Medical quality and operation performance evaluation method and system based on intelligent analysis

Abstract

The application provides a medical quality and operation performance evaluation method and system based on intelligent analysis, relates to the technical field of data processing, and comprises the following steps: constructing a multi-level influence network and calculating the influence weight of each node through time sequence correlation analysis, identifying the real contribution degree of key nodes to performance deviation based on counterfactual deduction, tracing a complete cause-effect chain along a cause-effect transmission path and generating a synergistic optimization strategy under resource constraints, updating the network and generating rules according to feedback data after the strategy is applied, and realizing continuous optimization.

Description

Technical Field

[0001] This invention relates to data processing technology, and more particularly to a method and system for evaluating medical quality and operational performance based on intelligent analysis. Background Technology

[0002] With the continuous development of the healthcare service system, healthcare quality management and operational efficiency optimization have become core challenges for medical institutions. Existing methods for evaluating healthcare quality and operational performance have significant shortcomings: they struggle to reveal the true causal relationships between various stages of the healthcare process and quality indicators, often only identifying superficial correlations and failing to accurately determine the actual impact of each stage on the final quality assessment result, leading to unclear optimization directions. Furthermore, existing evaluation systems lack scientific counterfactual analysis mechanisms, failing to accurately quantify the true contribution of abnormal states at specific operational nodes to overall performance. This results in a lack of objective basis for responsibility definition and resource allocation decisions, often leading to a mismatch between resource investment and actual problems. Summary of the Invention

[0003] This invention provides a method and system for evaluating medical quality and operational performance based on intelligent analysis, which can solve the problems in the prior art.

[0004] A first aspect of this invention provides a method for evaluating medical quality and operational performance based on intelligent analysis, comprising:

[0005] We acquire operational and quality assessment data from medical institutions, identify causal transmission paths between various business process nodes and changes in quality indicators through time-series correlation analysis, construct a multi-level influence network, and calculate the influence weight of each node on the quality assessment results.

[0006] Key nodes whose influence weight exceeds a preset influence threshold are selected. The current abnormal execution state is replaced with the historical normal execution state. The counterfactual performance benchmark is re-derived using the multi-level influence network. The difference between the actual performance state and the counterfactual performance benchmark is quantified to obtain the true contribution of each key node to the performance deviation.

[0007] For key nodes whose actual contribution exceeds a preset level, the complete causal chain is traced along the causal transmission path of the multi-level influence network, and a collaborative optimization strategy for each intermediate node in the complete causal chain is generated under resource constraints;

[0008] The collaborative optimization strategy is applied to the operation and management process, and the generation rules of the multi-level influence network and counterfactual performance benchmark are updated based on the feedback data after execution.

[0009] The steps involved in identifying the causal transmission paths between various business process nodes and changes in quality indicators through time-series correlation analysis, constructing a multi-level influence network, and calculating the influence weight of each node on the quality assessment results include:

[0010] Acquire business operation data and quality assessment data from medical institutions and divide them into time windows. Within each time window, extract the execution status characteristics and quality indicator change characteristics of business process nodes.

[0011] Calculate the correlation strength between the execution state features and the change features under different time lag conditions, and identify the direct influence relationship based on the time decay pattern of the correlation strength;

[0012] For business process nodes with direct impact relationships, trace the execution status characteristics of their preceding nodes along the time sequence to identify indirect impact relationships and their transmission paths transmitted through intermediate nodes;

[0013] A multi-level influence network is constructed based on the aforementioned direct and indirect influence relationships, with each business process node serving as a network node and the direct and indirect influence relationships serving as network edges;

[0014] For each business process node in the multi-level influence network, the direct influence of the node is calculated based on the direct influence relationship, and the indirect influence of the node is calculated based on the transmission path. The direct influence and indirect influence are weighted and summed to obtain the influence weight of the node on the quality assessment result.

[0015] The steps of selecting key nodes whose influence weight exceeds a preset influence threshold, replacing the current abnormal execution state with the historical normal execution state, re-deriving the counterfactual performance benchmark using the multi-level influence network, and quantifying the difference between the actual performance state and the counterfactual performance benchmark to obtain the true contribution of each key node to the performance deviation include:

[0016] Based on the influence weight, business process nodes that exceed the preset influence threshold are marked as key nodes, and the current execution status and historical normal execution status of each key node are extracted;

[0017] For each critical node, a historical state similar to the business characteristics of the current time window is selected from its historical normal execution states as a replacement benchmark, and the current abnormal execution state of the critical node is replaced with the replacement benchmark;

[0018] Based on the execution status of the replaced key node, the causal transmission path containing the key node is traced along the multi-level influence network, and the transmission direction and strength of the network edges are used to extrapolate to the quality indicator node, thus obtaining the counterfactual quality indicator value corresponding to the key node; the counterfactual quality indicator values ​​corresponding to all key nodes are summarized to obtain the counterfactual performance benchmark.

[0019] Calculate the numerical difference between the actual performance status and the counterfactual performance benchmark, and decompose the numerical difference to each key node to obtain the true contribution of each key node to the performance deviation.

[0020] Based on the execution state of the replaced key node, the steps of extrapolating to the quality indicator node along the causal transmission path containing the key node in the multi-level influence network, according to the transmission direction and strength of the network edges, to obtain the counterfactual quality indicator value corresponding to the key node include:

[0021] For each key node, identify all causal transmission paths from that key node to the quality indicator node in the multi-level influence network, and extract the transmission strength of each network edge on each causal transmission path;

[0022] For each causal transmission path, the execution state of the replaced key node is used as the starting input. The state of each intermediate node is deduced sequentially along the path according to the transmission direction and transmission strength until the quality index node is reached, thus obtaining the path counterfactual quality index value corresponding to the path.

[0023] Obtain the counterfactual quality index value and the true quality index value of each causal transmission path in the historical inference scenario, calculate the historical inference deviation of each path, and adjust the transmission strength of each network edge on the path based on the historical inference deviation;

[0024] Based on the adjusted transmission strength, the path deduction is re-executed to obtain the updated path counterfactual quality index value for each causal transmission path;

[0025] Calculate the path weight for each causal transmission path, and then weight and fuse the updated counterfactual quality index values ​​of all causal transmission paths according to the path weights to obtain the final counterfactual quality index value corresponding to the key node.

[0026] For key nodes whose actual contribution exceeds a preset level, the steps of tracing the complete causal chain along the causal transmission path of the multi-level influence network and generating collaborative optimization strategies for each intermediate node in the complete causal chain under resource constraints include:

[0027] Key nodes whose actual contribution exceeds a preset level are selected, and their predecessor nodes are traced back along the causal transmission path in the multi-level influence network to identify the complete causal chain and record the current execution status of each node in the complete causal chain;

[0028] Obtain the current resource constraints, and for each intermediate node in the complete causal chain, calculate the feasible optimization action space based on the node's current execution state and resource constraints;

[0029] Starting from the predecessor node, optimization actions are selected sequentially for each intermediate node from its optimization action space along the complete causal chain. The optimization action of each intermediate node is selected from the optimization action space of that node based on the optimization action of its predecessor node and resource constraints, generating multiple candidate collaborative optimization strategies.

[0030] For each candidate collaborative optimization strategy, the optimization actions of each intermediate node are taken as input, and the optimization is derived to the quality indicator node along the causal transmission path to obtain the optimized quality indicator value. The expected performance improvement of the candidate collaborative optimization strategy is quantified based on the difference between the optimized quality indicator value and the current quality indicator value. The strategy with the largest expected performance improvement is selected as the final collaborative optimization strategy.

[0031] The steps of generating multiple candidate collaborative optimization strategies by sequentially selecting optimization actions from the optimization action space of each intermediate node along the complete causal chain, starting from the predecessor node, include:

[0032] Initialize the policy generation tree, taking the first predecessor node of the complete causal chain as the root node, and select multiple optimization actions as branches from the optimization action space of this node; for each branch, based on the optimization action and resource constraints corresponding to the branch, dynamically reduce the optimization action space of the next intermediate node to obtain the constrained optimization action space;

[0033] From the constrained optimization action space, select multiple optimization action expansion branches, and repeatedly execute the dynamic reduction and branch expansion process until all intermediate nodes on the complete causal chain have been traversed, resulting in multiple complete paths from the root node to the leaf node;

[0034] Check whether the total resource consumption of optimization actions across all nodes on each complete path exceeds the resource constraint; if it does, prune the corresponding path.

[0035] All unpruned complete paths are converted into candidate collaborative optimization strategies, where the optimization actions of each node on each path constitute a candidate collaborative optimization strategy.

[0036] The steps of applying the collaborative optimization strategy to the operation and management process and updating the generation rules of the multi-level influence network and counterfactual performance benchmark based on the feedback data after execution include:

[0037] The collaborative optimization strategy is distributed to the business process execution units corresponding to each intermediate node in the complete causal chain, and business operation data and quality assessment data after the strategy execution are collected as feedback data.

[0038] Based on the feedback data, the correlation strength between each business process node is recalculated. For network edges where the correlation strength change exceeds a preset strength threshold, the original propagation strength is replaced with the recalculated correlation strength, and the multi-level influence network is updated.

