An emergency critical patient first-aid path optimization method and system based on big data

Abstract

The application relates to the technical field of treatment scheme optimization, and provides an emergency critical patient first-aid path optimization method and system based on big data, which comprises the following steps: constructing a standardized event sequence by extracting historical diagnosis and treatment data of a hospital information system, then analyzing the time sequence correlation between events by using a graph structure mining technology, identifying key diagnosis and treatment nodes and high and low frequency paths, and constructing a hierarchical first-aid path state transition network. When processing real-time patients, the system matches the electronic health records of the patients with the network core state, dynamically plans the subsequent standard transition path, and avoids low-frequency risk paths marked as contraindications, so that the first-aid path optimization scheme of risk avoidance and path dynamic optimization is realized, the treatment efficiency of emergency critical patients is improved, the treatment effect is improved, and the risk of adverse prognosis is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of treatment plan optimization, in particular to an emergency critical patient first-aid path optimization method and system based on big data. BACKGROUND

[0002] The content of this part only provides background information related to the present application, which may not constitute prior art.

[0003] The emergency critical patient's condition has the characteristics of sudden onset, rapid progression and significant individual differences. The scientificity and adaptability of the first-aid path directly determine the success rate of patient treatment and the prognosis effect, and is the core concern in the field of emergency medical care. At present, the formulation of emergency critical first-aid path mostly depends on the accumulation of clinical experience of medical staff, lacks systematic integration and in-depth mining of massive historical diagnosis and treatment data, resulting in that the existing path is mostly a universal fixed template, which is difficult to accurately match the specific clinical characteristics of different patients, and often has problems such as poor connection between diagnosis and treatment links, delayed key intervention opportunity or redundant unnecessary diagnosis and treatment steps, which affects the efficiency and safety of treatment. At the same time, in the process of formulating the traditional first-aid path, the internal time sequence correlation between diagnosis and treatment events is not fully sorted out, and it is also difficult to effectively identify high-risk diagnosis and treatment links that may lead to adverse prognosis, so as to provide targeted risk avoidance guidance for the first-aid process.

[0004] Although there are attempts to use big data to build treatment plan models in related technologies, there are still many drawbacks. For example, in the prior art, a Chinese patent with publication number CN120878211A discloses an intelligent emergency pre-examination triage method and system based on big data. The core is an intelligent emergency pre-examination triage method based on big data. By obtaining pre-examination data of patients including real-time physiological monitoring data and main statement information, determining the pre-triage department and the probability of disease deterioration, and then evaluating the disposal ability coefficient of the pre-triage department, the final triage strategy is determined to solve the problems of inaccurate department matching, untimely risk assessment of disease deterioration and insufficient consideration of hospital disposal capacity in traditional emergency pre-examination triage. However, this technical solution only focuses on the department matching and transfer decision of emergency pre-examination triage, and does not extend to the optimization of specific diagnosis and treatment process after triage. It evaluates the disease condition through data similarity matching, does not mine the time sequence correlation and key nodes of diagnosis and treatment events, and lacks adverse diagnosis and treatment path identification and avoidance mechanism. Moreover, the degree of individualization is insufficient, only meets the basic triage demand, and cannot dynamically generate an adaptive individualized diagnosis and treatment path according to the patient's clinical characteristics, making it difficult to cover the precise diagnosis and treatment and risk control needs of the whole first-aid process, and failing to fully play the core value of big data in optimizing the diagnosis and treatment process.

[0005] Therefore, there is an urgent need for an emergency critical patient first aid path optimization scheme that can adapt to the dynamic and urgent characteristics of emergency critical care, effectively mine key information from historical diagnosis and treatment data, accurately identify high-risk diagnosis and treatment links, closely match individual patient clinical characteristics, and achieve risk avoidance and dynamic optimization of the path, so as to improve the treatment efficiency of emergency critical patients, improve the treatment effect, and reduce the risk of adverse prognosis. SUMMARY

[0006] In order to solve the above technical problems, the purpose of the present application is to provide an emergency critical patient first aid path optimization method and system based on big data, which analyzes historical diagnosis and treatment data to construct a diagnosis and treatment event directed graph model, identifies key nodes and high and low frequency paths, and constructs a hierarchical first aid path state transition network based thereon, to achieve a first aid path optimization scheme of risk avoidance and dynamic optimization of the path, so as to improve the treatment efficiency of emergency critical patients, improve the treatment effect, and reduce the risk of adverse prognosis.

[0007] The purpose of the present application is achieved by the following technical solutions:

[0008] In a first aspect, the present application provides an emergency critical patient first aid path optimization method based on big data, comprising:

[0009] Obtain historical diagnosis and treatment data of emergency critical patients from a hospital information system, and generate a standardized patient diagnosis and treatment event sequence data set, which contains the event timestamp and diagnosis and treatment event type label of each patient;

[0010] An event sequence mining algorithm based on graph structure is used to analyze the patient diagnosis and treatment event sequence data set and construct a diagnosis and treatment event directed graph model, including: the nodes in the directed graph model represent diagnosis and treatment event types, and the edges represent the time sequence transition relationship between events; calculate the node degree centrality and frequent subsequence support, and determine the diagnosis and treatment events with node degree centrality higher than a first threshold value as key nodes; mark the diagnosis and treatment path fragments with frequent subsequence support higher than a second threshold value as high frequency paths, and mark the diagnosis and treatment path fragments with frequent subsequence support lower than the second threshold value and associated with adverse rescue results as low frequency paths;

[0011] Based on the key nodes, high frequency paths and low frequency paths, a hierarchical first aid path state transition network is constructed; wherein the key nodes are the core states in the network, the high frequency paths are marked as standard transfer channels connecting the core states, and the low frequency paths are marked as forbidden transfer edges in the network;

[0012] For emergency patients to be treated, real-time access to their electronic health records and extract the current state information of the patient; match the current state information with the core state in the hierarchical emergency path state transition network, take the current state of the patient as the starting point, select the subsequent path segment based on the standard transition path, and combine the forbidden transition edge to constrain and correct the selected path, real-time generate personalized emergency path matched with the clinical characteristics of the patient and avoid known diagnosis and treatment risks.

