability of downstream risk events triggered by changes in physiological state is queried and calculated, and a dynamic risk evolution trajectory curve with time as the horizontal axis and risk probability as the vertical axis is output.

7. The basic nursing risk intelligent identification and early warning system according to claim 6, characterized in that, The specific method for identifying potential care risks is as follows: The system retrieves preset risk probability thresholds and risk trend slope thresholds from the database. It compares the risk probability value at any moment in the dynamic risk evolution trajectory curve with the risk probability thresholds to identify instantaneous high-risk points that exceed the thresholds. Simultaneously, it calculates the instantaneous slope of the dynamic risk evolution trajectory curve and compares it with the risk trend slope thresholds to identify warning windows where the risk probability is on the rise. Based on the risk transmission knowledge graph, it performs source tracing analysis on the identified high-risk points and warning windows to determine the main risk factors and key physiological intermediate states that lead to the risks. Finally, it outputs specific potential nursing risk types, risk levels, and critical paths of risk evolution.

8. The basic nursing risk intelligent identification and early warning system according to claim 1, characterized in that, The specific method for simulating the effects of candidate intervention measures is as follows: A set of candidate nursing interventions is obtained from the database and represented as constraints on the parameter control logic of the high-fidelity physiological simulation model. The interventions are applied in parallel during the risk trajectory extrapolation process to simulate their inhibitory and regulatory effects on the risk trajectory curve. A comprehensive evaluation function is constructed, whose variables include at least the risk reduction area, the cost of implementing the intervention, and the patient's expected comfort. Based on the simulation results, a multi-objective optimization algorithm is used to perform Pareto front analysis on all candidate strategies, and the non-dominated solution set is output as the optimal intervention strategy recommendation.

9. A method for implementing the basic nursing risk intelligent identification and early warning system according to any one of claims 1-8, characterized in that, include: S1, Multi-source data perception and fusion, is used to acquire and fuse multi-source heterogeneous patient data in real time and output a structured patient state feature set; S2, the core processing for intelligent risk identification, is used to identify risks through simulation and deduction based on a structured set of patient status features; S3, Intelligent Early Warning and Intervention, is used to execute graded early warning and proactive intervention based on the output of the core processing of the intelligent risk identification; S4, Human-Machine Collaborative Feedback and Self-Evolution, is used to collect user feedback on the risk simulation results and the execution and effect of intervention strategies, and to use the feedback data to synchronously optimize the model and knowledge base during the process.

Images