[0039] Based on the updated multi-level influence network, the counterfactual quality index values ​​of the original key nodes are re-inferred. If the deviation between the inferred values ​​and the actual quality index values ​​after execution exceeds the preset range, the selection rules for historical normal execution states are adjusted, and the generation rules for counterfactual performance benchmarks are updated.

[0040] The updated rules for generating multi-level influence networks and counterfactual performance benchmarks will be used in the next round of performance evaluation and optimization.

[0041] A second aspect of the present invention provides a medical quality and operational performance evaluation system based on intelligent analysis, comprising:

[0042] The data acquisition and impact analysis module is used to acquire business operation data and quality assessment data from medical institutions. Through time-series correlation analysis, it identifies the causal transmission paths between various business process nodes and changes in quality indicators, constructs a multi-level impact network, and calculates the impact weight of each node on the quality assessment results.

[0043] The counterfactual analysis and contribution calculation module is used to select key nodes whose influence weight exceeds a preset influence threshold, replace the current abnormal execution state with the historical normal execution state, re-derive the counterfactual performance benchmark using the multi-level influence network, quantify the difference between the actual performance state and the counterfactual performance benchmark, and obtain the true contribution of each key node to the performance deviation.

[0044] The collaborative optimization strategy generation module is used to trace the complete causal chain along the causal transmission path of the multi-level influence network for key nodes whose actual contribution exceeds a preset level, and generate collaborative optimization strategies for each intermediate node in the complete causal chain under resource constraints;

[0045] The strategy execution and dynamic update module is used to apply the collaborative optimization strategy to the operation and management process, and update the generation rules of the multi-level influence network and counterfactual performance benchmark based on the feedback data after execution.

[0046] A third aspect of the present invention provides an electronic device, comprising:

[0047] processor;

[0048] Memory used to store processor-executable instructions;

[0049] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0050] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0051] This invention constructs a multi-level influence network through temporal correlation analysis, enabling accurate identification of causal relationships between business process nodes and quality indicators in medical institutions, overcoming the limitations of traditional evaluation methods in identifying key influencing factors. Employing counterfactual reasoning techniques, it replaces the current abnormal execution state with historical normal execution states, objectively quantifying the true contribution of each key node to performance deviations, avoiding evaluation biases caused by subjective judgments in traditional methods. By tracing the complete causal chain along the causal transmission path, it achieves precise localization of the root causes of deep-seated problems and generates collaborative optimization strategies under resource constraints, improving the targeting of optimization measures and the efficiency of resource utilization. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the medical quality and operational performance evaluation method based on intelligent analysis, as described in an embodiment of the present invention.

[0053] Figure 2 Flowchart for assessing the true contribution of key nodes. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0055] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0056] Figure 1 This is a flowchart illustrating the medical quality and operational performance evaluation method based on intelligent analysis, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0057] We acquire operational and quality assessment data from medical institutions, identify causal transmission paths between various business process nodes and changes in quality indicators through time-series correlation analysis, construct a multi-level influence network, and calculate the influence weight of each node on the quality assessment results.

[0058] Key nodes whose influence weight exceeds a preset influence threshold are selected. The current abnormal execution state is replaced with the historical normal execution state. The counterfactual performance benchmark is re-derived using the multi-level influence network. The difference between the actual performance state and the counterfactual performance benchmark is quantified to obtain the true contribution of each key node to the performance deviation.

[0059] For key nodes whose actual contribution exceeds a preset level, the complete causal chain is traced along the causal transmission path of the multi-level influence network, and a collaborative optimization strategy for each intermediate node in the complete causal chain is generated under resource constraints;

[0060] The collaborative optimization strategy is applied to the operation and management process, and the generation rules of the multi-level influence network and counterfactual performance benchmark are updated based on the feedback data after execution.

[0061] In one optional implementation, the steps of identifying the causal transmission path between each business process node and changes in quality indicators through time-series correlation analysis, constructing a multi-level influence network, and calculating the influence weight of each node on the quality assessment result include:

[0062] Acquire business operation data and quality assessment data from medical institutions and divide them into time windows. Within each time window, extract the execution status characteristics and quality indicator change characteristics of business process nodes.

[0063] Calculate the correlation strength between the execution state features and the change features under different time lag conditions, and identify the direct influence relationship based on the time decay pattern of the correlation strength;

[0064] For business process nodes with direct impact relationships, trace the execution status characteristics of their preceding nodes along the time sequence to identify indirect impact relationships and their transmission paths transmitted through intermediate nodes;

[0065] A multi-level influence network is constructed based on the aforementioned direct and indirect influence relationships, with each business process node serving as a network node and the direct and indirect influence relationships serving as network edges;

[0066] For each business process node in the multi-level influence network, the direct influence of the node is calculated based on the direct influence relationship, and the indirect influence of the node is calculated based on the transmission path. The direct influence and indirect influence are weighted and summed to obtain the influence weight of the node on the quality assessment result.

[0067] For example, operational data from healthcare institutions typically includes time records, execution records, and result records for each step of the patient's treatment process, such as registration time, waiting time, treatment time, laboratory and examination time, and medication dispensing time. Quality assessment data includes indicators related to medical quality, service quality, and operational efficiency, such as patient satisfaction, medical error rate, and average length of stay. The acquired data is segmented according to fixed time windows (e.g., daily, weekly, or monthly), and within each time window, the execution status characteristics of business process nodes and the change characteristics of quality indicators are extracted. Execution status characteristics may include node completion rate, average execution time, and exception rate; the change characteristics of quality indicators may include the absolute value, year-on-year change rate, and standard deviation of the quality indicators.

[0068] Taking the inpatient nursing process as an example, we can extract characteristics such as ward handover completion rate, nursing documentation standardization rate, and timely implementation rate of nursing measures as execution status features of business process nodes, and use patient satisfaction scores, pressure ulcer incidence rate, and hospital infection rate as quality indicator change features. Within each monthly time window, we calculate the statistical values ​​and trends of these features.

[0069] For each pair of business process nodes and quality indicators, calculate the Pearson correlation coefficient, Spearman's rank correlation coefficient, or mutual information value between the execution status characteristics of the node at time t and the change characteristics of the quality indicator at time t+Δt, where Δt is a time lag parameter, ranging from 0 to the maximum possible impact delay time (e.g., 7 days). The calculated correlation coefficient or mutual information value represents the association strength between the business process node and the quality indicator. If the association strength reaches its maximum at a certain Δt value and is significantly higher than the random level (e.g., the p-value of the statistical test of the correlation coefficient is less than 0.05), then the business process node and the quality indicator are considered to have a direct influence relationship. The association strength of this direct influence relationship is the absolute value of the maximum correlation coefficient, and the impact delay time is the corresponding Δt value.

[0070] For example, the analysis found that the negative correlation between the timeliness of nursing intervention implementation and the incidence of pressure ulcers was strongest at Δt=3 days (Pearson correlation coefficient was -0.78, p value was 0.002), indicating that after the timeliness of nursing intervention implementation was improved, the incidence of pressure ulcers would be significantly reduced after 3 days, and there was a direct influence between the two. The correlation strength of this direct influence was 0.78 (taking the absolute value of the correlation coefficient).

[0071] For a business process node (denoted as node A) with a direct impact relationship, trace its preceding node (denoted as node B) along the time series to identify the indirect impact relationship and its transmission path through intermediate nodes. Specifically, find node B whose execution state changes earlier than node A and has a significant temporal correlation with the execution state characteristics of node A. If the execution state characteristics of node B change at time t1, and the execution state characteristics of node A change at time t2 (t2>t1), and the absolute value of the correlation coefficient between the execution state characteristics of node B and node A exceeds a preset threshold (e.g., 0.6) and the p-value of the statistical test is less than 0.05, then node B and node A are considered to have a significant correlation. If a significant correlation is confirmed between node B and node A, and node A has a direct impact on the quality indicator, it can be identified that node B indirectly affects the quality indicator through node A, forming a transmission path of "node B → node A → quality indicator". Repeating this tracing process can identify more preceding nodes and longer transmission paths.

[0072] Continuing with the above example, if a significant positive correlation is found between the execution status characteristics of the rationality of nursing staff allocation and the execution status characteristics of the timeliness of nursing intervention execution at Δt=1 day (Pearson correlation coefficient of 0.82, p-value of 0.001), and the change time of the rationality of nursing staff allocation precedes the change time of the timeliness of nursing intervention execution, then the transmission path of "rationality of nursing staff allocation → timeliness of nursing intervention execution → pressure ulcer incidence" can be identified. This transmission path indicates that the rationality of nursing staff allocation indirectly affects the pressure ulcer incidence by influencing the timeliness of nursing intervention execution, with an association strength of 0.82 for "rationality of nursing staff allocation → timeliness of nursing intervention execution" and 0.78 for "timeliness of nursing intervention execution → pressure ulcer incidence".

[0073] A multi-level influence network is constructed based on the identified direct and indirect influence relationships. In this network, each business process node serves as a network node, and direct and indirect influence relationships are represented as network edges. The direction of the edge indicates the direction of influence transmission, and the weight of the edge is set to the corresponding association strength. Quality indicators also serve as nodes in the network; specifically, the state of a quality indicator node is characterized by the change characteristics of the quality indicator. When constructing the network, nodes can be organized into multiple levels according to the actual execution order and temporal relationship of the business processes, from upstream business process nodes (such as personnel allocation and resource preparation) to midstream nodes (such as specific operation execution), downstream nodes (such as operation result recording), and finally pointing to quality indicator nodes.