[0013] Further, a standardized patient diagnosis and treatment event sequence dataset is generated, specifically including:

[0014] Data cleaning is performed on the diagnosis and treatment data to remove duplicate records and invalid data; the diagnosis and treatment event type is mapped to a unified event type coding table, and arranged in ascending order of event timestamp to generate a diagnosis and treatment event sequence for each patient; the diagnosis and treatment event sequences of all patients are combined to generate a patient diagnosis and treatment event sequence dataset.

[0015] Further, the patient diagnosis and treatment event sequence dataset is analyzed and a directed graph model of diagnosis and treatment events is constructed, specifically including:

[0016] The prefix projection algorithm is used to extract frequent subsequences from the patient diagnosis and treatment event sequence dataset, and the support of the frequent subsequence is determined by calculating the frequency of the subsequence in the dataset;

[0017] Based on the patient diagnosis and treatment event sequence dataset, a directed graph model of diagnosis and treatment events is constructed, with diagnosis and treatment event types as nodes and time sequence transition relationships between events as edges;

[0018] In the directed graph model, the edge weight between nodes is calculated based on the transition probability between events, and the transition probability is determined by the ratio of the number of event transitions to the total number of transitions;

[0019] When calculating the node degree centrality, the in-degree centrality and out-degree centrality of each node are calculated respectively, and the value corresponding to the node degree centrality is obtained by weighted sum of the in-degree centrality and out-degree centrality.

[0020] Further, a hierarchical emergency path state transition network is constructed, specifically including:

[0021] The key nodes are taken as the core states in the network, the standard transition path between the core states is constructed based on the high-frequency path, and the weight of the standard transition path is determined by combining the frequent subsequence support corresponding to the path and the transition probability between events; the low-frequency path is marked as a forbidden transition edge in the network, and is highlighted in the network presentation;

[0022] The network is structured in layers, including a core layer and a detailed layer. The core layer contains the core status and standard transfer pathways for rapid emergency route planning. The detailed layer contains all diagnostic and treatment event nodes and transfer edges between events, and clearly marks contraindicated transfer edges.

[0023] Furthermore, the weighting expression for the standard transfer pathway includes:

[0024]

[0025] In the formula, Standard transfer path weights; These are the weighting coefficients for the support of frequent subsequences; The number of patient sequences containing frequent subsequence L; This represents the total number of patient sequences in the dataset. The weighting coefficients for the event transition probability; The number of event types in the frequent subsequence L; For the first subsequence L The event to the 1 +1 event transition probability.

[0026] Furthermore, real-time access to their electronic health records and extraction of the patient's current status information specifically includes:

[0027] Natural language processing algorithms are used to parse unstructured text in electronic health records and extract key clinical features, including symptoms, signs, and laboratory indicators. Key clinical features are mapped into numerical vectors through a pre-trained feature embedding model and matched with the core states in the hierarchical emergency pathway state transition network. The similarity matching uses a cosine similarity algorithm, and when the similarity is higher than a third threshold, the patient's state is considered to be successfully matched with the core state.

[0028] Furthermore, it also includes:

[0029] During emergency treatment, the generated personalized emergency treatment path is displayed in real time, and the treatment suggestions in the path are dynamically adjusted based on the patient's continuously updated vital signs information. The dynamic adjustment includes: comparing the real-time acquired vital signs data with the preset key node status thresholds of the path. If there is an abnormal deviation, the path adjustment mechanism is triggered. The system automatically matches the most relevant standard transfer path from the hierarchical emergency treatment path status transfer network, avoids prohibited transfer edges, and generates updated treatment suggestions.

[0030] Simultaneously, the actual sequence of diagnostic and treatment events is recorded and fed back as new diagnostic and treatment data to the patient diagnostic and treatment event sequence dataset for subsequent model optimization and update iteration.

[0031] Secondly, the present invention provides an emergency care pathway optimization system for critically ill patients based on big data, comprising:

[0032] The data acquisition module is used to obtain historical emergency and critical care patient diagnosis and treatment data from the hospital information system and generate a standardized patient diagnosis and treatment event sequence dataset, which contains event timestamps and diagnosis and treatment event type labels for each patient.

[0033] The data analysis module is used to analyze patient treatment event sequence datasets and construct directed graph models of treatment events based on graph-structured event sequence mining algorithms. This includes: nodes in the directed graph model representing treatment event types, and edges representing temporal transition relationships between events; calculating the degree centrality and frequent subsequence support of each node, identifying treatment events with a degree centrality higher than a first threshold as key nodes; marking treatment path segments with frequent subsequence support higher than a second threshold as high-frequency paths, and marking treatment path segments with frequent subsequence support lower than the second threshold and associated with adverse rescue outcomes as low-frequency paths.

[0034] The transfer network construction module is used to construct a hierarchical emergency path state transfer network based on key nodes, high-frequency paths, and low-frequency paths. Key nodes are the core states in the network, high-frequency paths are marked as standard transfer paths connecting the core states, and low-frequency paths are marked as forbidden transfer edges in the network.

[0035] The emergency route generation module is used to obtain the electronic health records of critically ill patients awaiting treatment in real time and extract the patient's current status information. It matches the current status information with the core status in the hierarchical emergency route state transition network. Starting from the patient's current status, it selects subsequent route segments based on the standard transfer pathway and combines prohibited transfer edges to constrain and modify the selected path, thereby generating a personalized emergency route in real time that matches the patient's clinical characteristics and avoids known treatment risks.

[0036] Thirdly, the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps corresponding to the method in the first aspect.

[0037] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps corresponding to the method in the first aspect.

