An artificial intelligence driven method for prioritizing tasks in intensive care nursing

Abstract

The application discloses an artificial intelligence driven intensive care nursing task priority sequencing method, and relates to the technical field of nursing data processing.The method comprises the following steps: acquiring real-time physiological monitoring data, historical illness records and diagnosis labels of intensive care patients, and evaluating illness grades and fluctuation scores; performing resource dependency analysis, and constructing resource occupation graphs and mutual exclusion conflict graphs; generating an initial task queue based on the resource occupation graphs, the mutual exclusion conflict graphs, the illness grades and the fluctuation scores, and combining a weighted directed graph sequencing algorithm; modeling the ability of nursing staff, and constructing a nursing ability vector; performing multi-objective sequencing optimization on nursing tasks by combining the initial task queue, the resource occupation graphs, the mutual exclusion conflict graphs and the nursing ability vector, and acquiring a staff-task sequence; generating a plurality of nursing task priority sequences correspondingly, and correspondingly issuing the sequences to the client end of the nursing staff.The application solves the technical problem of inaccurate nursing task allocation in the prior art.

Description

An AI-driven method for prioritizing critical care nursing tasks Technical Field

[0001] This invention relates to the field of nursing data processing technology, specifically to an artificial intelligence-driven method for prioritizing critical care nursing tasks. Background Technology

[0002] In the field of intensive care, the dynamic prioritization of nursing tasks directly impacts patient survival rates and the efficiency of medical resource utilization. Current technologies primarily rely on manual experience or static task allocation, such as judging task urgency based on simple threshold alarms of patient vital signs and fixed nursing procedure manuals, combined with subjective experience. This approach has significant drawbacks: firstly, manual assessment struggles to quantify the risk of sudden changes in patient condition in real time; secondly, while static allocation can achieve basic prioritization, it lacks optimal utilization of various factors, such as nursing resources. Especially in emergency rescue scenarios, existing solutions are prone to delays in critical nursing tasks or staff shortages, increasing nursing workload. Summary of the Invention

[0003] This application provides an artificial intelligence-driven method for prioritizing critical care nursing tasks, which addresses the technical problem of inaccurate allocation of nursing tasks in existing technologies.

[0004] Acquire real-time physiological monitoring data, historical medical records and diagnostic labels of intensive care patients, and conduct corresponding assessments of disease severity and disease fluctuation to obtain disease severity and fluctuation scores;

[0005] Obtain nursing task information to perform resource dependency analysis, and construct resource occupancy graph and mutual exclusion conflict graph based on the resource dependency analysis results;

[0006] Based on the resource occupancy graph, the mutual exclusion conflict graph, the disease severity level, and the fluctuation score, an initial task queue is generated using a weighted directed graph sorting algorithm.

[0007] A competency model is constructed for nursing staff to create a nursing competency vector. The nursing competency vector includes information that at least represents the staff’s skill proficiency, operational authority level, work fatigue status, current task load, and the spatial area they are responsible for.

[0008] By combining the initial task queue, the resource occupancy graph, the mutual exclusion conflict graph, and the nursing capability vector, a multi-objective sorting optimization based on resource-capability matching is performed on nursing tasks to obtain a personnel-task sequence.

[0009] Multiple nursing task priority sequences are generated based on the personnel-task sequence and then distributed to the nursing staff's client.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] This application proposes an AI-driven method for prioritizing critical care nursing tasks. Through a collaborative optimization mechanism that dynamically integrates multi-dimensional patient condition assessment, resource conflict modeling, and nurse staff capability adaptation, it significantly improves the real-time performance, resource utilization efficiency, and accuracy of personnel adaptation in critical care settings. Compared to traditional methods, the technical solution provided in this application significantly overcomes the subjective lag of human experience and the rigidity of static allocation, achieving precise allocation of nursing tasks with patient safety as the core orientation, minimizing resource conflicts as the constraint, and optimal nurse staff capability adaptation as the execution guarantee. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0013] Figure 1 is a flowchart illustrating an artificial intelligence-driven method for prioritizing critical care nursing tasks, as provided in an embodiment of this application.