[0074] For each business process node in the multi-level influence network, its influence weight on the quality assessment result is calculated. The direct influence of a node can be directly obtained through the correlation strength between the node and the quality indicator. If there is no direct influence relationship between the node and the quality indicator, its direct influence is 0. The indirect influence of a node needs to be calculated along the transmission path: identify all transmission paths from the node to the quality indicator node; for each transmission path, multiply the weights (i.e., correlation strengths) of all edges on the path to obtain the transmission coefficient of the path; sum the transmission coefficients of all transmission paths from the node to the quality indicator node to obtain the total indirect influence of the node. Finally, the direct and indirect influence of the node are weighted and accumulated according to a preset ratio (e.g., 50% each) to obtain the comprehensive influence weight of the node on the quality assessment result.

[0075] For example, the direct impact of the timely implementation rate of nursing interventions on the incidence of pressure ulcers is 0.78 (the strength of the direct impact relationship between it and the incidence of pressure ulcers); the rationality of nursing staff allocation has no direct impact on the incidence of pressure ulcers, so its direct impact is 0. However, this node indirectly affects the incidence of pressure ulcers through the path "rationality of nursing staff allocation → timely implementation rate of nursing interventions → incidence of pressure ulcers," and the transmission coefficient of this path is 0.82 × 0.78 = 0.64. Therefore, the indirect impact of the rationality of nursing staff allocation is 0.64. If the weights of the direct and indirect impacts are each set to 0.5, then the overall impact weight of the rationality of nursing staff allocation on the incidence of pressure ulcers is 0 × 0.5 + 0.64 × 0.5 = 0.32.

[0076] Using the methods described above, medical institutions can clearly identify the degree of impact and transmission path of each business process node on quality indicators, thereby enabling them to make targeted quality improvements, optimize resource allocation, and enhance the quality of medical services.

[0077] In one optional implementation, the steps of selecting key nodes whose influence weight exceeds a preset influence threshold, replacing the current abnormal execution state with the historical normal execution state, re-deriving the counterfactual performance benchmark using the multi-level influence network, and quantifying the difference between the actual performance state and the counterfactual performance benchmark to obtain the true contribution of each key node to the performance deviation include:

[0078] Based on the influence weight, business process nodes that exceed the preset influence threshold are marked as key nodes, and the current execution status and historical normal execution status of each key node are extracted;

[0079] For each critical node, a historical state similar to the business characteristics of the current time window is selected from its historical normal execution states as a replacement benchmark, and the current abnormal execution state of the critical node is replaced with the replacement benchmark;

[0080] Based on the execution status of the replaced key node, the causal transmission path containing the key node is traced along the multi-level influence network, and the transmission direction and strength of the network edges are used to extrapolate to the quality indicator node, thus obtaining the counterfactual quality indicator value corresponding to the key node; the counterfactual quality indicator values ​​corresponding to all key nodes are summarized to obtain the counterfactual performance benchmark.

[0081] Calculate the numerical difference between the actual performance status and the counterfactual performance benchmark, and decompose the numerical difference to each key node to obtain the true contribution of each key node to the performance deviation.

[0082] Combination Figure 2 The flowchart for assessing the true contribution of key nodes is explained. Based on the previously calculated comprehensive impact weight of each business process node on quality indicators, a preset impact threshold is set. For example, nodes with the top 20% impact weight or nodes with an impact weight greater than 0.3 can be marked as key nodes. For each marked key node, two types of execution status data are extracted: current execution status and historical normal execution status. The current execution status includes the actual operating parameters of the node within the assessment period, such as the actual nursing staff ratio, drug delivery response time, average utilization rate of examination equipment, and timely execution rate of medical orders; the historical normal execution status includes the execution parameter records of the node during periods of good performance in the past.

[0083] The selection of historical normal performance status follows these criteria: First, select time windows from the most recent 6 to 12 months of historical data where the quality indicator values ​​are within the range of the historical mean plus or minus one standard deviation. These time windows represent periods of normal or good medical quality performance. Second, ensure that the selected historical data includes at least 5 time windows to guarantee statistical validity. Finally, among the historical periods that meet the above conditions, prioritize the period with the best quality indicator performance as the historical normal performance status.

[0084] For each key node, a suitable historical state is selected as the replacement benchmark. This step employs a similarity matching method to ensure that the selected historical state is closest to the current business environment. Specifically, a business feature vector for the current time window is first constructed. Business features include seasonal factors (such as flu peaks, summer peak seasons for outpatient visits, etc.), patient visits, the proportion of hospitalized patients with different diseases, average patient age, and the proportion of severe cases. These business features are normalized to form the current feature vector. Similarly, a business feature vector is constructed for each candidate historical normal period. The cosine similarity between the current business feature vector and the business feature vectors of each historical period is calculated. The specific calculation method is as follows: perform a dot product operation between the current feature vector and the historical feature vector, and then divide by the product of the magnitude of the current feature vector and the magnitude of the historical feature vector. The result is the cosine similarity value. The historical period with the highest similarity and a similarity value greater than a preset threshold (e.g., 0.8) is selected as the replacement benchmark. If multiple historical states with close similarities exist, the quality indicator performance of these historical states is further compared, and the historical state with the best quality indicator value is selected as the final replacement benchmark.

[0085] For example, when assessing nursing staffing during the peak influenza season in winter, a period in historical data that is also a peak influenza season, with similar patient visits and disease composition should be selected as the replacement benchmark, rather than summer or a period with significantly different patient composition, to ensure the rationality of the comparison.

[0086] For each critical node, its current execution state is replaced with the selected historical normal execution state, while keeping the states of other nodes unchanged. The state change of the critical node is calculated, defined as the difference between the historical normal execution state parameter value and the current execution state parameter value. Following the causal propagation path in the multi-level influence network, the influence is extrapolated from the replaced critical node along the propagation direction of the network edges.

[0087] During the deduction process, the influence propagation follows these rules: Let the state change of key node A be ΔA, and the weight of the edge between node A and its direct downstream node B (i.e., the association strength, determined by the absolute value of the aforementioned correlation coefficient) be W_AB. Then, the influence of node A's state change on node B is equal to ΔA × W_AB. If node B is simultaneously affected by multiple upstream nodes, the influence propagated from all upstream nodes to node B is summed to obtain the total state change of node B. This process continues, propagating the influence layer by layer along the network path until it reaches the quality indicator node, yielding the quality indicator change ΔQ corresponding to the replacement of the key node. Adding this change ΔQ to the current actual quality indicator value gives the counterfactual quality indicator value corresponding to the key node.

[0088] In practical implementation, the extrapolation calculation can be implemented recursively. Starting from the key node to be replaced, the impact propagation value is calculated using the state change amount × edge weight based on the relationship with its directly connected downstream nodes. These impacts are then used as state changes for downstream nodes and propagated to even more downstream nodes until the impact reaches the quality indicator node. For example, if the state of the nursing staff allocation rationality node is replaced, replacing the current allocation (e.g., nurse-patient ratio 1:0.4) with the historical normal nursing staff ratio (e.g., nurse-patient ratio 1:0.6), the state change amount is -0.2. Assuming the edge weight between nursing staff allocation rationality and the timeliness of nursing intervention implementation is 0.82, the change in the timeliness of nursing intervention implementation is -0.2 × 0.82 = -0.164 (a lower nursing staff ratio indicates more sufficient nursing staff, thus improving the timeliness of implementation). Furthermore, assuming the edge weight between the timeliness of nursing intervention implementation and the pressure ulcer incidence rate is 0.78, the change in the pressure ulcer incidence rate is -0.164 × 0.78 = -0.128. If the current actual pressure ulcer incidence rate is 3%, then the counterfactual pressure ulcer incidence rate corresponding to this critical point is 3% - 0.128% = 2.872%, indicating that if the nursing staff allocation is restored to the historical normal level, the pressure ulcer incidence rate can be reduced to 2.872%.

[0089] This deduction process is repeated for all critical nodes to obtain the counterfactual quality index value for each critical node. All critical nodes are then simultaneously restored to their historical normal execution state, and the impact propagation is calculated comprehensively according to all transmission paths in the network to obtain the complete counterfactual performance benchmark. This benchmark represents the achievable quality index level assuming all critical nodes execute under historical normal conditions.

[0090] The actual performance status is compared with the counterfactual performance benchmark, and the numerical difference between the two is calculated. This difference equals the counterfactual performance benchmark value minus the actual performance value. This difference represents the potential performance improvement space achievable if all key nodes perform according to historical normal conditions. The overall difference is further decomposed into individual key nodes, and the true contribution of each key node to the performance deviation is calculated. A marginal contribution calculation method is used: each key node's state is individually replaced with its historical normal state, while keeping the states of other key nodes unchanged. The change in quality indicators is calculated using the aforementioned extrapolation method; this change is the marginal contribution of that node. The marginal contributions of all key nodes are normalized. The percentage contribution of a given key node is calculated by dividing that node's marginal contribution by the sum of the marginal contributions of all key nodes, and then multiplying by 100%.

[0091] Through the above steps, this method realizes the transformation from a multi-level influence network to precise performance attribution, providing an actionable analytical tool for quality management and operational performance improvement in medical institutions.