[0038] In summary, the technical solutions of the embodiments of the present invention have at least the following advantages and beneficial effects:

[0039] The application first extracts historical emergency critical patient diagnosis and treatment data from a hospital information system to generate a standardized patient diagnosis and treatment event sequence dataset containing patient event timestamps and diagnosis and treatment event type labels; then analyzes the dataset using a graph structure-based event sequence mining algorithm to construct a directed graph model with diagnosis and treatment event types as nodes and time sequence transfer relationships between events as edges, and by calculating node degree centrality and frequent subsequence support, screens out key nodes, high-frequency paths and low-frequency paths associated with adverse rescue results; and based on these elements, constructs a hierarchical emergency path state transfer network, taking key nodes as core states, high-frequency paths as standard transfer paths and low-frequency paths as forbidden transfer edges; for patients to be treated, real-time acquisition of their electronic health records and extraction of current state information, matching the information with the network core state, and selecting subsequent path segments through the standard transfer path starting from the current state, while using the forbidden transfer edge for constraint and correction, realizing the emergency path optimization scheme of risk avoidance and path dynamic optimization, so as to improve the rescue efficiency of emergency critical patients, improve the rescue effect and reduce the risk of adverse prognosis. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 A flowchart of an emergency critical patient emergency path optimization method based on big data provided by the application;

[0041] Figure 2 A structural schematic diagram of an emergency critical patient emergency path optimization system based on big data provided by the application;

[0042] Figure 3 A schematic diagram of an electronic device provided by the application. DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions and advantages of the embodiments of the application clearer, the technical solutions in the embodiments of the application will be described below in connection with the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. The components of the embodiments of the application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0044] As shown in Figure 1 A method for optimizing emergency critical patient emergency path based on big data provided by the application, comprising:

[0045] S101: Obtain the diagnosis and treatment data of historical emergency critical patients from the hospital information system, and generate a standardized patient diagnosis and treatment event sequence dataset, which contains the event timestamps and diagnosis and treatment event type labels of each patient.

[0046] Specifically, the hospital information system (HIS) is a core carrier for integrating the whole process data of patient diagnosis and treatment, covering the whole dimension data of emergency critical patients from admission triage to rescue outcome. The historical data in the HIS contains the inherent correlation between diagnosis and treatment behavior and patient prognosis, which is the core data source for path optimization. The beneficial effect is to ensure that the data covers the whole scene of diagnosis and treatment, and to provide comprehensive materials for mining effective diagnosis and treatment rules. The data screening needs to meet the "emergency critical" condition, which is defined by critical score (such as APACHE II score ≥ 15 points) and diagnosis coding (such as ICD-10 critical disease coding). The principle is to avoid non-critical data interference and ensure the relevance of the sample. The beneficial effect is to improve the adaptability of data and research goals. For example, extracting the diagnosis and treatment records (including gender, signs, test results, etc.) of patients with acute myocardial infarction and severe trauma in the emergency rescue room in the past three years can accurately focus on the critical diagnosis and treatment scene.

[0047] Then a standardized patient diagnosis and treatment event sequence data set is generated, which contains the event timestamp and diagnosis and treatment event type label of each patient. The principle is that the diagnosis and treatment data in the original hospital information system has problems such as format heterogeneity and inconsistent expression (such as different doctors recording the same operation differently, and some data lacking time annotation), which cannot be directly used for subsequent sequence mining analysis, so it is necessary to convert heterogeneous data into structured event sequence form through standardization processing. The event timestamp refers to the accurate time of each diagnosis and treatment event (accurate to the minute level to meet the timeliness requirements of emergency critical diagnosis and treatment), which ensures the time sequence of diagnosis and treatment events and provides a basis for analyzing the time sequence transfer relationship between events. The diagnosis and treatment event type label refers to the unified classification and labeling of different diagnosis and treatment behaviors, which realizes the standardization of diagnosis and treatment events and avoids event recognition bias caused by expression differences. For example, in the diagnosis and treatment events of a patient with acute myocardial infarction, "2025-06-1008:30 Measure Blood Pressure" corresponds to the timestamp "202506100830", and the diagnosis and treatment event type label is "Signs Monitoring-Blood Pressure"; "2025-06-1008:45 Intravenous Injection of Nitroglycerin" corresponds to the timestamp "202506100845", and the diagnosis and treatment event type label is "Drug Therapy-Nitroglycerin Injection". Through the above standardization processing, the diagnosis and treatment process of each patient can be converted into an ordered combination of "timestamp-event type label", and then a standardized patient diagnosis and treatment event sequence data set is formed.

[0048] Further, the patient diagnosis and treatment event sequence data set is standardized by the following steps:

[0049] Firstly, the data cleaning is performed on the diagnosis and treatment data to remove duplicate records and invalid data. The principle is that in the process of collecting and entering the original diagnosis and treatment data, duplicate records (such as multiple entries of the same test result) or invalid data (such as records lacking key information, logically inconsistent data) may be generated due to system failure, human operation error, etc. Such data will interfere with the accuracy of subsequent sequence mining, so data cleaning is needed to remove them. The identification of duplicate records can be achieved by matching the key fields, for example, using "patient ID + diagnosis and treatment event type + event timestamp" as the joint key. If there are multiple records that match exactly, it is determined as a duplicate record, and only one valid record is retained. The identification of invalid data requires the setting of screening rules in combination with clinical diagnosis and treatment logic, for example, removing diagnosis and treatment records lacking event timestamps, records with physical data beyond the clinical reasonable range (such as blood pressure values of 0 or 200 mmHg or above without special explanation), records with ambiguous diagnosis and treatment event type labels (such as records only labeled as "treatment operation" without specifying the specific operation type).

[0050] Secondly, "map the diagnosis and treatment event type to a unified event type coding table, and arrange it in ascending order of event timestamp to generate the diagnosis and treatment event sequence of each patient". The principle is that the same diagnosis and treatment event may be described differently by different departments and doctors (such as "intravenous injection of adrenaline" and "intravenous injection of adrenaline", which are essentially the same diagnosis and treatment event), so if the original description is directly used for analysis, it will cause confusion in event recognition, so the standardized mapping of diagnosis and treatment events needs to be realized through a unified event type coding table; at the same time, the diagnosis and treatment process of emergency critical patients has strict time sequence, and the order of diagnosis and treatment events directly affects the rescue effect, so the diagnosis and treatment process of patients needs to be restored by arranging the events in ascending order of event timestamp to generate structured diagnosis and treatment event sequences. Among them, the unified event type coding table needs to be formulated in combination with the clinical characteristics of emergency critical diagnosis and treatment, covering the core diagnosis and treatment event types in the emergency critical rescue process, including physical sign monitoring (such as blood pressure monitoring, heart rate monitoring, blood oxygen saturation monitoring, etc.), laboratory examination (such as blood routine, coagulation function, myocardial enzyme spectrum, etc.), imaging examination (such as electrocardiogram, head CT, chest X-ray, etc.), drug treatment (such as adrenaline injection, nitroglycerin injection, antibiotic use, etc.), invasive operation (such as tracheal intubation, central venous catheterization, cardiopulmonary resuscitation, etc.) and other categories. Each specific diagnosis and treatment event under each category corresponds to a unique code (such as "blood pressure monitoring" coded as 001, "heart rate monitoring" coded as 002, "intravenous injection of adrenaline" coded as 101, "tracheal intubation" coded as 201, etc.). The mapping process of diagnosis and treatment event types is to match the original diagnosis and treatment event description in the cleaned data with the standard description in the event type coding table through natural language matching technology, and then assign it a unique event type code; after completing the coding mapping, arrange all the diagnosis and treatment events of the patient in ascending order of event timestamp to form a diagnosis and treatment event sequence with time as the axis, taking the patient ID as the unique identifier. Finally, combine all the diagnosis and treatment event sequences of the patients to generate the patient diagnosis and treatment event sequence dataset.