[0014] Figure 2 is a schematic diagram of the process of constructing a resource occupancy graph and a mutual exclusion conflict graph in an artificial intelligence-driven intensive care nursing task priority ranking method provided in an embodiment of this application. Detailed Implementation

[0015] This application provides an artificial intelligence-driven method for prioritizing critical care nursing tasks, which addresses the technical problem of inaccurate allocation of nursing tasks in existing technologies.

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0017] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0018] Example 1, as shown in Figure 1, this application provides an artificial intelligence-driven method for prioritizing critical care nursing tasks, wherein the method includes:

[0019] S10: Obtain real-time physiological monitoring data, historical medical records and diagnostic labels of critically ill patients, and conduct corresponding assessments of disease severity and disease fluctuation to obtain disease severity and fluctuation scores.

[0020] In dynamic decision-making in intensive care, current technologies rely on human experience to integrate real-time physiological data and historical records of patients, making it difficult to quantify the risk of sudden changes in the patient's condition. The lack of a framework for fusing and analyzing multi-source heterogeneous data can lead to delays in responding to nursing tasks.

[0021] Step S10 in the method provided in this application embodiment includes:

[0022] Construct a multimodal hybrid evaluation model, wherein the multimodal hybrid evaluation model includes at least:

[0023] The first evaluation channel based on the rule engine is used to calculate the indicator deviation entropy, treatment dependence intensity and vital sign chaos degree based on the real-time physiological monitoring data and the historical medical records, and to generate a first evaluation level by weighted calculation.

[0024] A second evaluation channel based on a deep learning model is used to generate a second evaluation level based on the real-time physiological monitoring data.

[0025] Input the real-time physiological monitoring data and the historical medical records into the multimodal hybrid assessment model, obtain the first assessment level and the second assessment level, and merge the first assessment level, the second assessment level and the diagnostic label to generate the medical condition level;

[0026] Based on the historical medical records, historical indicator fluctuation features and historical fluctuation scores are extracted, wherein the historical indicator fluctuation features and the historical fluctuation scores correspond one-to-one.

[0027] Using the historical indicator fluctuation characteristics and the historical fluctuation scores as sample datasets, a fluctuation assessment model is constructed and trained.

[0028] The real-time physiological monitoring data is subjected to feature engineering processing, and the result of the feature engineering processing is input into the fluctuation assessment model to perform fluctuation assessment and obtain the fluctuation score.

[0029] In this embodiment of the application, a multimodal hybrid evaluation model is constructed, wherein the multimodal hybrid evaluation model includes at least: a first evaluation channel based on a rule engine and a second evaluation channel based on a deep learning model.

[0030] The first assessment channel, based on a rule engine, can be exemplarily constructed using the Drools rule engine. Inputs include real-time physiological monitoring data such as heart rate and blood oxygen saturation, as well as historical medical records, including historical physiological monitoring data and medication records. Calculations are performed to obtain indicator deviation entropy, treatment dependence intensity, and vital sign chaos degree. For example, heart rate indicator deviation entropy = Σ|current heart rate - historical mean heart rate| / historical standard deviation of heart rate, used to quantify the abnormality of physiological monitoring data. Treatment dependence intensity = current number of medication types ÷ historical average number of medication types, where the historical average number of medication types is the average number of medication types used by critically ill patients in history; a higher treatment dependence intensity indicates higher nursing complexity. Vital sign chaos degree = number of abnormal real-time monitored vital sign indicators ÷ total number of real-time monitored vital sign indicators; a higher vital sign chaos degree indicates more abnormal vital sign indicators, requiring more nursing care. Weighted calculations are then performed to obtain the first assessment level. For example, assign a weight of 0.3 to the indicator deviation entropy, a weight of 0.3 to the treatment dependence intensity, and a weight of 0.4 to the vital sign chaos degree. First assessment level = 0.3 × indicator deviation entropy + 0.3 × treatment dependence intensity + 0.4 × vital sign chaos degree.