[0092] In one optional implementation, the step of obtaining the counterfactual quality index value corresponding to the key node by extrapolating to the quality index node along the causal transmission path containing the key node in the multi-level influence network based on the execution state of the replaced key node includes:

[0093] For each key node, identify all causal transmission paths from that key node to the quality indicator node in the multi-level influence network, and extract the transmission strength of each network edge on each causal transmission path;

[0094] For each causal transmission path, the execution state of the replaced key node is used as the starting input. The state of each intermediate node is deduced sequentially along the path according to the transmission direction and transmission strength until the quality index node is reached, thus obtaining the path counterfactual quality index value corresponding to the path.

[0095] Obtain the counterfactual quality index value and the true quality index value of each causal transmission path in the historical inference scenario, calculate the historical inference deviation of each path, and adjust the transmission strength of each network edge on the path based on the historical inference deviation;

[0096] Based on the adjusted transmission strength, the path deduction is re-executed to obtain the updated path counterfactual quality index value for each causal transmission path;

[0097] Calculate the path weight for each causal transmission path, and then weight and fuse the updated counterfactual quality index values ​​of all causal transmission paths according to the path weights to obtain the final counterfactual quality index value corresponding to the key node.

[0098] For example, for each key node in a replaced state, all possible transmission paths from that key node to the quality indicator node are identified by traversing a multi-level influence network. In the influence network, the connections between nodes have directional and strength attributes, representing the direction and degree of influence of the causal relationship. The transmission strength of an edge is determined by the absolute value of the correlation coefficient calculated in the preceding steps, i.e., the absolute value of the Pearson correlation coefficient or Spearman rank correlation coefficient of the historical data between two nodes, ranging from 0 to 1; a larger value indicates a stronger influence transmission. For example, for a key node K1, there may be multiple paths to the quality indicator node Q, such as K1→M1→Q, K1→M2→M3→Q, etc., where M1, M2, and M3 are intermediate nodes. Assuming the absolute value of the historical correlation coefficient between K1 and M1 is 0.7, and the absolute value of the historical correlation coefficient between M1 and Q is 0.8, then the edge transmission strength from K1 to M1 is 0.7, and the edge transmission strength from M1 to Q is 0.8. During the identification process, a depth-first search algorithm is used to recursively explore all possible paths to the quality index nodes along the direction of the network edges, starting from the key nodes, while recording the transmission strength value of each network edge on each path.

[0099] After determining all transmission paths, a state deduction is performed for each path. Here, quality indicator nodes refer to nodes in the multi-level influence network that represent the final quality evaluation indicators, such as pressure ulcer incidence, patient satisfaction, and product qualification rate nodes. Taking the path K1→M1→Q as an example, where Q is the quality indicator node, assuming the execution state of K1 after replacement is 0.85 (indicating a good level of execution), and the edge transmission strength from K1 to M1 is 0.7, and the edge transmission strength from M1 to Q is 0.8. First, using the state value of K1 (0.85) as the initial input, the state value of M1 is calculated based on the transmission strength of 0.7: 0.85 × 0.7 = 0.595. Then, using the state value of M1 (0.595) as the input, the state value of Q is calculated based on the transmission strength of 0.8: 0.595 × 0.8 = 0.476. Thus, 0.476 is the path counterfactual quality indicator value for this path. For paths containing more intermediate nodes, the state of each node is deduced sequentially using the same principle until the value of the quality indicator node is obtained.

[0100] To improve the accuracy of the simulation, the transmission intensity needs to be adjusted based on historical data. The difference between the simulation results and the actual quality index values ​​for each transmission path is extracted from historical simulation scenarios, and the historical simulation bias is calculated. For example, if for a path K1→M1→Q in historical scenario A, the execution state of K1 is 0.8, and the simulation is performed using initial transmission intensities of 0.7 and 0.8, the simulation result is 0.8×0.7×0.8=0.448. However, the actual quality index value in this scenario is 0.52, so the bias is 0.52-0.448=0.072, and the bias rate is 0.072÷0.52×100%=13.8%. This indicates that the current transmission intensity underestimates the actual impact transmission effect.

[0101] Based on this deviation, an adaptive adjustment strategy can be used to correct the conduction strength. Specifically, for a positive deviation (the projected value is less than the actual value), it indicates that the conduction strength is underestimated, and the conduction strength on each side of the path should be appropriately increased; for a negative deviation (the projected value is greater than the actual value), it indicates that the conduction strength is overestimated, and the conduction strength should be appropriately decreased. The adjustment range can be achieved using a proportional correction method, that is, allocating a certain proportion (e.g., 50%) of the deviation rate to each side of the path.

[0102] Taking the aforementioned path K1→M1→Q as an example, the deviation rate is 13.8%. Taking 50% as the adjustment coefficient, the adjustment range is 13.8%×50%=6.9%. This path contains two edges, and the adjustment range can be distributed evenly or proportionally to the initial strength of each edge. Using the even distribution method, the adjustment range for each edge is 6.9%÷2=3.45%. Therefore, the conduction strength from K1 to M1 is adjusted from 0.7 to 0.7×(1+3.45%)=0.72415, and the conduction strength from M1 to Q is adjusted from 0.8 to 0.8×(1+3.45%)=0.8276. Through deviation statistics and adjustments in multiple historical scenarios, the conduction strength values ​​are gradually made closer to the actual influence transmission pattern, thereby improving the accuracy of counterfactual inference.

[0103] Subsequently, the path deduction is re-executed using the adjusted conduction strength. Taking the aforementioned path K1→M1→Q as an example, if the conduction strength from K1 to M1 is adjusted to 0.72415, and the conduction strength from M1 to Q is adjusted to 0.8276, then the updated path counterfactual quality index value is: 0.85 × 0.72415 × 0.8276 = 0.5095. In this way, a more accurate counterfactual quality index estimate can be obtained for each path.

[0104] After obtaining the updated index values ​​for each transmission path, they need to be fused to obtain the final counterfactual quality index value for the key node. During fusion, the weight of each path is considered, and the weight can be determined based on several factors: path length (generally, the shorter the path, the smaller the transmission loss, and the larger the weight), historical prediction accuracy (paths with higher accuracy are given greater weight), and the overall level of transmission intensity along the path. Assuming the weight of path K1→M1→Q is 0.6, and the weight of path K1→M2→M3→Q is 0.4, the corresponding updated counterfactual quality index values ​​for these paths are 0.5095 and 0.48, respectively. Then, the final counterfactual quality index value for the key node K1 is: 0.5095×0.6+0.48×0.4=0.4977.

[0105] In practical applications, a path importance assessment mechanism can also be introduced. For example, in the context of hospital nursing quality management, multiple influence paths are identified through a multi-level influence network for the quality indicator of pressure ulcer incidence. The path "rationality of nursing staff allocation → timeliness of nursing intervention implementation → pressure ulcer incidence" has a historical prediction accuracy of 92%, a path length of 2, and a total transmission strength of 0.7 × 0.8 = 0.56. Meanwhile, the path "rationality of nursing staff allocation → nursing training completion rate → nursing skills mastery → timeliness of nursing intervention implementation → pressure ulcer incidence" has a historical prediction accuracy of 78%, a path length of 4, and a total transmission strength of 0.6 × 0.7 × 0.65 × 0.8 = 0.2184. Based on a comprehensive assessment of accuracy, path length, and transmission strength, the first path is assigned a higher weight of 0.7, and the second path a weight of 0.3. By using the counterfactual reasoning method described above, key insights such as "if the nursing staff ratio is optimized from the current 1:0.6 to the historical normal level of 1:0.4, the incidence of pressure ulcers will decrease from 3% to 2.1%" can be identified, providing data support for optimizing the allocation of nursing resources.

[0106] By using the aforementioned counterfactual quality indicator deduction method, combined with historical data calibration and multi-path fusion strategies, we can accurately assess the impact of key node status changes on quality indicators, providing a scientific basis for quality improvement decisions.

[0107] In one optional implementation, the step of tracing the complete causal chain along the causal transmission path of the multi-level influence network for key nodes whose actual contribution exceeds a preset level, and generating a collaborative optimization strategy for each intermediate node in the complete causal chain under resource constraints, includes:

[0108] Key nodes whose actual contribution exceeds a preset level are selected, and their predecessor nodes are traced back along the causal transmission path in the multi-level influence network to identify the complete causal chain and record the current execution status of each node in the complete causal chain;

[0109] Obtain the current resource constraints, and for each intermediate node in the complete causal chain, calculate the feasible optimization action space based on the node's current execution state and resource constraints;

[0110] Starting from the predecessor node, optimization actions are selected sequentially for each intermediate node from its optimization action space along the complete causal chain. The optimization action of each intermediate node is selected from the optimization action space of that node based on the optimization action of its predecessor node and resource constraints, generating multiple candidate collaborative optimization strategies.

[0111] For each candidate collaborative optimization strategy, the optimization actions of each intermediate node are taken as input, and the optimization is derived to the quality indicator node along the causal transmission path to obtain the optimized quality indicator value. The expected performance improvement of the candidate collaborative optimization strategy is quantified based on the difference between the optimized quality indicator value and the current quality indicator value. The strategy with the largest expected performance improvement is selected as the final collaborative optimization strategy.