[0051] S102: Based on the event sequence mining algorithm of the graph structure, analyze the patient diagnosis and treatment event sequence dataset and construct a directed graph model of diagnosis and treatment events, including: the nodes in the directed graph model represent the diagnosis and treatment event types, and the edges represent the time sequence transfer relationship between events; calculate the node degree centrality and frequent subsequence support, and determine the diagnosis and treatment events with node degree centrality higher than the first threshold value as key nodes; mark the diagnosis and treatment path fragments with frequent subsequence support higher than the second threshold value as high-frequency paths, and mark the diagnosis and treatment path fragments with frequent subsequence support lower than the second threshold value and associated with adverse rescue results as low-frequency paths.

[0052] Specifically, the diagnosis and treatment process of emergency critical patients is essentially a combination of a series of ordered diagnosis and treatment events, and there is a clear sequence triggering logic between different diagnosis and treatment events. The directed graph model can convert the abstract time sequence association into a concrete node-edge structure. The node corresponds to the standardized diagnosis and treatment event type in S101 (such as "vital sign monitoring-blood pressure", "drug treatment-nitroglycerin injection", "invasive operation-tracheal intubation", etc.), which ensures consistency with the event type label of the standardized data set, and the direction of the edge strictly follows the occurrence order of the diagnosis and treatment event (such as "abnormal blood pressure monitoring" event triggers "intravenous injection of adrenaline" event, and there is a directed edge from "vital sign monitoring-blood pressure" node to "drug treatment-adrenaline injection" node). The association context of the global diagnosis and treatment event is directly presented through the graph structure, avoiding the omission of time sequence information, and laying a foundation for subsequent quantitative analysis of event association strength.

[0053] Then, the node degree centrality and frequent subsequence support are calculated, and the diagnosis and treatment event with node degree centrality higher than the first threshold value is determined as a key node. The principle is that the node degree centrality reflects the core degree of the diagnosis and treatment event in the global diagnosis and treatment process, and the frequent subsequence support reflects the clinical application frequency of a certain diagnosis and treatment path segment. The node degree centrality needs to be calculated separately for the in-degree centrality and the out-degree centrality. The in-degree centrality represents the number of events pointing to the node (i.e., the frequency of the event being triggered by other events), and the out-degree centrality represents the number of events pointing to other nodes from the node (i.e., the ability of the event to trigger subsequent events). The final degree centrality value is obtained by weighted summation (the weight can be set in combination with the emergency clinical priority, such as slightly higher out-degree weight to highlight the guiding nature of the previous event to rescue). The first threshold value can be set based on the statistical characteristics of the data set (such as taking the 75th percentile of all node degree centralities). For example, the "tracheal intubation" event is often triggered by multiple events such as "abnormal blood oxygen saturation monitoring" and "cardiopulmonary resuscitation", and can trigger subsequent events such as "ventilator use" and "central venous catheterization". Its degree centrality value is easy to be higher than the threshold value, and it is determined as a key node.

[0054] The labeling of high-frequency paths and low-frequency paths is based on the principle that high-frequency paths represent effective diagnosis and treatment process segments verified by a large number of clinical practices, and the higher the support, the stronger the applicability and reliability. Low-frequency paths have low application frequency, and if they are also associated with poor rescue results (such as the "antibiotic use without blood routine test" path segment corresponding to a higher infection control failure rate), it indicates that the path has potential diagnosis and treatment risks. The second threshold value can be determined by statistics of the distribution of subsequence support in the data set (such as taking the 60th percentile of the support), and the poor rescue result can be matched by the rescue outcome label in S101 (such as death, disease deterioration, etc.).

[0055] The specific implementation method is as follows:

[0056] Firstly, the frequent sub-sequence is extracted from the patient diagnosis and treatment event sequence data set using the prefix projection algorithm, and the support of the frequent sub-sequence is determined by calculating the frequency of the sub-sequence in the data set. The principle is that the prefix projection algorithm can avoid redundant calculation by focusing on the prefix of the sequence, efficiently mine the frequent sub-structure in the long time sequence, and is suitable for the analysis of massive emergency diagnosis and treatment sequence data. The support calculation method is the ratio of the number of patient sequences containing the sub-sequence to the total number of patient sequences. Thus, the efficiency and accuracy of extracting frequent sub-sequences are improved, and the processing needs of large-scale historical diagnosis and treatment data are adapted.

[0057] In this step, the frequent sub-sequence support formula is:

[0058] (1)

[0059] In the formula, is the support of the frequent sub-sequence L (reflecting the clinical application frequency of the diagnosis and treatment path fragment); is the number of patient sequences containing the frequent sub-sequence L; is the total number of patient sequences in the data set D.

[0060] Secondly, based on the patient diagnosis and treatment event sequence data set, a directed graph model of diagnosis and treatment events is constructed, with diagnosis and treatment event types as nodes and time sequence transfer relationships between events as edges. The principle is to combine the ordered combination of "time stamp-event type label" in S101, aggregate all patient time sequence transfer relationships, and ensure that the graph model covers the global diagnosis and treatment scene. The integrity and representativeness of the graph model are guaranteed, and the deviation of the local diagnosis and treatment scene is avoided.