[0031] A second evaluation channel based on a deep learning model is used to generate a second evaluation level based on real-time physiological monitoring data. An exemplary three-layer deep learning model is employed: the input layer receives the real-time physiological monitoring data, the hidden layer uses 64 nodes activated by the ReLU function, and the output layer outputs the second evaluation level. Real-time physiological monitoring data is collected as sample data, and the second evaluation level is manually labeled using a score from 0 to 10. 0 represents the least severe condition, and 10 represents the most severe condition. The second evaluation channel is trained using the sample real-time physiological monitoring data and the sample second evaluation levels until convergence. For example, if the error of the output second evaluation level is within ±1% when inputting real-time physiological monitoring data, the training of the second evaluation channel is considered complete.

[0032] Input real-time physiological monitoring data and historical medical records into the multimodal hybrid assessment model to obtain a first assessment level and a second assessment level. Then, merge the first and second assessment levels with the diagnostic label to generate the disease severity level. For example, the sum of the first and second assessment levels is used as the disease severity level. The diagnostic label refers to the expert's assessment label for the disease severity level, such as critical, severe, or moderate. Specifically, the critical label adds 1.5 points to the disease severity level, the severe label adds 1 point, and the moderate label adds 0.5 points.

[0033] Based on historical medical records, historical indicator fluctuation characteristics and historical fluctuation scores are extracted, with a one-to-one correspondence between the historical indicator fluctuation characteristics and the historical fluctuation scores. Historical indicator fluctuation characteristics are feature data characterizing the magnitude of historical indicator fluctuations, and can be expressed as the mean, variance, and range of indicators such as heart rate, blood oxygen, and blood pressure over a continuous 6-hour period. Historical fluctuation scores are obtained beforehand by manually assessing the degree of historical fluctuation, with a score range of 0-10, where 0 indicates slight fluctuation and 10 indicates severe fluctuation.

[0034] A volatility assessment model is constructed and trained using historical indicator volatility features and historical volatility scores as sample datasets. For example, a three-layer deep learning model is used to construct the volatility assessment model. The input layer receives historical indicator volatility features, the hidden layer uses 64 nodes activated by the ReLU function, and the output layer outputs the historical volatility scores. The volatility assessment model is trained using the historical indicator volatility features and historical volatility scores as sample datasets until convergence. For example, if the input historical indicator volatility features and the output historical volatility scores have an error within ±1, the volatility assessment model is considered successfully trained.

[0035] Feature engineering is performed on the real-time physiological monitoring data. Specifically, feature data is extracted from the real-time physiological monitoring data, such as the mean, variance, and range of heart rate, blood oxygen, and blood pressure indicators over a continuous 6-hour period. The feature engineering results are then input into the fluctuation assessment model to perform fluctuation assessment and obtain a fluctuation score.

[0036] By combining a multimodal hybrid assessment model with a rule engine and deep learning, the objective quantification of patient condition levels and fluctuation scores is achieved: on the one hand, the indicator deviation entropy and treatment dependence intensity calculated by the rule channel can identify hidden deterioration trends, while the deep learning channel enhances the ability to capture real-time data patterns; on the other hand, the fluctuation assessment model is trained based on historical features and can characterize parameter fluctuations, providing a standardized benchmark for subsequent ranking.

[0037] S20: Obtain nursing task information, perform resource dependency analysis, and construct a resource occupancy graph and a mutual exclusion conflict graph based on the results of the resource dependency analysis.

[0038] Traditional nursing task allocation only considers equipment availability and does not establish a topological model of resource dependencies. When multiple tasks compete for limited resources, such as sharing a ventilator or sterile operating table, the lack of explicit expression of resource occupation paths and mutual exclusion relationships makes it difficult for manual scheduling to predict cross-task resource conflicts, forcing frequent adjustments to task queues.

[0039] Step S20 in the method provided in this application embodiment is shown in Figure 2, including:

[0040] All nursing tasks within the intensive care unit environment are analyzed in a structured manner to extract resource dependency information;

[0041] A directed graph is constructed based on the resource dependency information, where graph nodes represent tasks and edges represent resource occupancy relationships, forming the resource occupancy graph;

[0042] Perform resource mutual exclusion analysis based on the resource occupancy graph of multiple tasks, and construct the mutual exclusion conflict graph based on the results of the resource mutual exclusion analysis.