[0112] For example, after calculating the true contribution value of each key node to the performance deviation according to the aforementioned steps, a preset threshold is set to screen the key nodes. For instance, a true contribution value greater than 0.25 is used as a screening condition, indicating that the node's contribution to the performance deviation exceeds 25%. In a nursing quality assessment scenario of a medical institution, counterfactual calculations revealed that the true contribution value of the "timeliness of nursing measure implementation" node was 0.32, and the true contribution value of the "rationality of nursing staff allocation" node was 0.28. Both nodes exceeded the preset level and were screened as key nodes requiring collaborative optimization.

[0113] For the selected key nodes, a breadth-first search algorithm is used to trace backwards in a multi-level influence network. Starting from the key node, all predecessor edges pointing to that node are searched in the network. The starting nodes of these edges are taken as direct predecessor nodes. The search process is repeated for each direct predecessor node, tracing upstream layer by layer until the root node with no predecessor edges is reached. Taking "timeliness of nursing measures implementation" as an example, the backward tracing reveals that its direct predecessor nodes include "rationality of nursing staff allocation" and "completion rate of nursing skills training". Continuing to trace upwards, the predecessor nodes of "rationality of nursing staff allocation" are "implementation degree of nursing staff recruitment plan", and the predecessor nodes of "completion rate of nursing skills training" are "implementation rate of nursing training budget" and "arrival rate of nursing training instructors". Thus, a complete causal chain from the root node to the key node is identified. During the tracing process, the current execution status of each node in the chain is recorded simultaneously, including the current value of "implementation degree of nursing staff recruitment plan" (72%), the current nurse-patient ratio of "rationality of nursing staff allocation" (1:0.55), the current value of "completion rate of nursing skills training" (68%), and the current value of "timeliness of nursing measures implementation" (76%).

[0114] After identifying the complete causal chain, the current resource constraints are obtained to determine the feasible boundaries of optimization actions. Resource constraints include dimensions such as human resource quotas, financial budgets, and time window limits. In this scenario, the currently available nursing human resources are the allocation of 8 additional nurses, the available training budget is 120,000 yuan, and the available time window is 3 months. Based on the current execution status of each intermediate node and its performance under different execution states in historical data, the feasible optimization action space for each node is calculated in conjunction with resource constraints. The current nurse-patient ratio for the "Nursing Staff Allocation Rationality" node is 1:0.55. Historical data shows that the nurse-patient ratio varies from 1:0.35 to 1:0.65. Combining the currently available 8 nurses, the optimization action space for this node includes: Action A0: Maintain the current nurse-patient ratio of 1:0.55, consuming 0 nurses; Action A1: Add 4 nurses to optimize the nurse-patient ratio to 1:0.6, consuming 4 nurses; Action A2: Add 8 nurses to optimize the nurse-patient ratio to 1:0.65, consuming 8 nurses. The current value of the "Nursing Skills Training Completion Rate" node is 68%. Historical data shows that when the completion rate is above 85%, the timeliness of nursing interventions is significantly improved. The optimization options for this node are: Action B0: Maintain the current completion rate of 68% (0 budget); Action B1: Invest 50,000 yuan in specialized training to increase the completion rate to 80%; Action B2: Invest 100,000 yuan in comprehensive training to increase the completion rate to 90%. The current value of the preceding node, "Nursing Staff Recruitment Plan Execution," is 72%. Its optimization options are: Action C0: Maintain the current completion rate of 72%; Action C1: Invest additional recruitment resources to increase the completion rate to 85% (1 month); Action C2: Fully promote recruitment to increase the completion rate to 95% (2 months).

[0115] After determining the optimization action space for each node, a dynamic programming method combined with a resource constraint pruning strategy is used to generate candidate collaborative optimization strategies. A strategy generation tree is constructed, with the root node corresponding to the first predecessor node of the causal chain. Actions are selected from the optimization action space of this node as branches, and each branch records the amount of resources consumed. Starting from the predecessor node "Nursing staff recruitment plan execution degree", if action C0 is selected, no resources are consumed, and the remaining resources are 8 nurses, 120,000 yuan budget, and 3 months. Continue to select actions for the next level node "Nursing staff allocation rationality". If action A1 is selected, 4 nurses are consumed, and the remaining resources are 4 nurses, 120,000 yuan budget, and 3 months. Then, actions are selected for the node "Nursing skills training completion rate". If action B1 is selected, 4 nurses and 50,000 yuan budget are consumed. Checking that the remaining resources are 4 nurses, 70,000 yuan budget, and 3 months still meet the constraints, this path forms candidate strategy S1. If selecting action C1 from the predecessor node consumes 1 month's time, the remaining resources are 8 nurses, a budget of 120,000 yuan, and 2 months' time. Selecting action A2 for the next-level node consumes 8 nurses, and selecting action B0 for the training node consumes no budget. The total consumption is 8 nurses + 1 month's time, forming candidate strategy S2. If selecting action C0 from the predecessor node, selecting action A0 for the configuration node consumes no nurses, and selecting action B2 for the training node consumes 100,000 yuan, the total consumption is 100,000 yuan's budget, forming candidate strategy S3. Through layer-by-layer expansion and resource constraint pruning, multiple candidate collaborative optimization strategies are generated. Each strategy clearly records the optimization actions and cumulative resource consumption of each intermediate node.

[0116] For each generated candidate collaborative optimization strategy, the optimized execution state corresponding to the optimization action of each intermediate node is used as input, and the optimized quality index value is calculated by forward deduction along the causal transmission path of the multi-level influence network. The deduction mechanism is the same as the aforementioned counterfactual deduction, using the transmission strength of the edge multiplied by the improvement amount of the node to transmit layer by layer. For candidate strategy S1, the improvement amount of optimizing the nurse-patient ratio from 1:0.55 to 1:0.6 in the "rationality of nursing staff allocation" node is 0.05. When the transmission along the network edge to the "timeliness rate of nursing measures execution" node, the transmission strength of the edge is 0.82. According to the conversion relationship that every 0.01 improvement in the nurse-patient ratio corresponds to a 1 percentage point increase in the timeliness rate, the improvement amount of the timeliness rate is 0.05 × 100 × 0.82 = 4.1 percentage points. At the same time, the improvement amount of "completion rate of nursing skills training" from 68% to 80% is 12 percentage points. When the transmission along the edge to the execution timeliness rate node, the transmission strength of the edge is 0.75, and the contribution improvement amount is 12 × 0.75 = 9 percentage points. The combined improvement from both paths results in the "timely implementation rate of nursing interventions" improving from the current 76% to 76% + 4.1% + 9% = 89.1%. Continuing along the path, the improvement in the timeliness of implementation from 76% to 89.1% is 13.1%. When this improvement is transmitted along the path to the quality indicator node "pressure ulcer incidence," the transmission strength is 0.78, indicating a negative impact. Therefore, the improvement in pressure ulcer incidence is 13.1 × 0.78 = 10.2 percentage points, reducing the current 3.2% to 3.2% × (1 - 10.2%) = 2.87%. The same deduction process was applied to candidate strategy S2. The "implementation rate of nursing staff recruitment plan" improved from 72% to 85%, an improvement of 13 percentage points. Through edge propagation and optimization combined with "rationality of nursing staff allocation," the nurse-patient ratio improved to 1:0.65. After multi-path propagation calculation, the "timeliness rate of nursing intervention implementation" was optimized to 92.3%. Further deduction showed that the "incidence rate of pressure ulcers" decreased to 2.58%. For candidate strategy S3, the "completion rate of nursing skills training" improved from 68% to 90%, an improvement of 22 percentage points. Through path propagation calculation, the "timeliness rate of nursing intervention implementation" was optimized to 85.6%, and the "incidence rate of pressure ulcers" decreased to 2.94%.

[0117] Based on the current monitoring data of the quality indicator node "pressure ulcer incidence" in the multi-level influence network, the current quality indicator value is 3.2%. The expected performance improvement is quantified by comparing the optimized quality indicator values ​​derived from each candidate strategy with this current value. Candidate strategy S1 reduces the pressure ulcer incidence from 3.2% to 2.87%, an absolute improvement of 0.33 percentage points, with an improvement margin of 0.33 ÷ 3.2 × 100% = 10.3%. Candidate strategy S2 reduces the pressure ulcer incidence from 3.2% to 2.58%, an absolute improvement of 0.62 percentage points, with an improvement margin of 19.4%. Candidate strategy S3 achieves an improvement margin of 8.1%. Comparing the expected performance improvement margins of each candidate strategy, candidate strategy S2 shows the largest improvement margin of 19.4%, and is therefore selected as the final co-optimization strategy.

[0118] In practical applications, if there are many resource constraints or a large number of causal chain nodes, leading to an exponential increase in the candidate strategy space, heuristic search methods such as genetic algorithms or simulated annealing algorithms can be used. With a population size of 100 to 200, a crossover probability of 0.7 to 0.9, and a mutation probability of 0.01 to 0.05, iterative optimization can yield an approximate optimal solution within an acceptable computational time. For scenarios requiring simultaneous optimization of multiple quality indicators, the improvement magnitude of each quality indicator is weighted and summed according to preset weights to obtain the overall performance improvement magnitude. The weights are determined based on the clinical importance and management priority of the quality indicators; for example, pressure ulcer incidence rate has a weight of 0.4, patient satisfaction has a weight of 0.3, and nursing error rate has a weight of 0.3.