[0061] Thirdly, in the directed graph model, the edge weight between nodes is calculated based on the transfer probability between events. The transfer probability is determined by the ratio of the number of event transfers to the total number of transfers. The principle is that the edge weight quantifies the possibility of event transfer, and the transfer probability = the total number of times from event A to event B / the total number of all subsequent transfer events of event A. For example, event A (abnormal blood pressure monitoring) is transferred to event B (nitroglycerin injection) 120 times, and to other events 80 times, so the transfer probability from A to B is 0.6, and the edge weight is set to 0.6. The weight depicts the strength of the transfer association, making the graph model more consistent with the probability distribution of clinical diagnosis and treatment decisions.

[0062] In this step, the event transfer probability formula is:

[0063] (2)

[0064] In the formula, is the probability of transferring from event to event (quantifying the possibility of event transfer); the total number of times of transferring from the event to the event ; the total number of times of transferring from the event to all subsequent transfer events.

[0065] Fourthly, when calculating the node degree centrality, the in-degree centrality and the out-degree centrality of each node are calculated respectively, and the node degree centrality corresponding value is obtained by weighted sum of the in-degree centrality and the out-degree centrality. The principle is to consider the input and output association of the node, and to avoid the misjudgment of key nodes caused by single dimension evaluation. The objectivity and clinical relevance of key node identification are ensured, and accurate core anchor points are provided for subsequent first aid path planning.

[0066] In this step, the in-degree centrality formula is:

[0067] (3)

[0068] In the formula, is the in-degree centrality of the event type , which reflects the comprehensive probability of transferring from all event types to , and embodies the core degree of as the "pre-event target"; is the probability of transferring from the event to the event ; is the total number of diagnosis and treatment event types.

[0069] The out-degree centrality of the event type is:

[0070] (4)

[0071] The weighted sum of the node degree centrality is:

[0072] (5)

[0073] The constraint condition is: wherein, is the weight coefficient of the in-degree centrality, is the weight coefficient of the out-degree centrality.

[0074] S103: Based on the key nodes, high-frequency paths and low-frequency paths, a hierarchical first aid path state transfer network is constructed; wherein the key nodes are the core states in the network, the high-frequency paths are marked as standard transfer paths connecting the core states, and the low-frequency paths are marked as forbidden transfer edges in the network.

[0075] ​Specifically, the core logic of constructing the hierarchical emergency path state transition network is to convert the diagnosis and treatment core elements identified in step S102 into the basic constituent units of the network, wherein "the key nodes are taken as the core states in the network". The principle is that the key nodes correspond to diagnosis and treatment events with node degree centrality higher than the first threshold. Such events have strong correlation and clinical guidance in the global diagnosis and treatment process, and are the core link of determining the emergency direction and rescue effect. Setting them as core states can focus on the key process of emergency, avoid decision deviation caused by redundant events, anchor the network to the core link of emergency, and provide clear core anchor for subsequent path planning. For example, "tracheal intubation" and "abnormal blood pressure monitoring" are key nodes as core states, which can directly associate a series of key diagnosis and treatment behaviors and provide clear core guidance for clinical decision-making.

[0076] The specific implementation is as follows:

[0077] Firstly, the standard transition path between the core states is constructed based on the high-frequency path. The weight of the standard transition path is determined by combining the support degree of the corresponding frequent subsequence and the transition probability between events. The principle is that the high-frequency path is a diagnosis and treatment path fragment with a support degree of frequent subsequence higher than the second threshold, which has been verified by a large number of clinical practices to have good applicability and reliability, and is the core embodiment of the clinical routine effective diagnosis and treatment process. Taking it as the standard transition path between the core states can ensure the clinical rationality of the path. The dual consideration logic of the weight is that the support degree of the frequent subsequence reflects the clinical application popularity of the path, and the transition probability between events reflects the inevitability of event triggering. The combination of the two can realize the accurate quantification of the priority of the standard transition path, avoiding the misjudgment of the path priority caused by a single indicator. The standard transition path has both a wide clinical practice basis and priority sorting through weight differentiation, which is convenient for subsequent dynamic selection of the optimal path according to the patient's condition. For example, the high-frequency path "abnormal blood oxygen saturation monitoring → tracheal intubation → use of breathing machine" has high support degree and transition probability, and the corresponding weight is set higher, which can be selected preferentially in path matching.

[0078] In this step, the weight formula of the standard transition path is:

[0079] (6)

[0080] wherein, is the mean value of the transition probability of all adjacent events in the frequent subsequence L, and the calculation formula is:

[0081] (7)

[0082] The weight formula of the standard transition path is obtained by combining formulas (1), (6) and (7):

[0083] (8)

[0084] wherein, is the weight of the standard transition path; is the weight coefficient of the frequent subsequence support; is the number of patient sequences containing the frequent subsequence L; is the total number of patient sequences in the dataset; is the weight coefficient of the event transition probability; is the number of event types in the frequent subsequence L; is the transition probability from the i-th event to the i+1-th event in the subsequence L.

[0085] Secondly, low-frequency paths are marked as forbidden transition edges in the network, which are highlighted in the network presentation. The principle is that low-frequency paths specifically refer to diagnosis and treatment path fragments with frequent subsequence support below the second threshold and associated with adverse rescue outcomes. Such paths have low clinical application frequency and have been proven to have potential diagnosis and treatment risks. Marking them as forbidden transition edges can achieve clear avoidance of high-risk diagnosis and treatment behaviors. Highlighting is to meet the timeliness requirements of emergency and first aid, so as to facilitate clinicians to quickly identify risk paths and avoid misselection.