[0043] In this embodiment of the application, all nursing tasks in the intensive care unit are analyzed in a structured manner to extract resource dependency information. Specifically, resource elements are analyzed for all nursing tasks in the intensive care unit, including human resources such as the number of nurses required and the level of nurses required, equipment requirements such as ventilators and monitors, material requirements such as syringes and sterilization packs, and space requirements such as aseptic operating areas and bedside space, as resource dependency information.

[0044] Using the NetworkX graph computation library, a directed graph is constructed based on resource dependency information. Graph nodes represent tasks, and edges represent resource occupancy relationships, forming a resource occupancy graph. The NetworkX graph computation library is an open-source Python library for creating, manipulating, and studying complex networks, providing efficient and simple tools for handling graph structures.

[0045] Resource mutual exclusion analysis is performed based on the resource occupancy graphs of multiple tasks, and a mutual exclusion conflict graph is constructed based on the results of the resource mutual exclusion analysis. This graph is used to characterize the dependencies and resource competition relationships between tasks in terms of equipment, manpower, materials, and spatial location. In this graph, nodes represent tasks, and edge nodes represent conflict relationships.

[0046] By structurally analyzing task resource dependencies, resource occupancy graphs and mutual exclusion conflict graphs are constructed, enabling visual modeling of device conflicts. This provides a prior knowledge base for conflict avoidance in sorting algorithms, reducing task interruption rates.

[0047] S30: Based on the resource occupancy graph, the mutual exclusion conflict graph, the disease level, and the fluctuation score, an initial task queue is generated using a weighted directed graph sorting algorithm.

[0048] In this embodiment, an initial task queue is generated based on a resource occupancy graph, a mutual exclusion conflict graph, and disease severity and fluctuation scores, combined with a weighted directed graph sorting algorithm. For example, node priority weights are calculated as follows: node priority weight = disease severity × 0.6 + fluctuation score × 0.4. Nodes are sorted from highest to lowest weight. Adjacent tasks in the queue are traversed; if an edge exists in the mutual exclusion conflict graph (i.e., a conflict exists), the positions of adjacent tasks are swapped. If a conflict still exists, the positions of the conflicting tasks are swapped until the conflict is resolved. The initial task queue is then output, providing a reliable data foundation for subsequent optimization.

[0049] S40: Perform competency modeling on nursing staff and construct a nursing competency vector, wherein the nursing competency vector includes at least information representing the staff’s skill proficiency, operational authority level, work fatigue status, current task load, and the spatial area they are responsible for.

[0050] Traditional methods often rely solely on manual allocation of nursing tasks, which can easily overlook the capabilities and fatigue levels of nursing staff, thus hindering the optimal allocation of nursing resources.

[0051] In this embodiment, nursing staff are modeled to construct a nursing capability vector. This vector includes information representing at least the staff's skill proficiency, operational authority level, work fatigue status, current workload, and the spatial area they are responsible for. For example, the nursing capability vector could be (skill proficiency, operational authority level, work fatigue status, current workload, and responsible spatial area). Skill proficiency includes the nursing staff's level, such as levels 1-5, where level 1 is the least proficient and level 5 is the most proficient. Operational authority level includes the permissions to use available instruments, such as the level of instrument permissions that each level can operate according to their job rank and relevant regulations, such as levels 1-5, where level 1 can only use a few instruments and level 5 can use all instruments. Work fatigue status is represented by continuous working hours in hours. Current workload is represented by the number of current tasks in units. The responsible spatial area information is represented by the ward they are responsible for, such as Cardiovascular Ward 1.

[0052] Capability modeling for nursing staff can ensure that their nursing abilities are not wasted while laying the foundation for accurate matching between nursing staff and nursing tasks.

[0053] S50: Combine the initial task queue, the resource occupancy graph, the mutual exclusion conflict graph, and the nursing capability vector to perform multi-objective sorting optimization of nursing tasks based on resource-capability matching, and obtain the personnel-task sequence.

[0054] The initial queue did not incorporate personnel ability constraints, which may result in highly skilled nurses being assigned simple tasks or fatigued nurses performing high-precision operations continuously. In addition, cross-regional tasks may also cause path redundancy.