[0119] This invention identifies the impact path by tracing the causal chain, and collaboratively selects multiple node optimization actions under resource constraints to achieve systematic linkage optimization, significantly improving the quality improvement effect and resource utilization efficiency.

[0120] In one optional implementation, the step of selecting optimization actions from the optimization action space of each intermediate node sequentially along the complete causal chain, starting from the predecessor node, to generate multiple candidate collaborative optimization strategies includes:

[0121] Initialize the policy generation tree, taking the first predecessor node of the complete causal chain as the root node, and select multiple optimization actions as branches from the optimization action space of this node; for each branch, based on the optimization action and resource constraints corresponding to the branch, dynamically reduce the optimization action space of the next intermediate node to obtain the constrained optimization action space;

[0122] From the constrained optimization action space, select multiple optimization action expansion branches, and repeatedly execute the dynamic reduction and branch expansion process until all intermediate nodes on the complete causal chain have been traversed, resulting in multiple complete paths from the root node to the leaf node;

[0123] Check whether the total resource consumption of optimization actions across all nodes on each complete path exceeds the resource constraint; if it does, prune the corresponding path.

[0124] All unpruned complete paths are converted into candidate collaborative optimization strategies, where the optimization actions of each node on each path constitute a candidate collaborative optimization strategy.

[0125] For example, when generating multiple candidate collaborative optimization strategies by selecting optimization actions from the optimization action space of each intermediate node along the complete causal chain starting from the predecessor node, a strategy spanning tree structure is used for dynamic expansion and resource constraint pruning.

[0126] The construction of the strategy generation tree begins with initialization, using the first predecessor node of the complete causal chain as the root node. Multiple optimization actions are selected from the optimization action space of this node as the first-level branches of the tree. In a nursing quality optimization scenario at a medical institution, the first predecessor node of the complete causal chain is "the execution degree of the nursing staff recruitment plan." The optimization action space of this node includes action C0 maintaining the status quo at 72% without consuming resources, action C1 increasing to 85% consuming one month, and action C2 increasing to 95% consuming two months. These three optimization actions are created as three branch nodes of the root node in the strategy generation tree, with each branch node recording the identifier of the corresponding optimization action and the amount of resources consumed.

[0127] For each branch node in the strategy generation tree, based on the corresponding optimization action and the amount of resources consumed, the optimization action space of the next intermediate node is dynamically reduced in conjunction with the current total resource constraints. Resource constraints include 8 available nursing staff, a training budget of 120,000 yuan, and a 3-month time window. For the branch node selecting action C1, this action consumes 1 month of time, leaving 8 nurses, a budget of 120,000 yuan, and 2 months of time. The next intermediate node is "rationality of nursing staff allocation." The original optimization action space for this node includes action A0 (maintaining the current nurse-patient ratio of 1:0.55, consuming 0 nurses), action A1 (adding 4 nurses to optimize the nurse-patient ratio to 1:0.6, consuming 4 nurses), and action A2 (adding 8 nurses to optimize the nurse-patient ratio to 1:0.65, consuming 8 nurses). Based on the remaining resource constraints, we check whether the resource consumption of each optimization action is within the remaining resource range. Action A0 consumes 0 nurses, which is less than the remaining 8 nurses, and consumes no time or budget, satisfying the remaining constraints. Action A1 consumes 4 nurses, which is less than the remaining 8 nurses, satisfying the constraints. Action A2 consumes 8 nurses, which is equal to the remaining 8 nurses, satisfying the constraints. All three actions are retained in the post-constraint optimization action space. For the branch node of selected action C2, this action consumes 2 months of time, and the remaining resource constraints are 8 nurses, 120,000 yuan budget, and 1 month of time. If subsequent nodes have actions that require more than 1 month of time, they are removed from the post-constraint optimization action space during dynamic reduction.

[0128] Multiple optimized action extension branches are selected from the constrained optimized action space. For the "rationality of nursing staff allocation" node, the constrained optimized action space includes actions A0, A1, and A2. These three actions are connected as child nodes to their parent branch nodes to form new branch paths. Taking the branch node of action C1 as an example, three child branch nodes are created by selecting actions A0, A1, and A2 from its constrained optimized action space. Each child branch node records the complete path from the root node to the current node and the cumulative resource consumption. The cumulative resource consumption of the child branch node of action A1 is 1 month of time + 4 nurses. Continue to extend the branches for the next intermediate node "nursing skills training completion rate". The original optimized action space of this node includes action B0 maintaining the status quo with a completion rate of 68% and consuming 0 yuan of budget, action B1 investing 50,000 yuan in special training to increase the completion rate to 80%, and action B2 investing 100,000 yuan in comprehensive training to increase the completion rate to 90%. For the branch path that consumes a cumulative 1 month's time plus 4 nurses, the remaining resource constraints are 4 nurses, a budget of 120,000 yuan, and 2 months' time. Checking the optimization actions of the training node, action B0 consumes 0 yuan and satisfies the constraint; action B1 consumes 50,000 yuan, which is less than the remaining 120,000 yuan and satisfies the constraint; action B2 consumes 100,000 yuan, which is less than the remaining 120,000 yuan and satisfies the constraint. All three actions are retained in the post-constraint optimization action space, and corresponding sub-branch nodes are created. The dynamic reduction and branch expansion process is repeated, traversing all intermediate nodes on the complete causal chain layer by layer until the terminal node of the causal chain is reached. At this point, each complete path from the root node to the leaf node in the strategy generation tree represents a potential collaborative optimization strategy.

[0129] After the strategy spanning tree is constructed, resource constraint checks are performed on each complete path from the root node to a leaf node. All nodes on the path are traversed, and the resource consumption for each node's corresponding optimization action is accumulated to calculate the total resource consumption for the path, including the cumulative number of nurses consumed, the budget amount, and the time duration. Taking a complete path as an example, this path contains the following node sequence: "Nursing staff recruitment plan execution rate" action C1 consumes 1 month of time; "Nursing staff allocation rationality" action A2 consumes 8 nurses; and "Nursing skills training completion rate" action B1 consumes 50,000 yuan of budget. The total accumulated resource is 8 nurses + 50,000 yuan of budget + 1 month of time. The total accumulated resource is compared with the initial resource constraints. The nurse consumption of 8 nurses equals the resource constraint of 8 nurses, satisfying the constraint; the budget consumption of 50,000 yuan is less than the resource constraint of 120,000 yuan, satisfying the constraint; and the time consumption of 1 month is less than the resource constraint of 3 months, satisfying the constraint. This path passes the resource constraint check and is retained. For the other path, if its total accumulated resources are 9 nurses + 100,000 yuan budget + 2 months, and the number of nurses consumed is 9, which exceeds the resource constraint of 8 nurses, then the path violates the resource constraint and is removed from the strategy generation tree along with its corresponding branch nodes.

[0130] In practice, resource constraint detection can be performed in real time during branch expansion. When the cumulative resource consumption of a sub-branch node exceeds the total resource constraint, the further expansion of that branch is immediately stopped to avoid generating invalid paths and improve computational efficiency. Resource constraint detection needs to consider the joint constraints of resources in multiple dimensions, and adopts a dimension-by-dimensional comparison method. If the resource consumption in any dimension exceeds the constraint, the path is determined to violate the constraint.

[0131] All intact paths that were not pruned are converted into candidate collaborative optimization strategies. The optimization actions of each node on each path are combined in causal chain order to form a candidate strategy. For the retained paths, the node identifier and corresponding optimization action identifier of each node on the path are extracted and arranged in order of the node's position in the causal chain to form a strategy description data structure. The path corresponding to a certain candidate strategy S1 includes: action C0 for the node "execution rate of nursing staff recruitment plan", action A1 for the node "rationality of nursing staff allocation", and action B1 for the node "completion rate of nursing skills training". The strategy description is a sequence of node-action pairs [(node ​​ID1, action C0), (node ​​ID2, action A1), (node ​​ID3, action B1)], with a total resource consumption of 4 nurses + 50,000 yuan budget. The path corresponding to candidate strategy S2 includes an action sequence [(node ​​ID1, action C1), (node ​​ID2, action A2), (node ​​ID3, action B0)], with a total resource consumption of 8 nurses + 1 month. The path corresponding to candidate strategy S3 contains the action sequence [(node ​​ID1, action C0), (node ​​ID2, action A0), (node ​​ID3, action B2)], with a cumulative resource consumption of 100,000 yuan. All candidate strategies are stored in the strategy set for subsequent performance evaluation, and each strategy is associated with its corresponding node-action sequence and cumulative resource consumption information.

[0132] During policy generation, if the causal chain contains parallel branches or multiple independent paths, a policy generation tree needs to be constructed for each path and resource constraint checks need to be performed. Finally, the candidate policies of all paths are merged into a unified policy set. The policy generation tree is implemented using a tree data structure. Each node contains fields such as a parent node pointer, a list of child node pointers, an identifier of the intermediate node corresponding to the node, an identifier of the selected optimization action, and a cumulative resource consumption vector. It supports depth-first or breadth-first traversal algorithms for path enumeration and pruning operations.