[0086] Thirdly, the hierarchical setting of the network includes a core layer and a detailed layer. The core layer contains core states and standard transition paths, which are used for rapid development of emergency path planning. The principle is that emergency critical care has strong timeliness. The design of the core layer aims to strip redundant diagnosis and treatment information, focus on core diagnosis and treatment links and effective paths, achieve rapid path matching and decision-making, and meet the rapid rescue needs in emergency situations. The beneficial effect is to greatly shorten the path planning time, avoid decision delay caused by information overload, and gain valuable rescue time for critically ill patients. For example, when the patient's condition is critical, the doctor can quickly match the standard transition path corresponding to the patient's current core state (such as "sudden drop in blood pressure") through the core layer, and directly develop key diagnosis and treatment operations. The detailed layer contains all diagnosis and treatment event nodes and transition edges between events, and clearly marks forbidden transition edges

[0087] S104: For an emergency critically ill patient to be treated, the electronic health record of the patient is acquired in real time, and the current state information of the patient is extracted. The current state information is matched with the core states in the hierarchical emergency path state transition network. Taking the current state of the patient as the starting point, the subsequent path fragments are selected based on the standard transition path, and the selected path is constrained and corrected in combination with the forbidden transition edge, so as to generate a personalized emergency path that matches the clinical characteristics of the patient and avoids known diagnosis and treatment risks in real time.

[0088] ​​Specifically, first, the core principle of obtaining the electronic health record in real time is that the condition of emergency critical patients progresses rapidly, and static data cannot reflect the dynamic changes of patients. The electronic health record, as a carrier of integrating real-time diagnosis and treatment data of patients, covers triage records, current signs, implemented preliminary diagnosis and treatment operations, laboratory instant results and other information. The data is obtained through the real-time interface of the hospital information system (HIS), and the beneficial effect is to ensure the timeliness of the data and provide the latest basis for dynamic path planning. For example, the patient's blood oxygen saturation is 92% when admitted to the hospital, and 10 minutes later the real-time data shows that it has dropped to 85%. This change can be captured in time to adjust the subsequent path. The specific implementation is as follows:

[0089] The natural language processing algorithm is used to analyze the unstructured text in the electronic health record, and the key clinical features including symptoms, signs and laboratory indicators are extracted. The principle is that there is a lot of unstructured information in the electronic health record (such as the doctor's handwritten triage note "sudden chest tightness, profuse sweating", and the nurse's verbal shift record "shallow and rapid breathing"). Such information cannot be directly used for state matching. Natural language processing algorithm can convert unstructured text into structured features through named entity recognition, semantic analysis and other technologies. The beneficial effect is to fully mine the clinical information of patients and avoid missing key condition signals. Among them, the extraction of "key clinical features" needs to focus on the core elements related to emergency path, such as symptoms (chest tightness, confusion), signs (blood pressure, heart rate, blood oxygen saturation), and laboratory indicators (instant blood glucose, arterial blood gas analysis results). For example, through algorithm analysis of the text "male patient, 56 years old, acute myocardial infarction admitted to hospital, blood pressure 80 / 50mmHg, heart rate 120 beats / min, blood oxygen saturation 88%", the key features can be extracted as "acute myocardial infarction, blood pressure 80 / 50mmHg (hypotension), heart rate 120 beats / min (tachycardia), blood oxygen saturation 88% (hypoxemia)". Then the key clinical features are mapped to a numerical vector as the current state information of the patient. The principle is that the core state in the hierarchical emergency path state transition network exists in the form of standardized coding (corresponding to the diagnosis and treatment event type label in S101 and the key node coding in S102). Text-based clinical features need to be converted into machine-recognizable numerical vectors to realize quantitative matching with network core states, thereby improving the efficiency and accuracy of state matching and adapting the algorithm logic of subsequent path selection. For example, set the feature mapping rule: hypotension code 01, tachycardia code 02, hypoxemia code 03, acute myocardial infarction code 04. The key clinical features of the patient can be mapped to the numerical vector [01, 02, 03, 04], which is the current state information of the patient.

[0090] Secondly, match the current state information with the core state in the hierarchical first aid path state transition network. The principle is that the core state (S102 determines the node degree centrality of the diagnosis and treatment event higher than the first threshold, such as "abnormal blood pressure monitoring", "abnormal blood oxygen saturation monitoring", "acute myocardial infarction emergency intervention") is the core anchor point of the first aid process, and its characteristics have been quantified as a standard numerical vector through historical data. By calculating the similarity (such as cosine similarity) between the patient's current state vector and each core state vector, the initial position of the patient in the first aid path network can be located. The beneficial effect is to quickly lock the first aid core direction and avoid redundant information interference, which meets the rapid treatment needs of emergency department. For example, the state vector [01, 02, 03, 04] of the aforementioned patient can be accurately matched with the core state "acute myocardial infarction combined with blood pressure-oxygen abnormality".

[0091] Next, based on the standard transition path, the subsequent path segment is selected starting from the current state of the patient. The principle is that the standard transition path (S103 is constructed based on high-frequency path) has been verified by a large number of clinical practices, and has high applicability and reliability. Its weight is determined by the support degree of frequent subsequence and the transition probability between events, which can reflect the clinical popularity and event triggering necessity of the path. Based on this weight, the path can be dynamically selected, which can balance "clinical effectiveness" and "individual adaptability". The beneficial effect is to ensure that the selected path not only conforms to the mainstream effective diagnosis and treatment standard, but also can adjust the priority according to the specific state of the patient. For example, starting from "acute myocardial infarction combined with blood pressure-oxygen abnormality", the corresponding standard transition path has "→tracheal intubation→ventilator assisted breathing→intravenous injection of nitroglycerin" (weight 0.85) and "→intravenous infusion→injection of pressor drugs→electrocardiogram review" (weight 0.6). If the patient also has the "shallow and fast breathing" feature, the path segment with higher weight is preferentially selected.

[0092] Then, in combination with the forbidden transition edge, the selected path is constrained and corrected. The principle is that the forbidden transition edge (S103 marks the low-frequency path segment associated with adverse outcomes, such as "using thrombolytic drugs without electrocardiogram review" and "performing invasive operation without monitoring coagulation function") represents known diagnosis and treatment risks. By checking whether the selected path contains such edges, high-risk links can be removed and corrected to a safe path, thereby avoiding potential medical errors from the source and improving the safety of the first aid path. For example, the initially selected path contains "→monitoring coagulation function→central venous catheterization" (forbidden transition edge), which is automatically removed and corrected to a safe path of "→monitoring coagulation function→central venous catheterization".