[0055] Step S50 in the method provided in this application embodiment includes:

[0056] The task matching degree is calculated by traversing the resource occupancy graph and the nursing ability vector, and a task allocation weight matrix is ​​constructed based on the task matching degree.

[0057] Using the initial task queue as the matching order and the task allocation weight matrix as the prior probability, iterative resource-capability matching is performed to determine the initial personnel-task sequence set;

[0058] Define time cost factor, labor intensity factor and mobilization cost factor to form a multi-objective cost function, and combine the mutual exclusion conflict graph to perform iterative multi-objective optimization on the initial personnel-task sequence set to obtain the personnel-task sequence;

[0059] Specifically, a time cost factor, a labor intensity factor, and a mobilization cost factor are defined to form a multi-objective cost function. This function, combined with the mutual exclusion conflict graph, is used to iteratively optimize the initial personnel-task sequence set to obtain the personnel-task sequence, including:

[0060] The time cost factor is defined by the total completion time of nursing tasks, the labor intensity factor is defined by the variance of nursing staff labor intensity, and the transfer cost factor is defined by the number of times nursing staff cross nursing zones.

[0061] The time cost factor, labor intensity factor, and mobilization cost factor are dimensionless, and the multi-objective cost function is constructed by combining a weighting method.

[0062] Based on the mutually exclusive conflict graph, a penalty factor is defined, and combined with the multi-objective cost function and the penalty factor, the initial personnel-task sequence set is iteratively updated based on the multi-objective cost function value calculation and random mutation.

[0063] If the number of iterations meets the preset iteration constraints, then the optimal multi-objective cost function value is the personnel-task sequence.

[0064] In this embodiment, the task matching degree is calculated by traversing the resource occupancy map and the nursing ability vector, and a task allocation weight matrix is ​​constructed based on the task matching degree. Specifically, the matching degree is equal to the number of matchings between nursing ability and resource occupancy map in the nursing ability vector, plus fatigue penalty, load penalty, and cross-region penalty. For example, the number of matchings between nursing ability and resource occupancy map is: matching degree + 2 when skill proficiency is greater than or equal to resource occupancy map, then the matching degree is 0; matching degree + 2 when operation permission is greater than or equal to resource occupancy map; fatigue penalty is: matching degree -0.6 when continuous working hours are greater than or equal to 6 hours, matching degree +0.2 when less than 6 hours but greater than 4 hours, matching degree +0.5 when less than or equal to 4 hours; load penalty is: matching degree -0.4 when there are 3 or more current tasks, matching degree +0.2 when there are less than 3 but greater than or equal to 1 current task, matching degree +0.5 when there are less than 1 current task; cross-region penalty is: matching degree -0.5 when the responsible space information and resource occupancy map are inconsistent, matching degree +0.5 when the responsible space information and resource occupancy map are consistent. When the skill proficiency or operation permission is less than that of the resource occupancy map, the matching degree is 0. For example, a task requires a skill proficiency level of 3 or higher, an operation permission level of 2 or higher, and the spatial region information is Cardiovascular Ward 1. A nurse's nursing ability vector is (Level 4, Level 4, 7 hours, 2 tasks, Cardiovascular Ward 1), then the matching degree = 2 + 2 - 0.6 + 0.2 + 0.5 = 4.1. The matching degrees of multiple resource occupancy maps and nursing abilities are calculated. Based on the task matching degree, a task allocation weight matrix is ​​constructed, where each row includes a resource occupancy map, a nursing ability, and a matching degree.

[0065] Using the initial task queue as the matching order and the task allocation weight matrix as the prior probability, iterative resource-capability matching is performed to determine the initial personnel-task sequence set. Specifically, for each task in the initial task queue, multiple nursing capabilities with a matching degree greater than 0 are matched to obtain the initial personnel-task sequence set.

[0066] A multi-objective cost function is formed by defining a time cost factor, a labor intensity factor, and a mobilization cost factor. This function is then used in conjunction with a mutual exclusion conflict graph to iteratively optimize the initial personnel-task sequence set to obtain the personnel-task sequence. Specifically, the time cost factor is defined by the total completion time of nursing tasks, the labor intensity factor is defined by the variance of nursing staff's labor intensity, and the mobilization cost factor is defined by the number of times nursing staff cross nursing zones.