[0133] This invention systematically generates multiple candidate collaborative optimization strategies that satisfy resource constraints through dynamic expansion of the strategy generation tree and resource constraint pruning, providing a complete candidate solution space for subsequent performance evaluation and strategy selection.

[0134] In one optional implementation, the step of applying the collaborative optimization strategy to the operation management process and updating the generation rules of the multi-level influence network and counterfactual performance benchmark based on the feedback data after execution includes:

[0135] The collaborative optimization strategy is distributed to the business process execution units corresponding to each intermediate node in the complete causal chain, and business operation data and quality assessment data after the strategy execution are collected as feedback data.

[0136] Based on the feedback data, the correlation strength between each business process node is recalculated. For network edges where the correlation strength change exceeds a preset strength threshold, the original propagation strength is replaced with the recalculated correlation strength, and the multi-level influence network is updated.

[0137] Based on the updated multi-level influence network, the counterfactual quality index values ​​of the original key nodes are re-inferred. If the deviation between the inferred values ​​and the actual quality index values ​​after execution exceeds the preset range, the selection rules for historical normal execution states are adjusted, and the generation rules for counterfactual performance benchmarks are updated.

[0138] The updated rules for generating multi-level influence networks and counterfactual performance benchmarks will be used in the next round of performance evaluation and optimization.

[0139] For example, the selected collaborative optimization strategy is distributed to the business process execution units corresponding to each intermediate node in the complete causal chain. When the strategy is distributed, the optimization actions of each node are converted into executable operation instructions for the business process execution units. For the "Nursing Staff Recruitment Plan Execution Rate" node, an instruction is distributed to the human resources management system to trigger the acceleration of the recruitment process, including specific measures such as increasing recruitment channel deployment, shortening the approval cycle, and increasing the frequency of interviews. The instruction clearly states the execution target as increasing the execution rate from the current 72% to 85%, with an execution period of one month. For the "Nursing Staff Allocation Rationality" node, an instruction is distributed to the nursing staff allocation system to execute a staffing adjustment of adding 8 nurses. The instruction includes parameters such as the target nurse-patient ratio of 1:0.65, the departmental distribution plan for the added staff, and the arrival time requirements. The strategy distribution uses a message queue mechanism, encapsulating the optimization instructions into message objects containing fields such as node identifier, action type, target parameters, execution period, and priority. These objects are pushed to the topic queues subscribed to by each business process execution unit. After receiving the message, the execution unit parses the instruction content and initiates the corresponding business process change operation.

[0140] During and after the strategy implementation, business operation data and quality assessment data were collected as feedback data. Business operation data was collected through the data reporting interfaces of each business process execution unit. The collected data included the actual status values ​​after each intermediate node performed optimization actions, the time progress during execution, resource consumption, and any abnormal events encountered. Due to actual limitations during strategy implementation, some nodes did not fully achieve their preset goals. One month after strategy implementation, the actual execution rate of the "Nursing Staff Recruitment Plan" reached 83%, not fully reaching the target of 85%; the actual nurse-patient ratio for "Nursing Staff Allocation Rationality" reached 1:0.63, with a total of 7 additional nurses; and the actual value of the "Nursing Skills Training Completion Rate" remained unchanged at 68%. Quality assessment data was collected through the periodic evaluation mechanism of the quality monitoring system. The collected data included the actual measured values ​​of quality indicator nodes and process indicator data related to the quality indicators. After strategy implementation, the actual value of the "Timeliness of Nursing Measures Implementation" was 90.5%, and the actual value of the "Pressure Ulcer Incidence Rate" was 2.65%, both showing significant improvement compared to the pre-implementation values ​​of 76% and 3.2%. The feedback data collection cycle is set according to business characteristics. For quality indicators that require a longer time to show results, a 3-month observation cycle is set, while for operational indicators with rapid response, a 1-week collection cycle is set. The collected feedback data undergoes data cleaning processing to remove outliers and missing values. Missing values ​​are filled using previous values ​​or linear interpolation, while outliers are identified and removed using a 3-standard-deviation rule.

[0141] Based on the collected feedback data, the correlation strength between each business process node is recalculated, using the same Pearson correlation coefficient calculation method as used in the aforementioned multi-level influence network construction. The node state values ​​collected after strategy execution are added to the historical data sequence, and the correlation coefficient between node pairs is recalculated as the updated correlation strength. For the network edge between "rationality of nursing staff allocation" and "timeliness of nursing intervention execution," the transmission strength calculated based on historical data before strategy execution was 0.82. After strategy execution, the newly collected nurse-patient ratio of 1:0.63 and the timeliness execution rate of 90.5% are added to the data sequence, and the correlation strength is recalculated to be 0.87. A preset strength threshold of 0.05 is set, and the change in correlation strength is 0.87 - 0.82 = 0.05, which equals the preset strength threshold. This network edge is then updated, replacing the original transmission strength of 0.82 with the recalculated correlation strength of 0.87. For the network edge between "Nursing Skills Training Completion Rate" and "Timely Implementation Rate of Nursing Measures," the transmission strength before strategy implementation was 0.75. After recalculation, the correlation strength was 0.77, a change of 0.02, which is less than the preset strength threshold of 0.05. Therefore, the transmission strength of this network edge remains unchanged. All network edges in the multi-level influence network are traversed, and the correlation strength is recalculated and the change is detected for each edge. The transmission strength field of network edges whose changes exceed the preset strength threshold is updated to the recalculated correlation strength value, completing the update of the multi-level influence network. The network update operation is executed through the batch update interface of the graph database, submitting all edges that need updating and their new transmission strength values ​​at once, ensuring the atomicity of the update operation.

[0142] The counterfactual quality index values ​​of the original key nodes were re-derived based on the updated multi-level influence network, using the same mechanism as the aforementioned counterfactual derivation. The original key nodes refer to the business process nodes with the most significant impact on quality indicators, selected through calculation and ranking of actual contribution before strategy execution. In the aforementioned scenario of optimizing nursing quality in medical institutions, the original key nodes included three nodes: "timeliness of nursing measure execution," "rationality of nursing staff allocation," and "completion rate of nursing skills training." Among them, "timeliness of nursing measure execution" had the highest actual contribution of 0.512. For the original key node "timeliness of nursing measure execution," its state was forcibly set to the historical normal execution state of 76%, and propagated along the updated network edges to the quality indicator node "pressure ulcer incidence rate." During the propagation process, the updated propagation strength of 0.87 was used for calculation, resulting in a counterfactual quality index value of 2.95%. Comparing the derived value of 2.95% with the actual measured quality index value of 2.65% after strategy execution, the calculated deviation was 2.95% - 2.65% = 0.3 percentage points. The preset deviation range was set at 0.2 percentage points. The actual deviation of 0.3 percentage points exceeded the preset range, indicating that the currently selected historical normal execution status of 76% deviated from the actual business environment, and the selection rules for historical normal execution status needed to be adjusted. The adjustment method was to shorten the selection time window for historical normal execution status from the past 6 months to the past 3 months, making the selected status value closer to the current business environment. The average value of "timeliness of nursing measures implementation" over the past 3 months was reselected as the historical normal execution status, and the new historical normal execution status was calculated to be 79%. Using the adjusted historical normal execution status of 79%, the counterfactual quality indicator value was re-derived, resulting in a derived value of 2.68%, which deviated from the actual value of 2.65% by 0.03 percentage points, within the preset range. This confirmed that the adjustment of the historical normal execution status selection rules was effective. The generation rules for the counterfactual performance benchmark were updated, and the selection time window parameter was changed from 6 months to 3 months and persistently stored.

[0143] In the next round of performance evaluation, counterfactual analysis will be performed using the updated network structure and propagation strength. Adjusted generation rules will be used to select historical normal execution states, calculate the true contribution of each node, identify new key nodes, and generate new collaborative optimization strategies. Through continuous closed-loop iteration of strategy execution, feedback collection, network updates, and rule adjustments, the multi-level influence network and counterfactual performance benchmarks gradually adapt to changes in the business environment, improving the accuracy of performance evaluation and the effectiveness of optimization strategies.

[0144] This invention achieves continuous optimization and self-adaptation of the performance evaluation model through a feedback data-driven network update and rule adjustment mechanism, significantly improving long-term operational management efficiency.

[0145] A second aspect of the present invention provides a medical quality and operational performance evaluation system based on intelligent analysis, comprising:

[0146] The data acquisition and impact analysis module is used to acquire business operation data and quality assessment data from medical institutions. Through time-series correlation analysis, it identifies the causal transmission paths between various business process nodes and changes in quality indicators, constructs a multi-level impact network, and calculates the impact weight of each node on the quality assessment results.

[0147] The counterfactual analysis and contribution calculation module is used to select key nodes whose influence weight exceeds a preset influence threshold, replace the current abnormal execution state with the historical normal execution state, re-derive the counterfactual performance benchmark using the multi-level influence network, quantify the difference between the actual performance state and the counterfactual performance benchmark, and obtain the true contribution of each key node to the performance deviation.

[0148] The collaborative optimization strategy generation module is used to trace the complete causal chain along the causal transmission path of the multi-level influence network for key nodes whose actual contribution exceeds a preset level, and generate collaborative optimization strategies for each intermediate node in the complete causal chain under resource constraints;

[0149] The strategy execution and dynamic update module is used to apply the collaborative optimization strategy to the operation and management process, and update the generation rules of the multi-level influence network and counterfactual performance benchmark based on the feedback data after execution.