[0093] Finally, a personalized emergency path that matches the patient's clinical characteristics and avoids known treatment risks is generated in real time. The principle is to integrate the whole process of "state matching-path selection-risk constraint", dynamically output the path based on the patient's real-time clinical characteristics, avoid the drawbacks of traditional fixed paths, and avoid known risks through taboo transfer edge constraints, realize the "personalized customization" and "safety bottom line" of the emergency path, shorten the doctor's decision-making time, save valuable treatment time for patients, and reduce the incidence of adverse rescue outcomes. The final personalized path is, for example: "Acute myocardial infarction with abnormal blood pressure and oxygen → tracheal intubation → breathing machine assisted breathing → monitoring of blood coagulation function → central venous catheter → intravenous injection of nitroglycerin → electrocardiogram review", which not only matches the patient's low blood pressure and low oxygen, but also avoids the taboo risk of invasive operation without monitoring blood coagulation function.

[0094] Further, in the emergency process, the generated personalized emergency path is displayed in real time, and the diagnosis and treatment suggestions in the path are dynamically adjusted according to the patient's continuously updated vital sign information. The principle is that emergency critical care has strong timeliness and uncertainty, and real-time display of the path can enable medical staff to quickly obtain standardized diagnosis and treatment guidelines, avoid missing key steps due to decision confusion in an emergency, and significantly improve the standardization and efficiency of diagnosis and treatment operations; and the continuously updated vital sign information is the core basis for reflecting the dynamic changes of the patient's condition, and a static path cannot adapt to the real-time evolution of the patient's condition, and dynamic adjustment can ensure that the diagnosis and treatment suggestions are accurately matched with the patient's current state.

[0095] Among them, dynamic adjustment specifically includes: comparing the real-time acquired vital sign data with the key node state threshold preset in the path, and if there is an abnormal deviation, triggering the path adjustment mechanism. The principle is that the key node state threshold is set based on the core sign range of the effective path in the historical diagnosis and treatment data (such as the blood oxygen saturation threshold preset for the "tracheal intubation" node is ≥90%), and abnormal deviation indicates that the original path adaptability is declining, and adjustment needs to be started to avoid diagnosis and treatment delay; to achieve timely response to changes in the patient's condition. After triggering the adjustment, the system automatically matches the most relevant standard transfer path from the hierarchical emergency path state transfer network, and avoids the taboo transfer edge, and generates updated diagnosis and treatment suggestions. The principle is that this network covers proven effective paths and risk taboo information, and automatically matching can quickly filter the optimal solution and avoid known risks; for example, if the patient's original path contains "breathing machine assisted breathing", real-time monitoring finds that the heart rate drops sharply, the system matches the standard path "→ injection of pressor drugs → heart rate retest", and eliminates the taboo transfer edge "without monitoring heart rate before adjusting breathing machine parameters". At the same time, the actual executed diagnosis and treatment event sequence is recorded as new diagnosis and treatment data feedback to the patient's diagnosis and treatment event sequence dataset for subsequent model optimization and update iteration.

[0096] Based on the same inventive concept, such as Figure 2As shown, this invention provides an emergency care pathway optimization system for critically ill patients based on big data, comprising:

[0097] The data acquisition module 201 is used to acquire historical emergency and critical care patient diagnosis and treatment data from the hospital information system and generate a standardized patient diagnosis and treatment event sequence dataset, which contains event timestamps and diagnosis and treatment event type labels for each patient.

[0098] Data analysis module 202 is used to analyze patient diagnosis and treatment event sequence datasets and construct a directed graph model of diagnosis and treatment events based on graph-structured event sequence mining algorithms. This includes: in the directed graph model, nodes represent diagnosis and treatment event types, and edges represent temporal transition relationships between events; calculating the degree centrality and frequent subsequence support of each node, identifying diagnosis and treatment events with a degree centrality higher than a first threshold as key nodes; marking diagnosis and treatment path segments with frequent subsequence support higher than a second threshold as high-frequency paths, and marking diagnosis and treatment path segments with frequent subsequence support lower than the second threshold and associated with adverse rescue outcomes as low-frequency paths.

[0099] The transfer network construction module 203 is used to construct a hierarchical emergency path state transfer network based on key nodes, high-frequency paths, and low-frequency paths; where key nodes are the core states in the network, high-frequency paths are marked as standard transfer paths connecting the core states, and low-frequency paths are marked as forbidden transfer edges in the network.

[0100] The emergency route generation module 204 is used to obtain the electronic health records of critically ill patients awaiting treatment in real time and extract the patient's current status information; match the current status information with the core status in the hierarchical emergency route state transition network; take the patient's current status as the starting point; select subsequent route segments based on the standard transfer pathway; and combine the prohibited transfer edges to constrain and correct the selected path, thereby generating a personalized emergency route in real time that matches the patient's clinical characteristics and avoids known treatment risks.

[0101] Based on the same inventive concept, such as Figure 3 As shown, the present invention provides an electronic device, including: a memory 302, a processor 301, and a computer program stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program, it implements a method for optimizing emergency medical pathways for critically ill patients based on big data.

[0102] Based on the same inventive concept, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for optimizing emergency care pathways for critically ill patients based on big data.

[0103] The above merely describes the preferred embodiments of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for optimizing emergency medical pathways for critically ill patients based on big data, characterized in that, include: Historical emergency and critical care patient diagnosis and treatment data are obtained from the hospital information system, and a standardized patient diagnosis and treatment event sequence dataset is generated. This dataset contains event timestamps and diagnosis and treatment event type labels for each patient. An event sequence mining algorithm based on graph structure analyzes the patient treatment event sequence dataset and constructs a directed graph model of treatment events. The algorithm includes: nodes representing treatment event types and edges representing temporal transition relationships between events; calculating the degree centrality and frequent subsequence support of each node; identifying treatment events with a degree centrality value higher than a first threshold as key nodes; marking treatment path segments with frequent subsequence support higher than a second threshold as high-frequency paths, and marking treatment path segments with frequent subsequence support lower than the second threshold and associated with adverse rescue outcomes as low-frequency paths; wherein, the node degree centrality includes in-degree centrality and out-degree centrality, where in-degree centrality represents the number of events pointing to the node, and out-degree centrality represents the number of events pointing to other nodes from the node, and the final node degree centrality value is obtained through weighted summation; the frequent subsequences are extracted from the patient treatment event sequence dataset using a prefix projection algorithm, and the support of the frequent subsequences is determined by calculating the frequency of occurrence of the subsequences in the dataset, where the support is calculated as the ratio of the number of patient sequences containing the subsequence to the total number of patient sequences; Based on the key nodes, high-frequency paths, and low-frequency paths, a hierarchical emergency path state transition network is constructed; wherein, the key nodes are the core states in the network, the high-frequency paths are marked as standard transition paths connecting the core states, and the low-frequency paths are marked as forbidden transition edges in the network. For critically ill emergency patients awaiting treatment, their electronic health records are acquired in real time, and their current status information is extracted. The current status information is matched with the core status in the hierarchical emergency path state transition network. Starting from the patient's current status, subsequent path segments are selected based on the standard transfer pathway. The selected path is constrained and modified in conjunction with the forbidden transfer edges, thereby generating a personalized emergency path that matches the patient's clinical characteristics and avoids known treatment risks in real time.