[0067] Dimensionless time cost factors, labor intensity factors, and mobilization cost factors are used, and a multi-objective cost function is constructed using a weighted approach. For example, the dimensionless time cost factor = total completion time of nursing tasks ÷ maximum possible duration, where the maximum possible duration is the maximum nursing duration in the past 365 days of historical nursing logs; the dimensionless labor intensity factor = labor intensity variance ÷ maximum possible variance, where the maximum possible variance is the maximum value of labor intensity variance in the past 365 days of historical nursing logs; and the dimensionless mobilization cost factor = number of times nurses cross nursing zoning ÷ maximum number of times nurses cross nursing zoning ... For example, the time cost factor is assigned a weight of 0.3, the labor intensity factor is assigned a weight of 0.4, and the mobilization cost factor is assigned a weight of 0.3. The cost function is calculated as: 0.3 × dimensionless time cost factor + 0.4 × dimensionless labor intensity factor + 0.3 × dimensionless mobilization cost factor.

[0068] Based on the mutual exclusion conflict graph, a penalty factor is defined. Combining the multi-objective cost function and the penalty factor, the initial personnel-task sequence set is iteratively updated using multi-objective cost function value calculation and random mutation. For example, the penalty factor is defined as follows: if adjacent tasks in the initial personnel-task sequence set are adjacent in the mutual exclusion conflict graph (i.e., mutually exclusive), then the multi-objective cost function result is reduced by 2 × the number of mutually exclusive task pairs. The multi-objective cost function value for each initial personnel-task in the initial personnel-task sequence set is calculated, and a random mutation algorithm is used for random iterative updates.

[0069] If the number of iterations meets the preset iteration constraints, the optimal multi-objective cost function value is output as the personnel-task sequence. For example, if the number of iterations is set to 100, the optimal multi-objective cost function value is output as the personnel-task sequence when the number of iterations reaches 100.

[0070] Based on resource occupancy graphs, capability vectors, and multi-objective cost functions, collaborative optimization of personnel-task sequences is achieved. The task matching weight matrix ensures skill adaptation, the labor intensity variance is minimized to avoid fatigue accumulation, and cost factors are mobilized to compress cross-regional paths.

[0071] S60: Generate multiple nursing task priority sequences according to the personnel-task sequence, and send them to the nursing staff's client accordingly.

[0072] Traditional methods, such as broadcast instructions, cannot support personalized dynamic sequence distribution, resulting in nurses having to manually integrate multi-source task information and queue updates being out of sync in case of emergencies.

[0073] Step S60 in the method provided in this application embodiment further includes:

[0074] Establish a task reordering trigger mechanism, which includes:

[0075] Based on the real-time physiological monitoring data and the preset anomaly discrimination matrix, abnormal events are identified and the level of abnormal events is determined.

[0076] The corresponding task impact range is determined based on the abnormal event and the abnormality discrimination matrix;

[0077] Update the disease severity assessment and disease fluctuation assessment for nursing tasks within the scope of the task's influence, and generate a corresponding local task queue.

[0078] Based on the local task queue, nursing tasks within the scope of the task's influence are locally reordered.

[0079] In this embodiment of the application, multiple nursing task priority sequences are generated according to the personnel-task sequence and distributed to the nursing staff client to guide the nursing staff in performing nursing tasks.

[0080] The method provided in this application embodiment also includes establishing a task rescheduling triggering mechanism.

[0081] Specifically, abnormal events are identified and their levels are determined based on real-time physiological monitoring data and a pre-defined anomaly discrimination matrix. For example, an anomaly discrimination matrix is ​​established where the first column represents abnormal parameters. For instance, a heart rate greater than 100 beats / min and less than or equal to 120 beats / min corresponds to an abnormal event level of 1; a heart rate greater than 120 beats / min and less than or equal to 160 beats / min corresponds to an abnormal event level of 2; and a heart rate greater than 160 beats / min corresponds to an abnormal event level of 3. Higher level numbers indicate more severe abnormal events that require more immediate attention. Based on the abnormal events and the anomaly discrimination matrix, the corresponding task impact range is determined. For example, the affected task range includes nursing tasks that share the same nursing resources as those required by the abnormal event.