[0150] A third aspect of the present invention provides an electronic device, comprising:

[0151] processor;

[0152] Memory used to store processor-executable instructions;

[0153] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0154] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0155] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating medical quality and operational performance based on intelligent analysis, characterized in that, include: We acquire operational and quality assessment data from medical institutions, identify causal transmission paths between various business process nodes and changes in quality indicators through time-series correlation analysis, construct a multi-level influence network, and calculate the influence weight of each node on the quality assessment results. Key nodes whose influence weight exceeds a preset influence threshold are selected. The current abnormal execution state is replaced with the historical normal execution state. The counterfactual performance benchmark is re-derived using the multi-level influence network. The difference between the actual performance state and the counterfactual performance benchmark is quantified to obtain the true contribution of each key node to the performance deviation. For key nodes whose actual contribution exceeds a preset level, the complete causal chain is traced along the causal transmission path of the multi-level influence network, and a collaborative optimization strategy for each intermediate node in the complete causal chain is generated under resource constraints; The collaborative optimization strategy is applied to the operation and management process, and the generation rules of the multi-level influence network and counterfactual performance benchmark are updated based on the feedback data after execution. The steps of selecting key nodes whose influence weight exceeds a preset influence threshold, replacing the current abnormal execution state with the historical normal execution state, re-deriving the counterfactual performance benchmark using the multi-level influence network, and quantifying the difference between the actual performance state and the counterfactual performance benchmark to obtain the true contribution of each key node to the performance deviation include: Based on the influence weight, business process nodes that exceed the preset influence threshold are marked as key nodes, and the current execution status and historical normal execution status of each key node are extracted; For each critical node, a historical state similar to the business characteristics of the current time window is selected from its historical normal execution states as a replacement benchmark, and the current abnormal execution state of the critical node is replaced with the replacement benchmark; Based on the execution state of the replaced key node, the causal transmission path containing the key node is traced along the multi-level influence network, and the transmission direction and strength of the network edges are used to extrapolate to the quality index node, thereby obtaining the counterfactual quality index value corresponding to the key node; The counterfactual performance benchmark is obtained by summing up the counterfactual quality indicator values ​​corresponding to all key nodes. Calculate the numerical difference between the actual performance status and the counterfactual performance benchmark, and decompose the numerical difference to each key node to obtain the true contribution of each key node to the performance deviation.

2. The method according to claim 1, characterized in that, The steps involved in identifying the causal transmission paths between various business process nodes and changes in quality indicators through time-series correlation analysis, constructing a multi-level influence network, and calculating the influence weight of each node on the quality assessment results include: Acquire business operation data and quality assessment data from medical institutions and divide them into time windows. Within each time window, extract the execution status characteristics and quality indicator change characteristics of business process nodes. Calculate the correlation strength between the execution state features and the change features under different time lag conditions, and identify the direct influence relationship based on the time decay pattern of the correlation strength; For business process nodes with direct impact relationships, trace the execution status characteristics of their preceding nodes along the time sequence to identify indirect impact relationships and their transmission paths transmitted through intermediate nodes; A multi-level influence network is constructed based on the aforementioned direct and indirect influence relationships, with each business process node serving as a network node and the direct and indirect influence relationships serving as network edges; For each business process node in the multi-level influence network, the direct influence of the node is calculated based on the direct influence relationship, and the indirect influence of the node is calculated based on the transmission path. The direct influence and indirect influence are weighted and summed to obtain the influence weight of the node on the quality assessment result.

3. The method according to claim 1, characterized in that, Based on the execution state of the replaced key node, the steps of extrapolating to the quality indicator node along the causal transmission path containing the key node in the multi-level influence network, according to the transmission direction and strength of the network edges, to obtain the counterfactual quality indicator value corresponding to the key node include: For each key node, identify all causal transmission paths from that key node to the quality indicator node in the multi-level influence network, and extract the transmission strength of each network edge on each causal transmission path; For each causal transmission path, the execution state of the replaced key node is used as the starting input. The state of each intermediate node is deduced sequentially along the path according to the transmission direction and transmission strength until the quality index node is reached, thus obtaining the path counterfactual quality index value corresponding to the path. Obtain the counterfactual quality index value and the true quality index value of each causal transmission path in the historical inference scenario, calculate the historical inference deviation of each path, and adjust the transmission strength of each network edge on the path based on the historical inference deviation; Based on the adjusted transmission strength, the path deduction is re-executed to obtain the updated path counterfactual quality index value for each causal transmission path; Calculate the path weight for each causal transmission path, and then weight and fuse the updated counterfactual quality index values ​​of all causal transmission paths according to the path weights to obtain the final counterfactual quality index value corresponding to the key node.

4. The method according to claim 1, characterized in that, For key nodes whose actual contribution exceeds a preset level, the steps of tracing the complete causal chain along the causal transmission path of the multi-level influence network and generating collaborative optimization strategies for each intermediate node in the complete causal chain under resource constraints include: Key nodes whose actual contribution exceeds a preset level are selected, and their predecessor nodes are traced back along the causal transmission path in the multi-level influence network to identify the complete causal chain and record the current execution status of each node in the complete causal chain; Obtain the current resource constraints, and for each intermediate node in the complete causal chain, calculate the feasible optimization action space based on the node's current execution state and resource constraints; Starting from the predecessor node, optimization actions are selected sequentially for each intermediate node from its optimization action space along the complete causal chain. The optimization action of each intermediate node is selected from the optimization action space of that node based on the optimization action of its predecessor node and resource constraints, generating multiple candidate collaborative optimization strategies. For each candidate collaborative optimization strategy, the optimization actions of each intermediate node are taken as input, and the optimization is derived to the quality indicator node along the causal transmission path to obtain the optimized quality indicator value. The expected performance improvement of the candidate collaborative optimization strategy is quantified based on the difference between the optimized quality indicator value and the current quality indicator value. The strategy with the greatest expected performance improvement is selected as the final collaborative optimization strategy.

5. The method according to claim 4, characterized in that, The steps of generating multiple candidate collaborative optimization strategies by sequentially selecting optimization actions from the optimization action space of each intermediate node along the complete causal chain, starting from the predecessor node, include: Initialize the policy generation tree, taking the first predecessor node of the complete causal chain as the root node, and selecting multiple optimization actions as branches from the optimization action space of this node; For each branch, based on the optimization actions and resource constraints corresponding to that branch, the optimization action space of the next intermediate node is dynamically reduced to obtain the constraint-constrained optimization action space; From the constrained optimization action space, select multiple optimization action expansion branches, and repeatedly execute the dynamic reduction and branch expansion process until all intermediate nodes on the complete causal chain have been traversed, resulting in multiple complete paths from the root node to the leaf node; Check whether the total resource consumption of optimization actions across all nodes on each complete path exceeds the resource constraint; if it does, prune the corresponding path. All unpruned complete paths are converted into candidate collaborative optimization strategies, where the optimization actions of each node on each path constitute a candidate collaborative optimization strategy.

6. The method according to claim 1, characterized in that, The steps of applying the collaborative optimization strategy to the operation and management process and updating the generation rules of the multi-level influence network and counterfactual performance benchmark based on the feedback data after execution include: The collaborative optimization strategy is distributed to the business process execution units corresponding to each intermediate node in the complete causal chain, and business operation data and quality assessment data after the strategy execution are collected as feedback data. Based on the feedback data, the correlation strength between each business process node is recalculated. For network edges where the correlation strength change exceeds a preset strength threshold, the original propagation strength is replaced with the recalculated correlation strength, and the multi-level influence network is updated. Based on the updated multi-level influence network, the counterfactual quality index values ​​of the original key nodes are re-inferred. If the deviation between the inferred values ​​and the actual quality index values ​​after execution exceeds the preset range, the selection rules for historical normal execution states are adjusted, and the generation rules for counterfactual performance benchmarks are updated. The updated rules for generating multi-level influence networks and counterfactual performance benchmarks will be used in the next round of performance evaluation and optimization.

7. A medical quality and operational performance evaluation system based on intelligent analysis, used to implement the method of any one of claims 1-6, characterized in that, include: The data acquisition and impact analysis module is used to acquire business operation data and quality assessment data from medical institutions. Through time-series correlation analysis, it identifies the causal transmission paths between various business process nodes and changes in quality indicators, constructs a multi-level impact network, and calculates the impact weight of each node on the quality assessment results. The counterfactual analysis and contribution calculation module is used to select key nodes whose influence weight exceeds a preset influence threshold, replace the current abnormal execution state with the historical normal execution state, re-derive the counterfactual performance benchmark using the multi-level influence network, quantify the difference between the actual performance state and the counterfactual performance benchmark, and obtain the true contribution of each key node to the performance deviation. The collaborative optimization strategy generation module is used to trace the complete causal chain along the causal transmission path of the multi-level influence network for key nodes whose actual contribution exceeds a preset level, and generate collaborative optimization strategies for each intermediate node in the complete causal chain under resource constraints; The strategy execution and dynamic update module is used to apply the collaborative optimization strategy to the operation and management process, and update the generation rules of the multi-level influence network and counterfactual performance benchmark based on the feedback data after execution.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images