2. The method for optimizing emergency medical pathways for critically ill patients based on big data, as described in claim 1, is characterized in that... The generation of a standardized patient diagnosis and treatment event sequence dataset specifically includes: The diagnostic and treatment data is cleaned to remove duplicate records and invalid data; the diagnostic and treatment event types are mapped to a unified event type coding table and sorted in ascending order according to the event timestamp to generate a diagnostic and treatment event sequence for each patient; the diagnostic and treatment event sequences of all patients are merged to generate the patient diagnostic and treatment event sequence dataset.

3. The method for optimizing emergency medical pathways for critically ill patients based on big data, as described in claim 1, is characterized in that... The construction of the hierarchical emergency medical route state transition network specifically includes: Key nodes are treated as core states in the network. Standard transition paths between core states are constructed based on high-frequency paths. The weights of the standard transition paths are determined by combining the support of the frequent subsequences corresponding to the path and the transition probability between events. Low-frequency paths are marked as taboo transition edges in the network and highlighted in the network presentation. The network is structured in layers, including a core layer and a detailed layer. The core layer contains the core status and standard transfer pathways for rapid emergency route planning. The detailed layer contains all diagnostic and treatment event nodes and transfer edges between events, and clearly marks contraindicated transfer edges.

4. The method for optimizing emergency medical pathways for critically ill patients based on big data, as described in claim 3, is characterized in that... The weighting expressions for standard transfer pathways include: In the formula, Standard transfer path weights; These are the weighting coefficients for the support of frequent subsequences; The number of patient sequences containing frequent subsequence L; This represents the total number of patient sequences in the dataset. The weighting coefficients for the event transition probability; The number of event types in the frequent subsequence L; For the first subsequence L The event to the +1 event transition probability.

5. The method for optimizing emergency medical pathways for critically ill patients based on big data according to claim 1, characterized in that, The real-time acquisition of electronic health records and extraction of the patient's current status information specifically includes: Natural language processing algorithms are used to parse unstructured text in electronic health records and extract key clinical features, including symptoms, signs, and laboratory indicators. These key clinical features are then mapped into numerical vectors to serve as the patient's current status information.

6. The method for optimizing emergency medical pathways for critically ill patients based on big data according to claim 1, characterized in that, Also includes: During emergency treatment, the generated personalized emergency treatment path is displayed in real time, and the treatment suggestions in the path are dynamically adjusted based on the patient's continuously updated vital signs information. The dynamic adjustment includes: comparing the real-time acquired vital signs data with the preset key node status thresholds of the path. If an abnormal deviation occurs, the path adjustment mechanism is triggered. The system automatically matches the most relevant standard transfer pathway from the hierarchical emergency path status transfer network, avoids contraindicated transfer edges, and generates updated treatment suggestions. Simultaneously, the actual sequence of diagnostic and treatment events is recorded and fed back as new diagnostic and treatment data to the patient diagnostic and treatment event sequence dataset for subsequent model optimization and update iteration.

7. A big data-based emergency medical pathway optimization system for critically ill patients, characterized in that, include: The data acquisition module is used to obtain historical emergency and critical care patient diagnosis and treatment data from the hospital information system and generate a standardized patient diagnosis and treatment event sequence dataset, which contains event timestamps and diagnosis and treatment event type labels for each patient. The data analysis module is used to analyze the patient treatment event sequence dataset and construct a directed graph model of treatment events based on a graph-structured event sequence mining algorithm. This includes: in the directed graph model, nodes represent treatment event types, and edges represent temporal transition relationships between events; calculating the degree centrality and frequent subsequence support of each node; identifying treatment events with a degree centrality value higher than a first threshold as key nodes; marking treatment path segments with frequent subsequence support higher than a second threshold as high-frequency paths; and identifying treatment path segments with frequent subsequence support lower than the second threshold that are associated with adverse rescue outcomes. Path segments are labeled as low-frequency paths; the node degree centrality includes in-degree centrality and out-degree centrality. The in-degree centrality represents the number of events pointing to the node, and the out-degree centrality represents the number of events pointing to other nodes from the node. The final node degree centrality value is obtained by weighted summation; the frequent subsequences are extracted from the patient diagnosis and treatment event sequence dataset using a prefix projection algorithm, and the support of the frequent subsequences is determined by calculating the frequency of occurrence of the subsequences in the dataset. The support is calculated as the ratio of the number of patient sequences containing the subsequence to the total number of patient sequences. The transfer network construction module is used to construct a hierarchical emergency path state transfer network based on the key nodes, high-frequency paths, and low-frequency paths; wherein, the key nodes are the core states in the network, the high-frequency paths are marked as standard transfer paths connecting the core states, and the low-frequency paths are marked as forbidden transfer edges in the network. The emergency route generation module is used to acquire the electronic health records of critically ill patients awaiting treatment in real time and extract the patient's current status information; match the current status information with the core status in the hierarchical emergency route state transition network; starting from the patient's current status, select subsequent route segments based on the standard transition pathway; and constrain and modify the selected path by combining the forbidden transition edges, thereby generating a personalized emergency route in real time that matches the patient's clinical characteristics and avoids known treatment risks.

Images