[0082] Update the disease severity assessment and disease fluctuation assessment for nursing tasks within the scope of the task's influence, and generate corresponding local task queues.

[0083] Based on a local task queue, nursing tasks within the scope of the task's influence are locally reordered, enabling abnormal events to be handled more quickly.

[0084] By generating personalized nursing task priority sequences and pushing them to the client in real time, the global sequence is broken down into an individual execution list, and the sequence triggered by abnormal events is updated in real time. Digital instructions eliminate information integration delays and can better guide nurses in performing nursing tasks.

[0085] In summary, the embodiments of this application have at least the following technical effects:

[0086] This application proposes an AI-driven method for prioritizing critical care nursing tasks. Through a collaborative optimization mechanism that dynamically integrates multi-dimensional patient condition assessment, resource conflict modeling, and nurse staff capability adaptation, it significantly improves the real-time performance, resource utilization efficiency, and accuracy of personnel adaptation in critical care scenarios. Specifically, based on the quantitative generation of patient condition levels and fluctuation scores using a multimodal hybrid assessment model, early warning of potential deterioration can be achieved, ensuring priority response to high-risk tasks and reducing the risk of rescue due to assessment delays. The collaborative construction of resource occupancy graphs and mutual exclusion conflict graphs automatically identifies equipment resource conflicts, avoiding resource deadlock during prioritization and reducing task execution interruption rates. Combined with multi-objective optimization prioritization using nursing capability vectors, it simultaneously balances nursing workload intensity and spatial movement costs, preventing efficiency losses caused by high-skilled nurses being occupied by low-value tasks or frequent cross-regional transfers. Furthermore, the task reordering trigger mechanism, through real-time abnormal event identification and local queue reconstruction, ensures the system can quickly respond to changes in patient condition during emergency rescue scenarios, maintaining the dynamic stability of the overall task queue while improving the efficiency of emergency event handling. Compared with traditional methods, the technical solution provided in this application significantly overcomes the subjective lag of human experience and the rigidity of static allocation, achieving the technical effect of precise allocation of nursing tasks with patient safety as the core orientation, minimization of resource conflicts as the constraint, and optimal matching of nursing staff capabilities as the execution guarantee.

[0087] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0088] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0089] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. An artificial intelligence-driven method for prioritizing critical care nursing tasks, characterized in that, include: Acquire real-time physiological monitoring data, historical medical records and diagnostic labels of intensive care patients, and conduct corresponding assessments of disease severity and disease fluctuation to obtain disease severity and fluctuation scores; Nursing task information is obtained for resource dependency analysis, and a resource occupancy graph and a mutual exclusion conflict graph are constructed based on the results of the resource dependency analysis. Based on the resource occupancy graph, the mutual exclusion conflict graph, the disease severity level, and the fluctuation score, an initial task queue is generated using a weighted directed graph sorting algorithm. Nursing staff competency modeling is performed to construct a nursing competency vector, which includes information representing at least the staff's skill proficiency, operational authority level, work fatigue status, current task load, and the spatial area they are responsible for. Combining the initial task queue, the resource occupancy graph, the mutual exclusion conflict graph, and the nursing competency vector, multi-objective ranking optimization of nursing tasks based on resource-competency matching is performed to obtain a staff-task sequence. Multiple nursing task priority sequences are generated according to the staff-task sequence and distributed to the nursing staff's client. Real-time physiological monitoring data, historical condition records, and diagnostic tags of intensive care patients are acquired, and corresponding condition severity assessments and condition fluctuation assessments are performed. Obtaining disease severity and fluctuation scores includes: constructing a multimodal hybrid assessment model, wherein the multimodal hybrid assessment model includes at least: a first assessment channel based on a rule engine, wherein the first assessment channel is used to calculate indicator deviation entropy, treatment dependence intensity, and vital sign chaos degree based on the real-time physiological monitoring data and the historical disease records, and weighted to generate a first assessment level; a second assessment channel based on a deep learning model, wherein the second assessment channel is used to generate a second assessment level based on the real-time physiological monitoring data; inputting the real-time physiological monitoring data and the historical disease records into the multimodal hybrid assessment model, obtaining the first assessment level and the second assessment level, and merging the first assessment level and the second assessment level. The assessment level, the second assessment level, and the diagnostic label are used to generate the condition level; the initial task queue, the resource occupancy graph, the mutual exclusion conflict graph, and the nursing capability vector are combined to perform multi-objective ranking optimization of nursing tasks based on resource-capability matching to obtain a personnel-task sequence, including: traversing the resource occupancy graph and the nursing capability vector to calculate the task matching degree, and constructing a task allocation weight matrix based on the task matching degree; iteratively performing resource-capability matching with the initial task queue as the matching order and the task allocation weight matrix as the prior probability to determine the initial personnel-task sequence set; defining time cost factor, labor intensity factor, and mobilization cost factor to form a multi-objective cost function. The initial personnel-task sequence set is iteratively optimized using the mutual exclusion conflict graph to obtain the personnel-task sequence. A time cost factor, a labor intensity factor, and a mobilization cost factor are defined to form a multi-objective cost function. The initial personnel-task sequence set is then iteratively optimized using the mutual exclusion conflict graph to obtain the personnel-task sequence. This includes: defining the time cost factor based on the total completion time of nursing tasks; defining the labor intensity factor based on the variance of nursing staff's labor intensity; and defining the mobilization cost factor based on the number of times nursing staff cross nursing zones. The time cost factor, labor intensity factor, and mobilization cost factor are dimensionless, and the multi-objective cost function is constructed using a weighted method.Based on the mutually exclusive conflict graph, a penalty factor is defined. Combined with the multi-objective cost function and the penalty factor, the initial personnel-task sequence set is iteratively updated using multi-objective cost function value calculation and random mutation. If the number of iterations satisfies a preset iteration constraint, the optimal multi-objective cost function value is output as the personnel-task sequence.

2. The artificial intelligence-driven intensive care nursing task prioritization method as described in claim 1, characterized in that, The method involves acquiring real-time physiological monitoring data, historical medical records, and diagnostic labels of patients in intensive care, and performing corresponding assessments of disease severity and fluctuation to obtain disease severity and fluctuation scores. The method further includes: extracting historical indicator fluctuation features and historical fluctuation scores based on the historical medical records, wherein the historical indicator fluctuation features and historical fluctuation scores correspond one-to-one; constructing and training a fluctuation assessment model using the historical indicator fluctuation features and historical fluctuation scores as a sample dataset; performing feature engineering processing on the real-time physiological monitoring data, and inputting the feature engineering processing results into the fluctuation assessment model to perform fluctuation assessment and obtain the fluctuation score.

3. The AI-driven intensive care nursing task prioritization method as described in claim 2, characterized in that, The process involves acquiring nursing task information, performing resource dependency analysis, and constructing a resource occupancy graph and a mutual exclusion conflict graph based on the results of the resource dependency analysis. This includes: performing structured parsing of all nursing tasks within the intensive care unit environment to extract resource dependency information; constructing a directed graph based on the resource dependency information, where graph nodes represent tasks and edges represent resource occupancy relationships, thus forming the resource occupancy graph; performing resource mutual exclusion analysis on the resource occupancy graphs of multiple tasks; and constructing the mutual exclusion conflict graph based on the results of the resource mutual exclusion analysis.

4. The AI-driven intensive care nursing task prioritization method as described in claim 1, characterized in that, Also includes: A task rescheduling triggering mechanism is established, which includes: identifying abnormal events based on the real-time physiological monitoring data and a preset abnormality discrimination matrix, and determining the level of the abnormal event; The corresponding task impact range is determined based on the abnormal event and the abnormality discrimination matrix; the nursing tasks within the task impact range are subject to updated disease level assessment and disease fluctuation assessment, and a local task queue is generated accordingly; the nursing tasks within the task impact range are locally reordered based on the local task queue.

Images